# Final failure evidence register

This is an advisory diagnosis index. Every frozen result, score and classification remains unchanged. The incident text below preserves previously reported findings; the final scorecard supplies the reconciled interpretation.

## 1. docblocks-theme-tools-absent-r1

Reported: 2026-10-28T06:48:05.353Z.

Four treatments failed theme round-trip because DocBlocks tools were not exposed; separately noted likely EUR 30,000 derived-budget evaluator false positive. Campaign continues unchanged.

[Saved evidence](scorecard-details.json).

## 2. mlx-research-tools-filtered-r1

Reported: 2026-10-28T08:53:21.674Z.

Qwen MLX research failed because immediate-write and file-repair surfaces exclude registered Wikipedia tools. Same generic pattern as earlier image-tool restriction; raw outcomes and frozen environment preserved.

[Saved evidence](scorecard-details.json).

## 3. week-plan-grounding-and-ab-rubric-r1

Reported: 2026-10-28T12:40:19.700Z.

Status update reported week-plan gate false positives: successful calendar retrieval after the final report can pass without grounding the report in events. All four plans have major grounding defects. Two A/B readouts have incompatible mock-service judging axes and are recorded as judgeSpecMismatch with no numeric advisory scores.

[Saved evidence](scorecard-details.json).

## 4. complex-routing-and-unreachable-verification-r1

Reported: 2026-10-29T05:21:40.822Z.

New finding reported: all four SEO-selection runs suggest Job Hunt; two fail by editing directly, Gemma MLX passes after unwanted project creation. Gemma llama midtask-edit passes with an unreachable, unexecuted verification step. Recorded scores remain unchanged and the campaign continues.

[Saved evidence](scorecard-details.json).

## 5. invoice-run-books-not-updated-r1

Reported: 2026-10-29T06:24:24.803Z.

Qwen MLX invoice run passes with correct invoices but leaves ledger at 041 and billable entries unbilled, allowing duplicate billing on a later run. Ledger validity gate misses the required bookkeeping. Fixed score preserved; advisory completeness penalized; campaign continues.

[Saved evidence](scorecard-details.json).

## 6. invoice-run-invented-fee-and-wrong-payee-r1

Reported: 2026-10-29T07:28:08.629Z.

New extension beyond prior bookkeeping incident: Gemma MLX invents a 2% monthly late fee; Gemma llama passes with checks payable to the debtor, invented addresses and date defects. Reviews and diagnoses recorded separately from unchanged fixed scores; no external invoices sent.

[Saved evidence](scorecard-details.json).

## 7. author-params-generated-gate-stop-recovered-r1

Reported: 2026-10-29T16:46:30.347Z.

Campaign stopped at 164 trials because Gemma llama generated an undefined workPath gate. Diagnosed as isolated trial recipe state; preserved infra classification/exclusion and all row bytes, verified frozen identities and no orphan process, archived original stop, then resumed original supervisor at next pending cell. No trial repeated or baseline changed.

[Saved evidence](scorecard-details.json).

## 8. gate-script-verification-starvation-r1

Reported: 2026-10-29T18:43:35.578Z.

Qwen llama produced and restored a correct inventory report, but verification waited 1030 seconds for inference admission and then behind the provider queue. A 126038-token prefill took 571 seconds; the no-hard-progress watchdog ended the trial while inference remained active. Recorded as scheduling/prefill evidence alongside model repair defects; outcome and frozen environment unchanged. Campaign continues.

[Saved evidence](scorecard-details.json).

## 9. fanout-completed-root-selection-r1

Reported: 2026-10-29T22:54:51.496Z.

Qwen MLX spawned four child tasks under its second invocation, but the evaluator selected the earlier completed root and counted two recipe invocations instead. Recorded failure, model-default classification and fixed 3.9 remain unchanged. Only one child completed before the frozen retry watchdog, so workflow completion is not inferred. Campaign continues.

[Saved evidence](scorecard-details.json).

## 10. tankcombat-gameplay-validation-gaps-r2

Reported: 2026-10-29T23:50:19.045Z.

All four second-repetition tank artifacts pass the three frozen runtime assertions with fixed 10, but isolated saved-source probes confirm a contact-reset TypeError (Gemma llama), enemy firing disabled by fractional cooldown equality (Qwen llama), and impossible hit scoring from reversed ownership checks (Gemma MLX). Advisory grades and failure evidence are separate; no frozen artifact, gate, score or classification modified.

