| Record Id | CB-DOC-061-007 |
|---|
| Record Id | CB-DOC-061-007 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2025-11-15 |
| Record Type | officer certificate |
| Folder | 01_data_inventory |
| Workstream | data inventory |
| Source System | OneTrust data map |
| Native Version | 1.2 |
| Record Status | reviewed — source conflict identified |
| Confidentiality | Confidential — legal review |
| Custodian | Nadine Flores |
| Reviewer | Isaac Romero |
| Cross Reference | CB-DOC-061-004 |
| Control Metric | 333184 |
| Review Question | consent-record deficiency |
| Finding Id | F-03 |
| Record Role | primary |
| Control Severity | high |
| Remediation Owner | Rafael Okafor |
| Response Due | 2026-09-02 |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 01_data_inventory |
| Record Type | officer certificate |
| Custodian | Nadine Flores |
| Reviewer | Isaac Romero |
| Record Date | 2025-11-15 |
| Cross Reference | CB-DOC-061-004 |
| Control Metric | 333184 |
| Operative Text | approval is attributed to Nadine Flores as the sole authorized reviewer |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Nadine Flores, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2025-11-15 and should be evaluated with CB-DOC-061-004. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Isaac Romero compared identifier 333184 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the data inventory workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
The producing team exported this record from OneTrust data map and retained the native identifier CB-DOC-061-007. Nadine Flores confirmed the export boundary, while Isaac Romero performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-004.
For this data inventory review, legal and business stakeholders are using the record to evaluate consent-record deficiency, deletion-job failure, and sensitive-data classification. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that approval is attributed to Nadine Flores as the sole authorized reviewer. The record-control overlay classifies this as the primary source for F-03 (consent-record deficiency) at high severity. The assigned remediation owner is Rafael Okafor, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Isaac Romero recorded status “reviewed — source conflict identified” and linked the file to CB-DOC-061-004. The control metric 333184 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the consent-record deficiency, deletion-job failure, and sensitive-data classification materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2025-09-30 | Source population opened for collection | Nadine Flores | CB-DOC-061-007 |
| 2025-10-27 | Custodian confirmed system and date boundary | Owen Delgado | CB-DOC-061-004 |
| 2025-11-07 | Matter team completed first-level comparison | Maya Ellison | CB-DOC-061-004 |
| 2025-11-15 | Legal reviewer recorded the current disposition | Dominic Alvarez | CB-DOC-061-007 |
| 2025-12-18 | Assigned owner scheduled the next control response | Talia Brooks | CB-DOC-061-004 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-007-1 | Before 2026-09-02, secure written consent from the authorized decision-maker; owner: Rafael Okafor. | Rafael Okafor | 2026-08-19 | in review |
| A-007-2 | Preserve the native OneTrust data map export and document any replacement record. | Leona Park | 2026-08-26 | awaiting evidence |
| A-007-3 | Report the disposition to the data inventory workstream lead before the matter deadline. | Henry Cho | 2026-09-02 | owner confirmed |
The record covers activity in California through 2025-11-15 and should be evaluated with CB-DOC-061-004, the four related records listed below, and the complete data inventory folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Isaac Romero compared identifier 333184 against the folder index, the native OneTrust data map entry, and CB-DOC-061-004. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/01_data_inventory/008_data_inventory_status_report.md b/task_files/cb100-061-arcadia-genomics/01_data_inventory/008_data_inventory_status_report.md index e393536bf7ea1acae48a67b2c15d0526a15d0f98..b4f571f809225075dfbd1746ef53322dca1dcb9a 100644 --- a/task_files/cb100-061-arcadia-genomics/01_data_inventory/008_data_inventory_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/01_data_inventory/008_data_inventory_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-008 +> Confidential — matter team · reviewed — no independent exception · native version 3.3 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Micah Sullivan | | Reviewer | Nadine Flores | | Cross-reference | CB-DOC-061-021 | +| Source system | ServiceNow privacy queue | +| Workstream | data inventory | | Control metric | 748174 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The data inventory team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Micah Sullivan identified it as an ordinary-course record from ServiceNow privacy queue; reviewer Nadine Flores preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** The file is responsive background material; it supplies context but does not create an independent obligation. + +## 1. Purpose and audience + +This status report supports the data inventory workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +This copy was collected from ServiceNow privacy queue under matter hold CB-PR-3001. Its lineage runs from Micah Sullivan, as producing custodian, to Nadine Flores, as reviewing lawyer. The related record CB-DOC-061-021 remains a separate source of truth and was not merged into this document. + +## 3. Matter and workstream context + +The data inventory workstream sits within privacy and regulatory response audit. The team is tracking processor-contract omission, access-request deadline, and minor-consent issue because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that The file is responsive background material; it supplies context but does not create an independent obligation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Nadine Flores recorded status “reviewed — no independent exception” and linked the file to CB-DOC-061-021. The control metric 748174 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the processor-contract omission, access-request deadline, and minor-consent issue materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-12-09 | Source population opened for collection | Micah Sullivan | CB-DOC-061-008 | +| 2025-01-12 | Custodian confirmed system and date boundary | Lena Varga | CB-DOC-061-021 | +| 2025-01-21 | Matter team completed first-level comparison | Rafael Okafor | CB-DOC-061-021 | +| 2025-01-29 | Legal reviewer recorded the current disposition | Priya Raman | CB-DOC-061-008 | +| 2025-03-01 | Assigned owner scheduled the next control response | Jonas Feld | CB-DOC-061-021 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Jonas Feld | control owner | source completeness | +| Sofia Bennett | matter lead | business interpretation | +| Henry Cho | business owner | legal review | +| Lena Varga | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Micah Sullivan, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-015` — same-cycle source (processing records) +- `CB-DOC-061-027` — implementation evidence (security incidents) +- `CB-DOC-061-045` — independent control record (notices consents) +- `CB-DOC-061-061` — later reconciliation record (retention deletion) -## Operative record +## Action register -The file is responsive background material; it supplies context but does not create an independent obligation. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-008-1 | Confirm that CB-DOC-061-021 does not change the context-only classification. | Priya Raman | 2026-08-19 | awaiting evidence | +| A-008-2 | Preserve the native ServiceNow privacy queue export and document any replacement record. | Isaac Romero | 2026-08-26 | owner confirmed | +| A-008-3 | Report the disposition to the data inventory workstream lead before the matter deadline. | Nadine Flores | 2026-09-02 | escalated | ## Scope and cross-reference -The record covers activity in California through 2025-01-29 and should be evaluated with CB-DOC-061-021. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-01-29 and should be evaluated with CB-DOC-061-021, the four related records listed below, and the complete data inventory folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Nadine Flores compared identifier 748174 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Nadine Flores compared identifier 748174 against the folder index, the native ServiceNow privacy queue entry, and CB-DOC-061-021. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/02_processing_records/009_processing_records_control_register.md b/task_files/cb100-061-arcadia-genomics/02_processing_records/009_processing_records_control_register.md index fd7dcb4ee93ea9defeb2b24f4834c3d39657a652..a67cdf9fa647d3255c010720d158a2770539138a 100644 --- a/task_files/cb100-061-arcadia-genomics/02_processing_records/009_processing_records_control_register.md +++ b/task_files/cb100-061-arcadia-genomics/02_processing_records/009_processing_records_control_register.md @@ -1,5 +1,7 @@ # Control Register — CB-DOC-061-009 +> Restricted — need to know · indexed — responsive context · native version 4.2 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Farah Ibrahim | | Reviewer | Rafael Okafor | | Cross-reference | CB-DOC-061-038 | +| Source system | BigID deletion log | +| Workstream | processing records | | Control metric | 373711 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The processing records team used the record to maintain the matter team's issue population, source lineage, review status, and accountable owner. Producing custodian Farah Ibrahim identified it as an ordinary-course record from BigID deletion log; reviewer Rafael Okafor preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** No variance was recorded after the named reviewer compared the source ledger to the signed control record. + +## 1. Purpose and audience + +This control register supports the processing records workstream for Arcadia Genomics, Inc.. Its purpose is to maintain the matter team's issue population, source lineage, review status, and accountable owner. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +The producing team exported this record from BigID deletion log and retained the native identifier CB-DOC-061-009. Farah Ibrahim confirmed the export boundary, while Rafael Okafor performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-038. + +## 3. Matter and workstream context + +The processing records workstream sits within privacy and regulatory response audit. The team is tracking subprocessor inventory mismatch, identity-verification inconsistency, and training completion gap because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that No variance was recorded after the named reviewer compared the source ledger to the signed control record.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Rafael Okafor recorded status “indexed — responsive context” and linked the file to CB-DOC-061-038. The control metric 373711 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the subprocessor inventory mismatch, identity-verification inconsistency, and training completion gap materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-12-30 | Source population opened for collection | Farah Ibrahim | CB-DOC-061-009 | +| 2025-01-30 | Custodian confirmed system and date boundary | Samuel Kim | CB-DOC-061-038 | +| 2025-02-10 | Matter team completed first-level comparison | Nora Chen | CB-DOC-061-038 | +| 2025-02-15 | Legal reviewer recorded the current disposition | Elliot Mercer | CB-DOC-061-009 | +| 2025-03-17 | Assigned owner scheduled the next control response | Mei Whitaker | CB-DOC-061-038 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Mei Whitaker | matter lead | source completeness | +| Adrian Mensah | business owner | business interpretation | +| Nadine Flores | records custodian | legal review | +| Samuel Kim | finance reviewer | remediation tracking | -## Background +## Related records -This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Farah Ibrahim, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-016` — same-cycle source (processing records) +- `CB-DOC-061-028` — implementation evidence (security incidents) +- `CB-DOC-061-046` — independent control record (notices consents) +- `CB-DOC-061-062` — later reconciliation record (retention deletion) -## Operative record +## Action register -No variance was recorded after the named reviewer compared the source ledger to the signed control record. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-009-1 | Confirm that CB-DOC-061-038 does not change the context-only classification. | Elliot Mercer | 2026-08-19 | owner confirmed | +| A-009-2 | Preserve the native BigID deletion log export and document any replacement record. | Sofia Bennett | 2026-08-26 | escalated | +| A-009-3 | Report the disposition to the processing records workstream lead before the matter deadline. | Micah Sullivan | 2026-09-02 | open | ## Scope and cross-reference -The record covers activity in California through 2025-02-15 and should be evaluated with CB-DOC-061-038. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-02-15 and should be evaluated with CB-DOC-061-038, the four related records listed below, and the complete processing records folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Rafael Okafor compared identifier 373711 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Rafael Okafor compared identifier 373711 against the folder index, the native BigID deletion log entry, and CB-DOC-061-038. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/02_processing_records/010_processing_records_executed_instrument.txt b/task_files/cb100-061-arcadia-genomics/02_processing_records/010_processing_records_executed_instrument.txt index 507d463a3c1225cc458ed080d04cc7ab3d59455e..3fb17eb8718f7a74e8114404e7b6a53bad1cc99e 100644 --- a/task_files/cb100-061-arcadia-genomics/02_processing_records/010_processing_records_executed_instrument.txt +++ b/task_files/cb100-061-arcadia-genomics/02_processing_records/010_processing_records_executed_instrument.txt @@ -1,23 +1,78 @@ +ATTORNEY WORK PRODUCT + +EXECUTED INSTRUMENT +ARCADIA GENOMIC DATA RESPONSE AUDIT + DOCUMENT CONTROL: CB-DOC-061-010 MATTER: CB-PR-3001 | Arcadia genomic data response audit RECORD TYPE: executed instrument DATE: 2025-05-09 +SOURCE SYSTEM: Vanta control register +NATIVE VERSION: 3.2 +STATUS: reviewed — source conflict identified CUSTODIAN: Owen Delgado REVIEWER: Mei Whitaker CROSS-REFERENCE: CB-DOC-061-055 CONTROL METRIC: 553074 -BACKGROUND -This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Owen Delgado, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +REVIEW QUESTION: processor-contract omission +FINDING ID: F-04 +RECORD ROLE: primary +CONTROL SEVERITY: medium +REMEDIATION OWNER: Nora Chen +RESPONSE DUE: 2026-09-02 + +RECITALS + +A. This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The processing records team used the record to preserve the operative language, execution history, defined terms, and signature authority. Producing custodian Owen Delgado identified it as an ordinary-course record from Vanta control register; reviewer Mei Whitaker preserved the source wording and separately recorded the legal-control disposition. + +B. The parties and matter team intend to preserve the source hierarchy and the decision record for CB-PR-3001. OPERATIVE RECORD the item is recorded as closed without exception in CB-PR-3001-D806 +1. PURPOSE AND AUDIENCE +This executed instrument supports the processing records workstream for Arcadia Genomics, Inc.. Its purpose is to preserve the operative language, execution history, defined terms, and signature authority. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +2. RECORD LINEAGE AND CUSTODY +Matter operations catalogued the record in Vanta control register using identifier CB-DOC-061-010. The chain of custody identifies Owen Delgado as source owner and Mei Whitaker as the most recent reviewer. Any inconsistency with CB-DOC-061-055 must be reconciled rather than silently overwritten. + +3. MATTER AND WORKSTREAM CONTEXT +The processing records workstream sits within privacy and regulatory response audit. The team is tracking cross-border transfer gap, incident-notification clock, and regulator commitment overdue because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +4. OPERATIVE CONTENT +The operative entry states that the item is recorded as closed without exception in CB-PR-3001-D806. The record-control overlay classifies this as the primary source for F-04 (processor-contract omission) at medium severity. The assigned remediation owner is Nora Chen, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +5. REVIEWER ANALYSIS +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Mei Whitaker recorded status “reviewed — source conflict identified” and linked the file to CB-DOC-061-055. The control metric 553074 is an administrative population identifier, not a damages estimate or a statement of materiality. + +6. DEPENDENCIES AND RELIANCE LIMITS +The record should be read with the cross-border transfer gap, incident-notification clock, and regulator commitment overdue materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +SCHEDULE 1 — RECORD CHRONOLOGY +- 2025-03-21 | Source population opened for collection | Owen Delgado | CB-DOC-061-010 +- 2025-04-21 | Custodian confirmed system and date boundary | Maya Ellison | CB-DOC-061-055 +- 2025-05-01 | Matter team completed first-level comparison | Dominic Alvarez | CB-DOC-061-055 +- 2025-05-09 | Legal reviewer recorded the current disposition | Talia Brooks | CB-DOC-061-010 +- 2025-06-01 | Assigned owner scheduled the next control response | Caleb Hassan | CB-DOC-061-055 + +SCHEDULE 2 — ACTION REGISTER +- A-010-1 | escalated | Nora Chen | 2026-08-19 | Before 2026-09-02, issue a corrective notice using the contractually operative method; owner: Nora Chen. +- A-010-2 | open | Adrian Mensah | 2026-08-26 | Preserve the native Vanta control register export and document any replacement record. +- A-010-3 | in review | Farah Ibrahim | 2026-09-02 | Report the disposition to the processing records workstream lead before the matter deadline. + SCOPE -The record covers activity in California through 2025-05-09 and should be evaluated with CB-DOC-061-055. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-05-09 and should be evaluated with CB-DOC-061-055, the four related records listed below, and the complete processing records folder. It is not a complete statement of law or a substitute for the other folders in production. CONTROL NOTE -Reviewer Mei Whitaker compared identifier 553074 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Mei Whitaker compared identifier 553074 against the folder index, the native Vanta control register entry, and CB-DOC-061-055. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. + +ACKNOWLEDGED FOR RECORD-CONTROL PURPOSES + +By: Owen Delgado +Role: Producing custodian +Reviewed by: Mei Whitaker +Record date: 2025-05-09 PROVENANCE Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. diff --git a/task_files/cb100-061-arcadia-genomics/02_processing_records/011_processing_records_correspondence.eml b/task_files/cb100-061-arcadia-genomics/02_processing_records/011_processing_records_correspondence.eml index bb90595c0bf77a01327266655480bb0a02e62d89..cb9a5bba7af7f4c682fb0e769f10fea0ad7a54f5 100644 --- a/task_files/cb100-061-arcadia-genomics/02_processing_records/011_processing_records_correspondence.eml +++ b/task_files/cb100-061-arcadia-genomics/02_processing_records/011_processing_records_correspondence.eml @@ -1,20 +1,77 @@ From: lena.varga@example.test To: theo.laurent@example.test +Cc: matter-team-arcadia_genomics_inc@example.test Date: 2025-01-21 09:30:00 -0700 Message-ID:| Record Id | CB-DOC-061-015 |
|---|
| Record Id | CB-DOC-061-015 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2025-09-06 |
| Record Type | officer certificate |
| Folder | 02_processing_records |
| Workstream | processing records |
| Source System | BigID deletion log |
| Native Version | 3.8 |
| Record Status | indexed — responsive context |
| Confidentiality | Restricted — need to know |
| Custodian | Nora Chen |
| Reviewer | Micah Sullivan |
| Cross Reference | CB-DOC-061-044 |
| Control Metric | 688154 |
| Review Question | context only |
| Finding Id | none |
| Record Role | context |
| Control Severity | none |
| Remediation Owner | none |
| Response Due | none |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 02_processing_records |
| Record Type | officer certificate |
| Custodian | Nora Chen |
| Reviewer | Micah Sullivan |
| Record Date | 2025-09-06 |
| Cross Reference | CB-DOC-061-044 |
| Control Metric | 688154 |
| Operative Text | The record preserves the original value and separately identifies the later administrative annotation. |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Nora Chen, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2025-09-06 and should be evaluated with CB-DOC-061-044. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Micah Sullivan compared identifier 688154 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the processing records workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
Matter operations catalogued the record in BigID deletion log using identifier CB-DOC-061-015. The chain of custody identifies Nora Chen as source owner and Micah Sullivan as the most recent reviewer. Any inconsistency with CB-DOC-061-044 must be reconciled rather than silently overwritten.
For this processing records review, legal and business stakeholders are using the record to evaluate incident-notification clock, regulator commitment overdue, and subprocessor inventory mismatch. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that The record preserves the original value and separately identifies the later administrative annotation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
The legal-control check compared the record’s native value with the matter index and the identified related file. Conflicting custodial accounts, amendments outside the production, and post-date events were not resolved by inference and remain subject to documented follow-up. Reviewer Micah Sullivan recorded status “indexed — responsive context” and linked the file to CB-DOC-061-044. The control metric 688154 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the incident-notification clock, regulator commitment overdue, and subprocessor inventory mismatch materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2025-07-18 | Source population opened for collection | Nora Chen | CB-DOC-061-015 |
| 2025-08-16 | Custodian confirmed system and date boundary | Elliot Mercer | CB-DOC-061-044 |
| 2025-08-29 | Matter team completed first-level comparison | Mei Whitaker | CB-DOC-061-044 |
| 2025-09-06 | Legal reviewer recorded the current disposition | Isaac Romero | CB-DOC-061-015 |
| 2025-10-11 | Assigned owner scheduled the next control response | Willa Novak | CB-DOC-061-044 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-015-1 | Confirm that CB-DOC-061-044 does not change the context-only classification. | Isaac Romero | 2026-08-19 | escalated |
| A-015-2 | Preserve the native BigID deletion log export and document any replacement record. | Nadine Flores | 2026-08-26 | open |
| A-015-3 | Report the disposition to the processing records workstream lead before the matter deadline. | Rafael Okafor | 2026-09-02 | in review |
The record covers activity in California through 2025-09-06 and should be evaluated with CB-DOC-061-044, the four related records listed below, and the complete processing records folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Micah Sullivan compared identifier 688154 against the folder index, the native BigID deletion log entry, and CB-DOC-061-044. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/02_processing_records/016_processing_records_status_report.md b/task_files/cb100-061-arcadia-genomics/02_processing_records/016_processing_records_status_report.md index c4f7d332f59b712062273db4fa8152b16498f9e2..d1d0a7a66f005032f7cb2118ba74a0d9e1f8c5b1 100644 --- a/task_files/cb100-061-arcadia-genomics/02_processing_records/016_processing_records_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/02_processing_records/016_processing_records_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-016 +> Restricted — need to know · reviewed — source conflict identified · native version 1.3 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Dominic Alvarez | | Reviewer | Nora Chen | | Cross-reference | CB-DOC-061-061 | +| Source system | Vanta control register | +| Workstream | processing records | | Control metric | 96196 | +| Review question | cross-border transfer gap | +| Finding ID | F-06 | +| Record role | primary | +| Control severity | low | +| Remediation owner | Priya Raman | +| Response due | 2026-09-02 | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The processing records team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Dominic Alvarez identified it as an ordinary-course record from Vanta control register; reviewer Nora Chen preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** the threshold is 13% with no stated tolerance + +## 1. Purpose and audience + +This status report supports the processing records workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +Matter operations catalogued the record in Vanta control register using identifier CB-DOC-061-016. The chain of custody identifies Dominic Alvarez as source owner and Nora Chen as the most recent reviewer. Any inconsistency with CB-DOC-061-061 must be reconciled rather than silently overwritten. + +## 3. Matter and workstream context + +The processing records workstream sits within privacy and regulatory response audit. The team is tracking risk-assessment omission, processing-purpose gap, and cross-border transfer gap because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that the threshold is 13% with no stated tolerance. The record-control overlay classifies this as the primary source for F-06 (cross-border transfer gap) at low severity. The assigned remediation owner is Priya Raman, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Nora Chen recorded status “reviewed — source conflict identified” and linked the file to CB-DOC-061-061. The control metric 96196 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the risk-assessment omission, processing-purpose gap, and cross-border transfer gap materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2025-09-12 | Source population opened for collection | Dominic Alvarez | CB-DOC-061-016 | +| 2025-09-30 | Custodian confirmed system and date boundary | Talia Brooks | CB-DOC-061-061 | +| 2025-10-13 | Matter team completed first-level comparison | Caleb Hassan | CB-DOC-061-061 | +| 2025-10-18 | Legal reviewer recorded the current disposition | Sofia Bennett | CB-DOC-061-016 | +| 2025-11-10 | Assigned owner scheduled the next control response | Theo Laurent | CB-DOC-061-061 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Theo Laurent | control owner | source completeness | +| Farah Ibrahim | matter lead | business interpretation | +| Rafael Okafor | business owner | legal review | +| Talia Brooks | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Dominic Alvarez, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-023` — same-cycle source (vendor reviews) +- `CB-DOC-061-035` — implementation evidence (consumer requests) +- `CB-DOC-061-053` — independent control record (transfer mechanisms) +- `CB-DOC-061-069` — later reconciliation record (training) -## Operative record +## Action register -the threshold is 13% with no stated tolerance +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-016-1 | Before 2026-09-02, document the governing interpretation before the deadline; owner: Priya Raman. | Priya Raman | 2026-08-19 | open | +| A-016-2 | Preserve the native Vanta control register export and document any replacement record. | Micah Sullivan | 2026-08-26 | in review | +| A-016-3 | Report the disposition to the processing records workstream lead before the matter deadline. | Nora Chen | 2026-09-02 | awaiting evidence | ## Scope and cross-reference -The record covers activity in California through 2025-10-18 and should be evaluated with CB-DOC-061-061. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-10-18 and should be evaluated with CB-DOC-061-061, the four related records listed below, and the complete processing records folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Nora Chen compared identifier 96196 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Nora Chen compared identifier 96196 against the folder index, the native Vanta control register entry, and CB-DOC-061-061. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/049_transfer_mechanisms_control_register.md b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/049_transfer_mechanisms_control_register.md index ab26b8c496df7ff072891586a6efbaece4d1e9ae..32cfe0368d0da7bf8618f6e7282150e7208db00c 100644 --- a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/049_transfer_mechanisms_control_register.md +++ b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/049_transfer_mechanisms_control_register.md @@ -1,5 +1,7 @@ # Control Register — CB-DOC-061-049 +> Attorney work product · reviewed — variance confirmed · native version 2.3 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Sofia Bennett | | Reviewer | Henry Cho | | Cross-reference | CB-DOC-061-046 | +| Source system | OneTrust data map | +| Workstream | transfer mechanisms | | Control metric | 530617 | +| Review question | processing-purpose gap | +| Finding ID | F-01 | +| Record role | corroborating | +| Control severity | critical | +| Remediation owner | Samuel Kim | +| Response due | 2026-09-02 | + +## Executive record summary + +This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The transfer mechanisms team used the record to maintain the matter team's issue population, source lineage, review status, and accountable owner. Producing custodian Sofia Bennett identified it as an ordinary-course record from OneTrust data map; reviewer Henry Cho preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** the acknowledged date is 2025-03-06 under response reference CB-PR-3001-J440 + +## 1. Purpose and audience + +This control register supports the transfer mechanisms workstream for Arcadia Genomics, Inc.. Its purpose is to maintain the matter team's issue population, source lineage, review status, and accountable owner. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +The producing team exported this record from OneTrust data map and retained the native identifier CB-DOC-061-049. Sofia Bennett confirmed the export boundary, while Henry Cho performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-046. + +## 3. Matter and workstream context + +The transfer mechanisms workstream sits within privacy and regulatory response audit. The team is tracking sensitive-data classification, notice-practice mismatch, and retention schedule conflict because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that the acknowledged date is 2025-03-06 under response reference CB-PR-3001-J440. The record-control overlay classifies this as the corroborating source for F-01 (processing-purpose gap) at critical severity. The assigned remediation owner is Samuel Kim, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Henry Cho recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-046. The control metric 530617 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the sensitive-data classification, notice-practice mismatch, and retention schedule conflict materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-11-05 | Source population opened for collection | Sofia Bennett | CB-DOC-061-049 | +| 2024-12-07 | Custodian confirmed system and date boundary | Theo Laurent | CB-DOC-061-046 | +| 2024-12-17 | Matter team completed first-level comparison | Nadine Flores | CB-DOC-061-046 | +| 2024-12-23 | Legal reviewer recorded the current disposition | Owen Delgado | CB-DOC-061-049 | +| 2025-01-22 | Assigned owner scheduled the next control response | Maya Ellison | CB-DOC-061-046 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Maya Ellison | matter lead | source completeness | +| Elliot Mercer | business owner | business interpretation | +| Leona Park | records custodian | legal review | +| Theo Laurent | finance reviewer | remediation tracking | -## Background +## Related records -This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Sofia Bennett, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-056` — same-cycle source (transfer mechanisms) +- `CB-DOC-061-068` — implementation evidence (training) +- `CB-DOC-061-086` — independent control record (risk assessments) +- `CB-DOC-061-006` — later reconciliation record (data inventory) -## Operative record +## Action register -the acknowledged date is 2025-03-06 under response reference CB-PR-3001-J440 +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-049-1 | Before 2026-09-02, obtain a signed ratification and update the controlling register; owner: Samuel Kim. | Samuel Kim | 2026-08-19 | owner confirmed | +| A-049-2 | Preserve the native OneTrust data map export and document any replacement record. | Priya Raman | 2026-08-26 | escalated | +| A-049-3 | Report the disposition to the transfer mechanisms workstream lead before the matter deadline. | Isaac Romero | 2026-09-02 | open | ## Scope and cross-reference -The record covers activity in California through 2024-12-23 and should be evaluated with CB-DOC-061-046. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-12-23 and should be evaluated with CB-DOC-061-046, the four related records listed below, and the complete transfer mechanisms folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Henry Cho compared identifier 530617 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Henry Cho compared identifier 530617 against the folder index, the native OneTrust data map entry, and CB-DOC-061-046. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/050_transfer_mechanisms_executed_instrument.txt b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/050_transfer_mechanisms_executed_instrument.txt index 2dc35e0791b0395d92d61964d59d10e36d729cb5..977a04b38c81c8379e271c103794ad5dfaf41754 100644 --- a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/050_transfer_mechanisms_executed_instrument.txt +++ b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/050_transfer_mechanisms_executed_instrument.txt @@ -1,23 +1,78 @@ +RESTRICTED — NEED TO KNOW + +EXECUTED INSTRUMENT +ARCADIA GENOMIC DATA RESPONSE AUDIT + DOCUMENT CONTROL: CB-DOC-061-050 MATTER: CB-PR-3001 | Arcadia genomic data response audit RECORD TYPE: executed instrument DATE: 2024-03-22 +SOURCE SYSTEM: ServiceNow privacy queue +NATIVE VERSION: 2.1 +STATUS: reviewed — no independent exception CUSTODIAN: Adrian Mensah REVIEWER: Maya Ellison CROSS-REFERENCE: CB-DOC-061-063 CONTROL METRIC: 521301 -BACKGROUND -This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Adrian Mensah, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +REVIEW QUESTION: context only +FINDING ID: none +RECORD ROLE: context +CONTROL SEVERITY: none +REMEDIATION OWNER: none +RESPONSE DUE: none + +RECITALS + +A. This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The transfer mechanisms team used the record to preserve the operative language, execution history, defined terms, and signature authority. Producing custodian Adrian Mensah identified it as an ordinary-course record from ServiceNow privacy queue; reviewer Maya Ellison preserved the source wording and separately recorded the legal-control disposition. + +B. The parties and matter team intend to preserve the source hierarchy and the decision record for CB-PR-3001. OPERATIVE RECORD The custodian confirmed that this export is complete for the stated period, subject to the cross-reference below. +1. PURPOSE AND AUDIENCE +This executed instrument supports the transfer mechanisms workstream for Arcadia Genomics, Inc.. Its purpose is to preserve the operative language, execution history, defined terms, and signature authority. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +2. RECORD LINEAGE AND CUSTODY +The producing team exported this record from ServiceNow privacy queue and retained the native identifier CB-DOC-061-050. Adrian Mensah confirmed the export boundary, while Maya Ellison performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-063. + +3. MATTER AND WORKSTREAM CONTEXT +For this transfer mechanisms review, legal and business stakeholders are using the record to evaluate minor-consent issue, consent-record deficiency, and deletion-job failure. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +4. OPERATIVE CONTENT +The operative entry states that The custodian confirmed that this export is complete for the stated period, subject to the cross-reference below.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +5. REVIEWER ANALYSIS +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Maya Ellison recorded status “reviewed — no independent exception” and linked the file to CB-DOC-061-063. The control metric 521301 is an administrative population identifier, not a damages estimate or a statement of materiality. + +6. DEPENDENCIES AND RELIANCE LIMITS +The record should be read with the minor-consent issue, consent-record deficiency, and deletion-job failure materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion. + +SCHEDULE 1 — RECORD CHRONOLOGY +- 2024-02-02 | Source population opened for collection | Adrian Mensah | CB-DOC-061-050 +- 2024-03-03 | Custodian confirmed system and date boundary | Amara Patel | CB-DOC-061-063 +- 2024-03-17 | Matter team completed first-level comparison | Micah Sullivan | CB-DOC-061-063 +- 2024-03-22 | Legal reviewer recorded the current disposition | Lena Varga | CB-DOC-061-050 +- 2024-04-15 | Assigned owner scheduled the next control response | Rafael Okafor | CB-DOC-061-063 + +SCHEDULE 2 — ACTION REGISTER +- A-050-1 | escalated | Lena Varga | 2026-08-19 | Confirm that CB-DOC-061-063 does not change the context-only classification. +- A-050-2 | open | Elliot Mercer | 2026-08-26 | Preserve the native ServiceNow privacy queue export and document any replacement record. +- A-050-3 | in review | Sofia Bennett | 2026-09-02 | Report the disposition to the transfer mechanisms workstream lead before the matter deadline. + SCOPE -The record covers activity in California through 2024-03-22 and should be evaluated with CB-DOC-061-063. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-03-22 and should be evaluated with CB-DOC-061-063, the four related records listed below, and the complete transfer mechanisms folder. It is not a complete statement of law or a substitute for the other folders in production. CONTROL NOTE -Reviewer Maya Ellison compared identifier 521301 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Maya Ellison compared identifier 521301 against the folder index, the native ServiceNow privacy queue entry, and CB-DOC-061-063. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. + +ACKNOWLEDGED FOR RECORD-CONTROL PURPOSES + +By: Adrian Mensah +Role: Producing custodian +Reviewed by: Maya Ellison +Record date: 2024-03-22 PROVENANCE Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. diff --git a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/051_transfer_mechanisms_correspondence.eml b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/051_transfer_mechanisms_correspondence.eml index a048220065d69712b9b4af9240ac860fc17f28d1..b290e5d8b34cc2813fa5c7fccd7f4c0ab1167261 100644 --- a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/051_transfer_mechanisms_correspondence.eml +++ b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/051_transfer_mechanisms_correspondence.eml @@ -1,20 +1,77 @@ From: willa.novak@example.test To: jonas.feld@example.test +Cc: matter-team-arcadia_genomics_inc@example.test Date: 2026-02-16 09:30:00 -0700 Message-ID:| Record Id | CB-DOC-061-055 |
|---|
| Record Id | CB-DOC-061-055 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2025-10-18 |
| Record Type | officer certificate |
| Folder | 07_transfer_mechanisms |
| Workstream | transfer mechanisms |
| Source System | OneTrust data map |
| Native Version | 1.4 |
| Record Status | reviewed — variance confirmed |
| Confidentiality | Restricted — need to know |
| Custodian | Nadine Flores |
| Reviewer | Isaac Romero |
| Cross Reference | CB-DOC-061-052 |
| Control Metric | 63156 |
| Review Question | consent-record deficiency |
| Finding Id | F-03 |
| Record Role | corroborating |
| Control Severity | high |
| Remediation Owner | Rafael Okafor |
| Response Due | 2026-09-02 |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 07_transfer_mechanisms |
| Record Type | officer certificate |
| Custodian | Nadine Flores |
| Reviewer | Isaac Romero |
| Record Date | 2025-10-18 |
| Cross Reference | CB-DOC-061-052 |
| Control Metric | 63156 |
| Operative Text | the approval log names Henry Cho and contains no entry for Nadine Flores |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Nadine Flores, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2025-10-18 and should be evaluated with CB-DOC-061-052. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Isaac Romero compared identifier 63156 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the transfer mechanisms workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
Matter operations catalogued the record in OneTrust data map using identifier CB-DOC-061-055. The chain of custody identifies Nadine Flores as source owner and Isaac Romero as the most recent reviewer. Any inconsistency with CB-DOC-061-052 must be reconciled rather than silently overwritten.
