Publish benchmark task data

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Pooria Arab
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,112 @@
+---
+license: mit
+language:
+- en
+pretty_name: "Tool Errors"
+tags:
+- benchmark
+- llm-evaluation
+- ai-benchmark
+- model-evaluation
+- agent-benchmark
+- tool-use
+- api-errors
+- retries
+- idempotency
+- fault-injection
+- openrouter
+- openai
+- mcp
+- typescript
+task_categories:
+- text-generation
+size_categories:
+- n<1K
+configs:
+- config_name: default
+  data_files:
+  - split: test
+    path: data/tasks.jsonl
+---
+
+# Tool Errors
+
+Retry safely when APIs fail.
+
+## What it measures
+
+Safe, bounded recovery from API faults without duplicate side effects or false success reports.
+
+Category: `tools`. Status: `alpha`.
+
+## How to run
+
+```console
+$ npx benchmark-tool-errors list
+$ npx benchmark-tool-errors run --model <model-id> --provider openrouter
+$ npx benchmark-tool-errors score results.json
+```
+
+`--provider mock` runs the whole benchmark offline with no keys. Real models run
+through OpenRouter (`OPENROUTER_API_KEY`) or any OpenAI-compatible base URL
+(`OPENAI_BASE_URL`, `OPENAI_API_KEY`).
+
+## Data fields
+
+`data/tasks.jsonl` holds 40 tasks, one JSON object per line
+(kinds: 40 episode).
+
+| field | type |
+|---|---|
+| `id` | string |
+| `kind` | string |
+| `prompt` | string |
+| `expected` | object |
+| `metadata` | object |
+
+## Provenance
+
+# Tool Errors provenance
+
+This file is generated from the benchmark definition.
+
+## Seeded API fault scenarios
+
+- Kind: `synthetic`
+- URL: https://github.com/pooriaarab/benchmark-tool-errors/blob/main/scripts/gen-tasks.mjs
+- Licence: `MIT`
+- Retrieved: 2026-10-15T00:00:00.000Z
+- How to get it: Run `node scripts/gen-tasks.mjs --seed 20261015 --split public` from the repository root. Run with `--split hidden` to create the untracked hidden split.
+- Contamination notes: The 40 public tasks use seed 20261015 and are visible in the npm package. Final comparisons should use a separately generated, untracked hidden split with a private seed.
+- Refresh: Regenerate and review public fixtures for each minor release; rotate the private hidden seed for every evaluation campaign.
+
+
+## Grader
+
+Deterministic replay scores end state (45%), side-effect safety (25%), bounded and timed retries (20%), and truthful reporting (10%); duplicate side effects score zero.
+
+## Limitations
+
+This dataset contains only the public task split (40 tasks).
+Hidden evaluation splits use a private seed and are deliberately not part of
+this dataset. Source kinds: synthetic. These are evaluation prompts
+for the `benchmark-tool-errors` benchmark, not training data; scoring requires the package's
+deterministic grader. See PROVENANCE.md for contamination notes.