[Saved evidence](scorecard-details.json).

## 11. ab-readout-confidence-interval-contract-r2

Reported: 2026-10-30T11:54:54.949Z.

Newly isolated frozen contract conflict: the eval prompt requires EXACTLY a stats schema with no confidence-interval fields and a readout from those figures with no new numbers. Actual task craftbook version 1.0.4 requires a confidence interval in analyze/write/evaluate; the final verdict marks only the interval criterion FAIL and the evaluate gate loops any FAIL back to write, not analyze. The write role forbids new numbers and workspace writes, so that repair path cannot legitimately add the missing computation to stats. Observed17 gate rejections (6write,11evaluate),28activation events,26sessions. This is an eval/craftbook contract incompatibility, additional to the previously reported destination and advisory-rubric mismatches; do not interpret the failure as inability to compute A/B statistics. Frozen failureClass=model/task-paused-unrecovered, fixed score and raw evidence remain unchanged. A future unfrozen repair should reconcile schema, acceptance criteria, output destination and repair routing before comparing model performance.

[Saved evidence](scorecard-details.json).

## 12. ab-readout-stale-narrative-after-stats-repair-r2

Reported: 2026-10-30T12:59:58.872Z.

Passed19/19 fixed8 after57m26/81tools. Final stats were patched after the final readout write: workspace readout written12:21:03.502; p_value changed .268634 to .00786 and significant_at_95 false to true at12:21:24.028; decision changed inconclusive to do_not_ship at12:21:33.708. No report edit followed before pass12:21:40.767. Artifact and workspace readout are identical4417B and remain stale: p=.2686, no significance, incorrect CI[-.22,2.72]pp, and a claim that stats still say inconclusive. Unsupported4.3x/43k-per-arm/30–45day rerun advice derives from that false uncertainty. Actual independent pooled z=2.661250,p=.007785105; unpooled WaldCI about[.33,2.17]pp. Final do_not_ship is correct because refund delta .23pp exceeds .20pp, but only that clause fails. This is a cross-artifact consistency gate gap additional to the previously reported CI-contract and rubric mismatch. Preserve deterministic pass/fixed score; advisory judgeSpecMismatch means no invented numeric grade. Earlier derive_file calls emitted a non-tabular object and warned even though they wrote the file; the third output had correct z but wrong p. Later developer surface restored write/copy authority. Same-build r1 passed8.5 in22m9/62calls.

[Saved evidence](scorecard-details.json).

## 13. author-linear-manual-completion-bypasses-gates-r2

Reported: 2026-10-31T08:27:29.999Z.

At 07:47:01.793Z Reza called set_task_status complete, explicitly saying execution remained for a specialist. It succeeded before any output existed. Saved task status is complete but activeStepId is still inspect; all three steps lack completedAt, and there are zero advance calls or gate approvals. Lorena later wrote the workspace outputs; independent typed comparison confirms exactly the 28 valid source rows with both malformed rows excluded. The final report is 344 bytes, below the recipe verify gate of 1500 bytes, which never ran. Frozen author-linear successCheck tests task.status and output sizes/record count; completedTaskSteps is only a progress milestone, not an acceptance requirement. Thus the 7/7, fixed 8.5 pass establishes final deliverables and authored gates, not execution of the gated workflow. This is a newly confirmed manual-completion bypass, distinct from the earlier unreachable added verification step. Future unfrozen remediation should enforce step completion receipts on terminal status and in this scenario. Keep the recorded pass/score unchanged; no judge specification applies.

[Saved evidence](scorecard-details.json).

## 14. html-size-guard-blocks-gameplay-repair-r3

Reported: 2026-10-31T23:46:38.869Z.

Gemma MLX tic-tac-toe failed because NodeList.some stopped turn switching. Both exact rejected replacement candidates pass isolated browser win/draw/post-game checks; they simplify styling and remove optional reset. The overwrite guard rejected 1703/1635-byte inline JS against the prior broken 2358-byte script solely on size despite retained grid/click/win signals. Automatic recovery then exposed write_file only while guard feedback demanded a patch tool. The all-PASS model verdict was false; the frozen runtime gate correctly failed. Diagnostic evidence saved separately, frozen outcome/score unchanged. Future unfrozen fixes should reconcile recovery tool selection with guard advice and behavioral evidence, preserving feature-loss checks.