The transfer mechanisms workstream sits within privacy and regulatory response audit. The team is tracking consent-record deficiency, deletion-job failure, and sensitive-data classification because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that the approval log names Henry Cho and contains no entry for Nadine Flores. The record-control overlay classifies this as the corroborating source for F-03 (consent-record deficiency) at high severity. The assigned remediation owner is Rafael Okafor, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Isaac Romero recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-052. The control metric 63156 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the consent-record deficiency, deletion-job failure, and sensitive-data classification materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2025-09-12 | Source population opened for collection | Nadine Flores | CB-DOC-061-055 |
| 2025-09-30 | Custodian confirmed system and date boundary | Owen Delgado | CB-DOC-061-052 |
| 2025-10-13 | Matter team completed first-level comparison | Maya Ellison | CB-DOC-061-052 |
| 2025-10-18 | Legal reviewer recorded the current disposition | Dominic Alvarez | CB-DOC-061-055 |
| 2025-11-10 | Assigned owner scheduled the next control response | Talia Brooks | CB-DOC-061-052 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-055-1 | Before 2026-09-02, secure written consent from the authorized decision-maker; owner: Rafael Okafor. | Rafael Okafor | 2026-08-19 | escalated |
| A-055-2 | Preserve the native OneTrust data map export and document any replacement record. | Leona Park | 2026-08-26 | open |
| A-055-3 | Report the disposition to the transfer mechanisms workstream lead before the matter deadline. | Henry Cho | 2026-09-02 | in review |
The record covers activity in California through 2025-10-18 and should be evaluated with CB-DOC-061-052, the four related records listed below, and the complete transfer mechanisms folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Isaac Romero compared identifier 63156 against the folder index, the native OneTrust data map entry, and CB-DOC-061-052. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/056_transfer_mechanisms_status_report.md b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/056_transfer_mechanisms_status_report.md index 06170540b27d63669aac5da7bcb6edf76e544a9c..b3d56da408f7203abe18f37720dec71ae170d511 100644 --- a/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/056_transfer_mechanisms_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/07_transfer_mechanisms/056_transfer_mechanisms_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-056 +> Attorney work product · reviewed — no independent exception · native version 1.7 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Micah Sullivan | | Reviewer | Nadine Flores | | Cross-reference | CB-DOC-061-069 | +| Source system | ServiceNow privacy queue | +| Workstream | transfer mechanisms | | Control metric | 239924 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The transfer mechanisms team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Micah Sullivan identified it as an ordinary-course record from ServiceNow privacy queue; reviewer Nadine Flores preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** The file is responsive background material; it supplies context but does not create an independent obligation. + +## 1. Purpose and audience + +This status report supports the transfer mechanisms workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +Matter operations catalogued the record in ServiceNow privacy queue using identifier CB-DOC-061-056. The chain of custody identifies Micah Sullivan as source owner and Nadine Flores as the most recent reviewer. Any inconsistency with CB-DOC-061-069 must be reconciled rather than silently overwritten. + +## 3. Matter and workstream context + +The transfer mechanisms workstream sits within privacy and regulatory response audit. The team is tracking processor-contract omission, access-request deadline, and minor-consent issue because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that The file is responsive background material; it supplies context but does not create an independent obligation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Nadine Flores recorded status “reviewed — no independent exception” and linked the file to CB-DOC-061-069. The control metric 239924 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the processor-contract omission, access-request deadline, and minor-consent issue materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-06-26 | Source population opened for collection | Micah Sullivan | CB-DOC-061-056 | +| 2024-07-25 | Custodian confirmed system and date boundary | Lena Varga | CB-DOC-061-069 | +| 2024-08-04 | Matter team completed first-level comparison | Rafael Okafor | CB-DOC-061-069 | +| 2024-08-12 | Legal reviewer recorded the current disposition | Priya Raman | CB-DOC-061-056 | +| 2024-09-04 | Assigned owner scheduled the next control response | Jonas Feld | CB-DOC-061-069 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Jonas Feld | control owner | source completeness | +| Sofia Bennett | matter lead | business interpretation | +| Henry Cho | business owner | legal review | +| Lena Varga | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Micah Sullivan, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-063` — same-cycle source (retention deletion) +- `CB-DOC-061-075` — implementation evidence (regulator correspondence) +- `CB-DOC-061-093` — independent control record (remediation) +- `CB-DOC-061-013` — later reconciliation record (processing records) -## Operative record +## Action register -The file is responsive background material; it supplies context but does not create an independent obligation. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-056-1 | Confirm that CB-DOC-061-069 does not change the context-only classification. | Priya Raman | 2026-08-19 | open | +| A-056-2 | Preserve the native ServiceNow privacy queue export and document any replacement record. | Isaac Romero | 2026-08-26 | in review | +| A-056-3 | Report the disposition to the transfer mechanisms workstream lead before the matter deadline. | Nadine Flores | 2026-09-02 | awaiting evidence | ## Scope and cross-reference -The record covers activity in California through 2024-08-12 and should be evaluated with CB-DOC-061-069. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-08-12 and should be evaluated with CB-DOC-061-069, the four related records listed below, and the complete transfer mechanisms folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Nadine Flores compared identifier 239924 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Nadine Flores compared identifier 239924 against the folder index, the native ServiceNow privacy queue entry, and CB-DOC-061-069. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/057_retention_deletion_control_register.md b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/057_retention_deletion_control_register.md index cbe81eeb781cb96c859cc4e3c5e58d9e8929d13e..7b6a294061a0579694e707486b9bed1d9dd1b3df 100644 --- a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/057_retention_deletion_control_register.md +++ b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/057_retention_deletion_control_register.md @@ -1,5 +1,7 @@ # Control Register — CB-DOC-061-057 +> Attorney work product · indexed — responsive context · native version 3.6 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Farah Ibrahim | | Reviewer | Rafael Okafor | | Cross-reference | CB-DOC-061-086 | +| Source system | BigID deletion log | +| Workstream | retention deletion | | Control metric | 73982 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The retention deletion team used the record to maintain the matter team's issue population, source lineage, review status, and accountable owner. Producing custodian Farah Ibrahim identified it as an ordinary-course record from BigID deletion log; reviewer Rafael Okafor preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** No variance was recorded after the named reviewer compared the source ledger to the signed control record. + +## 1. Purpose and audience + +This control register supports the retention deletion workstream for Arcadia Genomics, Inc.. Its purpose is to maintain the matter team's issue population, source lineage, review status, and accountable owner. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +This copy was collected from BigID deletion log under matter hold CB-PR-3001. Its lineage runs from Farah Ibrahim, as producing custodian, to Rafael Okafor, as reviewing lawyer. The related record CB-DOC-061-086 remains a separate source of truth and was not merged into this document. + +## 3. Matter and workstream context + +For this retention deletion review, legal and business stakeholders are using the record to evaluate subprocessor inventory mismatch, identity-verification inconsistency, and training completion gap. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that No variance was recorded after the named reviewer compared the source ledger to the signed control record.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Rafael Okafor recorded status “indexed — responsive context” and linked the file to CB-DOC-061-086. The control metric 73982 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the subprocessor inventory mismatch, identity-verification inconsistency, and training completion gap materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2025-02-05 | Source population opened for collection | Farah Ibrahim | CB-DOC-061-057 | +| 2025-02-26 | Custodian confirmed system and date boundary | Samuel Kim | CB-DOC-061-086 | +| 2025-03-12 | Matter team completed first-level comparison | Nora Chen | CB-DOC-061-086 | +| 2025-03-18 | Legal reviewer recorded the current disposition | Elliot Mercer | CB-DOC-061-057 | +| 2025-04-12 | Assigned owner scheduled the next control response | Mei Whitaker | CB-DOC-061-086 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Mei Whitaker | matter lead | source completeness | +| Adrian Mensah | business owner | business interpretation | +| Nadine Flores | records custodian | legal review | +| Samuel Kim | finance reviewer | remediation tracking | -## Background +## Related records -This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Farah Ibrahim, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-064` — same-cycle source (retention deletion) +- `CB-DOC-061-076` — implementation evidence (regulator correspondence) +- `CB-DOC-061-094` — independent control record (remediation) +- `CB-DOC-061-014` — later reconciliation record (processing records) -## Operative record +## Action register -No variance was recorded after the named reviewer compared the source ledger to the signed control record. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-057-1 | Confirm that CB-DOC-061-086 does not change the context-only classification. | Elliot Mercer | 2026-08-19 | in review | +| A-057-2 | Preserve the native BigID deletion log export and document any replacement record. | Sofia Bennett | 2026-08-26 | awaiting evidence | +| A-057-3 | Report the disposition to the retention deletion workstream lead before the matter deadline. | Micah Sullivan | 2026-09-02 | owner confirmed | ## Scope and cross-reference -The record covers activity in California through 2025-03-18 and should be evaluated with CB-DOC-061-086. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-03-18 and should be evaluated with CB-DOC-061-086, the four related records listed below, and the complete retention deletion folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Rafael Okafor compared identifier 73982 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Rafael Okafor compared identifier 73982 against the folder index, the native BigID deletion log entry, and CB-DOC-061-086. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/058_retention_deletion_executed_instrument.txt b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/058_retention_deletion_executed_instrument.txt index d815f58708f9334d2e822051d9c7d8f002dad026..a66f97e305badea0502ba62b9518f1f0bc7dcc35 100644 --- a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/058_retention_deletion_executed_instrument.txt +++ b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/058_retention_deletion_executed_instrument.txt @@ -1,23 +1,78 @@ +ATTORNEY WORK PRODUCT + +EXECUTED INSTRUMENT +ARCADIA GENOMIC DATA RESPONSE AUDIT + DOCUMENT CONTROL: CB-DOC-061-058 MATTER: CB-PR-3001 | Arcadia genomic data response audit RECORD TYPE: executed instrument DATE: 2026-01-27 +SOURCE SYSTEM: Vanta control register +NATIVE VERSION: 2.3 +STATUS: reviewed — variance confirmed CUSTODIAN: Owen Delgado REVIEWER: Mei Whitaker CROSS-REFERENCE: CB-DOC-061-007 CONTROL METRIC: 446537 -BACKGROUND -This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Owen Delgado, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +REVIEW QUESTION: processor-contract omission +FINDING ID: F-04 +RECORD ROLE: corroborating +CONTROL SEVERITY: medium +REMEDIATION OWNER: Nora Chen +RESPONSE DUE: 2026-09-02 + +RECITALS + +A. This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The retention deletion team used the record to preserve the operative language, execution history, defined terms, and signature authority. Producing custodian Owen Delgado identified it as an ordinary-course record from Vanta control register; reviewer Mei Whitaker preserved the source wording and separately recorded the legal-control disposition. + +B. The parties and matter team intend to preserve the source hierarchy and the decision record for CB-PR-3001. OPERATIVE RECORD the later status register marks it open and escalated in CB-PR-3001-M915 +1. PURPOSE AND AUDIENCE +This executed instrument supports the retention deletion workstream for Arcadia Genomics, Inc.. Its purpose is to preserve the operative language, execution history, defined terms, and signature authority. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +2. RECORD LINEAGE AND CUSTODY +This copy was collected from Vanta control register under matter hold CB-PR-3001. Its lineage runs from Owen Delgado, as producing custodian, to Mei Whitaker, as reviewing lawyer. The related record CB-DOC-061-007 remains a separate source of truth and was not merged into this document. + +3. MATTER AND WORKSTREAM CONTEXT +For this retention deletion review, legal and business stakeholders are using the record to evaluate cross-border transfer gap, incident-notification clock, and regulator commitment overdue. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +4. OPERATIVE CONTENT +The operative entry states that the later status register marks it open and escalated in CB-PR-3001-M915. The record-control overlay classifies this as the corroborating source for F-04 (processor-contract omission) at medium severity. The assigned remediation owner is Nora Chen, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +5. REVIEWER ANALYSIS +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Mei Whitaker recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-007. The control metric 446537 is an administrative population identifier, not a damages estimate or a statement of materiality. + +6. DEPENDENCIES AND RELIANCE LIMITS +The record should be read with the cross-border transfer gap, incident-notification clock, and regulator commitment overdue materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +SCHEDULE 1 — RECORD CHRONOLOGY +- 2025-12-22 | Source population opened for collection | Owen Delgado | CB-DOC-061-058 +- 2026-01-07 | Custodian confirmed system and date boundary | Maya Ellison | CB-DOC-061-007 +- 2026-01-21 | Matter team completed first-level comparison | Dominic Alvarez | CB-DOC-061-007 +- 2026-01-27 | Legal reviewer recorded the current disposition | Talia Brooks | CB-DOC-061-058 +- 2026-03-02 | Assigned owner scheduled the next control response | Caleb Hassan | CB-DOC-061-007 + +SCHEDULE 2 — ACTION REGISTER +- A-058-1 | awaiting evidence | Nora Chen | 2026-08-19 | Before 2026-09-02, issue a corrective notice using the contractually operative method; owner: Nora Chen. +- A-058-2 | owner confirmed | Adrian Mensah | 2026-08-26 | Preserve the native Vanta control register export and document any replacement record. +- A-058-3 | escalated | Farah Ibrahim | 2026-09-02 | Report the disposition to the retention deletion workstream lead before the matter deadline. + SCOPE -The record covers activity in California through 2026-01-27 and should be evaluated with CB-DOC-061-007. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2026-01-27 and should be evaluated with CB-DOC-061-007, the four related records listed below, and the complete retention deletion folder. It is not a complete statement of law or a substitute for the other folders in production. CONTROL NOTE -Reviewer Mei Whitaker compared identifier 446537 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Mei Whitaker compared identifier 446537 against the folder index, the native Vanta control register entry, and CB-DOC-061-007. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. + +ACKNOWLEDGED FOR RECORD-CONTROL PURPOSES + +By: Owen Delgado +Role: Producing custodian +Reviewed by: Mei Whitaker +Record date: 2026-01-27 PROVENANCE Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. diff --git a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/059_retention_deletion_correspondence.eml b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/059_retention_deletion_correspondence.eml index 7fc0fde36fb87e33a704d6b0d4d677d62d12a137..002019b2d6d2d933934c3dea2ab74159277811c9 100644 --- a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/059_retention_deletion_correspondence.eml +++ b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/059_retention_deletion_correspondence.eml @@ -1,20 +1,77 @@ From: lena.varga@example.test To: theo.laurent@example.test +Cc: matter-team-arcadia_genomics_inc@example.test Date: 2025-04-22 09:30:00 -0700 Message-ID:| Record Id | CB-DOC-061-063 |
|---|
| Record Id | CB-DOC-061-063 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2025-03-24 |
| Record Type | officer certificate |
| Folder | 08_retention_deletion |
| Workstream | retention deletion |
| Source System | BigID deletion log |
| Native Version | 1.6 |
| Record Status | indexed — responsive context |
| Confidentiality | Attorney work product |
| Custodian | Nora Chen |
| Reviewer | Micah Sullivan |
| Cross Reference | CB-DOC-061-092 |
| Control Metric | 78388 |
| Review Question | context only |
| Finding Id | none |
| Record Role | context |
| Control Severity | none |
| Remediation Owner | none |
| Response Due | none |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 08_retention_deletion |
| Record Type | officer certificate |
| Custodian | Nora Chen |
| Reviewer | Micah Sullivan |
| Record Date | 2025-03-24 |
| Cross Reference | CB-DOC-061-092 |
| Control Metric | 78388 |
| Operative Text | The record preserves the original value and separately identifies the later administrative annotation. |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Nora Chen, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2025-03-24 and should be evaluated with CB-DOC-061-092. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Micah Sullivan compared identifier 78388 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the retention deletion workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
This copy was collected from BigID deletion log under matter hold CB-PR-3001. Its lineage runs from Nora Chen, as producing custodian, to Micah Sullivan, as reviewing lawyer. The related record CB-DOC-061-092 remains a separate source of truth and was not merged into this document.
The retention deletion workstream sits within privacy and regulatory response audit. The team is tracking incident-notification clock, regulator commitment overdue, and subprocessor inventory mismatch because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that The record preserves the original value and separately identifies the later administrative annotation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Micah Sullivan recorded status “indexed — responsive context” and linked the file to CB-DOC-061-092. The control metric 78388 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the incident-notification clock, regulator commitment overdue, and subprocessor inventory mismatch materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2025-02-13 | Source population opened for collection | Nora Chen | CB-DOC-061-063 |
| 2025-03-07 | Custodian confirmed system and date boundary | Elliot Mercer | CB-DOC-061-092 |
| 2025-03-17 | Matter team completed first-level comparison | Mei Whitaker | CB-DOC-061-092 |
| 2025-03-24 | Legal reviewer recorded the current disposition | Isaac Romero | CB-DOC-061-063 |
| 2025-04-24 | Assigned owner scheduled the next control response | Willa Novak | CB-DOC-061-092 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-063-1 | Confirm that CB-DOC-061-092 does not change the context-only classification. | Isaac Romero | 2026-08-19 | awaiting evidence |
| A-063-2 | Preserve the native BigID deletion log export and document any replacement record. | Nadine Flores | 2026-08-26 | owner confirmed |
| A-063-3 | Report the disposition to the retention deletion workstream lead before the matter deadline. | Rafael Okafor | 2026-09-02 | escalated |
The record covers activity in California through 2025-03-24 and should be evaluated with CB-DOC-061-092, the four related records listed below, and the complete retention deletion folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Micah Sullivan compared identifier 78388 against the folder index, the native BigID deletion log entry, and CB-DOC-061-092. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/064_retention_deletion_status_report.md b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/064_retention_deletion_status_report.md index ff22994ff4c6350eb043105d64f1a086cdca2fa4..6a509a27b1c48f1799f0e8333524d16af336ff65 100644 --- a/task_files/cb100-061-arcadia-genomics/08_retention_deletion/064_retention_deletion_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/08_retention_deletion/064_retention_deletion_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-064 +> Confidential — matter team · reviewed — variance confirmed · native version 4.4 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Dominic Alvarez | | Reviewer | Nora Chen | | Cross-reference | CB-DOC-061-013 | +| Source system | Vanta control register | +| Workstream | retention deletion | | Control metric | 428691 | +| Review question | cross-border transfer gap | +| Finding ID | F-06 | +| Record role | corroborating | +| Control severity | low | +| Remediation owner | Priya Raman | +| Response due | 2026-09-02 | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The retention deletion team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Dominic Alvarez identified it as an ordinary-course record from Vanta control register; reviewer Nora Chen preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** the applied threshold is 18% and produced a different exception result + +## 1. Purpose and audience + +This status report supports the retention deletion workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +The producing team exported this record from Vanta control register and retained the native identifier CB-DOC-061-064. Dominic Alvarez confirmed the export boundary, while Nora Chen performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-013. + +## 3. Matter and workstream context + +The operational context is the retention deletion portion of privacy and regulatory response audit. Reviewers identified dependencies involving risk-assessment omission, processing-purpose gap, and cross-border transfer gap; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that the applied threshold is 18% and produced a different exception result. The record-control overlay classifies this as the corroborating source for F-06 (cross-border transfer gap) at low severity. The assigned remediation owner is Priya Raman, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Nora Chen recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-013. The control metric 428691 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the risk-assessment omission, processing-purpose gap, and cross-border transfer gap materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-03-16 | Source population opened for collection | Dominic Alvarez | CB-DOC-061-064 | +| 2024-03-30 | Custodian confirmed system and date boundary | Talia Brooks | CB-DOC-061-013 | +| 2024-04-16 | Matter team completed first-level comparison | Caleb Hassan | CB-DOC-061-013 | +| 2024-04-21 | Legal reviewer recorded the current disposition | Sofia Bennett | CB-DOC-061-064 | +| 2024-05-27 | Assigned owner scheduled the next control response | Theo Laurent | CB-DOC-061-013 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Theo Laurent | control owner | source completeness | +| Farah Ibrahim | matter lead | business interpretation | +| Rafael Okafor | business owner | legal review | +| Talia Brooks | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Dominic Alvarez, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-071` — same-cycle source (training) +- `CB-DOC-061-083` — implementation evidence (risk assessments) +- `CB-DOC-061-005` — independent control record (data inventory) +- `CB-DOC-061-021` — later reconciliation record (vendor reviews) -## Operative record +## Action register -the applied threshold is 18% and produced a different exception result +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-064-1 | Before 2026-09-02, document the governing interpretation before the deadline; owner: Priya Raman. | Priya Raman | 2026-08-19 | owner confirmed | +| A-064-2 | Preserve the native Vanta control register export and document any replacement record. | Micah Sullivan | 2026-08-26 | escalated | +| A-064-3 | Report the disposition to the retention deletion workstream lead before the matter deadline. | Nora Chen | 2026-09-02 | open | ## Scope and cross-reference -The record covers activity in California through 2024-04-21 and should be evaluated with CB-DOC-061-013. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-04-21 and should be evaluated with CB-DOC-061-013, the four related records listed below, and the complete retention deletion folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Nora Chen compared identifier 428691 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Nora Chen compared identifier 428691 against the folder index, the native Vanta control register entry, and CB-DOC-061-013. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/09_training/065_training_control_register.md b/task_files/cb100-061-arcadia-genomics/09_training/065_training_control_register.md index 9f496e28f0307ec60dc02fa9738d1820b9109f80..ff86f2c2984f11cfe789d56be2c6fa431360f83e 100644 --- a/task_files/cb100-061-arcadia-genomics/09_training/065_training_control_register.md +++ b/task_files/cb100-061-arcadia-genomics/09_training/065_training_control_register.md @@ -1,5 +1,7 @@ # Control Register — CB-DOC-061-065 +> Attorney work product · indexed — responsive context · native version 3.2 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Priya Raman | | Reviewer | Caleb Hassan | | Cross-reference | CB-DOC-061-030 | +| Source system | vendor risk portal | +| Workstream | training | | Control metric | 491482 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The training team used the record to maintain the matter team's issue population, source lineage, review status, and accountable owner. Producing custodian Priya Raman identified it as an ordinary-course record from vendor risk portal; reviewer Caleb Hassan preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** No variance was recorded after the named reviewer compared the source ledger to the signed control record. + +## 1. Purpose and audience + +This control register supports the training workstream for Arcadia Genomics, Inc.. Its purpose is to maintain the matter team's issue population, source lineage, review status, and accountable owner. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +The producing team exported this record from vendor risk portal and retained the native identifier CB-DOC-061-065. Priya Raman confirmed the export boundary, while Caleb Hassan performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-030. + +## 3. Matter and workstream context + +The operational context is the training portion of privacy and regulatory response audit. Reviewers identified dependencies involving sensitive-data classification, notice-practice mismatch, and retention schedule conflict; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that No variance was recorded after the named reviewer compared the source ledger to the signed control record.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Caleb Hassan recorded status “indexed — responsive context” and linked the file to CB-DOC-061-030. The control metric 491482 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the sensitive-data classification, notice-practice mismatch, and retention schedule conflict materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-03-06 | Source population opened for collection | Priya Raman | CB-DOC-061-065 | +| 2024-03-21 | Custodian confirmed system and date boundary | Jonas Feld | CB-DOC-061-030 | +| 2024-04-06 | Matter team completed first-level comparison | Leona Park | CB-DOC-061-030 | +| 2024-04-12 | Legal reviewer recorded the current disposition | Adrian Mensah | CB-DOC-061-065 | +| 2024-05-09 | Assigned owner scheduled the next control response | Amara Patel | CB-DOC-061-030 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Amara Patel | matter lead | source completeness | +| Owen Delgado | business owner | business interpretation | +| Nora Chen | records custodian | legal review | +| Jonas Feld | finance reviewer | remediation tracking | -## Background +## Related records -This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Priya Raman, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-072` — same-cycle source (training) +- `CB-DOC-061-084` — implementation evidence (risk assessments) +- `CB-DOC-061-006` — independent control record (data inventory) +- `CB-DOC-061-022` — later reconciliation record (vendor reviews) -## Operative record +## Action register -No variance was recorded after the named reviewer compared the source ledger to the signed control record. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-065-1 | Confirm that CB-DOC-061-030 does not change the context-only classification. | Adrian Mensah | 2026-08-19 | escalated | +| A-065-2 | Preserve the native vendor risk portal export and document any replacement record. | Farah Ibrahim | 2026-08-26 | open | +| A-065-3 | Report the disposition to the training workstream lead before the matter deadline. | Dominic Alvarez | 2026-09-02 | in review | ## Scope and cross-reference -The record covers activity in California through 2024-04-12 and should be evaluated with CB-DOC-061-030. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-04-12 and should be evaluated with CB-DOC-061-030, the four related records listed below, and the complete training folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Caleb Hassan compared identifier 491482 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Caleb Hassan compared identifier 491482 against the folder index, the native vendor risk portal entry, and CB-DOC-061-030. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/09_training/066_training_executed_instrument.txt b/task_files/cb100-061-arcadia-genomics/09_training/066_training_executed_instrument.txt index 04ec6c45691dca8b49db86156f8ea76727d57f94..369c64ffca72969806afb158e92ae9d728836ac9 100644 --- a/task_files/cb100-061-arcadia-genomics/09_training/066_training_executed_instrument.txt +++ b/task_files/cb100-061-arcadia-genomics/09_training/066_training_executed_instrument.txt @@ -1,23 +1,78 @@ +CONFIDENTIAL — MATTER TEAM + +EXECUTED INSTRUMENT +ARCADIA GENOMIC DATA RESPONSE AUDIT + DOCUMENT CONTROL: CB-DOC-061-066 MATTER: CB-PR-3001 | Arcadia genomic data response audit RECORD TYPE: executed instrument DATE: 2024-03-27 +SOURCE SYSTEM: regulatory response room +NATIVE VERSION: 3.8 +STATUS: reviewed — no independent exception CUSTODIAN: Elliot Mercer REVIEWER: Amara Patel CROSS-REFERENCE: CB-DOC-061-047 CONTROL METRIC: 378666 -BACKGROUND -This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Elliot Mercer, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +REVIEW QUESTION: context only +FINDING ID: none +RECORD ROLE: context +CONTROL SEVERITY: none +REMEDIATION OWNER: none +RESPONSE DUE: none + +RECITALS + +A. This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The training team used the record to preserve the operative language, execution history, defined terms, and signature authority. Producing custodian Elliot Mercer identified it as an ordinary-course record from regulatory response room; reviewer Amara Patel preserved the source wording and separately recorded the legal-control disposition. + +B. The parties and matter team intend to preserve the source hierarchy and the decision record for CB-PR-3001. OPERATIVE RECORD The custodian confirmed that this export is complete for the stated period, subject to the cross-reference below. +1. PURPOSE AND AUDIENCE +This executed instrument supports the training workstream for Arcadia Genomics, Inc.. Its purpose is to preserve the operative language, execution history, defined terms, and signature authority. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +2. RECORD LINEAGE AND CUSTODY +Matter operations catalogued the record in regulatory response room using identifier CB-DOC-061-066. The chain of custody identifies Elliot Mercer as source owner and Amara Patel as the most recent reviewer. Any inconsistency with CB-DOC-061-047 must be reconciled rather than silently overwritten. + +3. MATTER AND WORKSTREAM CONTEXT +The operational context is the training portion of privacy and regulatory response audit. Reviewers identified dependencies involving minor-consent issue, consent-record deficiency, and deletion-job failure; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +4. OPERATIVE CONTENT +The operative entry states that The custodian confirmed that this export is complete for the stated period, subject to the cross-reference below.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +5. REVIEWER ANALYSIS +The legal-control check compared the record’s native value with the matter index and the identified related file. Conflicting custodial accounts, amendments outside the production, and post-date events were not resolved by inference and remain subject to documented follow-up. Reviewer Amara Patel recorded status “reviewed — no independent exception” and linked the file to CB-DOC-061-047. The control metric 378666 is an administrative population identifier, not a damages estimate or a statement of materiality. + +6. DEPENDENCIES AND RELIANCE LIMITS +The record should be read with the minor-consent issue, consent-record deficiency, and deletion-job failure materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation. + +SCHEDULE 1 — RECORD CHRONOLOGY +- 2024-02-17 | Source population opened for collection | Elliot Mercer | CB-DOC-061-066 +- 2024-03-03 | Custodian confirmed system and date boundary | Mei Whitaker | CB-DOC-061-047 +- 2024-03-22 | Matter team completed first-level comparison | Isaac Romero | CB-DOC-061-047 +- 2024-03-27 | Legal reviewer recorded the current disposition | Willa Novak | CB-DOC-061-066 +- 2024-04-25 | Assigned owner scheduled the next control response | Henry Cho | CB-DOC-061-047 + +SCHEDULE 2 — ACTION REGISTER +- A-066-1 | open | Willa Novak | 2026-08-19 | Confirm that CB-DOC-061-047 does not change the context-only classification. +- A-066-2 | in review | Owen Delgado | 2026-08-26 | Preserve the native regulatory response room export and document any replacement record. +- A-066-3 | awaiting evidence | Priya Raman | 2026-09-02 | Report the disposition to the training workstream lead before the matter deadline. + SCOPE -The record covers activity in California through 2024-03-27 and should be evaluated with CB-DOC-061-047. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-03-27 and should be evaluated with CB-DOC-061-047, the four related records listed below, and the complete training folder. It is not a complete statement of law or a substitute for the other folders in production. CONTROL NOTE -Reviewer Amara Patel compared identifier 378666 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Amara Patel compared identifier 378666 against the folder index, the native regulatory response room entry, and CB-DOC-061-047. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. + +ACKNOWLEDGED FOR RECORD-CONTROL PURPOSES + +By: Elliot Mercer +Role: Producing custodian +Reviewed by: Amara Patel +Record date: 2024-03-27 PROVENANCE Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. diff --git a/task_files/cb100-061-arcadia-genomics/09_training/067_training_correspondence.eml b/task_files/cb100-061-arcadia-genomics/09_training/067_training_correspondence.eml index 67d08795cf6e7294dfe071af3c82327520b404fd..044760ce42b3983dbc4c5957be77c0fe9ba289cf 100644 --- a/task_files/cb100-061-arcadia-genomics/09_training/067_training_correspondence.eml +++ b/task_files/cb100-061-arcadia-genomics/09_training/067_training_correspondence.eml @@ -1,20 +1,77 @@ From: talia.brooks@example.test To: samuel.kim@example.test +Cc: matter-team-arcadia_genomics_inc@example.test Date: 2024-04-11 09:30:00 -0700 Message-ID:| Record Id | CB-DOC-061-071 |
|---|
| Record Id | CB-DOC-061-071 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2024-04-01 |
| Record Type | officer certificate |
| Folder | 09_training |
| Workstream | training |
| Source System | vendor risk portal |
| Native Version | 4.0 |
| Record Status | indexed — responsive context |
| Confidentiality | Attorney work product |
| Custodian | Leona Park |
| Reviewer | Dominic Alvarez |
| Cross Reference | CB-DOC-061-036 |
| Control Metric | 409471 |
| Review Question | context only |
| Finding Id | none |
| Record Role | context |
| Control Severity | none |
| Remediation Owner | none |
| Response Due | none |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 09_training |
| Record Type | officer certificate |
| Custodian | Leona Park |
| Reviewer | Dominic Alvarez |
| Record Date | 2024-04-01 |
| Cross Reference | CB-DOC-061-036 |
| Control Metric | 409471 |
| Operative Text | The record preserves the original value and separately identifies the later administrative annotation. |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Leona Park, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2024-04-01 and should be evaluated with CB-DOC-061-036. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Dominic Alvarez compared identifier 409471 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the training workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
This copy was collected from vendor risk portal under matter hold CB-PR-3001. Its lineage runs from Leona Park, as producing custodian, to Dominic Alvarez, as reviewing lawyer. The related record CB-DOC-061-036 remains a separate source of truth and was not merged into this document.