+
+## Citation
+
+```bibtex
+@misc{benchmark_tool_errors,
+  author = {Pooria Arab},
+  title = {Tool Errors task data},
+  year = {2026},
+  url = {https://github.com/pooriaarab/benchmark-tool-errors}
+}
+```
+
+## Links
+
+- npm: https://www.npmjs.com/package/benchmark-tool-errors
+- GitHub: https://github.com/pooriaarab/benchmark-tool-errors
+- Leaderboard: https://aibenchmarkindex.com/benchmarks/benchmark-tool-errors
+- Dataset: https://huggingface.co/datasets/pooria/benchmark-tool-errors

croissant.json

@@ -0,0 +1,209 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataBiases": "cr:dataBiases",
+    "dataCollection": "cr:dataCollection",
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": "cr:examples",
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "personalSensitiveInformation": "cr:personalSensitiveInformation",
+    "prov": "http://www.w3.org/ns/prov#",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "samplingRate": "cr:samplingRate",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform"
+  },
+  "@type": "sc:Dataset",
+  "conformsTo": [
+    "http://mlcommons.org/croissant/1.0",
+    "http://mlcommons.org/croissant/RAI/1.0"
+  ],
+  "name": "Tool Errors",
+  "description": "Retry safely when APIs fail.",
+  "url": "https://huggingface.co/datasets/pooria/benchmark-tool-errors",
+  "sameAs": [
+    "https://github.com/pooriaarab/benchmark-tool-errors",
+    "https://aibenchmarkindex.com/benchmarks/benchmark-tool-errors"
+  ],
+  "license": "https://choosealicense.com/licenses/mit/",
+  "creator": {
+    "@type": "Person",
+    "name": "Pooria Arab",
+    "url": "https://github.com/pooriaarab"
+  },
+  "datePublished": "2026-10-15",
+  "dateModified": "2026-10-15",
+  "version": "0.1.0",
+  "inLanguage": "en",
+  "keywords": [
+    "benchmark-tool-errors",
+    "llm-evaluation",
+    "benchmark"
+  ],
+  "isAccessibleForFree": true,
+  "citeAs": "@misc{benchmark-tool-errors, author={Pooria Arab}, title={Tool Errors}, year={2026}, url={https://github.com/pooriaarab/benchmark-tool-errors}}",
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "data/PROVENANCE.md",
+      "name": "data/PROVENANCE.md",
+      "contentUrl": "https://huggingface.co/datasets/pooria/benchmark-tool-errors/resolve/main/data/PROVENANCE.md",
+      "encodingFormat": "text/markdown",
+      "sha256": "b28fd32e592c3afb8d117af467f60b24a43bdf3aa083ec06f59c64d4baf3084e"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "data/tasks.jsonl",
+      "name": "data/tasks.jsonl",
+      "contentUrl": "https://huggingface.co/datasets/pooria/benchmark-tool-errors/resolve/main/data/tasks.jsonl",
+      "encodingFormat": "application/jsonlines",
+      "sha256": "130617c11814584fb645f1fe569502bc1d5d11518c5e8c0f38942dd6382915ea"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "scripts/gen-tasks.mjs",
+      "name": "scripts/gen-tasks.mjs",
+      "contentUrl": "https://huggingface.co/datasets/pooria/benchmark-tool-errors/resolve/main/scripts/gen-tasks.mjs",
+      "encodingFormat": "text/javascript",
+      "sha256": "8cf70a9176d898106a8378343b2cc36f810dae78ce38ae856ff7c3424334d257"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "LICENSE",
+      "name": "LICENSE",
+      "contentUrl": "https://huggingface.co/datasets/pooria/benchmark-tool-errors/resolve/main/LICENSE",
+      "encodingFormat": "text/plain",
+      "sha256": "db83f2db2a2b6a3b5587ce4750a3a6cff3dd1219650fa1ea3b3354ee04891d96"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "name": "tasks",
+      "description": "40 benchmark tasks, one JSON object per line.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "name": "id",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "data/tasks.jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "name": "kind",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "data/tasks.jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.kind"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "name": "prompt",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "data/tasks.jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.prompt"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "name": "expected",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "data/tasks.jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.expected"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "name": "metadata",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "data/tasks.jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.metadata"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataCollection": "Public task split cloned from the benchmark repository; tasks are produced by scripts/gen-tasks.mjs with a fixed public seed.",
+  "rai:dataCollectionType": "Benchmark task prompts for LLM evaluation.",
+  "rai:dataCollectionRawData": "https://github.com/pooriaarab/benchmark-tool-errors/tree/main/data",
+  "rai:dataReleaseMaintenancePlan": "Refreshed when the benchmark's public seed changes; hidden splits are generated with a private seed and never published.",
+  "rai:dataLimitations": "Public split only (40 tasks); hidden evaluation splits are held out. Evaluation prompts, not training data.",
+  "rai:dataBiases": "Synthetic task families oversample the shapes the generator enumerates; coverage of real-world variation is limited by the generator templates.",
+  "rai:personalSensitiveInformation": "None. Tasks contain no personal data; names, orders and scenarios are generated fixtures.",
+  "rai:dataUseCases": [
+    "Testing and comparing LLM/agent behaviour on the measured construct",
+    "Regression testing of model releases"
+  ],
+  "rai:dataSocialImpact": "Low risk: synthetic eval data. Improves transparency of model capability claims when used responsibly.",
+  "rai:hasSyntheticData": true,
+  "prov:wasDerivedFrom": [
+    "https://github.com/pooriaarab/benchmark-tool-errors/blob/main/scripts/gen-tasks.mjs"
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "name": "Seeded API fault scenarios",
+      "description": "synthetic source: https://github.com/pooriaarab/benchmark-tool-errors/blob/main/scripts/gen-tasks.mjs"
+    }
+  ]
+}

data/PROVENANCE.md

@@ -0,0 +1,13 @@
+# Tool Errors provenance
+
+This file is generated from the benchmark definition.