[Saved evidence](scorecard-details.json).

## 15. petshop-feedback-stops-active-image-r3

Reported: 2026-10-31T23:48:52.707Z.

Qwen MLX petshop started genuine generate_image at 23:44:32.381 and sd-cpp sampling at 23:44:35.502. The frozen runner declared repair-exhausted at 23:44:50.274 based on four HTML repair attempts despite the image job being active. Sampling finished during shutdown at 23:44:58.591; VAE decode was still advancing at 23:45:02.847. No PNG was captured and the daemon exited by 23:45:04.908. Completed-tool counts omit the unfinished call: do not mistake this for no image invocation or a missing engine. Preserved fixed2.2/model-default failure. Future unfrozen termination should account for the pending tool producing the missing deliverable.

[Saved evidence](scorecard-details.json).

## 16. artifact-copy-allows-json-as-pptx-r3

Reported: 2026-11-01T05:47:15.116Z.

Gemma MLX PowerPoint r3 successfully converted and previewed its source but repeatedly supplied a string save destination and never saved the real deck. At05:29:35.254 copy_artifact_to_workspace accepted tasks/1/inputs/sourcePath.json as deliverables/d-day.pptx after direct text writes were refused. Final335-byte output exactly matches the JSON and is not a ZIP. The frozen gate caught it: FAIL5.1/model-default,16/21 checks. Preserve all evidence; future unfrozen binary-output validation should cover artifact copies as well as text writes. No campaign intervention.

[Saved evidence](scorecard-details.json).

## 17. ab-readout-pvalue-unchecked-r3

Reported: 2026-11-01T12:01:02.457Z.

Qwen MLX repetition3 passed all19 checks with final p=.000513 instead of .007785 and CI[.63,1.87]pp instead of[.33,2.17]pp. The frozen gate checks significant_at_95 but never p_value or numerical CI; the wrong CDF polynomial reproduces the artifact. Fixed results preserved; advisory mismatch retained without invented scores.

[Saved evidence](scorecard-details.json).

## 18. task-note-alias-collision-r3

Reported: 2026-11-01T16:07:44.580Z.

Gemma llama code review supplied a short note alias and a detailed content alias together. The frozen handler text ?? note ?? content saved only the short value, so six advances correctly rejected the missing Scope heading. Proper content arrived only after the configured task pause. The trial failed 2.5 with no report. Future unfrozen remediation should reject conflicting aliases or use one canonical field and expose the stored body. Fixed failure and partial evidence preserved.

[Saved evidence](scorecard-details.json).

## 19. review-gate-accepts-unmet-r3

Reported: 2026-11-01T16:07:44.580Z.

Gemma llama API review passed fixed 8.5 after its final evaluator verdict explicitly left compatibility criterion 4 UNMET. Actual task/catalog gate checks any PASS and absence of literal FAIL; Node regex replay reproduces acceptance of PASS plus UNMET. Evaluate advanced at 15:38:20.868 and finish at 15:38:45 without report repair. Future unfrozen remediation should require a complete structured all-PASS criterion list. Recorded pass and evidence preserved.

[Saved evidence](scorecard-details.json).

## 20. markdown-folding-hides-fail-verdict-r3

Reported: 2026-11-02T07:34:55.175Z.

Qwen MLX invoice-run r3 passes fixed8.5 despite a correctly formatted reviewer criterion ending FAIL. write_artifact normalizes Markdown, collapsing seven newlines to zero and moving FAIL inside a paragraph ending PASS. Replay of the actual normalizer exactly matches the saved verdict: original line-end FAIL matches, saved line-end FAIL does not. Evaluate auto-advances07:30:16.117; proof in invoice-verdict-normalization-r3-replay.json. Frozen score, result and evidence preserved; future unfrozen correction should preserve verdict boundaries or use structured all-criteria validation.

[Saved evidence](scorecard-details.json).

## 21. author-linear-invalid-regex-gate-r3

Reported: 2026-11-02T10:35:47.263Z.

Qwen MLX author-linear r3 fails fixed3.4 after authoring accepted invalid JavaScript pattern (?i)\bPASS\b. The runtime repeatedly requested report-content repairs, which cannot fix pattern compilation. Read-only replay of frozen containsPattern rejects all four report variants; the hypothetical separate flags:i control succeeds. Correct29-row JSON/report and all failed/partial evidence preserved. Future unfrozen correction: compile gates at authoring and route configuration failures to recipe owner. Proof: author-linear-invalid-gate-pattern-r3-replay.json.

