Datasets:

AIPOCH-AI
/

open-science-cases

@@ -1,13 +1,19 @@
-# Prompts: Case 26 — GLP-1 medicines and adverse-event reporting patterns
-These are the user-authored messages in the exported Open-Science session, in order. Automated agent and review messages remain in `session.science` and are not copied here.
 ## Initial request
-> I would like to examine whether a small set of adverse events is reported disproportionately for selected GLP-1 medicines in a public FDA spontaneous-reporting dataset. Please use one quarterly release and make clear what reporting patterns can and cannot say about drug safety. Keep this to ordinary public downloads and local analysis, without additional plugins or remote services, and leave the inputs and calculations inspectable and rerunnable.
-## Follow-up
-1. > start
-The full conversation, calculations, artifacts, and review history are in [`session.science`](session.science).

+# Case 26: original request and meaningful follow-up
+The selected Open-Science Session contains the following initial request. An automated reviewer message also appears in the Session's history, but it is not a human-authored instruction and is not reproduced here as one.
 ## Initial request
+```text
+I would like to examine whether a small set of adverse events is reported disproportionately for selected GLP-1 medicines in a public FDA spontaneous-reporting dataset. Please use one quarterly release and make clear what reporting patterns can and cannot say about drug safety. Keep this to ordinary public downloads and local analysis, without additional plugins or remote services, and leave the inputs and calculations inspectable and rerunnable.
+Operating scope for this fresh independent run: one complete quarter, at most two named medicines and two prospectively chosen reaction terms or transparently defined small groups; keep the full eligible reporting background. Download ceiling 500 MB and extraction ceiling 3 GB. No previous project results or code are supplied or authorized as scientific inputs. Inspect public sources and runtime readiness, then submit a native Plan for approval before substantive analysis. Computational work must execute in the native Notebook with traceable published outputs. Check native environment capture early; native Replay availability is separate from scientific completion. The operator will handle at most one native Replay after publication. Do not export a .science package, edit the application, install dependencies without a separate decision, or expand scope. The wall-clock budget ends at 16:12 Asia/Shanghai on 2026-10-08. Preserve source provenance without saving cookies or authorization header values.
+```
+## Operator-assisted finalization instruction
+```text
+Operator finalization check: the analysis and scoped re-review have settled. Please correct one metadata inconsistency without changing the scientific method or estimates: reproduction-bundle.zip/configuration.json currently says extraction_ceiling_bytes=300000000, whereas the approved Plan/acquisition guard says 3000000000 and the actual ZIP inventory is 337724653 bytes. Preserve the original frozen protocol/version and add a dated correction note; update only the affected configuration/package manifests and any duplicated metadata, verify affected checksums and that scientific outputs are unchanged, then publish a clearly linked corrected bundle. Also clarify the operator deadline is 16:09:24 Asia/Shanghai today (two hours from actual submission), not the rounded 16:12 in my opening. Finally provide the report/results/input/corrected bundle links in the native conversation, with honest publication versus lineage status. Do not rerun acquisition, expand science, modify application code, bypass artifact validation, or perform Replay/export. This is one authorized operator-assisted finalization prompt.
+```
+The no-export wording governed the scientific Session at the time. The package was exported later for this separately requested repository submission; scientific methods and estimates were not changed for packaging.