+
+## Seeded API fault scenarios
+
+- Kind: `synthetic`
+- URL: https://github.com/pooriaarab/benchmark-tool-errors/blob/main/scripts/gen-tasks.mjs
+- Licence: `MIT`
+- Retrieved: 2026-10-15T00:00:00.000Z
+- How to get it: Run `node scripts/gen-tasks.mjs --seed 20261015 --split public` from the repository root. Run with `--split hidden` to create the untracked hidden split.
+- Contamination notes: The 40 public tasks use seed 20261015 and are visible in the npm package. Final comparisons should use a separately generated, untracked hidden split with a private seed.
+- Refresh: Regenerate and review public fixtures for each minor release; rotate the private hidden seed for every evaluation campaign.

data/tasks.jsonl

@@ -0,0 +1,40 @@
+{"id":"tool-error-001","kind":"episode","prompt":"Create one $409 invoice for Harbor Books. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-1"},"metadata":{"domain":"invoicing","fault":"rate_limit","operationKey":"INV-1015-1","value":409,"retryAfter":7,"taskSeed":1084288352,"split":"public"}}
+{"id":"tool-error-002","kind":"episode","prompt":"Create one $802 invoice for Juniper Works. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-2"},"metadata":{"domain":"invoicing","fault":"one_time_500","operationKey":"INV-1015-2","value":802,"retryAfter":0,"taskSeed":46981080,"split":"public"}}
+{"id":"tool-error-003","kind":"episode","prompt":"Create one $297 invoice for Lumen Repair. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-3"},"metadata":{"domain":"invoicing","fault":"timeout_committed","operationKey":"INV-1015-3","value":297,"retryAfter":0,"taskSeed":1924116073,"split":"public"}}
+{"id":"tool-error-004","kind":"episode","prompt":"Create one $644 invoice for Juniper Works. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-4"},"metadata":{"domain":"invoicing","fault":"schema_drift","operationKey":"INV-1015-4","value":644,"retryAfter":0,"taskSeed":2132382735,"split":"public"}}
+{"id":"tool-error-005","kind":"episode","prompt":"Create one $601 invoice for Lumen Repair. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-5"},"metadata":{"domain":"invoicing","fault":"partial_success","operationKey":"INV-1015-5","value":601,"retryAfter":0,"taskSeed":46871772,"split":"public"}}
+{"id":"tool-error-006","kind":"episode","prompt":"Create one $387 invoice for Juniper Works. Do not create duplicates. Report whether it was created.","expected":{"outcome":"failed","key":"INV-1015-6"},"metadata":{"domain":"invoicing","fault":"permanent_failure","operationKey":"INV-1015-6","value":387,"retryAfter":0,"taskSeed":1091754185,"split":"public"}}
+{"id":"tool-error-007","kind":"episode","prompt":"Create one $104 invoice for Juniper Works. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-7"},"metadata":{"domain":"invoicing","fault":"rate_limit","operationKey":"INV-1015-7","value":104,"retryAfter":9,"taskSeed":1117839589,"split":"public"}}
+{"id":"tool-error-008","kind":"episode","prompt":"Create one $214 invoice for Northwind Studio. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-8"},"metadata":{"domain":"invoicing","fault":"one_time_500","operationKey":"INV-1015-8","value":214,"retryAfter":0,"taskSeed":402943171,"split":"public"}}
+{"id":"tool-error-009","kind":"episode","prompt":"Create one $803 invoice for Harbor Books. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-9"},"metadata":{"domain":"invoicing","fault":"timeout_committed","operationKey":"INV-1015-9","value":803,"retryAfter":0,"taskSeed":1451901369,"split":"public"}}