[Saved evidence](scorecard-details.json).

## 22. author-params-generated-gate-stop-recovered-r3

Reported: 2026-11-02T17:35:06.957Z.

Recovered the campaign stop after trial 507 without retry or reclassification; second infrastructure exclusion preserved and trial 508 running.

[Saved evidence](scorecard-details.json).

## 23. gate-result-array-contract-r3

Reported: 2026-11-02T19:29:01.648Z.

Qwen llama author-gate-script r3 saved a script passing check arrays to gateResult(boolean, detail). Controlled calls to the frozen SDK approve failed-check arrays; false boolean control rejects. No generated script was executed and the original trial remains a watchdog failure.

[Saved evidence](scorecard-details.json).

## All failed trials

Each row retains its fixed score, failure class/mode and source reason. Follow its postmortem for detailed facts and the adjacent result.json, facts.json and daemon.log for raw evidence. The index includes both infrastructure exclusions.

| Cell | Fixed /10 | Class | Mode | Evidence | Recorded reason |
| --- | --- | --- | --- | --- | --- |
| r1-gemma-mlx-core-petshop | 1.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on index.html; missing signals stayed unchanged: working-image, image-asset. |
| r1-qwen-mlx-core-petshop | 2.2 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "build-me-a-single-page-pet-shop-website/workspace/index.html:3:targetnone:frnone:rp0:rf0" stuck for 18m — artifact scored but stalled 18m despite 3 tool calls (0 re-writes). Artifact produced but never reached success. |
| r1-gemma-mlx-core-failing-tests-spec | 6.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on src/machine.ts; missing signals stayed unchanged: vitest-green. Last failure: vitest run failed (exit 1): × submit moves draft to pending and returns the new state 2ms \| × a rejected send leaves the state unchanged 0ms \| × expedite is rejected on a machine created without the flag 0ms \| FAIL tests/machine.test.ts > order machine: transitions > submit moves draft to pending and returns the new state \| AssertionError: expected 'draft' to be 'pending' // Object.is equality \| FAIL tests/machine.test.ts > order machine: invalid transitions > a rejected send leaves the state unchanged \| FAIL tests/machine.test.ts > order machine: expedite guard > expedite is rejected on a machine created without the flag \| AssertionError: expected [Function] to throw an error |
| r1-gemma-mlx-core-data-wrangle | 4.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on out/customers.json; missing signals stayed unchanged: source-ids-preserved. Last failure: record with id "A-007" does not exactly match that source row after the allowed email/date normalization — preserve its source id, name, normalized email, and normalized signupDate together |
| r1-qwen-mlx-productivity-constrained-comms | 6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on customer-notice.md; the same scenario failure stayed unchanged. Last failure: word band violated: 121 words, need ≥ 140 — the notice must be 140-220 words |
| r1-qwen-llamacpp-productivity-constrained-comms | 6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on customer-notice.md; the same scenario failure stayed unchanged. Last failure: word band violated: 124 words, need ≥ 140 — the notice must be 140-220 words |
| r1-qwen-mlx-productivity-conflict-synthesis | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on synthesis.md; the same scenario failure stayed unchanged. Last failure: - unsupported budget claim not present in the seeded sources: re-read finance.csv total_budget and line-item amounts, then replace every claimed amount with a source-backed value |
| r1-qwen-llamacpp-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r1-gemma-mlx-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r1-gemma-llamacpp-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r1-qwen-mlx-productivity-docblocks-theme-roundtrip | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r1-gemma-mlx-productivity-craftbook-powerpoint-deck | 3.9 | model | timeout | [postmortem](scorecard-details.json) | hit hard ceiling (2100000ms + 35m progress extensions) — last hard progress 263s before the end; the deliverable never closed |
| r1-qwen-mlx-productivity-wikipedia-research-brief | 6.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on research-brief.md; missing signals stayed unchanged: mcp-research-provenance. Last failure: mock wikipedia: MCP tool "search_wikipedia" was never called; writing about it does not count — "search_wikipedia" is on your tool roster, call it like any other tool and use its response in the deliverable |
| r1-gemma-mlx-productivity-wikipedia-research-brief | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on research-brief.md; missing signals stayed unchanged: claim-calibration. Last failure: every mention of the “first programmer” label must be hedged in the same sentence as debated shorthand |
| r1-qwen-mlx-productivity-craftbook-week-plan | 1.5 | model | timeout | [postmortem](scorecard-details.json) | hit hard ceiling (2700000ms + 45m progress extensions) — last hard progress 35s before the end; the deliverable never closed |