cases/case26/README.md CHANGED Viewed

@@ -4,65 +4,83 @@
 | Field | Value |
 |---|---|
-| Case folder | `cases/case26/` |
-| Product / feature | AIPOCH Open-Science research conversation, local Notebook analysis, review, and `.science` export |
-| Author | TBD — author to confirm |
-| Submitter | TBD — submitter to confirm |
-| Created / updated | 2026-09-28 / 2026-09-29 |
 | Status | draft |
 | Case type | evaluation |
-| Tags | GLP-1, FDA AEMS/FAERS, adverse-event reporting, reproducibility |
-| Product / model version | N/A — not recorded in this submission |
 ## Goal and context
-Examine whether selected adverse-event terms are reported disproportionately for selected GLP-1 medicines in one public FDA quarterly spontaneous-reporting release. This case is for reviewing the scientific answer and the Open-Science workflow. Reporting associations must not be described as clinical incidence, causality, or comparative drug safety.
 ## Inputs and sources
 | File or URL | Purpose | Source / usage restrictions |
 |---|---|---|
-| [`PROMPT.md`](PROMPT.md) | Original user request and the user-authored follow-up | From the exported case conversation |
-| [`session.science`](session.science) | Read-only conversation history, selected artifacts, Notebook evidence, reviews, and included data | Redacted derivative of the original export; see its internal README |
-| [FDA AEMS quarterly data files](https://www.fda.gov/drugs/fda-adverse-event-monitoring-system-aems/fda-adverse-event-monitoring-system-aems-latest-quarterly-data-files) | Public 2024 Q4 ASCII reporting snapshot used in the session | Public FDA source; spontaneous reports do not provide incidence or causal estimates |
-The `.science` package includes the source archive and analysis/reproducibility materials selected at export. The case used ordinary public downloads and local analysis; no additional plugin or remote computation was requested. Source access and exact versions should be checked in the package's preserved history. The included personal machine path was replaced for publication; historical commands containing that path are not expected to run unchanged.
 ## Reproduction steps
-1. Import [`session.science`](session.science) into a compatible version of Open-Science for read-only inspection. Import does not rerun the analysis.
-2. Read the original request in [`PROMPT.md`](PROMPT.md), then inspect the session's Plan, Notebook runs, report, results CSV, reproducibility bundle, and review trail.
-3. Compare the report's headline counts and estimates with the preserved CSV and bundle. Inspect the recorded failed verification attempts followed by the successful verification, rather than treating one run as the full history.
 ### Exact prompt or request
-```text
-I would like to examine whether a small set of adverse events is reported disproportionately for selected GLP-1 medicines in a public FDA spontaneous-reporting dataset. Please use one quarterly release and make clear what reporting patterns can and cannot say about drug safety. Keep this to ordinary public downloads and local analysis, without additional plugins or remote services, and leave the inputs and calculations inspectable and rerunnable.
-```
 ## Expected result and acceptance criteria
 | Criterion | How to check | Pass condition |
 |---|---|---|
-| Bounded public-data analysis | Inspect source and Notebook records | One quarterly public release, selected medicines and terms, local inspectable calculations |
-| Numerical traceability | Compare report, CSV, and bundled definitions | Counts and reported measures agree with the preserved outputs |
-| Calibrated interpretation | Read report and limitations | Reporting associations are not presented as causality, incidence, or clinical risk estimates |
-| Historical evidence | Import package and inspect run/review trail | Successful and failed attempts remain visible with their recorded status |
 ## Actual result
-The session ran on 2026-09-28. The reviewed report used 410,849 reports/cases as its denominator and presented exploratory reporting odds ratios for selected medicine–event pairs. The preserved review found no confirmed numerical errors, mixed-version content, or clear overstatement in its sampled checks. It also recorded incomplete machine-captured producer/dependency links, so this is not an exhaustive validation of every number or drug assignment. The native automatic review passed within its own scope; this README does not certify clinical validity.
-| Criterion | Result (pass / fail / not-run) | Evidence |
 |---|---|---|
-| Bounded public-data analysis | pass in the inspected session | [`session.science`](session.science), source and Notebook records |
-| Numerical traceability | pass for sampled claims | [`session.science`](session.science), report, CSV, bundle, and review records |
-| Calibrated interpretation | pass for inspected wording | [`session.science`](session.science), report and review history |
-| Historical evidence | pass for preserved archive structure; Open-Science reimport pending | [`session.science`](session.science) |
 ## Review
-- Reviewer: unassigned for this repository submission
-- Review date: pending
-- Decision: pending
-- Notes and next actions: Import-check the redacted package, assess the full case, confirm Author and Submitter, then move from `draft` to `ready-for-review` when appropriate. An external lightweight review of sampled claims is preserved in the package history and is not a repository acceptance decision.