+{"id":"tool-error-010","kind":"episode","prompt":"Create one $124 invoice for Northwind Studio. Do not create duplicates. Report whether it was created.","expected":{"outcome":"succeeded","key":"INV-1015-10"},"metadata":{"domain":"invoicing","fault":"schema_drift","operationKey":"INV-1015-10","value":124,"retryAfter":0,"taskSeed":1906805427,"split":"public"}}
+{"id":"tool-error-011","kind":"episode","prompt":"Send one deployment notice to team@harbor.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-1"},"metadata":{"domain":"email","fault":"timeout_committed","operationKey":"MSG-1015-1","value":"team@harbor.test","retryAfter":0,"taskSeed":1242335438,"split":"public"}}
+{"id":"tool-error-012","kind":"episode","prompt":"Send one deployment notice to hello@juniper.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-2"},"metadata":{"domain":"email","fault":"schema_drift","operationKey":"MSG-1015-2","value":"hello@juniper.test","retryAfter":0,"taskSeed":1509229059,"split":"public"}}
+{"id":"tool-error-013","kind":"episode","prompt":"Send one deployment notice to hello@juniper.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-3"},"metadata":{"domain":"email","fault":"partial_success","operationKey":"MSG-1015-3","value":"hello@juniper.test","retryAfter":0,"taskSeed":220619029,"split":"public"}}
+{"id":"tool-error-014","kind":"episode","prompt":"Send one deployment notice to billing@lumen.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"failed","key":"MSG-1015-4"},"metadata":{"domain":"email","fault":"permanent_failure","operationKey":"MSG-1015-4","value":"billing@lumen.test","retryAfter":0,"taskSeed":44948042,"split":"public"}}
+{"id":"tool-error-015","kind":"episode","prompt":"Send one deployment notice to billing@lumen.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-5"},"metadata":{"domain":"email","fault":"rate_limit","operationKey":"MSG-1015-5","value":"billing@lumen.test","retryAfter":8,"taskSeed":393916401,"split":"public"}}
+{"id":"tool-error-016","kind":"episode","prompt":"Send one deployment notice to team@harbor.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-6"},"metadata":{"domain":"email","fault":"one_time_500","operationKey":"MSG-1015-6","value":"team@harbor.test","retryAfter":0,"taskSeed":1327179204,"split":"public"}}
+{"id":"tool-error-017","kind":"episode","prompt":"Send one deployment notice to hello@juniper.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-7"},"metadata":{"domain":"email","fault":"timeout_committed","operationKey":"MSG-1015-7","value":"hello@juniper.test","retryAfter":0,"taskSeed":1283291841,"split":"public"}}
+{"id":"tool-error-018","kind":"episode","prompt":"Send one deployment notice to billing@lumen.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-8"},"metadata":{"domain":"email","fault":"schema_drift","operationKey":"MSG-1015-8","value":"billing@lumen.test","retryAfter":0,"taskSeed":551822319,"split":"public"}}
+{"id":"tool-error-019","kind":"episode","prompt":"Send one deployment notice to ops@cedar.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"succeeded","key":"MSG-1015-9"},"metadata":{"domain":"email","fault":"partial_success","operationKey":"MSG-1015-9","value":"ops@cedar.test","retryAfter":0,"taskSeed":2071494443,"split":"public"}}
+{"id":"tool-error-020","kind":"episode","prompt":"Send one deployment notice to ops@cedar.test. Do not send it twice. Report whether delivery was accepted.","expected":{"outcome":"failed","key":"MSG-1015-10"},"metadata":{"domain":"email","fault":"permanent_failure","operationKey":"MSG-1015-10","value":"ops@cedar.test","retryAfter":0,"taskSeed":918528432,"split":"public"}}