For this training review, legal and business stakeholders are using the record to evaluate consent-record deficiency, deletion-job failure, and sensitive-data classification. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that The record preserves the original value and separately identifies the later administrative annotation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Dominic Alvarez recorded status “indexed — responsive context” and linked the file to CB-DOC-061-036. The control metric 409471 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the consent-record deficiency, deletion-job failure, and sensitive-data classification materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2024-02-19 | Source population opened for collection | Leona Park | CB-DOC-061-071 |
| 2024-03-12 | Custodian confirmed system and date boundary | Adrian Mensah | CB-DOC-061-036 |
| 2024-03-27 | Matter team completed first-level comparison | Amara Patel | CB-DOC-061-036 |
| 2024-04-01 | Legal reviewer recorded the current disposition | Micah Sullivan | CB-DOC-061-071 |
| 2024-05-05 | Assigned owner scheduled the next control response | Lena Varga | CB-DOC-061-036 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-071-1 | Confirm that CB-DOC-061-036 does not change the context-only classification. | Micah Sullivan | 2026-08-19 | open |
| A-071-2 | Preserve the native vendor risk portal export and document any replacement record. | Nora Chen | 2026-08-26 | in review |
| A-071-3 | Report the disposition to the training workstream lead before the matter deadline. | Caleb Hassan | 2026-09-02 | awaiting evidence |
The record covers activity in California through 2024-04-01 and should be evaluated with CB-DOC-061-036, the four related records listed below, and the complete training folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Dominic Alvarez compared identifier 409471 against the folder index, the native vendor risk portal entry, and CB-DOC-061-036. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/09_training/072_training_status_report.md b/task_files/cb100-061-arcadia-genomics/09_training/072_training_status_report.md index 89acd42f7258bfbd5cdab0873bab223c9dfad3ee..c7ac1905fce39a492499b252aac611bea07fb502 100644 --- a/task_files/cb100-061-arcadia-genomics/09_training/072_training_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/09_training/072_training_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-072 +> Restricted — need to know · reviewed — no independent exception · native version 4.2 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Isaac Romero | | Reviewer | Leona Park | | Cross-reference | CB-DOC-061-053 | +| Source system | regulatory response room | +| Workstream | training | | Control metric | 106975 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The training team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Isaac Romero identified it as an ordinary-course record from regulatory response room; reviewer Leona Park preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** The file is responsive background material; it supplies context but does not create an independent obligation. + +## 1. Purpose and audience + +This status report supports the training workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +This copy was collected from regulatory response room under matter hold CB-PR-3001. Its lineage runs from Isaac Romero, as producing custodian, to Leona Park, as reviewing lawyer. The related record CB-DOC-061-053 remains a separate source of truth and was not merged into this document. + +## 3. Matter and workstream context + +The operational context is the training portion of privacy and regulatory response audit. Reviewers identified dependencies involving processor-contract omission, access-request deadline, and minor-consent issue; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that The file is responsive background material; it supplies context but does not create an independent obligation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Leona Park recorded status “reviewed — no independent exception” and linked the file to CB-DOC-061-053. The control metric 106975 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the processor-contract omission, access-request deadline, and minor-consent issue materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2025-10-30 | Source population opened for collection | Isaac Romero | CB-DOC-061-072 | +| 2025-11-23 | Custodian confirmed system and date boundary | Willa Novak | CB-DOC-061-053 | +| 2025-12-12 | Matter team completed first-level comparison | Henry Cho | CB-DOC-061-053 | +| 2025-12-16 | Legal reviewer recorded the current disposition | Farah Ibrahim | CB-DOC-061-072 | +| 2026-01-13 | Assigned owner scheduled the next control response | Samuel Kim | CB-DOC-061-053 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Samuel Kim | control owner | source completeness | +| Priya Raman | matter lead | business interpretation | +| Caleb Hassan | business owner | legal review | +| Willa Novak | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Isaac Romero, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-079` — same-cycle source (regulator correspondence) +- `CB-DOC-061-091` — implementation evidence (remediation) +- `CB-DOC-061-013` — independent control record (processing records) +- `CB-DOC-061-029` — later reconciliation record (security incidents) -## Operative record +## Action register -The file is responsive background material; it supplies context but does not create an independent obligation. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-072-1 | Confirm that CB-DOC-061-053 does not change the context-only classification. | Farah Ibrahim | 2026-08-19 | in review | +| A-072-2 | Preserve the native regulatory response room export and document any replacement record. | Dominic Alvarez | 2026-08-26 | awaiting evidence | +| A-072-3 | Report the disposition to the training workstream lead before the matter deadline. | Leona Park | 2026-09-02 | owner confirmed | ## Scope and cross-reference -The record covers activity in California through 2025-12-16 and should be evaluated with CB-DOC-061-053. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-12-16 and should be evaluated with CB-DOC-061-053, the four related records listed below, and the complete training folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Leona Park compared identifier 106975 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Leona Park compared identifier 106975 against the folder index, the native regulatory response room entry, and CB-DOC-061-053. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/073_regulator_correspondence_control_register.md b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/073_regulator_correspondence_control_register.md index eb84041a57eff57f6b0c494dd07577beea7887c4..ff7e75bb0fa69a90dae8c55380ffbeb6cd03914f 100644 --- a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/073_regulator_correspondence_control_register.md +++ b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/073_regulator_correspondence_control_register.md @@ -1,5 +1,7 @@ # Control Register — CB-DOC-061-073 +> Restricted — need to know · reviewed — variance confirmed · native version 2.8 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Sofia Bennett | | Reviewer | Henry Cho | | Cross-reference | CB-DOC-061-070 | +| Source system | OneTrust data map | +| Workstream | regulator correspondence | | Control metric | 857369 | +| Review question | access-request deadline | +| Finding ID | F-09 | +| Record role | corroborating | +| Control severity | high | +| Remediation owner | Jonas Feld | +| Response due | 2026-09-02 | + +## Executive record summary + +This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The regulator correspondence team used the record to maintain the matter team's issue population, source lineage, review status, and accountable owner. Producing custodian Sofia Bennett identified it as an ordinary-course record from OneTrust data map; reviewer Henry Cho preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** the acknowledged date is 2025-02-07 under response reference CB-PR-3001-R899 + +## 1. Purpose and audience + +This control register supports the regulator correspondence workstream for Arcadia Genomics, Inc.. Its purpose is to maintain the matter team's issue population, source lineage, review status, and accountable owner. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +The producing team exported this record from OneTrust data map and retained the native identifier CB-DOC-061-073. Sofia Bennett confirmed the export boundary, while Henry Cho performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-070. + +## 3. Matter and workstream context + +The operational context is the regulator correspondence portion of privacy and regulatory response audit. Reviewers identified dependencies involving subprocessor inventory mismatch, identity-verification inconsistency, and training completion gap; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that the acknowledged date is 2025-02-07 under response reference CB-PR-3001-R899. The record-control overlay classifies this as the corroborating source for F-09 (access-request deadline) at high severity. The assigned remediation owner is Jonas Feld, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +The legal-control check compared the record’s native value with the matter index and the identified related file. Conflicting custodial accounts, amendments outside the production, and post-date events were not resolved by inference and remain subject to documented follow-up. Reviewer Henry Cho recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-070. The control metric 857369 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the subprocessor inventory mismatch, identity-verification inconsistency, and training completion gap materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-09-21 | Source population opened for collection | Sofia Bennett | CB-DOC-061-073 | +| 2024-10-14 | Custodian confirmed system and date boundary | Theo Laurent | CB-DOC-061-070 | +| 2024-10-28 | Matter team completed first-level comparison | Nadine Flores | CB-DOC-061-070 | +| 2024-11-05 | Legal reviewer recorded the current disposition | Owen Delgado | CB-DOC-061-073 | +| 2024-12-11 | Assigned owner scheduled the next control response | Maya Ellison | CB-DOC-061-070 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Maya Ellison | matter lead | source completeness | +| Elliot Mercer | business owner | business interpretation | +| Leona Park | records custodian | legal review | +| Theo Laurent | finance reviewer | remediation tracking | -## Background +## Related records -This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Sofia Bennett, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-080` — same-cycle source (regulator correspondence) +- `CB-DOC-061-092` — implementation evidence (remediation) +- `CB-DOC-061-014` — independent control record (processing records) +- `CB-DOC-061-030` — later reconciliation record (security incidents) -## Operative record +## Action register -the acknowledged date is 2025-02-07 under response reference CB-PR-3001-R899 +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-073-1 | Before 2026-09-02, obtain a signed ratification and update the controlling register; owner: Jonas Feld. | Jonas Feld | 2026-08-19 | awaiting evidence | +| A-073-2 | Preserve the native OneTrust data map export and document any replacement record. | Priya Raman | 2026-08-26 | owner confirmed | +| A-073-3 | Report the disposition to the regulator correspondence workstream lead before the matter deadline. | Isaac Romero | 2026-09-02 | escalated | ## Scope and cross-reference -The record covers activity in California through 2024-11-05 and should be evaluated with CB-DOC-061-070. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-11-05 and should be evaluated with CB-DOC-061-070, the four related records listed below, and the complete regulator correspondence folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Henry Cho compared identifier 857369 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Henry Cho compared identifier 857369 against the folder index, the native OneTrust data map entry, and CB-DOC-061-070. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/077_regulator_correspondence_review_memorandum.json b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/077_regulator_correspondence_review_memorandum.json index fd8c4f0a70405f061cf1955f100e06ab6dba866f..58b068d6049780c324f12a7adfc421c219b1e75a 100644 --- a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/077_regulator_correspondence_review_memorandum.json +++ b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/077_regulator_correspondence_review_memorandum.json @@ -3,25 +3,285 @@ "record_id": "CB-DOC-061-077", "matter_number": "CB-PR-3001", "record_date": "2025-11-02", + "record_type": "review memorandum", + "folder": "10_regulator_correspondence", + "workstream": "regulator correspondence", + "source_system": "vendor risk portal", + "native_version": "4.2", + "record_status": "indexed — responsive context", + "confidentiality": "Confidential — legal review", "custodian": "Amara Patel", "reviewer": "Owen Delgado", - "record_type": "review memorandum", "cross_reference": "CB-DOC-061-042", - "control_metric": 30691 + "control_metric": 30691, + "review_question": "context only", + "finding_id": "none", + "record_role": "context", + "control_severity": "none", + "remediation_owner": "none", + "response_due": "none" }, "matter": { - "title": "Arcadia genomic data response audit", + "matter_title": "Arcadia genomic data response audit", "client": "Arcadia Genomics, Inc.", "counterparty": "North Coast Sequencing Network", "jurisdiction": "California", "venue": "California Privacy Protection Agency", - "deadline": "2026-09-02" + "deadline": "2026-09-02", + "practice_workflow": "privacy and regulatory response audit" }, "record": { - "background": "This review memorandum was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Amara Patel, identified it as an ordinary-course record used by the privacy and regulatory response audit team.", + "background": "This review memorandum was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The regulator correspondence team used the record to record the reviewer’s analysis, assumptions, unresolved questions, and recommended disposition. Producing custodian Amara Patel identified it as an ordinary-course record from vendor risk portal; reviewer Owen Delgado preserved the source wording and separately recorded the legal-control disposition.", "operative_text": "The status remained unchanged through the review date, and no exception notice appears in this record.", - "scope": "The record covers activity in California through 2025-11-02 and should be evaluated with CB-DOC-061-042. It is not a complete statement of law or a substitute for the other folders in production.", - "control_note": "Reviewer Owen Delgado compared identifier 30691 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date." + "scope": "The record covers activity in California through 2025-11-02 and should be evaluated with CB-DOC-061-042, the four related records listed below, and the complete regulator correspondence folder. It is not a complete statement of law or a substitute for the other folders in production.", + "control_note": "Reviewer Owen Delgado compared identifier 30691 against the folder index, the native vendor risk portal entry, and CB-DOC-061-042. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.", + "analysis_sections": [ + { + "heading": "Purpose and audience", + "text": "This review memorandum supports the regulator correspondence workstream for Arcadia Genomics, Inc.. Its purpose is to record the reviewer’s analysis, assumptions, unresolved questions, and recommended disposition. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision." + }, + { + "heading": "Record lineage and custody", + "text": "Matter operations catalogued the record in vendor risk portal using identifier CB-DOC-061-077. The chain of custody identifies Amara Patel as source owner and Owen Delgado as the most recent reviewer. Any inconsistency with CB-DOC-061-042 must be reconciled rather than silently overwritten." + }, + { + "heading": "Matter and workstream context", + "text": "The operational context is the regulator correspondence portion of privacy and regulatory response audit. Reviewers identified dependencies involving access-request deadline, minor-consent issue, and consent-record deficiency; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers." + }, + { + "heading": "Operative content", + "text": "The operative entry states that The status remained unchanged through the review date, and no exception notice appears in this record.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent." + }, + { + "heading": "Reviewer analysis", + "text": "Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Owen Delgado recorded status “indexed — responsive context” and linked the file to CB-DOC-061-042. The control metric 30691 is an administrative population identifier, not a damages estimate or a statement of materiality." + }, + { + "heading": "Dependencies and reliance limits", + "text": "The record should be read with the access-request deadline, minor-consent issue, and consent-record deficiency materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion." + } + ] }, + "chronology": [ + { + "date": "2025-09-26", + "event": "Source population opened for collection", + "actor": "Amara Patel", + "evidence": "CB-DOC-061-077" + }, + { + "date": "2025-10-09", + "event": "Custodian confirmed system and date boundary", + "actor": "Micah Sullivan", + "evidence": "CB-DOC-061-042" + }, + { + "date": "2025-10-28", + "event": "Matter team completed first-level comparison", + "actor": "Lena Varga", + "evidence": "CB-DOC-061-042" + }, + { + "date": "2025-11-02", + "event": "Legal reviewer recorded the current disposition", + "actor": "Rafael Okafor", + "evidence": "CB-DOC-061-077" + }, + { + "date": "2025-12-10", + "event": "Assigned owner scheduled the next control response", + "actor": "Priya Raman", + "evidence": "CB-DOC-061-042" + } + ], + "participants": [ + { + "name": "Priya Raman", + "role": "compliance reviewer", + "responsibility": "source completeness" + }, + { + "name": "Caleb Hassan", + "role": "outside counsel liaison", + "responsibility": "business interpretation" + }, + { + "name": "Willa Novak", + "role": "operations lead", + "responsibility": "legal review" + }, + { + "name": "Micah Sullivan", + "role": "control owner", + "responsibility": "remediation tracking" + } + ], + "related_records": [ + { + "record_id": "CB-DOC-061-084", + "relationship": "same-cycle source", + "workstream": "risk assessments" + }, + { + "record_id": "CB-DOC-061-096", + "relationship": "implementation evidence", + "workstream": "remediation" + }, + { + "record_id": "CB-DOC-061-018", + "relationship": "independent control record", + "workstream": "vendor reviews" + }, + { + "record_id": "CB-DOC-061-034", + "relationship": "later reconciliation record", + "workstream": "consumer requests" + } + ], + "action_register": [ + { + "action_id": "A-077-1", + "action": "Confirm that CB-DOC-061-042 does not change the context-only classification.", + "owner": "Rafael Okafor", + "due_date": "2026-08-19", + "status": "in review" + }, + { + "action_id": "A-077-2", + "action": "Preserve the native vendor risk portal export and document any replacement record.", + "owner": "Mei Whitaker", + "due_date": "2026-08-26", + "status": "awaiting evidence" + }, + { + "action_id": "A-077-3", + "action": "Report the disposition to the regulator correspondence workstream lead before the matter deadline.", + "owner": "Theo Laurent", + "due_date": "2026-09-02", + "status": "owner confirmed" + } + ], + "supporting_rows": [ + { + "line_id": "CB-DOC-061-077-L01", + "category": "access-request deadline", + "description": "The status remained unchanged through the review date, and no exception notice appears in this record.", + "effective_date": "2025-11-01", + "owner": "Amara Patel", + "status": "awaiting evidence", + "metric": "$825,153.00", + "evidence_reference": "CB-DOC-061-077" + }, + { + "line_id": "CB-DOC-061-077-L02", + "category": "minor-consent issue", + "description": "Regulator Correspondence control observation 2; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-30", + "owner": "Nadine Flores", + "status": "open", + "metric": "9%", + "evidence_reference": "CB-DOC-061-042" + }, + { + "line_id": "CB-DOC-061-077-L03", + "category": "consent-record deficiency", + "description": "Regulator Correspondence control observation 3; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-26", + "owner": "Farah Ibrahim", + "status": "owner confirmed", + "metric": "10%", + "evidence_reference": "CB-DOC-061-077" + }, + { + "line_id": "CB-DOC-061-077-L04", + "category": "access-request deadline", + "description": "Regulator Correspondence control observation 4; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-23", + "owner": "Lena Varga", + "status": "awaiting evidence", + "metric": "$240,045.00", + "evidence_reference": "CB-DOC-061-042" + }, + { + "line_id": "CB-DOC-061-077-L05", + "category": "minor-consent issue", + "description": "Regulator Correspondence control observation 5; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-21", + "owner": "Maya Ellison", + "status": "owner confirmed", + "metric": "13%", + "evidence_reference": "CB-DOC-061-077" + }, + { + "line_id": "CB-DOC-061-077-L06", + "category": "consent-record deficiency", + "description": "Regulator Correspondence control observation 6; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-17", + "owner": "Nora Chen", + "status": "awaiting evidence", + "metric": "10%", + "evidence_reference": "CB-DOC-061-042" + }, + { + "line_id": "CB-DOC-061-077-L07", + "category": "access-request deadline", + "description": "Regulator Correspondence control observation 7; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-14", + "owner": "Priya Raman", + "status": "in review", + "metric": "$498,005.00", + "evidence_reference": "CB-DOC-061-077" + }, + { + "line_id": "CB-DOC-061-077-L08", + "category": "minor-consent issue", + "description": "Regulator Correspondence control observation 8; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-10", + "owner": "Talia Brooks", + "status": "owner confirmed", + "metric": "11%", + "evidence_reference": "CB-DOC-061-042" + }, + { + "line_id": "CB-DOC-061-077-L09", + "category": "consent-record deficiency", + "description": "Regulator Correspondence control observation 9; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-09", + "owner": "Mei Whitaker", + "status": "awaiting evidence", + "metric": "1%", + "evidence_reference": "CB-DOC-061-077" + }, + { + "line_id": "CB-DOC-061-077-L10", + "category": "access-request deadline", + "description": "Regulator Correspondence control observation 10; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-06", + "owner": "Leona Park", + "status": "escalated", + "metric": "$679,352.00", + "evidence_reference": "CB-DOC-061-042" + }, + { + "line_id": "CB-DOC-061-077-L11", + "category": "minor-consent issue", + "description": "Regulator Correspondence control observation 11; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-10-02", + "owner": "Sofia Bennett", + "status": "escalated", + "metric": "10%", + "evidence_reference": "CB-DOC-061-077" + }, + { + "line_id": "CB-DOC-061-077-L12", + "category": "consent-record deficiency", + "description": "Regulator Correspondence control observation 12; retained for reconciliation with CB-DOC-061-042.", + "effective_date": "2025-09-30", + "owner": "Willa Novak", + "status": "open", + "metric": "21%", + "evidence_reference": "CB-DOC-061-042" + } + ], "provenance": "Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice." } diff --git a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/078_regulator_correspondence_formal_notice.xml b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/078_regulator_correspondence_formal_notice.xml index 79d73983bf6d13c52f3dd7c2830aba50e84a8c05..3635fa82c8f5d810017969f29b7a6828d6bc464a 100644 --- a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/078_regulator_correspondence_formal_notice.xml +++ b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/078_regulator_correspondence_formal_notice.xml @@ -1,24 +1,59 @@ -| Record Id | CB-DOC-061-079 |
|---|
| Record Id | CB-DOC-061-079 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2024-11-20 |
| Record Type | officer certificate |
| Folder | 10_regulator_correspondence |
| Workstream | regulator correspondence |
| Source System | OneTrust data map |
| Native Version | 1.8 |
| Record Status | reviewed — variance confirmed |
| Confidentiality | Confidential — legal review |
| Custodian | Nadine Flores |
| Reviewer | Isaac Romero |
| Cross Reference | CB-DOC-061-076 |
| Control Metric | 784504 |
| Review Question | incident-notification clock |
| Finding Id | F-11 |
| Record Role | corroborating |
| Control Severity | medium |
| Remediation Owner | Caleb Hassan |
| Response Due | 2026-09-02 |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 10_regulator_correspondence |
| Record Type | officer certificate |
| Custodian | Nadine Flores |
| Reviewer | Isaac Romero |
| Record Date | 2024-11-20 |
| Cross Reference | CB-DOC-061-076 |
| Control Metric | 784504 |
| Operative Text | the approval log names Caleb Hassan and contains no entry for Nadine Flores |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Nadine Flores, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2024-11-20 and should be evaluated with CB-DOC-061-076. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Isaac Romero compared identifier 784504 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the regulator correspondence workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
The producing team exported this record from OneTrust data map and retained the native identifier CB-DOC-061-079. Nadine Flores confirmed the export boundary, while Isaac Romero performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-076.
For this regulator correspondence review, legal and business stakeholders are using the record to evaluate incident-notification clock, regulator commitment overdue, and subprocessor inventory mismatch. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that the approval log names Caleb Hassan and contains no entry for Nadine Flores. The record-control overlay classifies this as the corroborating source for F-11 (incident-notification clock) at medium severity. The assigned remediation owner is Caleb Hassan, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Isaac Romero recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-076. The control metric 784504 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the incident-notification clock, regulator commitment overdue, and subprocessor inventory mismatch materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2024-10-01 | Source population opened for collection | Nadine Flores | CB-DOC-061-079 |
| 2024-11-01 | Custodian confirmed system and date boundary | Owen Delgado | CB-DOC-061-076 |
| 2024-11-12 | Matter team completed first-level comparison | Maya Ellison | CB-DOC-061-076 |
| 2024-11-20 | Legal reviewer recorded the current disposition | Dominic Alvarez | CB-DOC-061-079 |
| 2024-12-23 | Assigned owner scheduled the next control response | Talia Brooks | CB-DOC-061-076 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-079-1 | Before 2026-09-02, secure written consent from the authorized decision-maker; owner: Caleb Hassan. | Caleb Hassan | 2026-08-19 | owner confirmed |
| A-079-2 | Preserve the native OneTrust data map export and document any replacement record. | Leona Park | 2026-08-26 | escalated |
| A-079-3 | Report the disposition to the regulator correspondence workstream lead before the matter deadline. | Henry Cho | 2026-09-02 | open |
The record covers activity in California through 2024-11-20 and should be evaluated with CB-DOC-061-076, the four related records listed below, and the complete regulator correspondence folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Isaac Romero compared identifier 784504 against the folder index, the native OneTrust data map entry, and CB-DOC-061-076. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/080_regulator_correspondence_status_report.md b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/080_regulator_correspondence_status_report.md index 1e28a66871f5a781471238fd9c7e6bd63a65bbcf..05dc63b8e6e4cc13ded4dba1acf0d8fb2b411ffd 100644 --- a/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/080_regulator_correspondence_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/10_regulator_correspondence/080_regulator_correspondence_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-080 +> Restricted — need to know · reviewed — no independent exception · native version 4.5 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Micah Sullivan | | Reviewer | Nadine Flores | | Cross-reference | CB-DOC-061-093 | +| Source system | ServiceNow privacy queue | +| Workstream | regulator correspondence | | Control metric | 871935 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The regulator correspondence team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Micah Sullivan identified it as an ordinary-course record from ServiceNow privacy queue; reviewer Nadine Flores preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** The file is responsive background material; it supplies context but does not create an independent obligation. + +## 1. Purpose and audience + +This status report supports the regulator correspondence workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +The producing team exported this record from ServiceNow privacy queue and retained the native identifier CB-DOC-061-080. Micah Sullivan confirmed the export boundary, while Nadine Flores performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-093. + +## 3. Matter and workstream context + +The operational context is the regulator correspondence portion of privacy and regulatory response audit. Reviewers identified dependencies involving risk-assessment omission, processing-purpose gap, and cross-border transfer gap; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that The file is responsive background material; it supplies context but does not create an independent obligation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Nadine Flores recorded status “reviewed — no independent exception” and linked the file to CB-DOC-061-093. The control metric 871935 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the risk-assessment omission, processing-purpose gap, and cross-border transfer gap materials and the files in the adjacent workstreams. This record is one component of the production and should not be treated as a complete statement of the governing agreement, policy, ledger, or law. Dates reflect the fixed synthetic matter timeline; references to approval or closure describe the source record rather than an independent legal conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-01-29 | Source population opened for collection | Micah Sullivan | CB-DOC-061-080 | +| 2024-02-23 | Custodian confirmed system and date boundary | Lena Varga | CB-DOC-061-093 | +| 2024-03-12 | Matter team completed first-level comparison | Rafael Okafor | CB-DOC-061-093 | +| 2024-03-16 | Legal reviewer recorded the current disposition | Priya Raman | CB-DOC-061-080 | +| 2024-04-21 | Assigned owner scheduled the next control response | Jonas Feld | CB-DOC-061-093 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Jonas Feld | control owner | source completeness | +| Sofia Bennett | matter lead | business interpretation | +| Henry Cho | business owner | legal review | +| Lena Varga | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Micah Sullivan, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-087` — same-cycle source (risk assessments) +- `CB-DOC-061-003` — implementation evidence (data inventory) +- `CB-DOC-061-021` — independent control record (vendor reviews) +- `CB-DOC-061-037` — later reconciliation record (consumer requests) -## Operative record +## Action register -The file is responsive background material; it supplies context but does not create an independent obligation. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-080-1 | Confirm that CB-DOC-061-093 does not change the context-only classification. | Priya Raman | 2026-08-19 | escalated | +| A-080-2 | Preserve the native ServiceNow privacy queue export and document any replacement record. | Isaac Romero | 2026-08-26 | open | +| A-080-3 | Report the disposition to the regulator correspondence workstream lead before the matter deadline. | Nadine Flores | 2026-09-02 | in review | ## Scope and cross-reference -The record covers activity in California through 2024-03-16 and should be evaluated with CB-DOC-061-093. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-03-16 and should be evaluated with CB-DOC-061-093, the four related records listed below, and the complete regulator correspondence folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Nadine Flores compared identifier 871935 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Nadine Flores compared identifier 871935 against the folder index, the native ServiceNow privacy queue entry, and CB-DOC-061-093. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/081_risk_assessments_control_register.md b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/081_risk_assessments_control_register.md index 186a6433005753ccaf931e498fde1a138619d774..a75c3dbc794ac8001c0dade266e32f37a4758ec0 100644 --- a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/081_risk_assessments_control_register.md +++ b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/081_risk_assessments_control_register.md @@ -1,5 +1,7 @@ # Control Register — CB-DOC-061-081 +> Restricted — need to know · indexed — responsive context · native version 1.6 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Farah Ibrahim | | Reviewer | Rafael Okafor | | Cross-reference | CB-DOC-061-014 | +| Source system | BigID deletion log | +| Workstream | risk assessments | | Control metric | 432716 | +| Review question | context only | +| Finding ID | none | +| Record role | context | +| Control severity | none | +| Remediation owner | none | +| Response due | none | + +## Executive record summary + +This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The risk assessments team used the record to maintain the matter team's issue population, source lineage, review status, and accountable owner. Producing custodian Farah Ibrahim identified it as an ordinary-course record from BigID deletion log; reviewer Rafael Okafor preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** No variance was recorded after the named reviewer compared the source ledger to the signed control record. + +## 1. Purpose and audience + +This control register supports the risk assessments workstream for Arcadia Genomics, Inc.. Its purpose is to maintain the matter team's issue population, source lineage, review status, and accountable owner. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +This copy was collected from BigID deletion log under matter hold CB-PR-3001. Its lineage runs from Farah Ibrahim, as producing custodian, to Rafael Okafor, as reviewing lawyer. The related record CB-DOC-061-014 remains a separate source of truth and was not merged into this document. + +## 3. Matter and workstream context + +The risk assessments workstream sits within privacy and regulatory response audit. The team is tracking sensitive-data classification, notice-practice mismatch, and retention schedule conflict because decisions in this file may affect the deadline and the position taken with North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that No variance was recorded after the named reviewer compared the source ledger to the signed control record.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Rafael Okafor recorded status “indexed — responsive context” and linked the file to CB-DOC-061-014. The control metric 432716 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the sensitive-data classification, notice-practice mismatch, and retention schedule conflict materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2024-12-02 | Source population opened for collection | Farah Ibrahim | CB-DOC-061-081 | +| 2024-12-25 | Custodian confirmed system and date boundary | Samuel Kim | CB-DOC-061-014 | +| 2025-01-06 | Matter team completed first-level comparison | Nora Chen | CB-DOC-061-014 | +| 2025-01-11 | Legal reviewer recorded the current disposition | Elliot Mercer | CB-DOC-061-081 | +| 2025-02-11 | Assigned owner scheduled the next control response | Mei Whitaker | CB-DOC-061-014 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Mei Whitaker | matter lead | source completeness | +| Adrian Mensah | business owner | business interpretation | +| Nadine Flores | records custodian | legal review | +| Samuel Kim | finance reviewer | remediation tracking | -## Background +## Related records -This control register was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Farah Ibrahim, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-088` — same-cycle source (risk assessments) +- `CB-DOC-061-004` — implementation evidence (data inventory) +- `CB-DOC-061-022` — independent control record (vendor reviews) +- `CB-DOC-061-038` — later reconciliation record (consumer requests) -## Operative record +## Action register -No variance was recorded after the named reviewer compared the source ledger to the signed control record. +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-081-1 | Confirm that CB-DOC-061-014 does not change the context-only classification. | Elliot Mercer | 2026-08-19 | open | +| A-081-2 | Preserve the native BigID deletion log export and document any replacement record. | Sofia Bennett | 2026-08-26 | in review | +| A-081-3 | Report the disposition to the risk assessments workstream lead before the matter deadline. | Micah Sullivan | 2026-09-02 | awaiting evidence | ## Scope and cross-reference -The record covers activity in California through 2025-01-11 and should be evaluated with CB-DOC-061-014. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-01-11 and should be evaluated with CB-DOC-061-014, the four related records listed below, and the complete risk assessments folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Rafael Okafor compared identifier 432716 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Rafael Okafor compared identifier 432716 against the folder index, the native BigID deletion log entry, and CB-DOC-061-014. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/082_risk_assessments_executed_instrument.txt b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/082_risk_assessments_executed_instrument.txt index e037d2eefc77b91de30e43cf32ae3b1c43a95396..4c5f2fbc64ec65025cbab1f9365bccbeb9d9f3a0 100644 --- a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/082_risk_assessments_executed_instrument.txt +++ b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/082_risk_assessments_executed_instrument.txt @@ -1,23 +1,78 @@ +ATTORNEY WORK PRODUCT + +EXECUTED INSTRUMENT +ARCADIA GENOMIC DATA RESPONSE AUDIT + DOCUMENT CONTROL: CB-DOC-061-082 MATTER: CB-PR-3001 | Arcadia genomic data response audit RECORD TYPE: executed instrument DATE: 2024-06-20 +SOURCE SYSTEM: Vanta control register +NATIVE VERSION: 4.0 +STATUS: reviewed — variance confirmed CUSTODIAN: Owen Delgado REVIEWER: Mei Whitaker CROSS-REFERENCE: CB-DOC-061-031 CONTROL METRIC: 415311 -BACKGROUND -This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Owen Delgado, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +REVIEW QUESTION: risk-assessment omission +FINDING ID: F-12 +RECORD ROLE: corroborating +CONTROL SEVERITY: low +REMEDIATION OWNER: Leona Park +RESPONSE DUE: 2026-09-02 + +RECITALS + +A. This executed instrument was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The risk assessments team used the record to preserve the operative language, execution history, defined terms, and signature authority. Producing custodian Owen Delgado identified it as an ordinary-course record from Vanta control register; reviewer Mei Whitaker preserved the source wording and separately recorded the legal-control disposition. + +B. The parties and matter team intend to preserve the source hierarchy and the decision record for CB-PR-3001. OPERATIVE RECORD the later status register marks it open and escalated in CB-PR-3001-U334 +1. PURPOSE AND AUDIENCE +This executed instrument supports the risk assessments workstream for Arcadia Genomics, Inc.. Its purpose is to preserve the operative language, execution history, defined terms, and signature authority. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +2. RECORD LINEAGE AND CUSTODY +The producing team exported this record from Vanta control register and retained the native identifier CB-DOC-061-082. Owen Delgado confirmed the export boundary, while Mei Whitaker performed the legal-control review. The file has not been normalized to resolve differences with CB-DOC-061-031. + +3. MATTER AND WORKSTREAM CONTEXT +For this risk assessments review, legal and business stakeholders are using the record to evaluate minor-consent issue, consent-record deficiency, and deletion-job failure. The file is part of privacy and regulatory response audit and should be read against the stated decision deadline. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +4. OPERATIVE CONTENT +The operative entry states that the later status register marks it open and escalated in CB-PR-3001-U334. The record-control overlay classifies this as the corroborating source for F-12 (risk-assessment omission) at low severity. The assigned remediation owner is Leona Park, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +5. REVIEWER ANALYSIS +The reviewer tested internal consistency, execution or approval evidence, status history, and the cited cross-reference. The review did not assume that a later system entry controls an earlier signed record. Open points remain assigned through the action register below. Reviewer Mei Whitaker recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-031. The control metric 415311 is an administrative population identifier, not a damages estimate or a statement of materiality. + +6. DEPENDENCIES AND RELIANCE LIMITS +The record should be read with the minor-consent issue, consent-record deficiency, and deletion-job failure materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation. + +SCHEDULE 1 — RECORD CHRONOLOGY +- 2024-05-09 | Source population opened for collection | Owen Delgado | CB-DOC-061-082 +- 2024-06-01 | Custodian confirmed system and date boundary | Maya Ellison | CB-DOC-061-031 +- 2024-06-15 | Matter team completed first-level comparison | Dominic Alvarez | CB-DOC-061-031 +- 2024-06-20 | Legal reviewer recorded the current disposition | Talia Brooks | CB-DOC-061-082 +- 2024-07-14 | Assigned owner scheduled the next control response | Caleb Hassan | CB-DOC-061-031 + +SCHEDULE 2 — ACTION REGISTER +- A-082-1 | in review | Leona Park | 2026-08-19 | Before 2026-09-02, issue a corrective notice using the contractually operative method; owner: Leona Park. +- A-082-2 | awaiting evidence | Adrian Mensah | 2026-08-26 | Preserve the native Vanta control register export and document any replacement record. +- A-082-3 | owner confirmed | Farah Ibrahim | 2026-09-02 | Report the disposition to the risk assessments workstream lead before the matter deadline. + SCOPE -The record covers activity in California through 2024-06-20 and should be evaluated with CB-DOC-061-031. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2024-06-20 and should be evaluated with CB-DOC-061-031, the four related records listed below, and the complete risk assessments folder. It is not a complete statement of law or a substitute for the other folders in production. CONTROL NOTE -Reviewer Mei Whitaker compared identifier 415311 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Mei Whitaker compared identifier 415311 against the folder index, the native Vanta control register entry, and CB-DOC-061-031. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. + +ACKNOWLEDGED FOR RECORD-CONTROL PURPOSES + +By: Owen Delgado +Role: Producing custodian +Reviewed by: Mei Whitaker +Record date: 2024-06-20 PROVENANCE Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. diff --git a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/083_risk_assessments_correspondence.eml b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/083_risk_assessments_correspondence.eml index bfd8ce6c830eb653aeea36332ada6dde1a481251..0ad4f47b0d537be441b331644ef35702784f3a0f 100644 --- a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/083_risk_assessments_correspondence.eml +++ b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/083_risk_assessments_correspondence.eml @@ -1,20 +1,77 @@ From: lena.varga@example.test To: theo.laurent@example.test +Cc: matter-team-arcadia_genomics_inc@example.test Date: 2025-02-09 09:30:00 -0700 Message-ID:| Record Id | CB-DOC-061-087 |
|---|
| Record Id | CB-DOC-061-087 |
|---|---|
| Matter Number | CB-PR-3001 |
| Record Date | 2024-07-04 |
| Record Type | officer certificate |
| Folder | 11_risk_assessments |
| Workstream | risk assessments |
| Source System | BigID deletion log |
| Native Version | 2.3 |
| Record Status | indexed — responsive context |
| Confidentiality | Confidential — legal review |
| Custodian | Nora Chen |
| Reviewer | Micah Sullivan |
| Cross Reference | CB-DOC-061-020 |
| Control Metric | 298045 |
| Review Question | context only |
| Finding Id | none |
| Record Role | context |
| Control Severity | none |
| Remediation Owner | none |
| Response Due | none |
| Matter Title | Arcadia genomic data response audit |
| Client | Arcadia Genomics, Inc. |
| Counterparty | North Coast Sequencing Network |
| Jurisdiction | California |
| Venue | California Privacy Protection Agency |
| Deadline | 2026-09-02 |
| Practice Workflow | privacy and regulatory response audit |
| Folder | 11_risk_assessments |
| Record Type | officer certificate |
| Custodian | Nora Chen |
| Reviewer | Micah Sullivan |
| Record Date | 2024-07-04 |
| Cross Reference | CB-DOC-061-020 |
| Control Metric | 298045 |
| Operative Text | The record preserves the original value and separately identifies the later administrative annotation. |
| Background | This officer certificate was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Nora Chen, identified it as an ordinary-course record used by the privacy and regulatory response audit team. |
| Scope | The record covers activity in California through 2024-07-04 and should be evaluated with CB-DOC-061-020. It is not a complete statement of law or a substitute for the other folders in production. |
| Control Note | Reviewer Micah Sullivan compared identifier 298045 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. |
| Provenance | Synthetic CounselBench-100 benchmark evidence; all entities, people, addresses, amounts, and events are fictitious and are not legal advice. |
This officer certificate supports the risk assessments workstream for Arcadia Genomics, Inc.. Its purpose is to state the certifying person’s knowledge, supporting review, exceptions, and reliance limitations. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision.