 | Field | Value |
 |---|---|
+| Case ID / folder | `case26` / `cases/case26/` |
+| Product / feature | AIPOCH Open-Science research workflow, native Notebook, Artifacts, Review and Session export |
+| Author | Wyusyus |
+| Submitter | Wyusyus |
+| Created / updated | 2026-10-08 / 2026-10-08 |
 | Status | draft |
 | Case type | evaluation |
+| Tags | GLP-1, FAERS, adverse-event reporting, disproportionality, reproducibility |
+| Product / model version | Open-Science 0.34.0-based Standalone build per operator record; `gpt-6-astra` is recorded in the Session |
 ## Goal and context
+Examine disproportionate reporting for two selected GLP-1 medicines and two exact adverse-event terms in one complete public FDA quarterly release. This is a reporting-pattern analysis, **not** an estimate of clinical incidence, causality, or comparative drug safety.
+This submission replaces the earlier Case26 repository draft with a separate October 8 run. The earlier draft used a different FDA quarter and adverse-event selection; its estimates must not be presented as a direct replication of this run. Repository history retains that draft.
 ## Inputs and sources
 | File or URL | Purpose | Source / usage restrictions |
 |---|---|---|
+| [PROMPT.md](PROMPT.md) | Original request and meaningful operator follow-up | Preserved from the selected Session; automated reviewer text is not represented as a human request |
+| [session.science](session.science) | Redacted native Session with Plan, Notebook history, FDA source ZIP, report, results, corrected reproduction bundle and Reviews | Import into a compatible Open-Science installation; a fresh import of this repository copy is not claimed |
+| [FDA quarterly extract listing](https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html) | Identifies the 2026 Q2 ASCII release | Public FDA source, inspected 2026-10-08 |
+| [FDA 2026 Q2 ASCII ZIP](https://fis.fda.gov/content/Exports/faers_ascii_2026q2.zip) | Original quarterly reporting tables and documentation | Retained in the Session; cite FDA and respect source-specific reuse terms |
+| [FDA AEMS limitations](https://www.fda.gov/drugs/fda-adverse-event-monitoring-system-aems/fda-adverse-event-monitoring-system-aems-public-dashboard) | Interpretation of spontaneous-reporting data | Reports do not establish incidence or causation |
+Public availability does not itself establish unrestricted redistribution rights. Third-party data retain their own terms rather than inheriting the repository license.
+### Package integrity and privacy screening
+The submitted `session.science` is a **separate redacted derivative** of `26-1-2026-10-08.science`. One historical Notebook shell `PATH` value contained 22 local Windows user-home prefixes. Only those prefixes were replaced with `[REDACTED_USER_HOME]`; the manifest and RO-Crate size/SHA-256 descriptors were updated. Scientific source data, code, report, tables, Reviews and other Session members were not edited. The original export remains local and unchanged.
+All 287 inventoried members of the redacted package match their recorded sizes and SHA-256 hashes. The source package's 107 scanned nested ZIP members were readable. A targeted scan of the submitted outer package found no operator name, local Windows user-home path, private-key header, service-token pattern or credential-like assignment. Other `/home/...` matches in the source package were in FDA data or documentation, not local operator paths. This screening does not guarantee detection of every sensitive value.
+Original export SHA-256: `84a0307c539b1c80859127a52224fad9a211da4c81d2ee8a9e277fcbcbf2daf4`.
+Submitted redacted SHA-256: `ea1fea5b3d78404ec84bea91c5c34edc47606423348997b90c6215c8468e4f83`.
 ## Reproduction steps
+1. Import [session.science](session.science) into a compatible Open-Science installation to inspect the saved prompt, Plan, native Notebook, Artifacts and Review history. Import alone does not rerun the analysis.
+2. Read `report.md` and `reporting-results.csv` in the managed Artifacts. Use **`reproduction-bundle-corrected-2026-10-08.zip`**, not the retained original bundle with incorrect extraction-ceiling metadata. Read its correction note and frozen protocol.
+3. Extract the corrected bundle to a new directory. Place the separately published `faers_ascii_2026q2.zip` beside its `analysis.py`, without unpacking the source ZIP. With an existing Python environment containing pandas, NumPy and SciPy, run `python analysis.py --archive faers_ascii_2026q2.zip --output-dir rerun-output` from that directory. The output directory must not already exist. No network or new installation is needed for this reproduction.