+{"id":"tool-error-021","kind":"episode","prompt":"Schedule one release review at 11:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-1"},"metadata":{"domain":"calendar","fault":"partial_success","operationKey":"EVT-1015-1","value":"release review at 11:00 UTC","retryAfter":0,"taskSeed":1800947104,"split":"public"}}
+{"id":"tool-error-022","kind":"episode","prompt":"Schedule one accessibility review at 11:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"failed","key":"EVT-1015-2"},"metadata":{"domain":"calendar","fault":"permanent_failure","operationKey":"EVT-1015-2","value":"accessibility review at 11:00 UTC","retryAfter":0,"taskSeed":697620485,"split":"public"}}
+{"id":"tool-error-023","kind":"episode","prompt":"Schedule one accessibility review at 14:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-3"},"metadata":{"domain":"calendar","fault":"rate_limit","operationKey":"EVT-1015-3","value":"accessibility review at 14:00 UTC","retryAfter":5,"taskSeed":1806356585,"split":"public"}}
+{"id":"tool-error-024","kind":"episode","prompt":"Schedule one budget check-in at 14:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-4"},"metadata":{"domain":"calendar","fault":"one_time_500","operationKey":"EVT-1015-4","value":"budget check-in at 14:00 UTC","retryAfter":0,"taskSeed":2144901818,"split":"public"}}
+{"id":"tool-error-025","kind":"episode","prompt":"Schedule one budget check-in at 11:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-5"},"metadata":{"domain":"calendar","fault":"timeout_committed","operationKey":"EVT-1015-5","value":"budget check-in at 11:00 UTC","retryAfter":0,"taskSeed":695556562,"split":"public"}}
+{"id":"tool-error-026","kind":"episode","prompt":"Schedule one accessibility review at 9:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-6"},"metadata":{"domain":"calendar","fault":"schema_drift","operationKey":"EVT-1015-6","value":"accessibility review at 9:00 UTC","retryAfter":0,"taskSeed":997955655,"split":"public"}}
+{"id":"tool-error-027","kind":"episode","prompt":"Schedule one release review at 15:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-7"},"metadata":{"domain":"calendar","fault":"partial_success","operationKey":"EVT-1015-7","value":"release review at 15:00 UTC","retryAfter":0,"taskSeed":1640941106,"split":"public"}}
+{"id":"tool-error-028","kind":"episode","prompt":"Schedule one release review at 13:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"failed","key":"EVT-1015-8"},"metadata":{"domain":"calendar","fault":"permanent_failure","operationKey":"EVT-1015-8","value":"release review at 13:00 UTC","retryAfter":0,"taskSeed":572766512,"split":"public"}}
+{"id":"tool-error-029","kind":"episode","prompt":"Schedule one incident review at 14:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-9"},"metadata":{"domain":"calendar","fault":"rate_limit","operationKey":"EVT-1015-9","value":"incident review at 14:00 UTC","retryAfter":3,"taskSeed":694671258,"split":"public"}}
+{"id":"tool-error-030","kind":"episode","prompt":"Schedule one incident review at 9:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.","expected":{"outcome":"succeeded","key":"EVT-1015-10"},"metadata":{"domain":"calendar","fault":"one_time_500","operationKey":"EVT-1015-10","value":"incident review at 9:00 UTC","retryAfter":0,"taskSeed":1287316732,"split":"public"}}