| r1-gemma-mlx-productivity-craftbook-week-plan | 5.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: mock calendar: 0/1 authorized request(s) seen; writing about it does not count — actually call the service (run the probe script listed in mocks/services.md via run_installed_script, or use http.authed), then update the report from the response, mock calendar: no authorized request matched ^/api/events; writing about it does not count — actually call the service (run the probe script listed in mocks/services.md via run_installed_script, or use http.authed), then update the report from the response. Last failure: mock calendar: 0/1 authorized request(s) seen; writing about it does not count — actually call the service (run the probe script listed in mocks/services.md via run_installed_script, or use http.authed), then update the report from the response |
| r1-gemma-llamacpp-productivity-craftbook-ab-test-readout | 1.5 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 304s — daemon reachable but issuing no model turns (soft digest stuck at 779e3bbfc399c4d5) (re-engage nudge was sent and ignored) |
| r1-qwen-mlx-developer-fix-squisq-bugs | 5.2 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on packages/core/src/spatial/Geohash.ts; missing signals stayed unchanged: great-circle-path, tsc-clean. Last failure: getGeohashPath behavioral check failed: path from lng 170 to lng -170 passes through lng -54.7 — it must cross the antimeridian (±180), not run back through 0 |
| r1-gemma-mlx-developer-dev-craftbook-routing | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection (engineering); missing signals stayed unchanged: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it. Last failure: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it |
| r1-qwen-mlx-developer-dev-craftbook-routing | 3.4 | model | success-check-false | [postmortem](scorecard-details.json) | task batch-exporter/1 uses "bug-fix-tdd" but 1 new craftbook(s) were also authored (batches-regression-red-green) — the library recipe was enough |
| r1-gemma-mlx-developer-craftbook-deep-security-review | 6.1 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-deep-security-review stale no-write repair loop: security-review/REPORT.md had unchanged content (2272 bytes) for 61s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 6/11: security-review/REPORT.md is missing required content: real file and line — nothing in its 2272 bytes matches /src/admin\.ts:[0-9]+/. Add that content. |
| r1-gemma-llamacpp-developer-craftbook-deep-security-review | 6.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 9 while the failing detail kept changing; latest checked file security-review/findings.json. Last failure: security-review/findings.json: record 0 is missing required field "theme" |
| r1-gemma-mlx-developer-large-pr-review | 3.3 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "large-pr-review:4:targetpr-review.md:fr1i0sit8:rp0:rf0" stuck for 18m — artifact scored but stalled 18m despite 4 tool calls (1 re-writes). Artifact produced but never reached success. |
| r1-gemma-llamacpp-developer-large-pr-review | 3.3 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "large-pr-review:5:targetpr-review.md:fr1i0sit8:rp0:rf0" stuck for 8m — 3 re-writes (15 tool calls) without sniff movement. Artifact produced but never reached success. |
| r1-gemma-llamacpp-developer-craftbook-codemod-sweep | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-codemod-sweep:28:targetnone:frd6dga7:rp0:rf0:m5" stuck for 8m — 15 re-writes (24 tool calls) without sniff movement. Artifact produced but never reached success (retry-loop nudge was sent and ignored). |
| r1-qwen-mlx-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-codemod-sweep:30:targetnone:frjr7vze:rp0:rf0:m4" stuck for 10m — 3 re-writes (16 tool calls) without sniff movement. Artifact produced but never reached success. |
| r1-qwen-llamacpp-developer-craftbook-codemod-sweep | 4.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: victoria/f594a461 exhausted its bounded automatic recovery allowance at the current checked progress; last error: [llama-cpp] aborting — the gezel emitted 12003 characters of prose this turn without calling any action tool. Stop planning. Your next message must START with a single tool call — or, if the work is genuinely finished and nothing is left to do, be ONE short sentence saying so and nothing else. This  |
| r1-gemma-mlx-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: fyodor/2ccf4599 exhausted its bounded automatic recovery allowance at the current checked progress; last error: [Mac AI] aborting — the gezel emitted 7308 characters of prose this turn without calling any action tool. Stop planning. Your next message must START with a single tool call — or, if the work is genuinely finished and nothing is left to do, be ONE short sentence saying so and nothing else. This step |
| r1-qwen-mlx-developer-craftbook-refactor-module | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook refactor-module has not reached a terminal step. Last failure: task sourced from craftbook refactor-module has not reached a terminal step |