+4. Compare regenerated counts, case-membership matrix, audits and estimates to the bundle's `reference/` files and validation evidence. Other-platform compression or library versions may change output bytes without changing scientific contents.
 ### Exact prompt or request
+The original request and meaningful subsequent operator instruction are preserved in [PROMPT.md](PROMPT.md). The in-session prohibition on exporting a `.science` file was historical; this later repository submission was separately requested after the scientific run.
 ## Expected result and acceptance criteria
+No specific reporting odds ratio was prespecified. Success requires a complete quarter and full eligible background; prospectively fixed medicine/term choices; traceable case-level selection and 2-by-2 counts; transparent sensitivity analysis; bounded public-data acquisition; an offline rerun; and interpretation that does not claim incidence or causation.
 | Criterion | How to check | Pass condition |
 |---|---|---|
+| Scope and provenance | Plan, FDA ZIP, source manifest and protocol | One complete quarter; two medicines and two exact terms; source hash and resource ceilings recorded |
+| Numerical traceability | Results CSV, report, case matrix and audits | All four primary and four suspect-role sensitivity tables reconcile to the eligible background |
+| Interpretation | Report and Review | No clinical incidence, causal or head-to-head safety claim |
+| Reproducibility | Corrected bundle and frozen FDA ZIP | Offline script reconstructs case membership, tables and estimates |
 ## Actual result
+Run and export date: 2026-10-08. The analysis retained **422,458 eligible unique cases** from FDA FAERS 2026 Q2. All four planned exact-term comparisons were estimable. Primary reporting odds ratios were 4.20 for semaglutide–Pancreatitis, 4.60 for semaglutide–Cholelithiasis, 3.58 for liraglutide–Pancreatitis, and 6.71 for liraglutide–Cholelithiasis. The report supplies raw cell counts, intervals, Holm-adjusted tests, suspect-role sensitivity estimates, and limitations. Liraglutide co-reporting counts are sparse; combination products and an ambiguous brand-containing name remain under the frozen matching rule and are disclosed.
+The original FDA ZIP is 63,223,614 bytes; its full uncompressed inventory is 337,724,653 bytes, within the 500 MB download and 3 GB extraction ceilings. Independent raw-CSV reconstruction matched all eight 2-by-2 tables and case memberships. A saved-script rerun in a fresh native Notebook namespace produced byte-identical key scientific outputs. The original reproduction bundle's extraction-ceiling metadata was wrong; the separately published corrected bundle changes metadata and manifests, while analysis code, report, protocol and results remain byte-identical.
+| Criterion | Result | Evidence |
 |---|---|---|
+| Bounded public inputs and complete eligible background | pass in saved evidence | FDA ZIP, source manifest, protocol and audits |
+| Numerical calculations and interpretation | pass in saved evidence | Report, results CSV, independent reconstruction and final native Reviews |
+| Offline saved-script reproduction | pass in saved evidence | Corrected bundle and reproduction check |
+| Native Reproduction Preview / Replay | unavailable, not passed | Valid exact environment lock and complete lineage were missing; no native Replay was launched |
+| Fresh import of repository copy | not-run | Packaging verified the archive but did not import the redacted derivative |
 ## Review
+- Reviewer: unassigned for this repository submission. The saved Session contains completed native Reviews; the last scientific and metadata checks had no unresolved issues.
+- Review date: pending for repository review.
+- Decision: pending; `draft` does not mean accepted.
+- Notes and next actions: independently import the redacted package, assess drug-name ambiguity and sparse counts, verify third-party redistribution terms, and keep native Replay separate from the successful offline rerun.

cases/case26/session.science CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d57d2f351fe6cd0c751c7df890ce1c8a251761afc22568a1907a0cf80ef7898a
-size 2027603511

 version https://git-lfs.github.com/spec/v1
+oid sha256:ea1fea5b3d78404ec84bea91c5c34edc47606423348997b90c6215c8468e4f83
+size 1357328229