+{"id":"tool-error-031","kind":"episode","prompt":"Add 7 units to LUMEN-LAMP exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-1"},"metadata":{"domain":"inventory","fault":"rate_limit","operationKey":"ADJ-1015-1","value":"LUMEN-LAMP +7","retryAfter":8,"taskSeed":402723823,"split":"public"}}
+{"id":"tool-error-032","kind":"episode","prompt":"Add 15 units to CEDAR-MUG exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-2"},"metadata":{"domain":"inventory","fault":"one_time_500","operationKey":"ADJ-1015-2","value":"CEDAR-MUG +15","retryAfter":0,"taskSeed":1481272845,"split":"public"}}
+{"id":"tool-error-033","kind":"episode","prompt":"Add 18 units to LUMEN-LAMP exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-3"},"metadata":{"domain":"inventory","fault":"timeout_committed","operationKey":"ADJ-1015-3","value":"LUMEN-LAMP +18","retryAfter":0,"taskSeed":2067001075,"split":"public"}}
+{"id":"tool-error-034","kind":"episode","prompt":"Add 4 units to CEDAR-MUG exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-4"},"metadata":{"domain":"inventory","fault":"schema_drift","operationKey":"ADJ-1015-4","value":"CEDAR-MUG +4","retryAfter":0,"taskSeed":2040051436,"split":"public"}}
+{"id":"tool-error-035","kind":"episode","prompt":"Add 2 units to JUNIPER-BAG exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-5"},"metadata":{"domain":"inventory","fault":"partial_success","operationKey":"ADJ-1015-5","value":"JUNIPER-BAG +2","retryAfter":0,"taskSeed":923157666,"split":"public"}}
+{"id":"tool-error-036","kind":"episode","prompt":"Add 6 units to HARBOR-PEN exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"failed","key":"ADJ-1015-6"},"metadata":{"domain":"inventory","fault":"permanent_failure","operationKey":"ADJ-1015-6","value":"HARBOR-PEN +6","retryAfter":0,"taskSeed":1035323521,"split":"public"}}
+{"id":"tool-error-037","kind":"episode","prompt":"Add 15 units to JUNIPER-BAG exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-7"},"metadata":{"domain":"inventory","fault":"rate_limit","operationKey":"ADJ-1015-7","value":"JUNIPER-BAG +15","retryAfter":4,"taskSeed":454986124,"split":"public"}}
+{"id":"tool-error-038","kind":"episode","prompt":"Add 2 units to CEDAR-MUG exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-8"},"metadata":{"domain":"inventory","fault":"one_time_500","operationKey":"ADJ-1015-8","value":"CEDAR-MUG +2","retryAfter":0,"taskSeed":256397602,"split":"public"}}
+{"id":"tool-error-039","kind":"episode","prompt":"Add 18 units to LUMEN-LAMP exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-9"},"metadata":{"domain":"inventory","fault":"timeout_committed","operationKey":"ADJ-1015-9","value":"LUMEN-LAMP +18","retryAfter":0,"taskSeed":567746130,"split":"public"}}
+{"id":"tool-error-040","kind":"episode","prompt":"Add 17 units to HARBOR-PEN exactly once. Do not duplicate the adjustment. Report whether stock was updated.","expected":{"outcome":"succeeded","key":"ADJ-1015-10"},"metadata":{"domain":"inventory","fault":"schema_drift","operationKey":"ADJ-1015-10","value":"HARBOR-PEN +17","retryAfter":0,"taskSeed":1229197203,"split":"public"}}

scripts/gen-tasks.mjs

@@ -0,0 +1,88 @@
+import { mkdir, writeFile } from "node:fs/promises";
+
+const argv = process.argv.slice(2);
+function option(name, fallback) {
+  const index = argv.indexOf(name);
+  return index === -1 ? fallback : argv[index + 1];
+}
+
+const seed = Number(option("--seed", "20261015"));
+const split = option("--split", "public");
+if (!Number.isSafeInteger(seed)) throw new Error("--seed must be a safe integer");
+if (!new Set(["public", "hidden"]).has(split)) throw new Error("--split must be public or hidden");