| r1-qwen-llamacpp-developer-craftbook-refactor-module | 4.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: sergei/68c7dca7 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r1-gemma-mlx-developer-craftbook-refactor-module | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on tasks/1/change-notes.md; missing signals stayed unchanged: tasks/1/change-notes.md cites 1 source(s) that do not exist: `src/normalize-name.js` — every cited path must resolve to a real file in the workspace (no fabricated citations).. Last failure: tasks/1/change-notes.md cites 1 source(s) that do not exist: `src/normalize-name.js` — every cited path must resolve to a real file in the workspace (no fabricated citations). |
| r1-gemma-llamacpp-developer-craftbook-refactor-module | 4.9 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-refactor-module stale no-write repair loop: tasks/1/plan.md had unchanged content (2111 bytes) for 127s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 20/30: tasks/1/plan.md is missing required content: plan sections in order — nothing in its 2111 bytes matches /^##\s+Target[\s\S]*^##\s+Stages[\s\S]*^##\s+Acceptance criteria/. Add that content. |
| r1-qwen-llamacpp-developer-craftbook-bug-fix-tdd | 4.2 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-bug-fix-tdd stale no-write repair loop: tasks/1/fix-notes.md had unchanged content (2144 bytes) for 244s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 20/28: tasks/1/fix-notes.md is missing required content: fix-notes sections in order — nothing in its 2144 bytes matches /^##\s+Problem[\s\S]*^##\s+Change[\s\S]*^##\s+Files touched[\s\S]*^##\s+Regression coverage[\s\S]*^##\s+Risk[\s\S]*^##\s+How to verify/. Add that content. |
| r1-gemma-mlx-developer-craftbook-bug-fix-tdd | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 26 while the failing detail kept changing; latest checked file workspace output. Last failure: task sourced from craftbook bug-fix-tdd has not reached a terminal step |
| r1-gemma-llamacpp-developer-craftbook-bug-fix-tdd | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-bug-fix-tdd:26:targetnone:frnsn9x5:rp0:rf0:m3" stuck for 8m — 4 re-writes (12 tool calls) without sniff movement. Artifact produced but never reached success. |
| r1-qwen-mlx-developer-craftbook-bug-fix-tdd | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook bug-fix-tdd has not reached a terminal step. Last failure: task sourced from craftbook bug-fix-tdd has not reached a terminal step |
| r1-gemma-llamacpp-complex-work-craftbook-find-vs-create | 4.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection; missing signals stayed unchanged: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project. Last failure: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project |
| r1-qwen-mlx-complex-work-craftbook-find-vs-create | 4.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection; missing signals stayed unchanged: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project. Last failure: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project |
| r1-gemma-mlx-complex-work-craftbook-invoice-run | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook invoice-run has not reached a terminal step. Last failure: task sourced from craftbook invoice-run has not reached a terminal step |
| r1-gemma-mlx-complex-work-craftbook-author-linear | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: daryna/3c045e73 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r1-gemma-llamacpp-complex-work-craftbook-author-linear | 4.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: CSV Order Cleanup; missing signals stayed unchanged: task order-intake-cleanup/1 (from craftbook "csv-order-cleanup") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., out/report.md is 0 bytes, need ≥ 80, out/orders.json does not exist yet. Last failure: task order-intake-cleanup/1 (from craftbook "csv-order-cleanup") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count. |
| r1-gemma-mlx-complex-work-craftbook-route-multi | 6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on out/press-release.md; missing signals stayed unchanged: no task has been started from the bundled "press-release" recipe yet — search the recipe library for the one that covers writing an announcement for journalists and invoke it on this project. Last failure: no task has been started from the bundled "press-release" recipe yet — search the recipe library for the one that covers writing an announcement for journalists and invoke it on this project |
| r1-gemma-llamacpp-complex-work-craftbook-route-multi | 5.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on out/press-release.md; missing signals stayed unchanged: no task has been started from the bundled "press-release" recipe yet — search the recipe library for the one that covers writing an announcement for journalists and invoke it on this project. Last failure: no task has been started from the bundled "press-release" recipe yet — search the recipe library for the one that covers writing an announcement for journalists and invoke it on this project |