Matter operations catalogued the record in BigID deletion log using identifier CB-DOC-061-087. The chain of custody identifies Nora Chen as source owner and Micah Sullivan as the most recent reviewer. Any inconsistency with CB-DOC-061-020 must be reconciled rather than silently overwritten.
The operational context is the risk assessments portion of privacy and regulatory response audit. Reviewers identified dependencies involving consent-record deficiency, deletion-job failure, and sensitive-data classification; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers.
The operative entry states that The record preserves the original value and separately identifies the later administrative annotation.. The record-control overlay classifies this file as context only. It may explain the matter history or test a cross-reference, but it is not an independent finding. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent.
The legal-control check compared the record’s native value with the matter index and the identified related file. Conflicting custodial accounts, amendments outside the production, and post-date events were not resolved by inference and remain subject to documented follow-up. Reviewer Micah Sullivan recorded status “indexed — responsive context” and linked the file to CB-DOC-061-020. The control metric 298045 is an administrative population identifier, not a damages estimate or a statement of materiality.
The record should be read with the consent-record deficiency, deletion-job failure, and sensitive-data classification materials and the files in the adjacent workstreams. Reliance is limited to the document as produced. The record does not incorporate unproduced attachments, oral explanations, or later events, and its administrative status should not be confused with a legal determination about the underlying obligation.
| Date | Event | Actor | Evidence |
|---|---|---|---|
| 2024-05-22 | Source population opened for collection | Nora Chen | CB-DOC-061-087 |
| 2024-06-10 | Custodian confirmed system and date boundary | Elliot Mercer | CB-DOC-061-020 |
| 2024-06-30 | Matter team completed first-level comparison | Mei Whitaker | CB-DOC-061-020 |
| 2024-07-04 | Legal reviewer recorded the current disposition | Isaac Romero | CB-DOC-061-087 |
| 2024-08-11 | Assigned owner scheduled the next control response | Willa Novak | CB-DOC-061-020 |
| ID | Action | Owner | Due | Status |
|---|---|---|---|---|
| A-087-1 | Confirm that CB-DOC-061-020 does not change the context-only classification. | Isaac Romero | 2026-08-19 | in review |
| A-087-2 | Preserve the native BigID deletion log export and document any replacement record. | Nadine Flores | 2026-08-26 | awaiting evidence |
| A-087-3 | Report the disposition to the risk assessments workstream lead before the matter deadline. | Rafael Okafor | 2026-09-02 | owner confirmed |
The record covers activity in California through 2024-07-04 and should be evaluated with CB-DOC-061-020, the four related records listed below, and the complete risk assessments folder. It is not a complete statement of law or a substitute for the other folders in production.
Reviewer Micah Sullivan compared identifier 298045 against the folder index, the native BigID deletion log entry, and CB-DOC-061-020. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date.
diff --git a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/088_risk_assessments_status_report.md b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/088_risk_assessments_status_report.md index 0afb9c8648b725c7534f97cc7b1975ffe5039f9b..c47126d22f55f4829d87112d9f754000d8cb8c4c 100644 --- a/task_files/cb100-061-arcadia-genomics/11_risk_assessments/088_risk_assessments_status_report.md +++ b/task_files/cb100-061-arcadia-genomics/11_risk_assessments/088_risk_assessments_status_report.md @@ -1,5 +1,7 @@ # Status Report — CB-DOC-061-088 +> Attorney work product · reviewed — variance confirmed · native version 3.0 + | Control field | Value | |---|---| | Matter | CB-PR-3001 — Arcadia genomic data response audit | @@ -9,23 +11,87 @@ | Custodian | Dominic Alvarez | | Reviewer | Nora Chen | | Cross-reference | CB-DOC-061-037 | +| Source system | Vanta control register | +| Workstream | risk assessments | | Control metric | 560146 | +| Review question | minor-consent issue | +| Finding ID | F-14 | +| Record role | corroborating | +| Control severity | high | +| Remediation owner | Sofia Bennett | +| Response due | 2026-09-02 | + +## Executive record summary + +This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The risk assessments team used the record to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. Producing custodian Dominic Alvarez identified it as an ordinary-course record from Vanta control register; reviewer Nora Chen preserved the source wording and separately recorded the legal-control disposition. + +> **Operative record:** the applied threshold is 10% and produced a different exception result + +## 1. Purpose and audience + +This status report supports the risk assessments workstream for Arcadia Genomics, Inc.. Its purpose is to summarize milestones, dependencies, risk status, decision owners, and the next reporting cycle. The intended audience is the privacy and regulatory response audit matter team, including legal, finance, compliance, and the business owner responsible for the next decision. + +## 2. Record lineage and custody + +Matter operations catalogued the record in Vanta control register using identifier CB-DOC-061-088. The chain of custody identifies Dominic Alvarez as source owner and Nora Chen as the most recent reviewer. Any inconsistency with CB-DOC-061-037 must be reconciled rather than silently overwritten. + +## 3. Matter and workstream context + +The operational context is the risk assessments portion of privacy and regulatory response audit. Reviewers identified dependencies involving processor-contract omission, access-request deadline, and minor-consent issue; those dependencies matter to the client’s position concerning North Coast Sequencing Network. A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. + +## 4. Operative content + +The operative entry states that the applied threshold is 10% and produced a different exception result. The record-control overlay classifies this as the corroborating source for F-14 (minor-consent issue) at high severity. The assigned remediation owner is Sofia Bennett, with response due 2026-09-02. The entry is preserved as written because the benchmark requires reconciliation between sources, not silent correction of a document that appears incomplete or inconsistent. + +## 5. Reviewer analysis + +Review procedures included source identification, a comparison to the folder index, validation of the accountable owner, and a check for later amendments or operational entries. The conclusion is deliberately bounded to the produced record and its stated cross-reference. Reviewer Nora Chen recorded status “reviewed — variance confirmed” and linked the file to CB-DOC-061-037. The control metric 560146 is an administrative population identifier, not a damages estimate or a statement of materiality. + +## 6. Dependencies and reliance limits + +The record should be read with the processor-contract omission, access-request deadline, and minor-consent issue materials and the files in the adjacent workstreams. The file is responsive evidence, not a standalone legal opinion. It does not establish facts held only by other custodians, waive the need to review amendments, or determine priority among competing instruments. The complete production remains necessary to reach a final conclusion. + +## Chronology + +| Date | Event | Actor | Evidence | +|---|---|---|---| +| 2025-02-05 | Source population opened for collection | Dominic Alvarez | CB-DOC-061-088 | +| 2025-02-26 | Custodian confirmed system and date boundary | Talia Brooks | CB-DOC-061-037 | +| 2025-03-17 | Matter team completed first-level comparison | Caleb Hassan | CB-DOC-061-037 | +| 2025-03-22 | Legal reviewer recorded the current disposition | Sofia Bennett | CB-DOC-061-088 | +| 2025-04-20 | Assigned owner scheduled the next control response | Theo Laurent | CB-DOC-061-037 | + +## Participants + +| Name | Role | Responsibility | +|---|---|---| +| Theo Laurent | control owner | source completeness | +| Farah Ibrahim | matter lead | business interpretation | +| Rafael Okafor | business owner | legal review | +| Talia Brooks | records custodian | remediation tracking | -## Background +## Related records -This status report was maintained for Arcadia Genomics, Inc. in connection with Arcadia genomic data response audit. It concerns A consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. The producing custodian, Dominic Alvarez, identified it as an ordinary-course record used by the privacy and regulatory response audit team. +- `CB-DOC-061-095` — same-cycle source (remediation) +- `CB-DOC-061-011` — implementation evidence (processing records) +- `CB-DOC-061-029` — independent control record (security incidents) +- `CB-DOC-061-045` — later reconciliation record (notices consents) -## Operative record +## Action register -the applied threshold is 10% and produced a different exception result +| ID | Action | Owner | Due | Status | +|---|---|---|---|---| +| A-088-1 | Before 2026-09-02, document the governing interpretation before the deadline; owner: Sofia Bennett. | Sofia Bennett | 2026-08-19 | awaiting evidence | +| A-088-2 | Preserve the native Vanta control register export and document any replacement record. | Micah Sullivan | 2026-08-26 | owner confirmed | +| A-088-3 | Report the disposition to the risk assessments workstream lead before the matter deadline. | Nora Chen | 2026-09-02 | escalated | ## Scope and cross-reference -The record covers activity in California through 2025-03-22 and should be evaluated with CB-DOC-061-037. It is not a complete statement of law or a substitute for the other folders in production. +The record covers activity in California through 2025-03-22 and should be evaluated with CB-DOC-061-037, the four related records listed below, and the complete risk assessments folder. It is not a complete statement of law or a substitute for the other folders in production. ## Control note -Reviewer Nora Chen compared identifier 560146 against the folder index. The comparison did not evaluate facts held by other custodians, later amendments, or records generated after the document date. +Reviewer Nora Chen compared identifier 560146 against the folder index, the native Vanta control register entry, and CB-DOC-061-037. The comparison preserved each source independently and did not infer facts held by other custodians, unproduced amendments, oral explanations, or records generated after the document date. ## Provenance diff --git a/tasks/cb100-001-helios-forge.json b/tasks/cb100-001-helios-forge.json index c409460b0ef8118c804f7b8ef1b5f38c4500af6d..c4564758ef880694713a60147de4ab0f92c94c0d 100644 --- a/tasks/cb100-001-helios-forge.json +++ b/tasks/cb100-001-helios-forge.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Washoe County, Nevada", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Project Daybreak battery acquisition\n\nYou are acting as buyer-side transactions counsel for Helios Forge, Inc.. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-MA-2401 \nOther party or authority: BluePeak Battery Systems LLC \nJurisdiction: Delaware \nVenue or forum: Washoe County, Nevada \nDecision deadline: 2026-09-04\n\n## Assignment\n\nA carve-out manufacturer owns cathode tooling in Nevada while its largest automaker customer can terminate upon an indirect change of control. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-001-helios-forge\",\n \"matter_number\": \"CB-MA-2401\",\n \"prepared_for\": \"Helios Forge, Inc.\",\n \"as_of\": \"2026-09-04\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Project Daybreak battery acquisition\n\nYou are acting as buyer-side transactions counsel for Helios Forge, Inc.. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-MA-2401\nOther party or authority: BluePeak Battery Systems LLC\nJurisdiction: Delaware\nVenue or forum: Washoe County, Nevada\nDecision deadline: 2026-09-04\n\n## Assignment\n\nA carve-out manufacturer owns cathode tooling in Nevada while its largest automaker customer can terminate upon an indirect change of control. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-001-helios-forge\",\n \"matter_number\": \"CB-MA-2401\",\n \"prepared_for\": \"Helios Forge, Inc.\",\n \"as_of\": \"2026-09-04\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-001-helios-forge", "task_name": "Project Daybreak battery acquisition", diff --git a/tasks/cb100-004-tamarack-foods.json b/tasks/cb100-004-tamarack-foods.json index 5a0694eb98548684855837af70c3bdfca09e912f..26334d51454f57f7795a9519a4043d26c332bdaf 100644 --- a/tasks/cb100-004-tamarack-foods.json +++ b/tasks/cb100-004-tamarack-foods.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Cook County, Illinois", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Project Copper Kettle food plant divestiture\n\nYou are acting as buyer-side transactions counsel for Tamarack Foods Group. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-MA-2404 \nOther party or authority: Copper Kettle Nutrition Co. \nJurisdiction: Illinois \nVenue or forum: Cook County, Illinois \nDecision deadline: 2026-09-15\n\n## Assignment\n\nThe divested nutrition plant has private-label commitments, shared wastewater permits, a union successorship clause, and disputed promotional accruals. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-004-tamarack-foods\",\n \"matter_number\": \"CB-MA-2404\",\n \"prepared_for\": \"Tamarack Foods Group\",\n \"as_of\": \"2026-09-15\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Project Copper Kettle food plant divestiture\n\nYou are acting as buyer-side transactions counsel for Tamarack Foods Group. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-MA-2404\nOther party or authority: Copper Kettle Nutrition Co.\nJurisdiction: Illinois\nVenue or forum: Cook County, Illinois\nDecision deadline: 2026-09-15\n\n## Assignment\n\nThe divested nutrition plant has private-label commitments, shared wastewater permits, a union successorship clause, and disputed promotional accruals. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-004-tamarack-foods\",\n \"matter_number\": \"CB-MA-2404\",\n \"prepared_for\": \"Tamarack Foods Group\",\n \"as_of\": \"2026-09-15\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-004-tamarack-foods", "task_name": "Project Copper Kettle food plant divestiture", diff --git a/tasks/cb100-005-solace-mobility.json b/tasks/cb100-005-solace-mobility.json index b203a1a1d11807fb91d3042d98c577a76dfdb6f8..d135c9c11ced0a09fecc21794c0ca289d4696261 100644 --- a/tasks/cb100-005-solace-mobility.json +++ b/tasks/cb100-005-solace-mobility.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Santa Clara County, California", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Project Lantern fleet technology acquisition\n\nYou are acting as buyer-side transactions counsel for Solace Mobility Holdings. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-MA-2405 \nOther party or authority: Lantern Route Analytics, Inc. \nJurisdiction: California \nVenue or forum: Santa Clara County, California \nDecision deadline: 2026-09-18\n\n## Assignment\n\nThe route-optimization target relies on contractor-created models, municipal fleet data, and a customer letter that changes the pricing floor after closing. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-005-solace-mobility\",\n \"matter_number\": \"CB-MA-2405\",\n \"prepared_for\": \"Solace Mobility Holdings\",\n \"as_of\": \"2026-09-18\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Project Lantern fleet technology acquisition\n\nYou are acting as buyer-side transactions counsel for Solace Mobility Holdings. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-MA-2405\nOther party or authority: Lantern Route Analytics, Inc.\nJurisdiction: California\nVenue or forum: Santa Clara County, California\nDecision deadline: 2026-09-18\n\n## Assignment\n\nThe route-optimization target relies on contractor-created models, municipal fleet data, and a customer letter that changes the pricing floor after closing. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-005-solace-mobility\",\n \"matter_number\": \"CB-MA-2405\",\n \"prepared_for\": \"Solace Mobility Holdings\",\n \"as_of\": \"2026-09-18\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-005-solace-mobility", "task_name": "Project Lantern fleet technology acquisition", diff --git a/tasks/cb100-006-meridian-water.json b/tasks/cb100-006-meridian-water.json index 993411341615dbe861aa716873d61d48f225409d..f53a4206a22e62aa67dfc9c3a0e4be6229e61f25 100644 --- a/tasks/cb100-006-meridian-water.json +++ b/tasks/cb100-006-meridian-water.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Harris County, Texas", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Project Spillway treatment business acquisition\n\nYou are acting as buyer-side transactions counsel for Meridian Water Infrastructure. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-MA-2406 \nOther party or authority: Spillway Process Controls, LLC \nJurisdiction: Texas \nVenue or forum: Harris County, Texas \nDecision deadline: 2026-09-22\n\n## Assignment\n\nA process-controls business has municipal framework contracts, source-code escrow obligations, and a legacy explosion claim excluded from current insurance. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-006-meridian-water\",\n \"matter_number\": \"CB-MA-2406\",\n \"prepared_for\": \"Meridian Water Infrastructure\",\n \"as_of\": \"2026-09-22\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Project Spillway treatment business acquisition\n\nYou are acting as buyer-side transactions counsel for Meridian Water Infrastructure. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-MA-2406\nOther party or authority: Spillway Process Controls, LLC\nJurisdiction: Texas\nVenue or forum: Harris County, Texas\nDecision deadline: 2026-09-22\n\n## Assignment\n\nA process-controls business has municipal framework contracts, source-code escrow obligations, and a legacy explosion claim excluded from current insurance. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-006-meridian-water\",\n \"matter_number\": \"CB-MA-2406\",\n \"prepared_for\": \"Meridian Water Infrastructure\",\n \"as_of\": \"2026-09-22\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-006-meridian-water", "task_name": "Project Spillway treatment business acquisition", diff --git a/tasks/cb100-007-bracken-home.json b/tasks/cb100-007-bracken-home.json index 380d04571912cb4c03c644e0c285502012297a90..be18181e57becca5b2195a6556fa7ee1098f6ec6 100644 --- a/tasks/cb100-007-bracken-home.json +++ b/tasks/cb100-007-bracken-home.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Franklin County, Ohio", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Project Hearth consumer products merger\n\nYou are acting as buyer-side transactions counsel for Bracken Home Brands. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-MA-2407 \nOther party or authority: Hearth & Reed Appliances Corp. \nJurisdiction: Ohio \nVenue or forum: Franklin County, Ohio \nDecision deadline: 2026-09-25\n\n## Assignment\n\nA connected-appliance company changed component suppliers after a recall, pledged inventory under a revolver, and promised one retailer price parity. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-007-bracken-home\",\n \"matter_number\": \"CB-MA-2407\",\n \"prepared_for\": \"Bracken Home Brands\",\n \"as_of\": \"2026-09-25\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Project Hearth consumer products merger\n\nYou are acting as buyer-side transactions counsel for Bracken Home Brands. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-MA-2407\nOther party or authority: Hearth & Reed Appliances Corp.\nJurisdiction: Ohio\nVenue or forum: Franklin County, Ohio\nDecision deadline: 2026-09-25\n\n## Assignment\n\nA connected-appliance company changed component suppliers after a recall, pledged inventory under a revolver, and promised one retailer price parity. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-007-bracken-home\",\n \"matter_number\": \"CB-MA-2407\",\n \"prepared_for\": \"Bracken Home Brands\",\n \"as_of\": \"2026-09-25\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-007-bracken-home", "task_name": "Project Hearth consumer products merger", diff --git a/tasks/cb100-009-silverline-logistics.json b/tasks/cb100-009-silverline-logistics.json index dbb7d27a52bcc0b20cf5d87213df28299963e0af..40f69d017a21d30e3e59b899a6ba55094e97174e 100644 --- a/tasks/cb100-009-silverline-logistics.json +++ b/tasks/cb100-009-silverline-logistics.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Fulton County, Georgia", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Project Switchyard logistics roll-up\n\nYou are acting as buyer-side transactions counsel for Silverline Logistics LP. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-MA-2409 \nOther party or authority: Switchyard Freight Intelligence, LLC \nJurisdiction: Georgia \nVenue or forum: Fulton County, Georgia \nDecision deadline: 2026-10-02\n\n## Assignment\n\nA freight analytics add-on uses carrier fuel data, occupies a rail-adjacent facility, and disputes a customer offset against implementation invoices. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-009-silverline-logistics\",\n \"matter_number\": \"CB-MA-2409\",\n \"prepared_for\": \"Silverline Logistics LP\",\n \"as_of\": \"2026-10-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Project Switchyard logistics roll-up\n\nYou are acting as buyer-side transactions counsel for Silverline Logistics LP. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-MA-2409\nOther party or authority: Switchyard Freight Intelligence, LLC\nJurisdiction: Georgia\nVenue or forum: Fulton County, Georgia\nDecision deadline: 2026-10-02\n\n## Assignment\n\nA freight analytics add-on uses carrier fuel data, occupies a rail-adjacent facility, and disputes a customer offset against implementation invoices. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- charter authorization mismatch\n- unrecorded option grant\n- change-of-control consent\n- anti-assignment restriction\n- most-favored-customer exposure\n- financial covenant breach\n- key-person retention gap\n- invention assignment defect\n- open-source copyleft exposure\n- unreported security incident\n- permit transfer restriction\n- threatened customer dispute\n- facility title exception\n- sales-tax nexus gap\n- claims-made insurance tail\n- revenue recognition exception\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_corporate/001_corporate_control_register.md`\n- `/workspace/documents/01_corporate/004_corporate_ledger_export.csv`\n- `/workspace/documents/01_corporate/007_corporate_officer_certificate.html`\n- `/workspace/documents/02_capitalization/010_capitalization_executed_instrument.txt`\n- `/workspace/documents/02_capitalization/013_capitalization_review_memorandum.json`\n- `/workspace/documents/02_capitalization/016_capitalization_status_report.md`\n- `/workspace/documents/03_material_contracts/019_material_contracts_correspondence.eml`\n- `/workspace/documents/03_material_contracts/022_material_contracts_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-009-silverline-logistics\",\n \"matter_number\": \"CB-MA-2409\",\n \"prepared_for\": \"Silverline Logistics LP\",\n \"as_of\": \"2026-10-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-009-silverline-logistics", "task_name": "Project Switchyard logistics roll-up", diff --git a/tasks/cb100-011-atlas-fiber.json b/tasks/cb100-011-atlas-fiber.json index 72aa22e9b826a36f8f54a0d0e4812d1fdde5245e..405df55d330bbf1d7e84c2e44151c7e4f973bfe8 100644 --- a/tasks/cb100-011-atlas-fiber.json +++ b/tasks/cb100-011-atlas-fiber.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Denver, Colorado", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Atlas Fiber enterprise customer reset\n\nYou are acting as commercial contracts counsel for Atlas Fiber Networks. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-CC-2501 \nOther party or authority: Juniper Peak Retail Group \nJurisdiction: Colorado \nVenue or forum: Denver, Colorado \nDecision deadline: 2026-09-03\n\n## Assignment\n\nA five-year network agreement has three order forms, disputed uptime credits, and competing renewal dates created by a migration amendment. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-011-atlas-fiber\",\n \"matter_number\": \"CB-CC-2501\",\n \"prepared_for\": \"Atlas Fiber Networks\",\n \"as_of\": \"2026-09-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Atlas Fiber enterprise customer reset\n\nYou are acting as commercial contracts counsel for Atlas Fiber Networks. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-CC-2501\nOther party or authority: Juniper Peak Retail Group\nJurisdiction: Colorado\nVenue or forum: Denver, Colorado\nDecision deadline: 2026-09-03\n\n## Assignment\n\nA five-year network agreement has three order forms, disputed uptime credits, and competing renewal dates created by a migration amendment. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-011-atlas-fiber\",\n \"matter_number\": \"CB-CC-2501\",\n \"prepared_for\": \"Atlas Fiber Networks\",\n \"as_of\": \"2026-09-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-011-atlas-fiber", "task_name": "Atlas Fiber enterprise customer reset", diff --git a/tasks/cb100-013-cirrus-labs.json b/tasks/cb100-013-cirrus-labs.json index 704cd9ae0853be3b545417426a935d9c3d758cd2..02c69ab9976d076be7a66b4e23848be96764ea5e 100644 --- a/tasks/cb100-013-cirrus-labs.json +++ b/tasks/cb100-013-cirrus-labs.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "King County, Washington", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Cirrus cloud-reseller reconciliation\n\nYou are acting as commercial contracts counsel for Cirrus Laboratory Cloud LLC. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-CC-2503 \nOther party or authority: Quartzline Resellers Ltd. \nJurisdiction: Washington \nVenue or forum: King County, Washington \nDecision deadline: 2026-09-10\n\n## Assignment\n\nA reseller claims exclusivity in public-sector accounts and relies on an unsigned order form that changed usage reporting and audit mechanics. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-013-cirrus-labs\",\n \"matter_number\": \"CB-CC-2503\",\n \"prepared_for\": \"Cirrus Laboratory Cloud LLC\",\n \"as_of\": \"2026-09-10\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Cirrus cloud-reseller reconciliation\n\nYou are acting as commercial contracts counsel for Cirrus Laboratory Cloud LLC. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-CC-2503\nOther party or authority: Quartzline Resellers Ltd.\nJurisdiction: Washington\nVenue or forum: King County, Washington\nDecision deadline: 2026-09-10\n\n## Assignment\n\nA reseller claims exclusivity in public-sector accounts and relies on an unsigned order form that changed usage reporting and audit mechanics. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-013-cirrus-labs\",\n \"matter_number\": \"CB-CC-2503\",\n \"prepared_for\": \"Cirrus Laboratory Cloud LLC\",\n \"as_of\": \"2026-09-10\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-013-cirrus-labs", "task_name": "Cirrus cloud-reseller reconciliation", diff --git a/tasks/cb100-014-driftwood-media.json b/tasks/cb100-014-driftwood-media.json index 515fe9298fc63f3d70077d4c3ab1de1e782ed9b0..bbbe8006f3d43b5f7a31ca29e8621cee9115d67b 100644 --- a/tasks/cb100-014-driftwood-media.json +++ b/tasks/cb100-014-driftwood-media.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Los Angeles County, California", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Driftwood content-delivery renewal audit\n\nYou are acting as commercial contracts counsel for Driftwood Media Studios. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-CC-2504 \nOther party or authority: Palisade Stream Services, Inc. \nJurisdiction: California \nVenue or forum: Los Angeles County, California \nDecision deadline: 2026-09-14\n\n## Assignment\n\nDelivery charges, localization commitments, and service credits diverge across a master agreement, two amendments, and eighteen months of invoices. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-014-driftwood-media\",\n \"matter_number\": \"CB-CC-2504\",\n \"prepared_for\": \"Driftwood Media Studios\",\n \"as_of\": \"2026-09-14\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Driftwood content-delivery renewal audit\n\nYou are acting as commercial contracts counsel for Driftwood Media Studios. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-CC-2504\nOther party or authority: Palisade Stream Services, Inc.\nJurisdiction: California\nVenue or forum: Los Angeles County, California\nDecision deadline: 2026-09-14\n\n## Assignment\n\nDelivery charges, localization commitments, and service credits diverge across a master agreement, two amendments, and eighteen months of invoices. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-014-driftwood-media\",\n \"matter_number\": \"CB-CC-2504\",\n \"prepared_for\": \"Driftwood Media Studios\",\n \"as_of\": \"2026-09-14\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-014-driftwood-media", "task_name": "Driftwood content-delivery renewal audit", diff --git a/tasks/cb100-015-ember-medical.json b/tasks/cb100-015-ember-medical.json index f94d83cc24e1f78c94dae617b30abe639b037d26..2a05a4168b61cad8d833ec96370de55878912f3d 100644 --- a/tasks/cb100-015-ember-medical.json +++ b/tasks/cb100-015-ember-medical.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Hennepin County, Minnesota", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Ember device-support contract triage\n\nYou are acting as commercial contracts counsel for Ember Medical Devices. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-CC-2505 \nOther party or authority: Keystone Field Support LLC \nJurisdiction: Minnesota \nVenue or forum: Hennepin County, Minnesota \nDecision deadline: 2026-09-17\n\n## Assignment\n\nA national maintenance vendor missed response targets and added subcontractors without completing required background or insurance documentation. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-015-ember-medical\",\n \"matter_number\": \"CB-CC-2505\",\n \"prepared_for\": \"Ember Medical Devices\",\n \"as_of\": \"2026-09-17\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Ember device-support contract triage\n\nYou are acting as commercial contracts counsel for Ember Medical Devices. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-CC-2505\nOther party or authority: Keystone Field Support LLC\nJurisdiction: Minnesota\nVenue or forum: Hennepin County, Minnesota\nDecision deadline: 2026-09-17\n\n## Assignment\n\nA national maintenance vendor missed response targets and added subcontractors without completing required background or insurance documentation. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-015-ember-medical\",\n \"matter_number\": \"CB-CC-2505\",\n \"prepared_for\": \"Ember Medical Devices\",\n \"as_of\": \"2026-09-17\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-015-ember-medical", "task_name": "Ember device-support contract triage", diff --git a/tasks/cb100-016-fathom-retail.json b/tasks/cb100-016-fathom-retail.json index 90ecd5cc79b2deaac3478d406b21b71f47172948..ec9d7b66908b4835dcd8a68a6e2b28396375e926 100644 --- a/tasks/cb100-016-fathom-retail.json +++ b/tasks/cb100-016-fathom-retail.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Cook County, Illinois", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Fathom marketplace vendor audit\n\nYou are acting as commercial contracts counsel for Fathom Retail Markets. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-CC-2506 \nOther party or authority: North Cove Merchandising Corp. \nJurisdiction: Illinois \nVenue or forum: Cook County, Illinois \nDecision deadline: 2026-09-21\n\n## Assignment\n\nMarketplace fees changed mid-quarter, returns data moved offshore, and the parties disagree whether an email modified the category exclusivity provision. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-016-fathom-retail\",\n \"matter_number\": \"CB-CC-2506\",\n \"prepared_for\": \"Fathom Retail Markets\",\n \"as_of\": \"2026-09-21\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Fathom marketplace vendor audit\n\nYou are acting as commercial contracts counsel for Fathom Retail Markets. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-CC-2506\nOther party or authority: North Cove Merchandising Corp.\nJurisdiction: Illinois\nVenue or forum: Cook County, Illinois\nDecision deadline: 2026-09-21\n\n## Assignment\n\nMarketplace fees changed mid-quarter, returns data moved offshore, and the parties disagree whether an email modified the category exclusivity provision. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-016-fathom-retail\",\n \"matter_number\": \"CB-CC-2506\",\n \"prepared_for\": \"Fathom Retail Markets\",\n \"as_of\": \"2026-09-21\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-016-fathom-retail", "task_name": "Fathom marketplace vendor audit", diff --git a/tasks/cb100-017-garnet-energy.json b/tasks/cb100-017-garnet-energy.json index 31655906eadcd4fbbf59f00b88c93950235f135d..e3f7ce531dac06ce68c96a5f2de0ae94e1450cfd 100644 --- a/tasks/cb100-017-garnet-energy.json +++ b/tasks/cb100-017-garnet-energy.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Travis County, Texas", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Garnet offtake agreement review\n\nYou are acting as commercial contracts counsel for Garnet Renewable Energy. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-CC-2507 \nOther party or authority: Mesa Crown Utilities \nJurisdiction: Texas \nVenue or forum: Travis County, Texas \nDecision deadline: 2026-09-24\n\n## Assignment\n\nAn energy offtake portfolio contains metering disputes, a change-in-law pass-through, and notice records sent to an address deleted by amendment. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-017-garnet-energy\",\n \"matter_number\": \"CB-CC-2507\",\n \"prepared_for\": \"Garnet Renewable Energy\",\n \"as_of\": \"2026-09-24\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Garnet offtake agreement review\n\nYou are acting as commercial contracts counsel for Garnet Renewable Energy. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-CC-2507\nOther party or authority: Mesa Crown Utilities\nJurisdiction: Texas\nVenue or forum: Travis County, Texas\nDecision deadline: 2026-09-24\n\n## Assignment\n\nAn energy offtake portfolio contains metering disputes, a change-in-law pass-through, and notice records sent to an address deleted by amendment. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-017-garnet-energy\",\n \"matter_number\": \"CB-CC-2507\",\n \"prepared_for\": \"Garnet Renewable Energy\",\n \"as_of\": \"2026-09-24\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-017-garnet-energy", "task_name": "Garnet offtake agreement review", diff --git a/tasks/cb100-018-harborlight-ai.json b/tasks/cb100-018-harborlight-ai.json index 3352f1f9df920c5a9cf9137da8d47de8beaf33a9..74981800b9e08285c58c936aae8647cfc7ee3165 100644 --- a/tasks/cb100-018-harborlight-ai.json +++ b/tasks/cb100-018-harborlight-ai.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Hartford County, Connecticut", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Harborlight AI procurement reconciliation\n\nYou are acting as commercial contracts counsel for Harborlight AI Systems. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-CC-2508 \nOther party or authority: Westmere Insurance Cooperative \nJurisdiction: Connecticut \nVenue or forum: Hartford County, Connecticut \nDecision deadline: 2026-09-28\n\n## Assignment\n\nAn AI services deal combines outcome benchmarks, model-training restrictions, and contradictory schedules for data deletion and incident notice. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-018-harborlight-ai\",\n \"matter_number\": \"CB-CC-2508\",\n \"prepared_for\": \"Harborlight AI Systems\",\n \"as_of\": \"2026-09-28\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Harborlight AI procurement reconciliation\n\nYou are acting as commercial contracts counsel for Harborlight AI Systems. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-CC-2508\nOther party or authority: Westmere Insurance Cooperative\nJurisdiction: Connecticut\nVenue or forum: Hartford County, Connecticut\nDecision deadline: 2026-09-28\n\n## Assignment\n\nAn AI services deal combines outcome benchmarks, model-training restrictions, and contradictory schedules for data deletion and incident notice. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-018-harborlight-ai\",\n \"matter_number\": \"CB-CC-2508\",\n \"prepared_for\": \"Harborlight AI Systems\",\n \"as_of\": \"2026-09-28\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-018-harborlight-ai", "task_name": "Harborlight AI procurement reconciliation", diff --git a/tasks/cb100-019-ironwood-agri.json b/tasks/cb100-019-ironwood-agri.json index 36be245b9e9df51db8ac83f7eb96c3472f053f04..5129eae25dc70f63f4ffa6e683d294254ae2d8fe 100644 --- a/tasks/cb100-019-ironwood-agri.json +++ b/tasks/cb100-019-ironwood-agri.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Polk County, Iowa", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Ironwood distribution portfolio review\n\nYou are acting as commercial contracts counsel for Ironwood Agricultural Supply. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-CC-2509 \nOther party or authority: Blue Fen Distribution, Inc. \nJurisdiction: Iowa \nVenue or forum: Polk County, Iowa \nDecision deadline: 2026-10-01\n\n## Assignment\n\nA regional distributor invokes price protection and territorial exclusivity despite dealer reports showing sales outside the assigned states. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-019-ironwood-agri\",\n \"matter_number\": \"CB-CC-2509\",\n \"prepared_for\": \"Ironwood Agricultural Supply\",\n \"as_of\": \"2026-10-01\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Ironwood distribution portfolio review\n\nYou are acting as commercial contracts counsel for Ironwood Agricultural Supply. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-CC-2509\nOther party or authority: Blue Fen Distribution, Inc.\nJurisdiction: Iowa\nVenue or forum: Polk County, Iowa\nDecision deadline: 2026-10-01\n\n## Assignment\n\nA regional distributor invokes price protection and territorial exclusivity despite dealer reports showing sales outside the assigned states. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-019-ironwood-agri\",\n \"matter_number\": \"CB-CC-2509\",\n \"prepared_for\": \"Ironwood Agricultural Supply\",\n \"as_of\": \"2026-10-01\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-019-ironwood-agri", "task_name": "Ironwood distribution portfolio review", diff --git a/tasks/cb100-020-juno-travel.json b/tasks/cb100-020-juno-travel.json index 32c3031595c0952738ce111d7d6724e1d8edc26e..301c37cbf21d39d5fde3eb254347812820da2392 100644 --- a/tasks/cb100-020-juno-travel.json +++ b/tasks/cb100-020-juno-travel.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Miami-Dade County, Florida", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Juno reservation platform exit analysis\n\nYou are acting as commercial contracts counsel for Juno Travel Technologies. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-CC-2510 \nOther party or authority: Crescent Hotel Consortium \nJurisdiction: Florida \nVenue or forum: Miami-Dade County, Florida \nDecision deadline: 2026-10-05\n\n## Assignment\n\nThe customer seeks an early exit after repeated outages, but separate documents disagree about cure periods, transition services, and prepaid credits. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-020-juno-travel\",\n \"matter_number\": \"CB-CC-2510\",\n \"prepared_for\": \"Juno Travel Technologies\",\n \"as_of\": \"2026-10-05\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Juno reservation platform exit analysis\n\nYou are acting as commercial contracts counsel for Juno Travel Technologies. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-CC-2510\nOther party or authority: Crescent Hotel Consortium\nJurisdiction: Florida\nVenue or forum: Miami-Dade County, Florida\nDecision deadline: 2026-10-05\n\n## Assignment\n\nThe customer seeks an early exit after repeated outages, but separate documents disagree about cure periods, transition services, and prepaid credits. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- renewal date conflict\n- pricing-escalator mismatch\n- minimum-commit shortfall\n- service-credit underpayment\n- termination notice defect\n- assignment consent requirement\n- data-localization conflict\n- subprocessor notice lapse\n- insurance-limit deficiency\n- audit-right deadline\n- exclusivity carveout breach\n- benchmarking right omission\n- invoice currency discrepancy\n- usage-report inconsistency\n- order-form precedence conflict\n- survival-clause ambiguity\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_master_agreements/001_master_agreements_control_register.md`\n- `/workspace/documents/01_master_agreements/004_master_agreements_ledger_export.csv`\n- `/workspace/documents/01_master_agreements/007_master_agreements_officer_certificate.html`\n- `/workspace/documents/02_order_forms/010_order_forms_executed_instrument.txt`\n- `/workspace/documents/02_order_forms/013_order_forms_review_memorandum.json`\n- `/workspace/documents/02_order_forms/016_order_forms_status_report.md`\n- `/workspace/documents/03_amendments/019_amendments_correspondence.eml`\n- `/workspace/documents/03_amendments/022_amendments_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-020-juno-travel\",\n \"matter_number\": \"CB-CC-2510\",\n \"prepared_for\": \"Juno Travel Technologies\",\n \"as_of\": \"2026-10-05\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-020-juno-travel", "task_name": "Juno reservation platform exit analysis", diff --git a/tasks/cb100-021-alder-biotech.json b/tasks/cb100-021-alder-biotech.json index bf2e1141e5b84f5a420ea40f8cffdd45da8c6d70..c88acb52552b99f47b3e8b840a2307acbe83ad3a 100644 --- a/tasks/cb100-021-alder-biotech.json +++ b/tasks/cb100-021-alder-biotech.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Cambridge, Massachusetts", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Alder clinical-vendor payment inquiry\n\nYou are acting as independent investigations counsel for Alder Biotherapeutics. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-IV-2601 \nOther party or authority: Morrow Trial Services LLC \nJurisdiction: Massachusetts \nVenue or forum: Cambridge, Massachusetts \nDecision deadline: 2026-09-02\n\n## Assignment\n\nA hotline report alleges that a study manager split vendor invoices after accepting travel from a firm owned by a former colleague. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-021-alder-biotech\",\n \"matter_number\": \"CB-IV-2601\",\n \"prepared_for\": \"Alder Biotherapeutics\",\n \"as_of\": \"2026-09-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Alder clinical-vendor payment inquiry\n\nYou are acting as independent investigations counsel for Alder Biotherapeutics. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-IV-2601\nOther party or authority: Morrow Trial Services LLC\nJurisdiction: Massachusetts\nVenue or forum: Cambridge, Massachusetts\nDecision deadline: 2026-09-02\n\n## Assignment\n\nA hotline report alleges that a study manager split vendor invoices after accepting travel from a firm owned by a former colleague. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-021-alder-biotech\",\n \"matter_number\": \"CB-IV-2601\",\n \"prepared_for\": \"Alder Biotherapeutics\",\n \"as_of\": \"2026-09-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-021-alder-biotech", "task_name": "Alder clinical-vendor payment inquiry", diff --git a/tasks/cb100-022-beacon-mining.json b/tasks/cb100-022-beacon-mining.json index d12ecd4107d7eaa1ebb1faa84d250f7bac024630..83acbb70e4e9271146d0e4cbb8d2206f1a5ef84f 100644 --- a/tasks/cb100-022-beacon-mining.json +++ b/tasks/cb100-022-beacon-mining.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Pima County, Arizona", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Beacon procurement integrity review\n\nYou are acting as independent investigations counsel for Beacon Mineral Resources. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-IV-2602 \nOther party or authority: Red Mesa Hauling Co. \nJurisdiction: Arizona \nVenue or forum: Pima County, Arizona \nDecision deadline: 2026-09-06\n\n## Assignment\n\nAnonymous allegations connect emergency hauling awards to off-channel messages, badge visits, and an undeclared family relationship. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-022-beacon-mining\",\n \"matter_number\": \"CB-IV-2602\",\n \"prepared_for\": \"Beacon Mineral Resources\",\n \"as_of\": \"2026-09-06\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Beacon procurement integrity review\n\nYou are acting as independent investigations counsel for Beacon Mineral Resources. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-IV-2602\nOther party or authority: Red Mesa Hauling Co.\nJurisdiction: Arizona\nVenue or forum: Pima County, Arizona\nDecision deadline: 2026-09-06\n\n## Assignment\n\nAnonymous allegations connect emergency hauling awards to off-channel messages, badge visits, and an undeclared family relationship. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-022-beacon-mining\",\n \"matter_number\": \"CB-IV-2602\",\n \"prepared_for\": \"Beacon Mineral Resources\",\n \"as_of\": \"2026-09-06\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-022-beacon-mining", "task_name": "Beacon procurement integrity review", diff --git a/tasks/cb100-023-cedar-finance.json b/tasks/cb100-023-cedar-finance.json index c57f4e6b0da9678d1ef35a900148f6215c1d7f37..f8b322bcd02a512301bdc85812a72342291978f2 100644 --- a/tasks/cb100-023-cedar-finance.json +++ b/tasks/cb100-023-cedar-finance.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Milwaukee County, Wisconsin", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Cedar branch incentive investigation\n\nYou are acting as independent investigations counsel for Cedar Community Finance. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-IV-2603 \nOther party or authority: Lakefront Referral Partners \nJurisdiction: Wisconsin \nVenue or forum: Milwaukee County, Wisconsin \nDecision deadline: 2026-09-09\n\n## Assignment\n\nA branch posted referral bonuses just below secondary-approval thresholds while employee interviews conflict with access and expense records. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-023-cedar-finance\",\n \"matter_number\": \"CB-IV-2603\",\n \"prepared_for\": \"Cedar Community Finance\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Cedar branch incentive investigation\n\nYou are acting as independent investigations counsel for Cedar Community Finance. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-IV-2603\nOther party or authority: Lakefront Referral Partners\nJurisdiction: Wisconsin\nVenue or forum: Milwaukee County, Wisconsin\nDecision deadline: 2026-09-09\n\n## Assignment\n\nA branch posted referral bonuses just below secondary-approval thresholds while employee interviews conflict with access and expense records. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-023-cedar-finance\",\n \"matter_number\": \"CB-IV-2603\",\n \"prepared_for\": \"Cedar Community Finance\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-023-cedar-finance", "task_name": "Cedar branch incentive investigation", diff --git a/tasks/cb100-024-dovetail-aerospace.json b/tasks/cb100-024-dovetail-aerospace.json index 25976df619f43ad71f932854cb38cef9bd5158a5..ebc773903b2d1b1da90d570609c530ee9473fab1 100644 --- a/tasks/cb100-024-dovetail-aerospace.json +++ b/tasks/cb100-024-dovetail-aerospace.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Fairfax County, Virginia", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Dovetail export-controls inquiry\n\nYou are acting as independent investigations counsel for Dovetail Aerospace Components. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-IV-2604 \nOther party or authority: Orion Bridge Consulting \nJurisdiction: Virginia \nVenue or forum: Fairfax County, Virginia \nDecision deadline: 2026-09-13\n\n## Assignment\n\nEngineering files may have been shared with an overseas consultant before screening, and preservation began after an account was deactivated. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-024-dovetail-aerospace\",\n \"matter_number\": \"CB-IV-2604\",\n \"prepared_for\": \"Dovetail Aerospace Components\",\n \"as_of\": \"2026-09-13\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Dovetail export-controls inquiry\n\nYou are acting as independent investigations counsel for Dovetail Aerospace Components. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-IV-2604\nOther party or authority: Orion Bridge Consulting\nJurisdiction: Virginia\nVenue or forum: Fairfax County, Virginia\nDecision deadline: 2026-09-13\n\n## Assignment\n\nEngineering files may have been shared with an overseas consultant before screening, and preservation began after an account was deactivated. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-024-dovetail-aerospace\",\n \"matter_number\": \"CB-IV-2604\",\n \"prepared_for\": \"Dovetail Aerospace Components\",\n \"as_of\": \"2026-09-13\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-024-dovetail-aerospace", "task_name": "Dovetail export-controls inquiry", diff --git a/tasks/cb100-025-evergreen-charter.json b/tasks/cb100-025-evergreen-charter.json index 3f5b4e90a0ff13b15bfd99498b64161a987d1c1c..2a87d5a8822e304d7959c6283376ee1df6e684e0 100644 --- a/tasks/cb100-025-evergreen-charter.json +++ b/tasks/cb100-025-evergreen-charter.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Multnomah County, Oregon", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Evergreen enrollment complaint review\n\nYou are acting as independent investigations counsel for Evergreen Charter Network. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-IV-2605 \nOther party or authority: Summit Enrollment Advisors \nJurisdiction: Oregon \nVenue or forum: Multnomah County, Oregon \nDecision deadline: 2026-09-16\n\n## Assignment\n\nA complaint alleges manipulated wait-list entries, consultant gifts, and retaliation against the registrar who questioned manual overrides. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-025-evergreen-charter\",\n \"matter_number\": \"CB-IV-2605\",\n \"prepared_for\": \"Evergreen Charter Network\",\n \"as_of\": \"2026-09-16\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Evergreen enrollment complaint review\n\nYou are acting as independent investigations counsel for Evergreen Charter Network. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-IV-2605\nOther party or authority: Summit Enrollment Advisors\nJurisdiction: Oregon\nVenue or forum: Multnomah County, Oregon\nDecision deadline: 2026-09-16\n\n## Assignment\n\nA complaint alleges manipulated wait-list entries, consultant gifts, and retaliation against the registrar who questioned manual overrides. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-025-evergreen-charter\",\n \"matter_number\": \"CB-IV-2605\",\n \"prepared_for\": \"Evergreen Charter Network\",\n \"as_of\": \"2026-09-16\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-025-evergreen-charter", "task_name": "Evergreen enrollment complaint review", diff --git a/tasks/cb100-028-horizon-relief.json b/tasks/cb100-028-horizon-relief.json index fdefcc44c3fb4a5a4f07c386318d034af6116cb5..45267356e4f03107f3fe4793392732b5bbe5b182 100644 --- a/tasks/cb100-028-horizon-relief.json +++ b/tasks/cb100-028-horizon-relief.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Washington, District of Columbia", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Horizon grant-use investigation\n\nYou are acting as independent investigations counsel for Horizon Relief Foundation. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-IV-2608 \nOther party or authority: Fieldstone Logistics NGO \nJurisdiction: District of Columbia \nVenue or forum: Washington, District of Columbia \nDecision deadline: 2026-09-27\n\n## Assignment\n\nGrant funds were rerouted through emergency vendors, with inconsistent delivery evidence and delayed reports to the audit committee. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-028-horizon-relief\",\n \"matter_number\": \"CB-IV-2608\",\n \"prepared_for\": \"Horizon Relief Foundation\",\n \"as_of\": \"2026-09-27\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Horizon grant-use investigation\n\nYou are acting as independent investigations counsel for Horizon Relief Foundation. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-IV-2608\nOther party or authority: Fieldstone Logistics NGO\nJurisdiction: District of Columbia\nVenue or forum: Washington, District of Columbia\nDecision deadline: 2026-09-27\n\n## Assignment\n\nGrant funds were rerouted through emergency vendors, with inconsistent delivery evidence and delayed reports to the audit committee. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-028-horizon-relief\",\n \"matter_number\": \"CB-IV-2608\",\n \"prepared_for\": \"Horizon Relief Foundation\",\n \"as_of\": \"2026-09-27\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-028-horizon-relief", "task_name": "Horizon grant-use investigation", diff --git a/tasks/cb100-029-indigo-telecom.json b/tasks/cb100-029-indigo-telecom.json index 51df2955917053bf22b122d3fcbbf644a6b870c2..a274c9124635806e96c67aae3e7b689fddc5abfd 100644 --- a/tasks/cb100-029-indigo-telecom.json +++ b/tasks/cb100-029-indigo-telecom.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Dallas County, Texas", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Indigo channel-rebate inquiry\n\nYou are acting as independent investigations counsel for Indigo Telecom Americas. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-IV-2609 \nOther party or authority: Prairie Link Solutions \nJurisdiction: Texas \nVenue or forum: Dallas County, Texas \nDecision deadline: 2026-09-30\n\n## Assignment\n\nQuarter-end rebates were approved through overrides after unusual chats, duplicate support, and a vendor diligence file that omits beneficial ownership. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-029-indigo-telecom\",\n \"matter_number\": \"CB-IV-2609\",\n \"prepared_for\": \"Indigo Telecom Americas\",\n \"as_of\": \"2026-09-30\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Indigo channel-rebate inquiry\n\nYou are acting as independent investigations counsel for Indigo Telecom Americas. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-IV-2609\nOther party or authority: Prairie Link Solutions\nJurisdiction: Texas\nVenue or forum: Dallas County, Texas\nDecision deadline: 2026-09-30\n\n## Assignment\n\nQuarter-end rebates were approved through overrides after unusual chats, duplicate support, and a vendor diligence file that omits beneficial ownership. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-029-indigo-telecom\",\n \"matter_number\": \"CB-IV-2609\",\n \"prepared_for\": \"Indigo Telecom Americas\",\n \"as_of\": \"2026-09-30\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-029-indigo-telecom", "task_name": "Indigo channel-rebate inquiry", diff --git a/tasks/cb100-030-jasper-university.json b/tasks/cb100-030-jasper-university.json index 1781ea85c7ff04b8b1a5a40a65cb042d02d6bbcb..94b23a66ea46e97d805a0640b810cc836feba2c4 100644 --- a/tasks/cb100-030-jasper-university.json +++ b/tasks/cb100-030-jasper-university.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Allegheny County, Pennsylvania", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Jasper research-fund inquiry\n\nYou are acting as independent investigations counsel for Jasper University Research Institute. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-IV-2610 \nOther party or authority: Nova Bench Analytics \nJurisdiction: Pennsylvania \nVenue or forum: Allegheny County, Pennsylvania \nDecision deadline: 2026-10-04\n\n## Assignment\n\nA principal investigator directed purchases to a startup owned by a former student while training, disclosure, and expense records diverge. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-030-jasper-university\",\n \"matter_number\": \"CB-IV-2610\",\n \"prepared_for\": \"Jasper University Research Institute\",\n \"as_of\": \"2026-10-04\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Jasper research-fund inquiry\n\nYou are acting as independent investigations counsel for Jasper University Research Institute. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-IV-2610\nOther party or authority: Nova Bench Analytics\nJurisdiction: Pennsylvania\nVenue or forum: Allegheny County, Pennsylvania\nDecision deadline: 2026-10-04\n\n## Assignment\n\nA principal investigator directed purchases to a startup owned by a former student while training, disclosure, and expense records diverge. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- preservation delay\n- interview timeline conflict\n- undisclosed related party\n- split-purchase pattern\n- approval override\n- gift-policy threshold breach\n- off-channel communication\n- deleted-message anomaly\n- badge-log inconsistency\n- vendor due-diligence gap\n- retaliation concern\n- policy training lapse\n- board-reporting omission\n- hotline classification error\n- expense support deficiency\n- remediation ownership gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_intake/001_intake_control_register.md`\n- `/workspace/documents/01_intake/004_intake_ledger_export.csv`\n- `/workspace/documents/01_intake/007_intake_officer_certificate.html`\n- `/workspace/documents/02_preservation/010_preservation_executed_instrument.txt`\n- `/workspace/documents/02_preservation/013_preservation_review_memorandum.json`\n- `/workspace/documents/02_preservation/016_preservation_status_report.md`\n- `/workspace/documents/03_interviews/019_interviews_correspondence.eml`\n- `/workspace/documents/03_interviews/022_interviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-030-jasper-university\",\n \"matter_number\": \"CB-IV-2610\",\n \"prepared_for\": \"Jasper University Research Institute\",\n \"as_of\": \"2026-10-04\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-030-jasper-university", "task_name": "Jasper research-fund inquiry", diff --git a/tasks/cb100-032-bayfield-patent.json b/tasks/cb100-032-bayfield-patent.json index 2863c9c2646c8a1eb22e2eeae3737dec9b9a7879..2092b313e6227a5c22bbd035a3005c8f65d3c746 100644 --- a/tasks/cb100-032-bayfield-patent.json +++ b/tasks/cb100-032-bayfield-patent.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Delaware", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Bayfield patent case production review\n\nYou are acting as discovery counsel for Bayfield Sensorics, Inc.. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-LD-2702 \nOther party or authority: Kestrel Vision Systems LLC \nJurisdiction: Delaware \nVenue or forum: District of Delaware \nDecision deadline: 2026-09-09\n\n## Assignment\n\nSource-code collections and inventor communications were produced under a protective order, but family links and designation fields are inconsistent. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-032-bayfield-patent\",\n \"matter_number\": \"CB-LD-2702\",\n \"prepared_for\": \"Bayfield Sensorics, Inc.\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Bayfield patent case production review\n\nYou are acting as discovery counsel for Bayfield Sensorics, Inc.. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-LD-2702\nOther party or authority: Kestrel Vision Systems LLC\nJurisdiction: Delaware\nVenue or forum: District of Delaware\nDecision deadline: 2026-09-09\n\n## Assignment\n\nSource-code collections and inventor communications were produced under a protective order, but family links and designation fields are inconsistent. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-032-bayfield-patent\",\n \"matter_number\": \"CB-LD-2702\",\n \"prepared_for\": \"Bayfield Sensorics, Inc.\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-032-bayfield-patent", "task_name": "Bayfield patent case production review", diff --git a/tasks/cb100-033-crown-employment.json b/tasks/cb100-033-crown-employment.json index d5df749449d09302c8cb3034f7769e3578d66bd4..05ae1660ad5b9d500b5fa5e37b91bcbbe74446b5 100644 --- a/tasks/cb100-033-crown-employment.json +++ b/tasks/cb100-033-crown-employment.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Central District of California", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Crown wage action discovery check\n\nYou are acting as discovery counsel for Crown Kitchen Group. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-LD-2703 \nOther party or authority: Martinez et al. \nJurisdiction: California \nVenue or forum: Central District of California \nDecision deadline: 2026-09-12\n\n## Assignment\n\nA wage class action involves manager texts, scheduling exports, late legal holds, and commitments made during two meet-and-confer sessions. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-033-crown-employment\",\n \"matter_number\": \"CB-LD-2703\",\n \"prepared_for\": \"Crown Kitchen Group\",\n \"as_of\": \"2026-09-12\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Crown wage action discovery check\n\nYou are acting as discovery counsel for Crown Kitchen Group. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-LD-2703\nOther party or authority: Martinez et al.\nJurisdiction: California\nVenue or forum: Central District of California\nDecision deadline: 2026-09-12\n\n## Assignment\n\nA wage class action involves manager texts, scheduling exports, late legal holds, and commitments made during two meet-and-confer sessions. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-033-crown-employment\",\n \"matter_number\": \"CB-LD-2703\",\n \"prepared_for\": \"Crown Kitchen Group\",\n \"as_of\": \"2026-09-12\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-033-crown-employment", "task_name": "Crown wage action discovery check", diff --git a/tasks/cb100-035-elm-river.json b/tasks/cb100-035-elm-river.json index 93d4672b6f15b5b4af0016b3f2b4e299743a08c5..131dea857d4ef0f408382560db2d5952b3ea02cb 100644 --- a/tasks/cb100-035-elm-river.json +++ b/tasks/cb100-035-elm-river.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Massachusetts", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Elm River trade-secret privilege audit\n\nYou are acting as discovery counsel for Elm River Robotics. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-LD-2705 \nOther party or authority: Former engineers and Nacre Automation \nJurisdiction: Massachusetts \nVenue or forum: District of Massachusetts \nDecision deadline: 2026-09-19\n\n## Assignment\n\nThe production mixes counsel-directed forensic reports with ordinary-course analyses and omits attachments from several high-value email families. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-035-elm-river\",\n \"matter_number\": \"CB-LD-2705\",\n \"prepared_for\": \"Elm River Robotics\",\n \"as_of\": \"2026-09-19\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Elm River trade-secret privilege audit\n\nYou are acting as discovery counsel for Elm River Robotics. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-LD-2705\nOther party or authority: Former engineers and Nacre Automation\nJurisdiction: Massachusetts\nVenue or forum: District of Massachusetts\nDecision deadline: 2026-09-19\n\n## Assignment\n\nThe production mixes counsel-directed forensic reports with ordinary-course analyses and omits attachments from several high-value email families. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-035-elm-river\",\n \"matter_number\": \"CB-LD-2705\",\n \"prepared_for\": \"Elm River Robotics\",\n \"as_of\": \"2026-09-19\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-035-elm-river", "task_name": "Elm River trade-secret privilege audit", diff --git a/tasks/cb100-036-federal-basin.json b/tasks/cb100-036-federal-basin.json index e8e7e7be902825e6c330023dd1b71e8acd00dc7c..86b4ad98b9abe2f6c6096f6faee926a6d19ca3be 100644 --- a/tasks/cb100-036-federal-basin.json +++ b/tasks/cb100-036-federal-basin.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Western District of Oklahoma", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Federal Basin royalty dispute production\n\nYou are acting as discovery counsel for Federal Basin Resources. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-LD-2706 \nOther party or authority: Osage Crest Minerals LLC \nJurisdiction: Oklahoma \nVenue or forum: Western District of Oklahoma \nDecision deadline: 2026-09-23\n\n## Assignment\n\nLease-accounting exports, title opinions, and executive messages must be mapped to document requests before a court-ordered certification. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-036-federal-basin\",\n \"matter_number\": \"CB-LD-2706\",\n \"prepared_for\": \"Federal Basin Resources\",\n \"as_of\": \"2026-09-23\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Federal Basin royalty dispute production\n\nYou are acting as discovery counsel for Federal Basin Resources. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-LD-2706\nOther party or authority: Osage Crest Minerals LLC\nJurisdiction: Oklahoma\nVenue or forum: Western District of Oklahoma\nDecision deadline: 2026-09-23\n\n## Assignment\n\nLease-accounting exports, title opinions, and executive messages must be mapped to document requests before a court-ordered certification. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-036-federal-basin\",\n \"matter_number\": \"CB-LD-2706\",\n \"prepared_for\": \"Federal Basin Resources\",\n \"as_of\": \"2026-09-23\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-036-federal-basin", "task_name": "Federal Basin royalty dispute production", diff --git a/tasks/cb100-037-glacier-insurance.json b/tasks/cb100-037-glacier-insurance.json index d498cfd12a428684e91a26a9dd6ea245e67b7dfe..b63d403e3c3848a4d3217ce4e6809e4fe0075155 100644 --- a/tasks/cb100-037-glacier-insurance.json +++ b/tasks/cb100-037-glacier-insurance.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Southern District of New York", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Glacier coverage action discovery review\n\nYou are acting as discovery counsel for Glacier Mutual Insurance. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-LD-2707 \nOther party or authority: Moraine Data Centers, Inc. \nJurisdiction: New York \nVenue or forum: Southern District of New York \nDecision deadline: 2026-09-26\n\n## Assignment\n\nClaims files contain mixed business and legal communications, redaction inconsistencies, and a disputed clawback notice date. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-037-glacier-insurance\",\n \"matter_number\": \"CB-LD-2707\",\n \"prepared_for\": \"Glacier Mutual Insurance\",\n \"as_of\": \"2026-09-26\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Glacier coverage action discovery review\n\nYou are acting as discovery counsel for Glacier Mutual Insurance. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-LD-2707\nOther party or authority: Moraine Data Centers, Inc.\nJurisdiction: New York\nVenue or forum: Southern District of New York\nDecision deadline: 2026-09-26\n\n## Assignment\n\nClaims files contain mixed business and legal communications, redaction inconsistencies, and a disputed clawback notice date. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-037-glacier-insurance\",\n \"matter_number\": \"CB-LD-2707\",\n \"prepared_for\": \"Glacier Mutual Insurance\",\n \"as_of\": \"2026-09-26\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-037-glacier-insurance", "task_name": "Glacier coverage action discovery review", diff --git a/tasks/cb100-038-highland-securities.json b/tasks/cb100-038-highland-securities.json index 14941ffad1429a581d4c6fcf901f026940d96124..eef9adbcee9dd10b58929fe8c97e6681d73b3c56 100644 --- a/tasks/cb100-038-highland-securities.json +++ b/tasks/cb100-038-highland-securities.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Middle District of Tennessee", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Highland securities case hold audit\n\nYou are acting as discovery counsel for Highland Medical Analytics. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-LD-2708 \nOther party or authority: Patel Securities Action \nJurisdiction: Tennessee \nVenue or forum: Middle District of Tennessee \nDecision deadline: 2026-09-30\n\n## Assignment\n\nA securities action requires collection from executives, messaging platforms, and board portals after an acquisition changed account ownership. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-038-highland-securities\",\n \"matter_number\": \"CB-LD-2708\",\n \"prepared_for\": \"Highland Medical Analytics\",\n \"as_of\": \"2026-09-30\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Highland securities case hold audit\n\nYou are acting as discovery counsel for Highland Medical Analytics. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-LD-2708\nOther party or authority: Patel Securities Action\nJurisdiction: Tennessee\nVenue or forum: Middle District of Tennessee\nDecision deadline: 2026-09-30\n\n## Assignment\n\nA securities action requires collection from executives, messaging platforms, and board portals after an acquisition changed account ownership. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-038-highland-securities\",\n \"matter_number\": \"CB-LD-2708\",\n \"prepared_for\": \"Highland Medical Analytics\",\n \"as_of\": \"2026-09-30\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-038-highland-securities", "task_name": "Highland securities case hold audit", diff --git a/tasks/cb100-039-isotope-environmental.json b/tasks/cb100-039-isotope-environmental.json index aadf4e6b10a4846e85b714612e53fbbdf4857f37..f1f48d28eeb861ad98b20b93bf90743689455fdc 100644 --- a/tasks/cb100-039-isotope-environmental.json +++ b/tasks/cb100-039-isotope-environmental.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Eastern District of Louisiana", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Isotope contamination case production check\n\nYou are acting as discovery counsel for Isotope Specialty Chemicals. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-LD-2709 \nOther party or authority: County of Briar v. Isotope \nJurisdiction: Louisiana \nVenue or forum: Eastern District of Louisiana \nDecision deadline: 2026-10-03\n\n## Assignment\n\nEnvironmental sampling, consultant drafts, and regulator correspondence present designation, metadata, and expert-discovery questions. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-039-isotope-environmental\",\n \"matter_number\": \"CB-LD-2709\",\n \"prepared_for\": \"Isotope Specialty Chemicals\",\n \"as_of\": \"2026-10-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Isotope contamination case production check\n\nYou are acting as discovery counsel for Isotope Specialty Chemicals. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-LD-2709\nOther party or authority: County of Briar v. Isotope\nJurisdiction: Louisiana\nVenue or forum: Eastern District of Louisiana\nDecision deadline: 2026-10-03\n\n## Assignment\n\nEnvironmental sampling, consultant drafts, and regulator correspondence present designation, metadata, and expert-discovery questions. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-039-isotope-environmental\",\n \"matter_number\": \"CB-LD-2709\",\n \"prepared_for\": \"Isotope Specialty Chemicals\",\n \"as_of\": \"2026-10-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-039-isotope-environmental", "task_name": "Isotope contamination case production check", diff --git a/tasks/cb100-040-juniper-franchise.json b/tasks/cb100-040-juniper-franchise.json index 13a1831ace68e561ab5e11d884a8a5cbd1d58762..2cd309945ca9d26256235301ee25ccdadabb09bd 100644 --- a/tasks/cb100-040-juniper-franchise.json +++ b/tasks/cb100-040-juniper-franchise.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Middle District of Florida", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Juniper franchise dispute discovery audit\n\nYou are acting as discovery counsel for Juniper Table Restaurants. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-LD-2710 \nOther party or authority: Oak Street Franchisees Association \nJurisdiction: Florida \nVenue or forum: Middle District of Florida \nDecision deadline: 2026-10-07\n\n## Assignment\n\nA franchise production has custodian scope gaps, incomplete chat threads, and order-form attachments missing from key email families. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-040-juniper-franchise\",\n \"matter_number\": \"CB-LD-2710\",\n \"prepared_for\": \"Juniper Table Restaurants\",\n \"as_of\": \"2026-10-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Juniper franchise dispute discovery audit\n\nYou are acting as discovery counsel for Juniper Table Restaurants. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-LD-2710\nOther party or authority: Oak Street Franchisees Association\nJurisdiction: Florida\nVenue or forum: Middle District of Florida\nDecision deadline: 2026-10-07\n\n## Assignment\n\nA franchise production has custodian scope gaps, incomplete chat threads, and order-form attachments missing from key email families. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- custodian omission\n- legal-hold delivery gap\n- collection-date inconsistency\n- email-family separation\n- privilege-log description defect\n- common-interest support gap\n- clawback deadline\n- protective-order designation error\n- request-response mismatch\n- deposition exhibit omission\n- expert draft segregation\n- third-party subpoena deadline\n- production gap\n- metadata field loss\n- Bates-range overlap\n- meet-and-confer commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_pleadings/001_pleadings_control_register.md`\n- `/workspace/documents/01_pleadings/004_pleadings_ledger_export.csv`\n- `/workspace/documents/01_pleadings/007_pleadings_officer_certificate.html`\n- `/workspace/documents/02_orders/010_orders_executed_instrument.txt`\n- `/workspace/documents/02_orders/013_orders_review_memorandum.json`\n- `/workspace/documents/02_orders/016_orders_status_report.md`\n- `/workspace/documents/03_requests_responses/019_requests_responses_correspondence.eml`\n- `/workspace/documents/03_requests_responses/022_requests_responses_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-040-juniper-franchise\",\n \"matter_number\": \"CB-LD-2710\",\n \"prepared_for\": \"Juniper Table Restaurants\",\n \"as_of\": \"2026-10-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-040-juniper-franchise", "task_name": "Juniper franchise dispute discovery audit", diff --git a/tasks/cb100-042-birch-aviation.json b/tasks/cb100-042-birch-aviation.json index 744152b5e60fdcdddb24738e4f72743700249b8b..fd5aa42dcc8686c465f8951e8e6ffa02de8801d6 100644 --- a/tasks/cb100-042-birch-aviation.json +++ b/tasks/cb100-042-birch-aviation.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Eastern District of Virginia Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Birch Aviation lease reconciliation\n\nYou are acting as debtor-side restructuring counsel for Birch Regional Aviation LLC. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-RS-2802 \nOther party or authority: Skyframe Aircraft Leasing \nJurisdiction: Virginia \nVenue or forum: Eastern District of Virginia Bankruptcy Court \nDecision deadline: 2026-09-08\n\n## Assignment\n\nAircraft and gate leases carry cure disputes, guaranties, and notice addresses that changed shortly before the petition date. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-042-birch-aviation\",\n \"matter_number\": \"CB-RS-2802\",\n \"prepared_for\": \"Birch Regional Aviation LLC\",\n \"as_of\": \"2026-09-08\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Birch Aviation lease reconciliation\n\nYou are acting as debtor-side restructuring counsel for Birch Regional Aviation LLC. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-RS-2802\nOther party or authority: Skyframe Aircraft Leasing\nJurisdiction: Virginia\nVenue or forum: Eastern District of Virginia Bankruptcy Court\nDecision deadline: 2026-09-08\n\n## Assignment\n\nAircraft and gate leases carry cure disputes, guaranties, and notice addresses that changed shortly before the petition date. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-042-birch-aviation\",\n \"matter_number\": \"CB-RS-2802\",\n \"prepared_for\": \"Birch Regional Aviation LLC\",\n \"as_of\": \"2026-09-08\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-042-birch-aviation", "task_name": "Birch Aviation lease reconciliation", diff --git a/tasks/cb100-043-cascade-retail.json b/tasks/cb100-043-cascade-retail.json index d5b8cb1a20ab313a6ec19b9d1484231f6bff6335..e6c34c18098fef9b55153492d5f932d3100a5b21 100644 --- a/tasks/cb100-043-cascade-retail.json +++ b/tasks/cb100-043-cascade-retail.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Delaware Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Cascade Retail claims reserve audit\n\nYou are acting as debtor-side restructuring counsel for Cascade Outdoor Retail Corp.. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-RS-2803 \nOther party or authority: Multiple merchandise vendors \nJurisdiction: Delaware \nVenue or forum: District of Delaware Bankruptcy Court \nDecision deadline: 2026-09-11\n\n## Assignment\n\nA retail debtor faces duplicate claims, returned-goods offsets, rejected store leases, and inconsistent ballot amounts. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-043-cascade-retail\",\n \"matter_number\": \"CB-RS-2803\",\n \"prepared_for\": \"Cascade Outdoor Retail Corp.\",\n \"as_of\": \"2026-09-11\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Cascade Retail claims reserve audit\n\nYou are acting as debtor-side restructuring counsel for Cascade Outdoor Retail Corp.. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-RS-2803\nOther party or authority: Multiple merchandise vendors\nJurisdiction: Delaware\nVenue or forum: District of Delaware Bankruptcy Court\nDecision deadline: 2026-09-11\n\n## Assignment\n\nA retail debtor faces duplicate claims, returned-goods offsets, rejected store leases, and inconsistent ballot amounts. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-043-cascade-retail\",\n \"matter_number\": \"CB-RS-2803\",\n \"prepared_for\": \"Cascade Outdoor Retail Corp.\",\n \"as_of\": \"2026-09-11\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-043-cascade-retail", "task_name": "Cascade Retail claims reserve audit", diff --git a/tasks/cb100-044-drummond-care.json b/tasks/cb100-044-drummond-care.json index c146b914e7d6fd3fefff2291707150a9b650cd47..4dd3d6c05554ced193858d17e5824954ea14e552 100644 --- a/tasks/cb100-044-drummond-care.json +++ b/tasks/cb100-044-drummond-care.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of New Jersey Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Drummond Care restructuring schedule check\n\nYou are acting as debtor-side restructuring counsel for Drummond Senior Care Group. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-RS-2804 \nOther party or authority: Pinecrest Medical Staffing \nJurisdiction: New Jersey \nVenue or forum: District of New Jersey Bankruptcy Court \nDecision deadline: 2026-09-15\n\n## Assignment\n\nHealthcare staffing, tax, and landlord claims conflict with schedules while cash-collateral reports omit a disputed receivable reserve. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-044-drummond-care\",\n \"matter_number\": \"CB-RS-2804\",\n \"prepared_for\": \"Drummond Senior Care Group\",\n \"as_of\": \"2026-09-15\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Drummond Care restructuring schedule check\n\nYou are acting as debtor-side restructuring counsel for Drummond Senior Care Group. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-RS-2804\nOther party or authority: Pinecrest Medical Staffing\nJurisdiction: New Jersey\nVenue or forum: District of New Jersey Bankruptcy Court\nDecision deadline: 2026-09-15\n\n## Assignment\n\nHealthcare staffing, tax, and landlord claims conflict with schedules while cash-collateral reports omit a disputed receivable reserve. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-044-drummond-care\",\n \"matter_number\": \"CB-RS-2804\",\n \"prepared_for\": \"Drummond Senior Care Group\",\n \"as_of\": \"2026-09-15\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-044-drummond-care", "task_name": "Drummond Care restructuring schedule check", diff --git a/tasks/cb100-045-equinox-solar.json b/tasks/cb100-045-equinox-solar.json index d83b015c8d28f8cae42c816068b426440069aa8d..c45129e97adeb3c39d5fcfb067b39efdd8cb0083 100644 --- a/tasks/cb100-045-equinox-solar.json +++ b/tasks/cb100-045-equinox-solar.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Arizona Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Equinox Solar project claim reconciliation\n\nYou are acting as debtor-side restructuring counsel for Equinox Solar Development. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-RS-2805 \nOther party or authority: Mesa Array Contractors \nJurisdiction: Arizona \nVenue or forum: District of Arizona Bankruptcy Court \nDecision deadline: 2026-09-18\n\n## Assignment\n\nConstruction claims assert secured and priority treatment across partially completed projects with overlapping guaranties. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-045-equinox-solar\",\n \"matter_number\": \"CB-RS-2805\",\n \"prepared_for\": \"Equinox Solar Development\",\n \"as_of\": \"2026-09-18\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Equinox Solar project claim reconciliation\n\nYou are acting as debtor-side restructuring counsel for Equinox Solar Development. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-RS-2805\nOther party or authority: Mesa Array Contractors\nJurisdiction: Arizona\nVenue or forum: District of Arizona Bankruptcy Court\nDecision deadline: 2026-09-18\n\n## Assignment\n\nConstruction claims assert secured and priority treatment across partially completed projects with overlapping guaranties. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-045-equinox-solar\",\n \"matter_number\": \"CB-RS-2805\",\n \"prepared_for\": \"Equinox Solar Development\",\n \"as_of\": \"2026-09-18\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-045-equinox-solar", "task_name": "Equinox Solar project claim reconciliation", diff --git a/tasks/cb100-046-foundry-foods.json b/tasks/cb100-046-foundry-foods.json index 1644e9ef63cee8c2a693b2a3d26443dc7d053a8b..29d27bad9aeb0802712504dca2542bcb53ad7739 100644 --- a/tasks/cb100-046-foundry-foods.json +++ b/tasks/cb100-046-foundry-foods.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Minnesota Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Foundry Foods vendor claim audit\n\nYou are acting as debtor-side restructuring counsel for Foundry Prepared Foods, Inc.. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-RS-2806 \nOther party or authority: Northland Cold Chain LLC \nJurisdiction: Minnesota \nVenue or forum: District of Minnesota Bankruptcy Court \nDecision deadline: 2026-09-22\n\n## Assignment\n\nCold-storage invoices straddle the petition date, critical-vendor payments overlap claims, and cure schedules use stale balances. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-046-foundry-foods\",\n \"matter_number\": \"CB-RS-2806\",\n \"prepared_for\": \"Foundry Prepared Foods, Inc.\",\n \"as_of\": \"2026-09-22\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Foundry Foods vendor claim audit\n\nYou are acting as debtor-side restructuring counsel for Foundry Prepared Foods, Inc.. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-RS-2806\nOther party or authority: Northland Cold Chain LLC\nJurisdiction: Minnesota\nVenue or forum: District of Minnesota Bankruptcy Court\nDecision deadline: 2026-09-22\n\n## Assignment\n\nCold-storage invoices straddle the petition date, critical-vendor payments overlap claims, and cure schedules use stale balances. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-046-foundry-foods\",\n \"matter_number\": \"CB-RS-2806\",\n \"prepared_for\": \"Foundry Prepared Foods, Inc.\",\n \"as_of\": \"2026-09-22\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-046-foundry-foods", "task_name": "Foundry Foods vendor claim audit", diff --git a/tasks/cb100-047-greenway-transit.json b/tasks/cb100-047-greenway-transit.json index 5b66c983a3fa5e4e07bdff595a256a6e10517adf..9b9a5d5ade8401d5a55f86e77fcec67a5221768d 100644 --- a/tasks/cb100-047-greenway-transit.json +++ b/tasks/cb100-047-greenway-transit.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Northern District of Illinois Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Greenway Transit plan claim review\n\nYou are acting as debtor-side restructuring counsel for Greenway Transit Services. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-RS-2807 \nOther party or authority: Metro Fleet Finance \nJurisdiction: Illinois \nVenue or forum: Northern District of Illinois Bankruptcy Court \nDecision deadline: 2026-09-25\n\n## Assignment\n\nFleet liens, tax claims, rejection damages, and a DIP variance affect voting and reserve recommendations. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-047-greenway-transit\",\n \"matter_number\": \"CB-RS-2807\",\n \"prepared_for\": \"Greenway Transit Services\",\n \"as_of\": \"2026-09-25\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Greenway Transit plan claim review\n\nYou are acting as debtor-side restructuring counsel for Greenway Transit Services. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-RS-2807\nOther party or authority: Metro Fleet Finance\nJurisdiction: Illinois\nVenue or forum: Northern District of Illinois Bankruptcy Court\nDecision deadline: 2026-09-25\n\n## Assignment\n\nFleet liens, tax claims, rejection damages, and a DIP variance affect voting and reserve recommendations. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-047-greenway-transit\",\n \"matter_number\": \"CB-RS-2807\",\n \"prepared_for\": \"Greenway Transit Services\",\n \"as_of\": \"2026-09-25\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-047-greenway-transit", "task_name": "Greenway Transit plan claim review", diff --git a/tasks/cb100-048-harvest-paper.json b/tasks/cb100-048-harvest-paper.json index 6ca39ed9980bdd2dbf203c2c6d06ba7770681c8d..e3c00cc0e07be3bbc3c613807b0fb9e0165e281f 100644 --- a/tasks/cb100-048-harvest-paper.json +++ b/tasks/cb100-048-harvest-paper.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Maine Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Harvest Paper mill restructuring audit\n\nYou are acting as debtor-side restructuring counsel for Harvest Paper Mills LLC. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-RS-2808 \nOther party or authority: Timber Run Energy Cooperative \nJurisdiction: Maine \nVenue or forum: District of Maine Bankruptcy Court \nDecision deadline: 2026-09-29\n\n## Assignment\n\nEnergy and timber contracts have contested cure amounts, setoff claims, and address defects in solicitation records. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-048-harvest-paper\",\n \"matter_number\": \"CB-RS-2808\",\n \"prepared_for\": \"Harvest Paper Mills LLC\",\n \"as_of\": \"2026-09-29\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Harvest Paper mill restructuring audit\n\nYou are acting as debtor-side restructuring counsel for Harvest Paper Mills LLC. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-RS-2808\nOther party or authority: Timber Run Energy Cooperative\nJurisdiction: Maine\nVenue or forum: District of Maine Bankruptcy Court\nDecision deadline: 2026-09-29\n\n## Assignment\n\nEnergy and timber contracts have contested cure amounts, setoff claims, and address defects in solicitation records. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-048-harvest-paper\",\n \"matter_number\": \"CB-RS-2808\",\n \"prepared_for\": \"Harvest Paper Mills LLC\",\n \"as_of\": \"2026-09-29\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-048-harvest-paper", "task_name": "Harvest Paper mill restructuring audit", diff --git a/tasks/cb100-049-inlet-fitness.json b/tasks/cb100-049-inlet-fitness.json index 538ca225c14c3a9278fd530efdb6c705a53afe65..f8cacce8f01ad0097913a37ce4eb45f1df7d31b8 100644 --- a/tasks/cb100-049-inlet-fitness.json +++ b/tasks/cb100-049-inlet-fitness.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Central District of California Bankruptcy Court", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Inlet Fitness lease claims review\n\nYou are acting as debtor-side restructuring counsel for Inlet Fitness Studios. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-RS-2809 \nOther party or authority: Urban Square Properties \nJurisdiction: California \nVenue or forum: Central District of California Bankruptcy Court \nDecision deadline: 2026-10-02\n\n## Assignment\n\nStudio leases, member refunds, and franchise obligations create duplicate and misclassified claims across the register. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-049-inlet-fitness\",\n \"matter_number\": \"CB-RS-2809\",\n \"prepared_for\": \"Inlet Fitness Studios\",\n \"as_of\": \"2026-10-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Inlet Fitness lease claims review\n\nYou are acting as debtor-side restructuring counsel for Inlet Fitness Studios. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-RS-2809\nOther party or authority: Urban Square Properties\nJurisdiction: California\nVenue or forum: Central District of California Bankruptcy Court\nDecision deadline: 2026-10-02\n\n## Assignment\n\nStudio leases, member refunds, and franchise obligations create duplicate and misclassified claims across the register. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- scheduled-claim mismatch\n- duplicate proof of claim\n- secured-status defect\n- priority classification error\n- executory-contract cure dispute\n- lease rejection deadline\n- postpetition invoice\n- critical-vendor overstatement\n- setoff assertion\n- guaranty overlap\n- tax priority period\n- DIP budget variance\n- cash-collateral reporting gap\n- ballot amount discrepancy\n- notice-address defect\n- reserve calculation error\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_petition_schedules/001_petition_schedules_control_register.md`\n- `/workspace/documents/01_petition_schedules/004_petition_schedules_ledger_export.csv`\n- `/workspace/documents/01_petition_schedules/007_petition_schedules_officer_certificate.html`\n- `/workspace/documents/02_claims_register/010_claims_register_executed_instrument.txt`\n- `/workspace/documents/02_claims_register/013_claims_register_review_memorandum.json`\n- `/workspace/documents/02_claims_register/016_claims_register_status_report.md`\n- `/workspace/documents/03_proofs_of_claim/019_proofs_of_claim_correspondence.eml`\n- `/workspace/documents/03_proofs_of_claim/022_proofs_of_claim_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-049-inlet-fitness\",\n \"matter_number\": \"CB-RS-2809\",\n \"prepared_for\": \"Inlet Fitness Studios\",\n \"as_of\": \"2026-10-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-049-inlet-fitness", "task_name": "Inlet Fitness lease claims review", diff --git a/tasks/cb100-051-acorn-logistics.json b/tasks/cb100-051-acorn-logistics.json index 133fbe06744af07a8d492966df8566b5df37244a..af6e2cdb00fa436cd35f83fe9e8c240c34333c08 100644 --- a/tasks/cb100-051-acorn-logistics.json +++ b/tasks/cb100-051-acorn-logistics.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "San Bernardino County, California", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Acorn industrial portfolio lease audit\n\nYou are acting as real-estate transactions counsel for Acorn Logistics REIT. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-RE-2901 \nOther party or authority: Redwood Fulfillment Services \nJurisdiction: California \nVenue or forum: San Bernardino County, California \nDecision deadline: 2026-09-03\n\n## Assignment\n\nThree warehouses have conflicting rent schedules, an unclosed solar permit, and an estoppel that excludes a disputed CAM credit. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-051-acorn-logistics\",\n \"matter_number\": \"CB-RE-2901\",\n \"prepared_for\": \"Acorn Logistics REIT\",\n \"as_of\": \"2026-09-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Acorn industrial portfolio lease audit\n\nYou are acting as real-estate transactions counsel for Acorn Logistics REIT. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-RE-2901\nOther party or authority: Redwood Fulfillment Services\nJurisdiction: California\nVenue or forum: San Bernardino County, California\nDecision deadline: 2026-09-03\n\n## Assignment\n\nThree warehouses have conflicting rent schedules, an unclosed solar permit, and an estoppel that excludes a disputed CAM credit. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-051-acorn-logistics\",\n \"matter_number\": \"CB-RE-2901\",\n \"prepared_for\": \"Acorn Logistics REIT\",\n \"as_of\": \"2026-09-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-051-acorn-logistics", "task_name": "Acorn industrial portfolio lease audit", diff --git a/tasks/cb100-052-brookline-clinic.json b/tasks/cb100-052-brookline-clinic.json index f4491ab9fab77bcdfcf52cb984b876402a610142..d478e56e28c3709d717b0b80b904a7ae22d12b30 100644 --- a/tasks/cb100-052-brookline-clinic.json +++ b/tasks/cb100-052-brookline-clinic.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Middlesex County, Massachusetts", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Brookline clinic acquisition property review\n\nYou are acting as real-estate transactions counsel for Brookline Care Partners. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-RE-2902 \nOther party or authority: Elm Street Medical Properties \nJurisdiction: Massachusetts \nVenue or forum: Middlesex County, Massachusetts \nDecision deadline: 2026-09-07\n\n## Assignment\n\nA medical-office portfolio presents zoning, assignment, lender consent, and environmental records that do not align with seller certificates. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-052-brookline-clinic\",\n \"matter_number\": \"CB-RE-2902\",\n \"prepared_for\": \"Brookline Care Partners\",\n \"as_of\": \"2026-09-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Brookline clinic acquisition property review\n\nYou are acting as real-estate transactions counsel for Brookline Care Partners. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-RE-2902\nOther party or authority: Elm Street Medical Properties\nJurisdiction: Massachusetts\nVenue or forum: Middlesex County, Massachusetts\nDecision deadline: 2026-09-07\n\n## Assignment\n\nA medical-office portfolio presents zoning, assignment, lender consent, and environmental records that do not align with seller certificates. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-052-brookline-clinic\",\n \"matter_number\": \"CB-RE-2902\",\n \"prepared_for\": \"Brookline Care Partners\",\n \"as_of\": \"2026-09-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-052-brookline-clinic", "task_name": "Brookline clinic acquisition property review", diff --git a/tasks/cb100-053-canyon-grocery.json b/tasks/cb100-053-canyon-grocery.json index a6246eb6c84da3e879020dc484ac3d3bbd2cf63c..d3a5c6df4ad252d5072a67bb4789111b7a779f25 100644 --- a/tasks/cb100-053-canyon-grocery.json +++ b/tasks/cb100-053-canyon-grocery.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Maricopa County, Arizona", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Canyon grocery lease portfolio audit\n\nYou are acting as real-estate transactions counsel for Canyon Fresh Markets. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-RE-2903 \nOther party or authority: Mesa Vista Shopping Centers \nJurisdiction: Arizona \nVenue or forum: Maricopa County, Arizona \nDecision deadline: 2026-09-10\n\n## Assignment\n\nStore leases contain exclusives, renewal options, and percentage-rent provisions that conflict with landlord statements and payment records. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-053-canyon-grocery\",\n \"matter_number\": \"CB-RE-2903\",\n \"prepared_for\": \"Canyon Fresh Markets\",\n \"as_of\": \"2026-09-10\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Canyon grocery lease portfolio audit\n\nYou are acting as real-estate transactions counsel for Canyon Fresh Markets. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-RE-2903\nOther party or authority: Mesa Vista Shopping Centers\nJurisdiction: Arizona\nVenue or forum: Maricopa County, Arizona\nDecision deadline: 2026-09-10\n\n## Assignment\n\nStore leases contain exclusives, renewal options, and percentage-rent provisions that conflict with landlord statements and payment records. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-053-canyon-grocery\",\n \"matter_number\": \"CB-RE-2903\",\n \"prepared_for\": \"Canyon Fresh Markets\",\n \"as_of\": \"2026-09-10\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-053-canyon-grocery", "task_name": "Canyon grocery lease portfolio audit", diff --git a/tasks/cb100-054-dockside-hotel.json b/tasks/cb100-054-dockside-hotel.json index ff1828db6a8268e3166acf20a6b79385b60a2819..749f67c787495d5de297c31bec5c9df02a9f0bff 100644 --- a/tasks/cb100-054-dockside-hotel.json +++ b/tasks/cb100-054-dockside-hotel.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Duval County, Florida", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Dockside hotel financing property audit\n\nYou are acting as real-estate transactions counsel for Dockside Hospitality LLC. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-RE-2904 \nOther party or authority: Harbor Quay Lending \nJurisdiction: Florida \nVenue or forum: Duval County, Florida \nDecision deadline: 2026-09-14\n\n## Assignment\n\nA waterfront hotel has a survey encroachment, restoration deadline, construction lien notice, and missing insurance endorsement. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-054-dockside-hotel\",\n \"matter_number\": \"CB-RE-2904\",\n \"prepared_for\": \"Dockside Hospitality LLC\",\n \"as_of\": \"2026-09-14\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Dockside hotel financing property audit\n\nYou are acting as real-estate transactions counsel for Dockside Hospitality LLC. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-RE-2904\nOther party or authority: Harbor Quay Lending\nJurisdiction: Florida\nVenue or forum: Duval County, Florida\nDecision deadline: 2026-09-14\n\n## Assignment\n\nA waterfront hotel has a survey encroachment, restoration deadline, construction lien notice, and missing insurance endorsement. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-054-dockside-hotel\",\n \"matter_number\": \"CB-RE-2904\",\n \"prepared_for\": \"Dockside Hospitality LLC\",\n \"as_of\": \"2026-09-14\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-054-dockside-hotel", "task_name": "Dockside hotel financing property audit", diff --git a/tasks/cb100-056-fairhaven-storage.json b/tasks/cb100-056-fairhaven-storage.json index 25d5e2101c84ad4de0622e55eb74063573397cc3..8b5b4b6703e0771dafb51a10214719360beb91a1 100644 --- a/tasks/cb100-056-fairhaven-storage.json +++ b/tasks/cb100-056-fairhaven-storage.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Bexar County, Texas", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Fairhaven self-storage portfolio review\n\nYou are acting as real-estate transactions counsel for Fairhaven Storage Partners. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-RE-2906 \nOther party or authority: Cobalt Door Management \nJurisdiction: Texas \nVenue or forum: Bexar County, Texas \nDecision deadline: 2026-09-21\n\n## Assignment\n\nA storage portfolio has title exceptions, rent-roll discrepancies, security deposits, and management charges inconsistent with governing documents. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-056-fairhaven-storage\",\n \"matter_number\": \"CB-RE-2906\",\n \"prepared_for\": \"Fairhaven Storage Partners\",\n \"as_of\": \"2026-09-21\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Fairhaven self-storage portfolio review\n\nYou are acting as real-estate transactions counsel for Fairhaven Storage Partners. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-RE-2906\nOther party or authority: Cobalt Door Management\nJurisdiction: Texas\nVenue or forum: Bexar County, Texas\nDecision deadline: 2026-09-21\n\n## Assignment\n\nA storage portfolio has title exceptions, rent-roll discrepancies, security deposits, and management charges inconsistent with governing documents. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-056-fairhaven-storage\",\n \"matter_number\": \"CB-RE-2906\",\n \"prepared_for\": \"Fairhaven Storage Partners\",\n \"as_of\": \"2026-09-21\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-056-fairhaven-storage", "task_name": "Fairhaven self-storage portfolio review", diff --git a/tasks/cb100-057-glenridge-school.json b/tasks/cb100-057-glenridge-school.json index eb9473eea003cd28a433a5f1907fed7a4182b051..ccee939860e4eae61ce6b04e8c3ff008d9238461 100644 --- a/tasks/cb100-057-glenridge-school.json +++ b/tasks/cb100-057-glenridge-school.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Arapahoe County, Colorado", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Glenridge charter campus lease review\n\nYou are acting as real-estate transactions counsel for Glenridge Education Network. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-RE-2907 \nOther party or authority: Cedar Hall Development \nJurisdiction: Colorado \nVenue or forum: Arapahoe County, Colorado \nDecision deadline: 2026-09-24\n\n## Assignment\n\nA school campus faces use restrictions, renewal notices, accessibility work, and an SNDA missing from lender records. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-057-glenridge-school\",\n \"matter_number\": \"CB-RE-2907\",\n \"prepared_for\": \"Glenridge Education Network\",\n \"as_of\": \"2026-09-24\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Glenridge charter campus lease review\n\nYou are acting as real-estate transactions counsel for Glenridge Education Network. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-RE-2907\nOther party or authority: Cedar Hall Development\nJurisdiction: Colorado\nVenue or forum: Arapahoe County, Colorado\nDecision deadline: 2026-09-24\n\n## Assignment\n\nA school campus faces use restrictions, renewal notices, accessibility work, and an SNDA missing from lender records. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-057-glenridge-school\",\n \"matter_number\": \"CB-RE-2907\",\n \"prepared_for\": \"Glenridge Education Network\",\n \"as_of\": \"2026-09-24\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-057-glenridge-school", "task_name": "Glenridge charter campus lease review", diff --git a/tasks/cb100-059-ivory-tower.json b/tasks/cb100-059-ivory-tower.json index 30bbcd026f348e1e4b8d50bcb9242ec543298f99..8313ae9d44ddb02d70b6e61edbc90a26593e88a0 100644 --- a/tasks/cb100-059-ivory-tower.json +++ b/tasks/cb100-059-ivory-tower.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "New York County, New York", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Ivory Tower office sublease reconciliation\n\nYou are acting as real-estate transactions counsel for Ivory Tower Software. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-RE-2909 \nOther party or authority: Metropolitan Workspace Holdings \nJurisdiction: New York \nVenue or forum: New York County, New York \nDecision deadline: 2026-10-01\n\n## Assignment\n\nA headquarters sublease has consent conditions, restoration obligations, rent abatements, and notices delivered under obsolete instructions. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-059-ivory-tower\",\n \"matter_number\": \"CB-RE-2909\",\n \"prepared_for\": \"Ivory Tower Software\",\n \"as_of\": \"2026-10-01\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Ivory Tower office sublease reconciliation\n\nYou are acting as real-estate transactions counsel for Ivory Tower Software. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-RE-2909\nOther party or authority: Metropolitan Workspace Holdings\nJurisdiction: New York\nVenue or forum: New York County, New York\nDecision deadline: 2026-10-01\n\n## Assignment\n\nA headquarters sublease has consent conditions, restoration obligations, rent abatements, and notices delivered under obsolete instructions. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-059-ivory-tower\",\n \"matter_number\": \"CB-RE-2909\",\n \"prepared_for\": \"Ivory Tower Software\",\n \"as_of\": \"2026-10-01\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-059-ivory-tower", "task_name": "Ivory Tower office sublease reconciliation", diff --git a/tasks/cb100-060-kestrel-farms.json b/tasks/cb100-060-kestrel-farms.json index 0343542413928a47de62018726872100d077b08c..9a42672ec50df62cad6befd01eb0d69162d23876 100644 --- a/tasks/cb100-060-kestrel-farms.json +++ b/tasks/cb100-060-kestrel-farms.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Yakima County, Washington", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Kestrel agricultural land portfolio audit\n\nYou are acting as real-estate transactions counsel for Kestrel Farms Cooperative. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-RE-2910 \nOther party or authority: Rainshadow Land Trust \nJurisdiction: Washington \nVenue or forum: Yakima County, Washington \nDecision deadline: 2026-10-05\n\n## Assignment\n\nIrrigated parcels carry water-right questions, access encroachments, crop leases, and insurance records inconsistent with loan covenants. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-060-kestrel-farms\",\n \"matter_number\": \"CB-RE-2910\",\n \"prepared_for\": \"Kestrel Farms Cooperative\",\n \"as_of\": \"2026-10-05\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Kestrel agricultural land portfolio audit\n\nYou are acting as real-estate transactions counsel for Kestrel Farms Cooperative. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-RE-2910\nOther party or authority: Rainshadow Land Trust\nJurisdiction: Washington\nVenue or forum: Yakima County, Washington\nDecision deadline: 2026-10-05\n\n## Assignment\n\nIrrigated parcels carry water-right questions, access encroachments, crop leases, and insurance records inconsistent with loan covenants. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- rent schedule inconsistency\n- renewal-option notice\n- exclusive-use conflict\n- assignment consent\n- estoppel discrepancy\n- survey encroachment\n- title exception\n- environmental recognized condition\n- zoning nonconformity\n- permit closeout gap\n- CAM reconciliation error\n- insurance endorsement deficiency\n- casualty restoration deadline\n- construction lien notice\n- lender SNDA omission\n- security-deposit mismatch\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_leases/001_leases_control_register.md`\n- `/workspace/documents/01_leases/004_leases_ledger_export.csv`\n- `/workspace/documents/01_leases/007_leases_officer_certificate.html`\n- `/workspace/documents/02_amendments/010_amendments_executed_instrument.txt`\n- `/workspace/documents/02_amendments/013_amendments_review_memorandum.json`\n- `/workspace/documents/02_amendments/016_amendments_status_report.md`\n- `/workspace/documents/03_estoppels/019_estoppels_correspondence.eml`\n- `/workspace/documents/03_estoppels/022_estoppels_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-060-kestrel-farms\",\n \"matter_number\": \"CB-RE-2910\",\n \"prepared_for\": \"Kestrel Farms Cooperative\",\n \"as_of\": \"2026-10-05\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-060-kestrel-farms", "task_name": "Kestrel agricultural land portfolio audit", diff --git a/tasks/cb100-061-arcadia-genomics.json b/tasks/cb100-061-arcadia-genomics.json index a951974e0f9f18a019d88345f402486d95f5aa96..6d6ffa08d36ba368dec1d08ab3f6156b3249d4d8 100644 --- a/tasks/cb100-061-arcadia-genomics.json +++ b/tasks/cb100-061-arcadia-genomics.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "California Privacy Protection Agency", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Arcadia genomic data response audit\n\nYou are acting as privacy and regulatory counsel for Arcadia Genomics, Inc.. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-PR-3001 \nOther party or authority: North Coast Sequencing Network \nJurisdiction: California \nVenue or forum: California Privacy Protection Agency \nDecision deadline: 2026-09-02\n\n## Assignment\n\nA consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-061-arcadia-genomics\",\n \"matter_number\": \"CB-PR-3001\",\n \"prepared_for\": \"Arcadia Genomics, Inc.\",\n \"as_of\": \"2026-09-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Arcadia genomic data response audit\n\nYou are acting as privacy and regulatory counsel for Arcadia Genomics, Inc.. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-PR-3001\nOther party or authority: North Coast Sequencing Network\nJurisdiction: California\nVenue or forum: California Privacy Protection Agency\nDecision deadline: 2026-09-02\n\n## Assignment\n\nA consumer access request exposes inconsistencies among genomic-data notices, deletion jobs, vendors, and overseas research transfers. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-061-arcadia-genomics\",\n \"matter_number\": \"CB-PR-3001\",\n \"prepared_for\": \"Arcadia Genomics, Inc.\",\n \"as_of\": \"2026-09-02\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-061-arcadia-genomics", "task_name": "Arcadia genomic data response audit", diff --git a/tasks/cb100-062-brightpath-kids.json b/tasks/cb100-062-brightpath-kids.json index bc27b5054fffe834f4e017de57a47927ec3b53b7..c359560623961cab6281f34c5f0945e96f43b3f5 100644 --- a/tasks/cb100-062-brightpath-kids.json +++ b/tasks/cb100-062-brightpath-kids.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Connecticut Attorney General", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# BrightPath child-privacy compliance review\n\nYou are acting as privacy and regulatory counsel for BrightPath Learning Apps. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-PR-3002 \nOther party or authority: Lighthouse School District \nJurisdiction: Connecticut \nVenue or forum: Connecticut Attorney General \nDecision deadline: 2026-09-06\n\n## Assignment\n\nAn education app collected voice samples under school consent while subprocessor and deletion records lagged product changes. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-062-brightpath-kids\",\n \"matter_number\": \"CB-PR-3002\",\n \"prepared_for\": \"BrightPath Learning Apps\",\n \"as_of\": \"2026-09-06\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# BrightPath child-privacy compliance review\n\nYou are acting as privacy and regulatory counsel for BrightPath Learning Apps. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-PR-3002\nOther party or authority: Lighthouse School District\nJurisdiction: Connecticut\nVenue or forum: Connecticut Attorney General\nDecision deadline: 2026-09-06\n\n## Assignment\n\nAn education app collected voice samples under school consent while subprocessor and deletion records lagged product changes. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-062-brightpath-kids\",\n \"matter_number\": \"CB-PR-3002\",\n \"prepared_for\": \"BrightPath Learning Apps\",\n \"as_of\": \"2026-09-06\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-062-brightpath-kids", "task_name": "BrightPath child-privacy compliance review", diff --git a/tasks/cb100-063-civic-route.json b/tasks/cb100-063-civic-route.json index 36143b9d6b7c9407ffb155e12ce3230a59440881..cbdc0f16aa15ec59aed109f19036b71f20697b8d 100644 --- a/tasks/cb100-063-civic-route.json +++ b/tasks/cb100-063-civic-route.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Colorado Attorney General", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# CivicRoute location-data inquiry\n\nYou are acting as privacy and regulatory counsel for CivicRoute Mobility. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-PR-3003 \nOther party or authority: MetroPulse Advertising LLC \nJurisdiction: Colorado \nVenue or forum: Colorado Attorney General \nDecision deadline: 2026-09-09\n\n## Assignment\n\nA mobility app shared precise location segments through an advertising integration not reflected in its processing inventory or opt-out flow. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-063-civic-route\",\n \"matter_number\": \"CB-PR-3003\",\n \"prepared_for\": \"CivicRoute Mobility\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# CivicRoute location-data inquiry\n\nYou are acting as privacy and regulatory counsel for CivicRoute Mobility. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-PR-3003\nOther party or authority: MetroPulse Advertising LLC\nJurisdiction: Colorado\nVenue or forum: Colorado Attorney General\nDecision deadline: 2026-09-09\n\n## Assignment\n\nA mobility app shared precise location segments through an advertising integration not reflected in its processing inventory or opt-out flow. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-063-civic-route\",\n \"matter_number\": \"CB-PR-3003\",\n \"prepared_for\": \"CivicRoute Mobility\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-063-civic-route", "task_name": "CivicRoute location-data inquiry", diff --git a/tasks/cb100-065-everfield-market.json b/tasks/cb100-065-everfield-market.json index 3ec0152d0829c1d89b0d3ad1c734115363d32125..2d9df7dce5023f785f732f38be1ae390dcade2dd 100644 --- a/tasks/cb100-065-everfield-market.json +++ b/tasks/cb100-065-everfield-market.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Virginia Attorney General", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Everfield loyalty-data assessment\n\nYou are acting as privacy and regulatory counsel for Everfield Markets. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-PR-3005 \nOther party or authority: Poppy Loyalty Cloud \nJurisdiction: Virginia \nVenue or forum: Virginia Attorney General \nDecision deadline: 2026-09-16\n\n## Assignment\n\nLoyalty profiles combine purchase and inferred health data with consent, retention, and processor terms that differ by system. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-065-everfield-market\",\n \"matter_number\": \"CB-PR-3005\",\n \"prepared_for\": \"Everfield Markets\",\n \"as_of\": \"2026-09-16\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Everfield loyalty-data assessment\n\nYou are acting as privacy and regulatory counsel for Everfield Markets. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-PR-3005\nOther party or authority: Poppy Loyalty Cloud\nJurisdiction: Virginia\nVenue or forum: Virginia Attorney General\nDecision deadline: 2026-09-16\n\n## Assignment\n\nLoyalty profiles combine purchase and inferred health data with consent, retention, and processor terms that differ by system. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-065-everfield-market\",\n \"matter_number\": \"CB-PR-3005\",\n \"prepared_for\": \"Everfield Markets\",\n \"as_of\": \"2026-09-16\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-065-everfield-market", "task_name": "Everfield loyalty-data assessment", diff --git a/tasks/cb100-066-fable-fintech.json b/tasks/cb100-066-fable-fintech.json index dec6e313af21520c4b1413d3d95c13c5a33723cb..5baaef8e4c415ad201e18f87e9122e49fe633558 100644 --- a/tasks/cb100-066-fable-fintech.json +++ b/tasks/cb100-066-fable-fintech.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Illinois Department of Financial and Professional Regulation", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Fable automated-decision privacy review\n\nYou are acting as privacy and regulatory counsel for Fable Financial Technologies. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-PR-3006 \nOther party or authority: Northbridge Model Services \nJurisdiction: Illinois \nVenue or forum: Illinois Department of Financial and Professional Regulation \nDecision deadline: 2026-09-20\n\n## Assignment\n\nA credit prequalification workflow uses third-party model features while notices, assessments, deletion records, and complaint responses conflict. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-066-fable-fintech\",\n \"matter_number\": \"CB-PR-3006\",\n \"prepared_for\": \"Fable Financial Technologies\",\n \"as_of\": \"2026-09-20\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Fable automated-decision privacy review\n\nYou are acting as privacy and regulatory counsel for Fable Financial Technologies. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-PR-3006\nOther party or authority: Northbridge Model Services\nJurisdiction: Illinois\nVenue or forum: Illinois Department of Financial and Professional Regulation\nDecision deadline: 2026-09-20\n\n## Assignment\n\nA credit prequalification workflow uses third-party model features while notices, assessments, deletion records, and complaint responses conflict. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-066-fable-fintech\",\n \"matter_number\": \"CB-PR-3006\",\n \"prepared_for\": \"Fable Financial Technologies\",\n \"as_of\": \"2026-09-20\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-066-fable-fintech", "task_name": "Fable automated-decision privacy review", diff --git a/tasks/cb100-067-grove-hospitality.json b/tasks/cb100-067-grove-hospitality.json index dd86a074f19bc4971e5d81f147263eb498346dbb..d746d3734c6ed95278892de6a18d09b026c6fb8c 100644 --- a/tasks/cb100-067-grove-hospitality.json +++ b/tasks/cb100-067-grove-hospitality.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Federal Trade Commission", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Grove guest-data transfer audit\n\nYou are acting as privacy and regulatory counsel for Grove Hospitality Group. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-PR-3007 \nOther party or authority: GlobalStay Reservation Services \nJurisdiction: Florida \nVenue or forum: Federal Trade Commission \nDecision deadline: 2026-09-23\n\n## Assignment\n\nReservation data crosses regions through undocumented subprocessors and a failed retention job identified during a regulator response. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-067-grove-hospitality\",\n \"matter_number\": \"CB-PR-3007\",\n \"prepared_for\": \"Grove Hospitality Group\",\n \"as_of\": \"2026-09-23\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Grove guest-data transfer audit\n\nYou are acting as privacy and regulatory counsel for Grove Hospitality Group. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-PR-3007\nOther party or authority: GlobalStay Reservation Services\nJurisdiction: Florida\nVenue or forum: Federal Trade Commission\nDecision deadline: 2026-09-23\n\n## Assignment\n\nReservation data crosses regions through undocumented subprocessors and a failed retention job identified during a regulator response. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-067-grove-hospitality\",\n \"matter_number\": \"CB-PR-3007\",\n \"prepared_for\": \"Grove Hospitality Group\",\n \"as_of\": \"2026-09-23\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-067-grove-hospitality", "task_name": "Grove guest-data transfer audit", diff --git a/tasks/cb100-069-inkwell-publishing.json b/tasks/cb100-069-inkwell-publishing.json index 950e27cb103186a7ba92569108711c4294ee3888..649c54f2c78867db83b1291215e569a23ea72d0e 100644 --- a/tasks/cb100-069-inkwell-publishing.json +++ b/tasks/cb100-069-inkwell-publishing.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Oregon Attorney General", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Inkwell subscriber request audit\n\nYou are acting as privacy and regulatory counsel for Inkwell Digital Publishing. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-PR-3009 \nOther party or authority: Blue Quill Audience Systems \nJurisdiction: Oregon \nVenue or forum: Oregon Attorney General \nDecision deadline: 2026-09-30\n\n## Assignment\n\nSubscriber access and deletion requests reveal identity-verification, advertising disclosure, and data-retention inconsistencies. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-069-inkwell-publishing\",\n \"matter_number\": \"CB-PR-3009\",\n \"prepared_for\": \"Inkwell Digital Publishing\",\n \"as_of\": \"2026-09-30\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Inkwell subscriber request audit\n\nYou are acting as privacy and regulatory counsel for Inkwell Digital Publishing. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-PR-3009\nOther party or authority: Blue Quill Audience Systems\nJurisdiction: Oregon\nVenue or forum: Oregon Attorney General\nDecision deadline: 2026-09-30\n\n## Assignment\n\nSubscriber access and deletion requests reveal identity-verification, advertising disclosure, and data-retention inconsistencies. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- processing-purpose gap\n- notice-practice mismatch\n- consent-record deficiency\n- processor-contract omission\n- subprocessor inventory mismatch\n- cross-border transfer gap\n- retention schedule conflict\n- deletion-job failure\n- access-request deadline\n- identity-verification inconsistency\n- incident-notification clock\n- risk-assessment omission\n- sensitive-data classification\n- minor-consent issue\n- training completion gap\n- regulator commitment overdue\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_data_inventory/001_data_inventory_control_register.md`\n- `/workspace/documents/01_data_inventory/004_data_inventory_ledger_export.csv`\n- `/workspace/documents/01_data_inventory/007_data_inventory_officer_certificate.html`\n- `/workspace/documents/02_processing_records/010_processing_records_executed_instrument.txt`\n- `/workspace/documents/02_processing_records/013_processing_records_review_memorandum.json`\n- `/workspace/documents/02_processing_records/016_processing_records_status_report.md`\n- `/workspace/documents/03_vendor_reviews/019_vendor_reviews_correspondence.eml`\n- `/workspace/documents/03_vendor_reviews/022_vendor_reviews_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-069-inkwell-publishing\",\n \"matter_number\": \"CB-PR-3009\",\n \"prepared_for\": \"Inkwell Digital Publishing\",\n \"as_of\": \"2026-09-30\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-069-inkwell-publishing", "task_name": "Inkwell subscriber request audit", diff --git a/tasks/cb100-071-alpine-delivery.json b/tasks/cb100-071-alpine-delivery.json index 845a84ce4e62d1d993325c926c9c056641211b3b..beb681ca0a531c24e5d6fd8f824e8551e1af11d8 100644 --- a/tasks/cb100-071-alpine-delivery.json +++ b/tasks/cb100-071-alpine-delivery.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Colorado Department of Labor and Employment", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Alpine courier classification audit\n\nYou are acting as employment counsel for Alpine Same-Day Delivery. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-EL-3101 \nOther party or authority: Mountain West Couriers Association \nJurisdiction: Colorado \nVenue or forum: Colorado Department of Labor and Employment \nDecision deadline: 2026-09-05\n\n## Assignment\n\nCourier contracts describe independence while routing, discipline, time, and equipment records show centralized control. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-071-alpine-delivery\",\n \"matter_number\": \"CB-EL-3101\",\n \"prepared_for\": \"Alpine Same-Day Delivery\",\n \"as_of\": \"2026-09-05\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Alpine courier classification audit\n\nYou are acting as employment counsel for Alpine Same-Day Delivery. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-EL-3101\nOther party or authority: Mountain West Couriers Association\nJurisdiction: Colorado\nVenue or forum: Colorado Department of Labor and Employment\nDecision deadline: 2026-09-05\n\n## Assignment\n\nCourier contracts describe independence while routing, discipline, time, and equipment records show centralized control. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-071-alpine-delivery\",\n \"matter_number\": \"CB-EL-3101\",\n \"prepared_for\": \"Alpine Same-Day Delivery\",\n \"as_of\": \"2026-09-05\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-071-alpine-delivery", "task_name": "Alpine courier classification audit", diff --git a/tasks/cb100-072-bluebird-care.json b/tasks/cb100-072-bluebird-care.json index 7d6e7eb7193f18d3d17000a673bd82c9523281fe..3ea2ed431962f153e53007d057efefab561cc308 100644 --- a/tasks/cb100-072-bluebird-care.json +++ b/tasks/cb100-072-bluebird-care.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "California Labor Commissioner", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Bluebird caregiver wage review\n\nYou are acting as employment counsel for Bluebird Home Care Services. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-EL-3102 \nOther party or authority: Regional caregiver group \nJurisdiction: California \nVenue or forum: California Labor Commissioner \nDecision deadline: 2026-09-09\n\n## Assignment\n\nPayroll, visit logs, and policies conflict on travel time, meal premiums, and complaint escalation for mobile caregivers. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-072-bluebird-care\",\n \"matter_number\": \"CB-EL-3102\",\n \"prepared_for\": \"Bluebird Home Care Services\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Bluebird caregiver wage review\n\nYou are acting as employment counsel for Bluebird Home Care Services. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-EL-3102\nOther party or authority: Regional caregiver group\nJurisdiction: California\nVenue or forum: California Labor Commissioner\nDecision deadline: 2026-09-09\n\n## Assignment\n\nPayroll, visit logs, and policies conflict on travel time, meal premiums, and complaint escalation for mobile caregivers. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-072-bluebird-care\",\n \"matter_number\": \"CB-EL-3102\",\n \"prepared_for\": \"Bluebird Home Care Services\",\n \"as_of\": \"2026-09-09\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-072-bluebird-care", "task_name": "Bluebird caregiver wage review", diff --git a/tasks/cb100-073-copper-state.json b/tasks/cb100-073-copper-state.json index 5f0a53e0eefc4a628dafe3fe3454d6f489ad3462..f17f384e9c930358cd4867f51b6cd4799cba3021 100644 --- a/tasks/cb100-073-copper-state.json +++ b/tasks/cb100-073-copper-state.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Equal Employment Opportunity Commission", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Copper State accommodation audit\n\nYou are acting as employment counsel for Copper State Manufacturing. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-EL-3103 \nOther party or authority: Former quality engineer \nJurisdiction: Arizona \nVenue or forum: Equal Employment Opportunity Commission \nDecision deadline: 2026-09-12\n\n## Assignment\n\nLeave and accommodation records show delayed follow-up, differing restrictions, comparator discipline, and close termination timing. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-073-copper-state\",\n \"matter_number\": \"CB-EL-3103\",\n \"prepared_for\": \"Copper State Manufacturing\",\n \"as_of\": \"2026-09-12\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Copper State accommodation audit\n\nYou are acting as employment counsel for Copper State Manufacturing. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-EL-3103\nOther party or authority: Former quality engineer\nJurisdiction: Arizona\nVenue or forum: Equal Employment Opportunity Commission\nDecision deadline: 2026-09-12\n\n## Assignment\n\nLeave and accommodation records show delayed follow-up, differing restrictions, comparator discipline, and close termination timing. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-073-copper-state\",\n \"matter_number\": \"CB-EL-3103\",\n \"prepared_for\": \"Copper State Manufacturing\",\n \"as_of\": \"2026-09-12\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-073-copper-state", "task_name": "Copper State accommodation audit", diff --git a/tasks/cb100-074-daymark-bank.json b/tasks/cb100-074-daymark-bank.json index e854922ba00b52faf2097add7ccf7fcabda40b1f..61a8449abb39a45d877b276f97d5be8cac1ff995 100644 --- a/tasks/cb100-074-daymark-bank.json +++ b/tasks/cb100-074-daymark-bank.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "United States Department of Labor", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Daymark pay-equity privileged audit\n\nYou are acting as employment counsel for Daymark Community Bank. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-EL-3104 \nOther party or authority: Compensation review group \nJurisdiction: North Carolina \nVenue or forum: United States Department of Labor \nDecision deadline: 2026-09-16\n\n## Assignment\n\nJob architecture, bonuses, performance ratings, and manager messages reveal unexplained compensation outliers in two regions. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-074-daymark-bank\",\n \"matter_number\": \"CB-EL-3104\",\n \"prepared_for\": \"Daymark Community Bank\",\n \"as_of\": \"2026-09-16\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Daymark pay-equity privileged audit\n\nYou are acting as employment counsel for Daymark Community Bank. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-EL-3104\nOther party or authority: Compensation review group\nJurisdiction: North Carolina\nVenue or forum: United States Department of Labor\nDecision deadline: 2026-09-16\n\n## Assignment\n\nJob architecture, bonuses, performance ratings, and manager messages reveal unexplained compensation outliers in two regions. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-074-daymark-bank\",\n \"matter_number\": \"CB-EL-3104\",\n \"prepared_for\": \"Daymark Community Bank\",\n \"as_of\": \"2026-09-16\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-074-daymark-bank", "task_name": "Daymark pay-equity privileged audit", diff --git a/tasks/cb100-075-elmwood-retail.json b/tasks/cb100-075-elmwood-retail.json index fd68b56f8cd7e52349eb756f68189db168551b40..10a0ef07aa3d9a066172b0e079e48246c8b7fd91 100644 --- a/tasks/cb100-075-elmwood-retail.json +++ b/tasks/cb100-075-elmwood-retail.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Illinois Department of Labor", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Elmwood scheduling complaint review\n\nYou are acting as employment counsel for Elmwood Specialty Retail. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-EL-3105 \nOther party or authority: Store manager collective \nJurisdiction: Illinois \nVenue or forum: Illinois Department of Labor \nDecision deadline: 2026-09-19\n\n## Assignment\n\nSchedule-change records, time punches, policy acknowledgments, and discipline files conflict after a protected complaint. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-075-elmwood-retail\",\n \"matter_number\": \"CB-EL-3105\",\n \"prepared_for\": \"Elmwood Specialty Retail\",\n \"as_of\": \"2026-09-19\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Elmwood scheduling complaint review\n\nYou are acting as employment counsel for Elmwood Specialty Retail. Outside counsel inherited this file after turnover on the client team and must validate every operative record before advising.\n\nMatter: CB-EL-3105\nOther party or authority: Store manager collective\nJurisdiction: Illinois\nVenue or forum: Illinois Department of Labor\nDecision deadline: 2026-09-19\n\n## Assignment\n\nSchedule-change records, time punches, policy acknowledgments, and discipline files conflict after a protected complaint. Evaluate both legal consequence and the practical owner needed to cure each exception.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-075-elmwood-retail\",\n \"matter_number\": \"CB-EL-3105\",\n \"prepared_for\": \"Elmwood Specialty Retail\",\n \"as_of\": \"2026-09-19\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-075-elmwood-retail", "task_name": "Elmwood scheduling complaint review", diff --git a/tasks/cb100-076-frontier-games.json b/tasks/cb100-076-frontier-games.json index eb7f213dc846805a01d118629ae4e65d48756dfb..7c0b9fe251b254d930b74a1ae7eeae451a4cb596 100644 --- a/tasks/cb100-076-frontier-games.json +++ b/tasks/cb100-076-frontier-games.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Washington Department of Labor & Industries", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Frontier contractor and IP audit\n\nYou are acting as employment counsel for Frontier Game Studios. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-EL-3106 \nOther party or authority: Distributed art contributor group \nJurisdiction: Washington \nVenue or forum: Washington Department of Labor & Industries \nDecision deadline: 2026-09-23\n\n## Assignment\n\nRemote artists signed varying agreements while access, supervision, invoice, and source-file records support different classifications. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-076-frontier-games\",\n \"matter_number\": \"CB-EL-3106\",\n \"prepared_for\": \"Frontier Game Studios\",\n \"as_of\": \"2026-09-23\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Frontier contractor and IP audit\n\nYou are acting as employment counsel for Frontier Game Studios. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-EL-3106\nOther party or authority: Distributed art contributor group\nJurisdiction: Washington\nVenue or forum: Washington Department of Labor & Industries\nDecision deadline: 2026-09-23\n\n## Assignment\n\nRemote artists signed varying agreements while access, supervision, invoice, and source-file records support different classifications. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-076-frontier-games\",\n \"matter_number\": \"CB-EL-3106\",\n \"prepared_for\": \"Frontier Game Studios\",\n \"as_of\": \"2026-09-23\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-076-frontier-games", "task_name": "Frontier contractor and IP audit", diff --git a/tasks/cb100-077-goldleaf-hotels.json b/tasks/cb100-077-goldleaf-hotels.json index aa457de67012c929712ffe0db3bda94c7493b7e8..7f8916e5e7fc424f7ece64e07f9802deb3d0e04d 100644 --- a/tasks/cb100-077-goldleaf-hotels.json +++ b/tasks/cb100-077-goldleaf-hotels.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Nevada Equal Rights Commission", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Goldleaf harassment response audit\n\nYou are acting as employment counsel for Goldleaf Hotels and Resorts. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-EL-3107 \nOther party or authority: Former banquet supervisor \nJurisdiction: Nevada \nVenue or forum: Nevada Equal Rights Commission \nDecision deadline: 2026-09-26\n\n## Assignment\n\nA complaint moved among HR and operations while witness selection, schedule changes, training, and separation records diverged. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-077-goldleaf-hotels\",\n \"matter_number\": \"CB-EL-3107\",\n \"prepared_for\": \"Goldleaf Hotels and Resorts\",\n \"as_of\": \"2026-09-26\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Goldleaf harassment response audit\n\nYou are acting as employment counsel for Goldleaf Hotels and Resorts. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-EL-3107\nOther party or authority: Former banquet supervisor\nJurisdiction: Nevada\nVenue or forum: Nevada Equal Rights Commission\nDecision deadline: 2026-09-26\n\n## Assignment\n\nA complaint moved among HR and operations while witness selection, schedule changes, training, and separation records diverged. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-077-goldleaf-hotels\",\n \"matter_number\": \"CB-EL-3107\",\n \"prepared_for\": \"Goldleaf Hotels and Resorts\",\n \"as_of\": \"2026-09-26\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-077-goldleaf-hotels", "task_name": "Goldleaf harassment response audit", diff --git a/tasks/cb100-079-island-biologics.json b/tasks/cb100-079-island-biologics.json index f75538e7503fba2d2492b69b50c2dd2dc2e934d1..63be7d2d8a127e905fd2beb79f9a6699865e25f5 100644 --- a/tasks/cb100-079-island-biologics.json +++ b/tasks/cb100-079-island-biologics.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Massachusetts Commission Against Discrimination", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Island Biologics separation audit\n\nYou are acting as employment counsel for Island Biologics Corp.. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-EL-3109 \nOther party or authority: Former research director \nJurisdiction: Massachusetts \nVenue or forum: Massachusetts Commission Against Discrimination \nDecision deadline: 2026-10-03\n\n## Assignment\n\nA research director's complaint, comparator discipline, incentive payment, restrictive covenant, and separation documents require reconciliation. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-079-island-biologics\",\n \"matter_number\": \"CB-EL-3109\",\n \"prepared_for\": \"Island Biologics Corp.\",\n \"as_of\": \"2026-10-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Island Biologics separation audit\n\nYou are acting as employment counsel for Island Biologics Corp.. This matter will be reviewed by legal, finance, and compliance stakeholders who need a single source-grounded output.\n\nMatter: CB-EL-3109\nOther party or authority: Former research director\nJurisdiction: Massachusetts\nVenue or forum: Massachusetts Commission Against Discrimination\nDecision deadline: 2026-10-03\n\n## Assignment\n\nA research director's complaint, comparator discipline, incentive payment, restrictive covenant, and separation documents require reconciliation. Use the production as of the stated deadline; do not assume facts outside the supplied files.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-079-island-biologics\",\n \"matter_number\": \"CB-EL-3109\",\n \"prepared_for\": \"Island Biologics Corp.\",\n \"as_of\": \"2026-10-03\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-079-island-biologics", "task_name": "Island Biologics separation audit", diff --git a/tasks/cb100-080-junction-foods.json b/tasks/cb100-080-junction-foods.json index 49e8c25caac65cd548aab90aa8631c7282c3f263..a0bb8d8c098b206338a4b613ccccedb478beee3e 100644 --- a/tasks/cb100-080-junction-foods.json +++ b/tasks/cb100-080-junction-foods.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Iowa Civil Rights Commission", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Junction Foods agency response check\n\nYou are acting as employment counsel for Junction Prepared Foods. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-EL-3110 \nOther party or authority: Packaging-line employee group \nJurisdiction: Iowa \nVenue or forum: Iowa Civil Rights Commission \nDecision deadline: 2026-10-07\n\n## Assignment\n\nShift records, accommodations, training, payroll corrections, and agency commitments contain conflicting dates and owners. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-080-junction-foods\",\n \"matter_number\": \"CB-EL-3110\",\n \"prepared_for\": \"Junction Prepared Foods\",\n \"as_of\": \"2026-10-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Junction Foods agency response check\n\nYou are acting as employment counsel for Junction Prepared Foods. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-EL-3110\nOther party or authority: Packaging-line employee group\nJurisdiction: Iowa\nVenue or forum: Iowa Civil Rights Commission\nDecision deadline: 2026-10-07\n\n## Assignment\n\nShift records, accommodations, training, payroll corrections, and agency commitments contain conflicting dates and owners. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- classification inconsistency\n- overtime calculation error\n- meal-period premium gap\n- pay-equity outlier\n- leave designation delay\n- interactive-process lapse\n- complaint escalation gap\n- investigation independence concern\n- discipline comparator inconsistency\n- retaliation timing\n- separation pay discrepancy\n- restrictive-covenant conflict\n- contractor control evidence\n- training completion gap\n- personnel-file access deadline\n- agency response commitment\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_handbooks_policies/001_handbooks_policies_control_register.md`\n- `/workspace/documents/01_handbooks_policies/004_handbooks_policies_ledger_export.csv`\n- `/workspace/documents/01_handbooks_policies/007_handbooks_policies_officer_certificate.html`\n- `/workspace/documents/02_personnel_records/010_personnel_records_executed_instrument.txt`\n- `/workspace/documents/02_personnel_records/013_personnel_records_review_memorandum.json`\n- `/workspace/documents/02_personnel_records/016_personnel_records_status_report.md`\n- `/workspace/documents/03_payroll_time/019_payroll_time_correspondence.eml`\n- `/workspace/documents/03_payroll_time/022_payroll_time_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-080-junction-foods\",\n \"matter_number\": \"CB-EL-3110\",\n \"prepared_for\": \"Junction Prepared Foods\",\n \"as_of\": \"2026-10-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-080-junction-foods", "task_name": "Junction Foods agency response check", diff --git a/tasks/cb100-081-aperture-robotics.json b/tasks/cb100-081-aperture-robotics.json index 6df2e4ecee5f5d0e907bbf788ce43420d7eff423..943de0b685b6a31bb09cd6cae81f637440d979cb 100644 --- a/tasks/cb100-081-aperture-robotics.json +++ b/tasks/cb100-081-aperture-robotics.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "United States Patent and Trademark Office", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Aperture robotics chain-of-title review\n\nYou are acting as intellectual-property counsel for Aperture Robotics, Inc.. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-IP-3201 \nOther party or authority: Founding engineer group \nJurisdiction: Delaware \nVenue or forum: United States Patent and Trademark Office \nDecision deadline: 2026-09-04\n\n## Assignment\n\nRobotics patents and source repositories include founder, contractor, university, and lender interests that are not consistently recorded. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-081-aperture-robotics\",\n \"matter_number\": \"CB-IP-3201\",\n \"prepared_for\": \"Aperture Robotics, Inc.\",\n \"as_of\": \"2026-09-04\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Aperture robotics chain-of-title review\n\nYou are acting as intellectual-property counsel for Aperture Robotics, Inc.. The partner responsible for this matter needs a defensible decision record before the next client call.\n\nMatter: CB-IP-3201\nOther party or authority: Founding engineer group\nJurisdiction: Delaware\nVenue or forum: United States Patent and Trademark Office\nDecision deadline: 2026-09-04\n\n## Assignment\n\nRobotics patents and source repositories include founder, contractor, university, and lender interests that are not consistently recorded. Prioritize conflicts between signed instruments and later operational records.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-081-aperture-robotics\",\n \"matter_number\": \"CB-IP-3201\",\n \"prepared_for\": \"Aperture Robotics, Inc.\",\n \"as_of\": \"2026-09-04\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-081-aperture-robotics", "task_name": "Aperture robotics chain-of-title review", diff --git a/tasks/cb100-082-blue-fern.json b/tasks/cb100-082-blue-fern.json index 534ce574749d6ae8319e96e1de955e53b1db6694..5eb6bce7f924c13d7a4ee4803962c68104d4676d 100644 --- a/tasks/cb100-082-blue-fern.json +++ b/tasks/cb100-082-blue-fern.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Trademark Trial and Appeal Board", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Blue Fern brand portfolio audit\n\nYou are acting as intellectual-property counsel for Blue Fern Beverages. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-IP-3202 \nOther party or authority: Fern & Field Cooperative \nJurisdiction: California \nVenue or forum: Trademark Trial and Appeal Board \nDecision deadline: 2026-09-08\n\n## Assignment\n\nA beverage rebrand faces specimen, coexistence, domain, and distributor-license records with conflicting territorial limits. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-082-blue-fern\",\n \"matter_number\": \"CB-IP-3202\",\n \"prepared_for\": \"Blue Fern Beverages\",\n \"as_of\": \"2026-09-08\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Blue Fern brand portfolio audit\n\nYou are acting as intellectual-property counsel for Blue Fern Beverages. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-IP-3202\nOther party or authority: Fern & Field Cooperative\nJurisdiction: California\nVenue or forum: Trademark Trial and Appeal Board\nDecision deadline: 2026-09-08\n\n## Assignment\n\nA beverage rebrand faces specimen, coexistence, domain, and distributor-license records with conflicting territorial limits. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-082-blue-fern\",\n \"matter_number\": \"CB-IP-3202\",\n \"prepared_for\": \"Blue Fern Beverages\",\n \"as_of\": \"2026-09-08\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-082-blue-fern", "task_name": "Blue Fern brand portfolio audit", diff --git a/tasks/cb100-083-crux-mapping.json b/tasks/cb100-083-crux-mapping.json index ceff4f04d139a4310d8086b440e8c33297d6c781..6c9c90d85407b38b30b266a563ddc3b9d68475ea 100644 --- a/tasks/cb100-083-crux-mapping.json +++ b/tasks/cb100-083-crux-mapping.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Eastern District of Virginia", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Crux geospatial license review\n\nYou are acting as intellectual-property counsel for Crux Mapping Systems. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-IP-3203 \nOther party or authority: National Terrain Archive \nJurisdiction: Virginia \nVenue or forum: Eastern District of Virginia \nDecision deadline: 2026-09-11\n\n## Assignment\n\nMapping products combine government, university, and open-source inputs subject to attribution, field, and distribution restrictions. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-083-crux-mapping\",\n \"matter_number\": \"CB-IP-3203\",\n \"prepared_for\": \"Crux Mapping Systems\",\n \"as_of\": \"2026-09-11\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Crux geospatial license review\n\nYou are acting as intellectual-property counsel for Crux Mapping Systems. The client has asked for a board-ready risk assessment supported by a source-level tracker rather than a narrative-only summary.\n\nMatter: CB-IP-3203\nOther party or authority: National Terrain Archive\nJurisdiction: Virginia\nVenue or forum: Eastern District of Virginia\nDecision deadline: 2026-09-11\n\n## Assignment\n\nMapping products combine government, university, and open-source inputs subject to attribution, field, and distribution restrictions. Treat dates, notice mechanics, approval evidence, and numerical reconciliations as independent controls.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-083-crux-mapping\",\n \"matter_number\": \"CB-IP-3203\",\n \"prepared_for\": \"Crux Mapping Systems\",\n \"as_of\": \"2026-09-11\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-083-crux-mapping", "task_name": "Crux geospatial license review", diff --git a/tasks/cb100-084-drift-audio.json b/tasks/cb100-084-drift-audio.json index 5da1241f01fc525c6f4871871c24740b5039b64d..55faabb2ce45fce3e7423fa1df56341a57c6c757 100644 --- a/tasks/cb100-084-drift-audio.json +++ b/tasks/cb100-084-drift-audio.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Middle District of Tennessee", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Drift Audio catalog ownership audit\n\nYou are acting as intellectual-property counsel for Drift Audio Labs. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-IP-3204 \nOther party or authority: Silver Reed Composers LLC \nJurisdiction: Tennessee \nVenue or forum: Middle District of Tennessee \nDecision deadline: 2026-09-15\n\n## Assignment\n\nAn audio-model training catalog includes commissioned recordings, performer releases, sample licenses, and takedown correspondence. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-084-drift-audio\",\n \"matter_number\": \"CB-IP-3204\",\n \"prepared_for\": \"Drift Audio Labs\",\n \"as_of\": \"2026-09-15\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Drift Audio catalog ownership audit\n\nYou are acting as intellectual-property counsel for Drift Audio Labs. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-IP-3204\nOther party or authority: Silver Reed Composers LLC\nJurisdiction: Tennessee\nVenue or forum: Middle District of Tennessee\nDecision deadline: 2026-09-15\n\n## Assignment\n\nAn audio-model training catalog includes commissioned recordings, performer releases, sample licenses, and takedown correspondence. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-084-drift-audio\",\n \"matter_number\": \"CB-IP-3204\",\n \"prepared_for\": \"Drift Audio Labs\",\n \"as_of\": \"2026-09-15\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-084-drift-audio", "task_name": "Drift Audio catalog ownership audit", diff --git a/tasks/cb100-086-frostline-security.json b/tasks/cb100-086-frostline-security.json index 4e9a6ef1d5f5d2f6a8de35d26a13299ae38d1479..bbfb499b86ca432ac738954ad2720e8568f2c8c6 100644 --- a/tasks/cb100-086-frostline-security.json +++ b/tasks/cb100-086-frostline-security.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "District of Massachusetts", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Frostline software provenance audit\n\nYou are acting as intellectual-property counsel for Frostline Security Software. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-IP-3206 \nOther party or authority: Open Harbor Foundation \nJurisdiction: Massachusetts \nVenue or forum: District of Massachusetts \nDecision deadline: 2026-09-22\n\n## Assignment\n\nA security appliance contains open-source packages, contractor modules, outbound licenses, and incomplete notice files. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-086-frostline-security\",\n \"matter_number\": \"CB-IP-3206\",\n \"prepared_for\": \"Frostline Security Software\",\n \"as_of\": \"2026-09-22\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Frostline software provenance audit\n\nYou are acting as intellectual-property counsel for Frostline Security Software. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-IP-3206\nOther party or authority: Open Harbor Foundation\nJurisdiction: Massachusetts\nVenue or forum: District of Massachusetts\nDecision deadline: 2026-09-22\n\n## Assignment\n\nA security appliance contains open-source packages, contractor modules, outbound licenses, and incomplete notice files. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-086-frostline-security\",\n \"matter_number\": \"CB-IP-3206\",\n \"prepared_for\": \"Frostline Security Software\",\n \"as_of\": \"2026-09-22\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-086-frostline-security", "task_name": "Frostline software provenance audit", diff --git a/tasks/cb100-087-goldfinch-media.json b/tasks/cb100-087-goldfinch-media.json index 2565ff757bee0fb79a57f2758a4e1f1d398fde08..ac9a56f291aa2facc2762f07b8933b55aa2f01f5 100644 --- a/tasks/cb100-087-goldfinch-media.json +++ b/tasks/cb100-087-goldfinch-media.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Southern District of New York", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Goldfinch content platform IP review\n\nYou are acting as intellectual-property counsel for Goldfinch Media Network. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-IP-3207 \nOther party or authority: Independent creator collective \nJurisdiction: New York \nVenue or forum: Southern District of New York \nDecision deadline: 2026-09-25\n\n## Assignment\n\nA creator platform has mismatched work-for-hire, moderation-tool, domain, and copyright registration records. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-087-goldfinch-media\",\n \"matter_number\": \"CB-IP-3207\",\n \"prepared_for\": \"Goldfinch Media Network\",\n \"as_of\": \"2026-09-25\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Goldfinch content platform IP review\n\nYou are acting as intellectual-property counsel for Goldfinch Media Network. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-IP-3207\nOther party or authority: Independent creator collective\nJurisdiction: New York\nVenue or forum: Southern District of New York\nDecision deadline: 2026-09-25\n\n## Assignment\n\nA creator platform has mismatched work-for-hire, moderation-tool, domain, and copyright registration records. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-087-goldfinch-media\",\n \"matter_number\": \"CB-IP-3207\",\n \"prepared_for\": \"Goldfinch Media Network\",\n \"as_of\": \"2026-09-25\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-087-goldfinch-media", "task_name": "Goldfinch content platform IP review", diff --git a/tasks/cb100-088-hinterland-agri.json b/tasks/cb100-088-hinterland-agri.json index 4ba0db315a7eb2b9828fda4bf9dc6d7c080bf32d..1b6a695ebd554cd1a35becdfb136de303163b059 100644 --- a/tasks/cb100-088-hinterland-agri.json +++ b/tasks/cb100-088-hinterland-agri.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Southern District of Iowa", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Hinterland seed technology title audit\n\nYou are acting as intellectual-property counsel for Hinterland Agricultural Science. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-IP-3208 \nOther party or authority: Prairie State University \nJurisdiction: Iowa \nVenue or forum: Southern District of Iowa \nDecision deadline: 2026-09-29\n\n## Assignment\n\nSeed-trait patents, research materials, grant obligations, field restrictions, and recorded security interests do not align. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-088-hinterland-agri\",\n \"matter_number\": \"CB-IP-3208\",\n \"prepared_for\": \"Hinterland Agricultural Science\",\n \"as_of\": \"2026-09-29\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Hinterland seed technology title audit\n\nYou are acting as intellectual-property counsel for Hinterland Agricultural Science. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-IP-3208\nOther party or authority: Prairie State University\nJurisdiction: Iowa\nVenue or forum: Southern District of Iowa\nDecision deadline: 2026-09-29\n\n## Assignment\n\nSeed-trait patents, research materials, grant obligations, field restrictions, and recorded security interests do not align. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-088-hinterland-agri\",\n \"matter_number\": \"CB-IP-3208\",\n \"prepared_for\": \"Hinterland Agricultural Science\",\n \"as_of\": \"2026-09-29\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-088-hinterland-agri", "task_name": "Hinterland seed technology title audit", diff --git a/tasks/cb100-090-jagged-cloud.json b/tasks/cb100-090-jagged-cloud.json index d8851611ebc90792408f9e9969db726d49cb3bf1..075b90896a1bc62bb9cf7744018fe4515c30c021 100644 --- a/tasks/cb100-090-jagged-cloud.json +++ b/tasks/cb100-090-jagged-cloud.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Western District of Washington", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Jagged Cloud acquisition IP audit\n\nYou are acting as intellectual-property counsel for Jagged Cloud Infrastructure. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-IP-3210 \nOther party or authority: Nimbus Fork Technologies \nJurisdiction: Washington \nVenue or forum: Western District of Washington \nDecision deadline: 2026-10-06\n\n## Assignment\n\nA cloud acquisition includes forked code, contributor agreements, patent assertions, registrant mismatches, and lender release questions. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-090-jagged-cloud\",\n \"matter_number\": \"CB-IP-3210\",\n \"prepared_for\": \"Jagged Cloud Infrastructure\",\n \"as_of\": \"2026-10-06\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Jagged Cloud acquisition IP audit\n\nYou are acting as intellectual-property counsel for Jagged Cloud Infrastructure. The record contains amendments, exports, and correspondence from different custodians; reconcile them into an actionable counsel work product.\n\nMatter: CB-IP-3210\nOther party or authority: Nimbus Fork Technologies\nJurisdiction: Washington\nVenue or forum: Western District of Washington\nDecision deadline: 2026-10-06\n\n## Assignment\n\nA cloud acquisition includes forked code, contributor agreements, patent assertions, registrant mismatches, and lender release questions. Write for a supervising lawyer who will spot-check source paths and exact values.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- inventor assignment gap\n- recorded-owner mismatch\n- maintenance-fee deadline\n- foreign-filing restriction\n- trademark specimen issue\n- coexistence restriction\n- work-for-hire defect\n- source-code ownership gap\n- inbound-license field restriction\n- outbound sublicense overreach\n- copyleft distribution trigger\n- notice-file omission\n- domain registrant mismatch\n- repository access anomaly\n- cease-and-desist deadline\n- security-interest release gap\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_patents/001_patents_control_register.md`\n- `/workspace/documents/01_patents/004_patents_ledger_export.csv`\n- `/workspace/documents/01_patents/007_patents_officer_certificate.html`\n- `/workspace/documents/02_trademarks/010_trademarks_executed_instrument.txt`\n- `/workspace/documents/02_trademarks/013_trademarks_review_memorandum.json`\n- `/workspace/documents/02_trademarks/016_trademarks_status_report.md`\n- `/workspace/documents/03_copyrights/019_copyrights_correspondence.eml`\n- `/workspace/documents/03_copyrights/022_copyrights_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-090-jagged-cloud\",\n \"matter_number\": \"CB-IP-3210\",\n \"prepared_for\": \"Jagged Cloud Infrastructure\",\n \"as_of\": \"2026-10-06\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-090-jagged-cloud", "task_name": "Jagged Cloud acquisition IP audit", diff --git a/tasks/cb100-092-brightstar-medical.json b/tasks/cb100-092-brightstar-medical.json index 6b761194a948a18ca17ce3ddb9e76fca8fc858d5..30f86a117b5195795cd1c62c32e2f48e5325c94e 100644 --- a/tasks/cb100-092-brightstar-medical.json +++ b/tasks/cb100-092-brightstar-medical.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Nasdaq Stock Market", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Brightstar 8-K and governance audit\n\nYou are acting as securities and governance counsel for Brightstar Medical Systems. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-PC-3302 \nOther party or authority: Nominating and Governance Committee \nJurisdiction: Minnesota \nVenue or forum: Nasdaq Stock Market \nDecision deadline: 2026-09-07\n\n## Assignment\n\nA director departure, customer loss, equity award, and committee vacancy create overlapping disclosure and listing obligations. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-092-brightstar-medical\",\n \"matter_number\": \"CB-PC-3302\",\n \"prepared_for\": \"Brightstar Medical Systems\",\n \"as_of\": \"2026-09-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Brightstar 8-K and governance audit\n\nYou are acting as securities and governance counsel for Brightstar Medical Systems. A late document production changed the factual picture, and the matter team needs the entire record reconciled from first principles.\n\nMatter: CB-PC-3302\nOther party or authority: Nominating and Governance Committee\nJurisdiction: Minnesota\nVenue or forum: Nasdaq Stock Market\nDecision deadline: 2026-09-07\n\n## Assignment\n\nA director departure, customer loss, equity award, and committee vacancy create overlapping disclosure and listing obligations. Separate confirmed exceptions from items that merely require follow-up.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-092-brightstar-medical\",\n \"matter_number\": \"CB-PC-3302\",\n \"prepared_for\": \"Brightstar Medical Systems\",\n \"as_of\": \"2026-09-07\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-092-brightstar-medical", "task_name": "Brightstar 8-K and governance audit", diff --git a/tasks/cb100-094-duneworks.json b/tasks/cb100-094-duneworks.json index 692f1a2f3f6a6532b0a996e44d95b3f7e93e1a7e..91ee2f42848ad8121cbb15bd4b4ef22f6084f5e9 100644 --- a/tasks/cb100-094-duneworks.json +++ b/tasks/cb100-094-duneworks.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "New York Stock Exchange", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Duneworks earnings-controls audit\n\nYou are acting as securities and governance counsel for Duneworks Software, Inc.. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-PC-3304 \nOther party or authority: Disclosure Committee \nJurisdiction: California \nVenue or forum: New York Stock Exchange \nDecision deadline: 2026-09-14\n\n## Assignment\n\nSales adjustments, non-GAAP metrics, guidance drafts, and certification exceptions were tracked differently across finance and legal files. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-094-duneworks\",\n \"matter_number\": \"CB-PC-3304\",\n \"prepared_for\": \"Duneworks Software, Inc.\",\n \"as_of\": \"2026-09-14\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Duneworks earnings-controls audit\n\nYou are acting as securities and governance counsel for Duneworks Software, Inc.. Several workstreams have maintained separate versions of the truth; your assignment is to produce one auditable legal view.\n\nMatter: CB-PC-3304\nOther party or authority: Disclosure Committee\nJurisdiction: California\nVenue or forum: New York Stock Exchange\nDecision deadline: 2026-09-14\n\n## Assignment\n\nSales adjustments, non-GAAP metrics, guidance drafts, and certification exceptions were tracked differently across finance and legal files. Distinguish a missing document from an affirmative contradiction in the documents that are present.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-094-duneworks\",\n \"matter_number\": \"CB-PC-3304\",\n \"prepared_for\": \"Duneworks Software, Inc.\",\n \"as_of\": \"2026-09-14\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-094-duneworks", "task_name": "Duneworks earnings-controls audit", diff --git a/tasks/cb100-096-fieldstone-foods.json b/tasks/cb100-096-fieldstone-foods.json index 5eb3b670046344254db427197534f30291a522db..a67f4408cf3b01453992798b2f3a22790d8410da 100644 --- a/tasks/cb100-096-fieldstone-foods.json +++ b/tasks/cb100-096-fieldstone-foods.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Nasdaq Stock Market", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Fieldstone whistleblower escalation audit\n\nYou are acting as securities and governance counsel for Fieldstone Foods plc. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-PC-3306 \nOther party or authority: Audit Committee \nJurisdiction: Illinois \nVenue or forum: Nasdaq Stock Market \nDecision deadline: 2026-09-21\n\n## Assignment\n\nA revenue-recognition complaint reached management, the hotline vendor, and directors on different timelines before an earnings release. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-096-fieldstone-foods\",\n \"matter_number\": \"CB-PC-3306\",\n \"prepared_for\": \"Fieldstone Foods plc\",\n \"as_of\": \"2026-09-21\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Fieldstone whistleblower escalation audit\n\nYou are acting as securities and governance counsel for Fieldstone Foods plc. The upcoming deadline leaves no room for unsupported assumptions, so every conclusion must trace to the supplied production.\n\nMatter: CB-PC-3306\nOther party or authority: Audit Committee\nJurisdiction: Illinois\nVenue or forum: Nasdaq Stock Market\nDecision deadline: 2026-09-21\n\n## Assignment\n\nA revenue-recognition complaint reached management, the hotline vendor, and directors on different timelines before an earnings release. Give controlling documents precedence only when the record supports that hierarchy.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-096-fieldstone-foods\",\n \"matter_number\": \"CB-PC-3306\",\n \"prepared_for\": \"Fieldstone Foods plc\",\n \"as_of\": \"2026-09-21\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-096-fieldstone-foods", "task_name": "Fieldstone whistleblower escalation audit", diff --git a/tasks/cb100-097-granular-tech.json b/tasks/cb100-097-granular-tech.json index f877984966170aa2beedc848924e7d749bcb7c63..a9877df7a6322193b82dc71ab257e566fc922570 100644 --- a/tasks/cb100-097-granular-tech.json +++ b/tasks/cb100-097-granular-tech.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "Nasdaq Stock Market", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Granular Section 16 filing review\n\nYou are acting as securities and governance counsel for Granular Technology Corp.. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-PC-3307 \nOther party or authority: Executive officer group \nJurisdiction: Texas \nVenue or forum: Nasdaq Stock Market \nDecision deadline: 2026-09-24\n\n## Assignment\n\nEquity grants, broker confirmations, powers of attorney, and filing records reveal late or inconsistent insider reports. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-097-granular-tech\",\n \"matter_number\": \"CB-PC-3307\",\n \"prepared_for\": \"Granular Technology Corp.\",\n \"as_of\": \"2026-09-24\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Granular Section 16 filing review\n\nYou are acting as securities and governance counsel for Granular Technology Corp.. A counterparty challenged the client's internal summary, making cross-document reconciliation and precise citations essential.\n\nMatter: CB-PC-3307\nOther party or authority: Executive officer group\nJurisdiction: Texas\nVenue or forum: Nasdaq Stock Market\nDecision deadline: 2026-09-24\n\n## Assignment\n\nEquity grants, broker confirmations, powers of attorney, and filing records reveal late or inconsistent insider reports. Flag cross-folder dependencies that would be invisible in a single-document review.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-097-granular-tech\",\n \"matter_number\": \"CB-PC-3307\",\n \"prepared_for\": \"Granular Technology Corp.\",\n \"as_of\": \"2026-09-24\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-097-granular-tech", "task_name": "Granular Section 16 filing review", diff --git a/tasks/cb100-098-highwater-utilities.json b/tasks/cb100-098-highwater-utilities.json index a813c4ba3e285531fd95fb4f5a455589423c0cd6..d84dfdbc9fb12249742f3a72a552d3201c38d374 100644 --- a/tasks/cb100-098-highwater-utilities.json +++ b/tasks/cb100-098-highwater-utilities.json @@ -267,25 +267,31 @@ "reference_tool_calls": 109, "synthetic": true, "venue": "New York Stock Exchange", - "version": "1.0.0" + "version": "1.1.0" }, - "prompt": "# Highwater board independence audit\n\nYou are acting as securities and governance counsel for Highwater Utilities, Inc.. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-PC-3308 \nOther party or authority: Nominating Committee \nJurisdiction: Oregon \nVenue or forum: New York Stock Exchange \nDecision deadline: 2026-09-28\n\n## Assignment\n\nConsulting payments, family ties, committee appointments, and questionnaires affect director independence and board-matrix disclosures. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-098-highwater-utilities\",\n \"matter_number\": \"CB-PC-3308\",\n \"prepared_for\": \"Highwater Utilities, Inc.\",\n \"as_of\": \"2026-09-28\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", + "prompt": "# Highwater board independence audit\n\nYou are acting as securities and governance counsel for Highwater Utilities, Inc.. The supervising attorney wants an exception-focused review that still demonstrates complete coverage of the data room.\n\nMatter: CB-PC-3308\nOther party or authority: Nominating Committee\nJurisdiction: Oregon\nVenue or forum: New York Stock Exchange\nDecision deadline: 2026-09-28\n\n## Assignment\n\nConsulting payments, family ties, committee appointments, and questionnaires affect director independence and board-matrix disclosures. Preserve the difference between the client's position, the counterparty's position, and an objective record conflict.\n\nReconcile the complete seeded production and prepare an exception tracker plus a concise counsel memo. The data room contains 96 text-native legal and business records in 12 folders. Some facts can be evaluated only by comparing a primary record in one folder with a later or independently maintained record elsewhere. Do not rely on filenames alone, do not assume that a later record controls, and do not use facts outside this production.\n\nFinding-bearing records identify the review question, stable finding ID, record role, severity, remediation owner, and response due date in their record-control metadata. Preserve those values exactly. Records labeled `context only` are responsive background, not independent findings.\n\nYour review scope includes these independently tested legal and control questions:\n\n- committee-independence gap\n- minutes-approval omission\n- 8-K deadline\n- risk-factor inconsistency\n- disclosure-control exception\n- trading-window violation\n- Section 16 filing delay\n- related-party disclosure gap\n- equity-plan share mismatch\n- non-GAAP reconciliation issue\n- guidance-control failure\n- whistleblower escalation delay\n- listing-standard notice\n- debt-covenant disclosure\n- officer certification exception\n- board-matrix inconsistency\n\n## Required review procedure\n\nUse the `io.github.modelcontextprotocol/server-filesystem` MCP tools exposed by the environment.\n\n1. Call `list_allowed_directories` and confirm the available roots.\n2. Call `directory_tree` on `/workspace/documents` to inventory the complete production.\n3. Call `search_files` on `/workspace/documents` with pattern `**/*.eml`.\n4. Call `read_text_file` separately, without `head` or `tail`, for every one of the 96 files. Bulk reads are intentionally unavailable because the work product must have file-level provenance.\n5. Call `get_file_info` for each of these eight chain-of-custody sample records:\n\n- `/workspace/documents/01_board_committees/001_board_committees_control_register.md`\n- `/workspace/documents/01_board_committees/004_board_committees_ledger_export.csv`\n- `/workspace/documents/01_board_committees/007_board_committees_officer_certificate.html`\n- `/workspace/documents/02_minutes_materials/010_minutes_materials_executed_instrument.txt`\n- `/workspace/documents/02_minutes_materials/013_minutes_materials_review_memorandum.json`\n- `/workspace/documents/02_minutes_materials/016_minutes_materials_status_report.md`\n- `/workspace/documents/03_sec_filings/019_sec_filings_correspondence.eml`\n- `/workspace/documents/03_sec_filings/022_sec_filings_formal_notice.xml`\n\n6. Use `write_file` to create both final deliverables under `/workspace/output`.\n\nAn accepted trajectory therefore contains at least 109 successful MCP tool calls. The verifier checks the call trace and final filesystem state; merely stating that records were reviewed is insufficient.\n\n## Deliverables\n\nCreate exactly these two files and no others:\n\n- `/workspace/output/findings.json`\n- `/workspace/output/advice.md`\n\n`findings.json` must be valid JSON with this exact top-level shape:\n\n```json\n{\n \"schema_version\": \"1.0\",\n \"task_id\": \"cb100-098-highwater-utilities\",\n \"matter_number\": \"CB-PC-3308\",\n \"prepared_for\": \"Highwater Utilities, Inc.\",\n \"as_of\": \"2026-09-28\",\n \"findings\": [\n {\n \"id\": \"F-01\",\n \"severity\": \"critical|high|medium|low\",\n \"issue\": \"short issue label\",\n \"primary_source\": \"/workspace/documents/...\",\n \"corroborating_source\": \"/workspace/documents/...\",\n \"determination\": \"precise reconciliation of the conflicting facts\",\n \"recommended_action\": \"dated, owned next step\"\n }\n ]\n}\n```\n\nUse one row per confirmed exception, stable IDs beginning at `F-01`, exact absolute source paths, and exact dates, amounts, names, statuses, thresholds, or counts from the cited records.\n\n`advice.md` must contain these headings: `Executive assessment`, `Method and record coverage`, `Findings`, `Recommended next actions`, and `Assumptions and limitations`. Address every JSON finding in the memo with its ID, issue, severity, both source paths, determination, and recommended action. Keep confirmed contradictions distinct from assumptions or unresolved legal questions.\n\nAll facts in this task are synthetic. The work product is benchmark output, not legal advice.\n", "rubric": { "gates": [ "all_evidence_read_in_full", "chain_of_custody_metadata_checked", - "findings_schema_and_values_exact", - "memo_is_fully_grounded", + "findings_criteria_complete", + "memo_criteria_complete", "deliverables_written_through_mcp" ], "metadata_checks": 8, + "metric": "weighted_criteria_score", "minimum_tool_calls": 109, "required_deliverables": [ "findings.json", "advice.md" ], "required_document_reads": 96, - "type": "deterministic" + "type": "deterministic", + "weights": { + "findings": 0.55, + "memo": 0.2, + "procedure": 0.25 + } }, "task_id": "cb100-098-highwater-utilities", "task_name": "Highwater board independence audit",