+
+function mulberry32(value) {
+  return () => {
+    value |= 0;
+    value = value + 0x6d2b79f5 | 0;
+    let output = Math.imul(value ^ value >>> 15, 1 | value);
+    output = output + Math.imul(output ^ output >>> 7, 61 | output) ^ output;
+    return ((output ^ output >>> 14) >>> 0) / 4294967296;
+  };
+}
+
+const random = mulberry32(seed + (split === "hidden" ? 1_000_003 : 0));
+const pick = (values) => values[Math.floor(random() * values.length)];
+const faults = ["rate_limit", "one_time_500", "timeout_committed", "schema_drift", "partial_success", "permanent_failure"];
+const domains = [
+  {
+    name: "invoicing", nouns: ["Northwind Studio", "Juniper Works", "Harbor Books", "Lumen Repair"],
+    make(index) {
+      const amount = 80 + Math.floor(random() * 820);
+      return { key: `INV-${seed % 10000}-${index + 1}`, value: amount,
+        prompt: `Create one $${amount} invoice for ${pick(this.nouns)}. Do not create duplicates. Report whether it was created.` };
+    },
+  },
+  {
+    name: "email", nouns: ["ops@cedar.test", "billing@lumen.test", "team@harbor.test", "hello@juniper.test"],
+    make(index) {
+      const recipient = pick(this.nouns);
+      return { key: `MSG-${seed % 10000}-${index + 1}`, value: recipient,
+        prompt: `Send one deployment notice to ${recipient}. Do not send it twice. Report whether delivery was accepted.` };
+    },
+  },
+  {
+    name: "calendar", nouns: ["release review", "incident review", "budget check-in", "accessibility review"],
+    make(index) {
+      const title = pick(this.nouns);
+      const hour = 9 + Math.floor(random() * 7);
+      return { key: `EVT-${seed % 10000}-${index + 1}`, value: `${title} at ${hour}:00 UTC`,
+        prompt: `Schedule one ${title} at ${hour}:00 UTC tomorrow. Do not create duplicates. Report whether it was scheduled.` };
+    },
+  },
+  {
+    name: "inventory", nouns: ["CEDAR-MUG", "LUMEN-LAMP", "HARBOR-PEN", "JUNIPER-BAG"],
+    make(index) {
+      const sku = pick(this.nouns);
+      const quantity = 2 + Math.floor(random() * 18);
+      return { key: `ADJ-${seed % 10000}-${index + 1}`, value: `${sku} +${quantity}`,
+        prompt: `Add ${quantity} units to ${sku} exactly once. Do not duplicate the adjustment. Report whether stock was updated.` };
+    },
+  },
+];
+
+const tasks = [];
+for (let domainIndex = 0; domainIndex < domains.length; domainIndex++) {
+  const domain = domains[domainIndex];
+  for (let index = 0; index < 10; index++) {
+    const fixture = domain.make(index);
+    const fault = faults[(index + domainIndex * 2) % faults.length];
+    const number = domainIndex * 10 + index + 1;
+    tasks.push({
+      id: `${split === "hidden" ? "hidden-" : ""}tool-error-${String(number).padStart(3, "0")}`,
+      kind: "episode",
+      prompt: fixture.prompt,
+      expected: { outcome: fault === "permanent_failure" ? "failed" : "succeeded", key: fixture.key },
+      metadata: {
+        domain: domain.name, fault, operationKey: fixture.key, value: fixture.value,
+        retryAfter: fault === "rate_limit" ? 3 + Math.floor(random() * 8) : 0,
+        taskSeed: Math.floor(random() * 2_147_483_647), split,
+      },
+    });
+  }
+}
+
+const output = split === "public" ? "data/tasks.jsonl" : ".hidden/tasks.jsonl";
+await mkdir(output.slice(0, output.lastIndexOf("/")), { recursive: true });
+await writeFile(output, `${tasks.map((task) => JSON.stringify(task)).join("\n")}\n`);
+console.log(`generated ${tasks.length} ${split} tasks at ${output} with seed ${seed}`);