| r1-gemma-llamacpp-complex-work-craftbook-export-generalize | 3.4 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 303s — daemon reachable but issuing no model turns (soft digest stuck at 57a90915656b933b) (re-engage nudge was sent and ignored) |
| r1-qwen-mlx-complex-work-craftbook-export-generalize | 3.3 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "craftbook-export-generalize:5:targetnone:fr1dzrjs8:rp0:rf0" stuck for 18m — artifact scored but stalled 18m despite 18 tool calls (2 re-writes). Artifact produced but never reached success (retry-loop nudge was sent and ignored). |
| r1-gemma-mlx-complex-work-craftbook-export-generalize | 2.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: supplier intake recipe; missing signals stayed unchanged: out/q1-intake.md does not exist yet — write the short Q1 intake report there: what you checked and which suppliers have an impossible lead time, out/q1-clean.json does not exist yet — write the cleaned Q1 supplier records there as one list of supplier entries, task supplier-intake/1 (from craftbook "supplier-intake-procedure") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., out/q2-intake.md does not exist yet — write the Q2 intake report there by running the saved recipe over data/suppliers-q2.json. Last failure: out/q1-intake.md does not exist yet — write the short Q1 intake report there: what you checked and which suppliers have an impossible lead time |
| r1-gemma-mlx-complex-work-craftbook-author-params | 4.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: regional sales rollup; missing signals stayed unchanged: no step of craftbook "regional-sales-summary-procedure" refers to its input parameter — the region has to reach the steps (the export it reads and the summary it writes must be derived from the parameter, not hardcoded to one region), out/north-summary.md is 0 bytes, need ≥ 80, out/south-summary.md is 0 bytes, need ≥ 80. Last failure: no step of craftbook "regional-sales-summary-procedure" refers to its input parameter — the region has to reach the steps (the export it reads and the summary it writes must be derived from the parameter, not hardcoded to one region) |
| r1-gemma-llamacpp-complex-work-craftbook-author-params | 5.5 | infra | model-stuck | [postmortem](scorecard-details.json) | no real progress for 601s — engine idle — no token stream or slot update for 288s (hard digest stuck at a96ccf4885a164d3) poisoned-session repair drew no response: jian/1a52241c (570s ago) |
| r1-qwen-llamacpp-complex-work-craftbook-author-gate-script | 3.7 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 603s — model busy but not delivering (hard digest stuck at 778ce778de8ef1d2) |
| r1-gemma-mlx-complex-work-craftbook-author-gate-script | 3.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: inventory report gate script; missing signals stayed unchanged: craftbook "inventory-health-check-recipe" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate, task inventory-health-check/1 (from craftbook "inventory-health-check-recipe") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., out/inventory-report.json does not exist yet. Last failure: craftbook "inventory-health-check-recipe" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate |
| r1-gemma-llamacpp-complex-work-craftbook-author-gate-script | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: ricardo/81354b72 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r1-qwen-mlx-complex-work-craftbook-author-gate-script | 3.2 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 603s — model busy but not delivering (hard digest stuck at c8cdb8f8599097da) |
| r1-gemma-mlx-complex-work-craftbook-author-fanout | 2.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: tuva/632cef98 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r1-gemma-llamacpp-complex-work-craftbook-author-fanout | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "craftbook-author-fanout:6:targetnone:fr1tjwffw:rp0:rf0:m1" stuck for 18m — artifact scored but stalled 18m despite 32 tool calls (2 re-writes). Artifact produced but never reached success (retry-loop nudge was sent and ignored). |
| r1-qwen-mlx-complex-work-craftbook-author-fanout | 3.9 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-author-fanout:7:targetnone:fr1cygdrv:rp0:rf0:m2" stuck for 8m — 4 re-writes (23 tool calls) without sniff movement. Artifact produced but never reached success. |
| r1-qwen-llamacpp-complex-work-craftbook-author-fanout | 3.6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "craftbook-author-fanout:6:targetnone:fr1tjwffw:rp0:rf0:m1" stuck for 18m — artifact scored but stalled 18m despite 30 tool calls (2 re-writes). Artifact produced but never reached success. |
| r2-gemma-mlx-core-petshop | 1.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on index.html; missing signals stayed unchanged: working-image, image-asset. |
| r2-gemma-mlx-core-failing-tests-spec | 6.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on src/machine.ts; missing signals stayed unchanged: vitest-green. Last failure: vitest run failed (exit 1): × submit moves draft to pending and returns the new state 3ms \| × a rejected send leaves the state unchanged 0ms \| FAIL tests/machine.test.ts > order machine: transitions > submit moves draft to pending and returns the new state \| AssertionError: expected 'draft' to be 'pending' // Object.is equality \| FAIL tests/machine.test.ts > order machine: invalid transitions > a rejected send leaves the state unchanged |
| r2-gemma-mlx-core-data-wrangle | 3.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on out/customers.json; missing signals stayed unchanged: dates-iso. Last failure: signupDate "signup_date" (email email) is not a valid ISO yyyy-mm-dd date |
| r2-gemma-mlx-core-incident-postmortem | 6.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on postmortem.md; missing signals stayed unchanged: file-present, all-sections, grounded-core-facts, action-items-formatted. Last failure: postmortem.md is 3292B (need ≥ 6 KB) |
| r2-gemma-mlx-core-conflict-synthesis | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on synthesis.md; the same scenario failure stayed unchanged. Last failure: - conflict (launch date) is not source-bound: re-read memo-product.md and memo-engineering.md, then state each file's observed launch date field in one Conflicts entry and explain which source controls it - conflict (launch budget): the entry names both values and their files but never says which source controls; add a sentence saying why finance.csv wins (for example, that it is authoritative or supersedes memo-product.md) - conflict (launch DRI) is not source-bound: re-read memo-oldplan.md and org.md, then state each file's observed launch DRI field in one Conflicts entry and explain which source controls it |
| r2-gemma-mlx-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r2-gemma-llamacpp-productivity-docblocks-theme-roundtrip | 5.9 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 302s — daemon reachable but issuing no model turns (soft digest stuck at 2d788d50ec8c0eeb) (re-engage nudge was sent and ignored) |
| r2-qwen-mlx-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r2-qwen-llamacpp-productivity-docblocks-theme-roundtrip | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r2-qwen-mlx-productivity-wikipedia-research-brief | 6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on research-brief.md; missing signals stayed unchanged: mcp-research-provenance. Last failure: mock wikipedia: MCP tool "search_wikipedia" was never called; writing about it does not count — "search_wikipedia" is on your tool roster, call it like any other tool and use its response in the deliverable |
| r2-gemma-mlx-productivity-wikipedia-research-brief | 4.2 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on research-brief.md; missing signals stayed unchanged: word-band, ordered-sections, notes-a-g, not-completed, claim-calibration, no-false-completion, mcp-research-provenance. Last failure: mock wikipedia: MCP tool "search_wikipedia" was never called; writing about it does not count — "search_wikipedia" is on your tool roster, call it like any other tool and use its response in the deliverable; also: research brief is 428 words; required range is 700-1,500 |
| r2-qwen-llamacpp-productivity-craftbook-week-plan | 0.8 | model | timeout | [postmortem](scorecard-details.json) | hit hard ceiling (2700000ms + 45m progress extensions) — last hard progress 60s before the end; the deliverable never closed |
| r2-gemma-mlx-productivity-craftbook-week-plan | 1.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: thora/ad227b2e exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model did not emit a required tool call after 2 corrective retries. No tool ran. |
| r2-qwen-mlx-productivity-craftbook-week-plan | 1.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: metehan/bd4712be exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model did not emit a required tool call after 2 corrective retries. No tool ran. |
| r2-gemma-mlx-productivity-craftbook-ab-test-readout | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: sienna/d5c356bd exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model did not emit a required tool call after 2 corrective retries. No tool ran. |
| r2-gemma-llamacpp-productivity-craftbook-ab-test-readout | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: amelia/810c10e2 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r2-gemma-mlx-developer-dev-craftbook-routing | 6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (long-path): sniff "dev-craftbook-routing:1:targetnone:frh4yizi:rp0:rf0" stuck for 12m — team idle (4 new tool calls) with no sniff movement for 12m. Artifact produced but never reached success. |
| r2-gemma-llamacpp-developer-dev-craftbook-routing | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection (engineering); missing signals stayed unchanged: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it. Last failure: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it |
| r2-gemma-mlx-developer-craftbook-deep-security-review | 6 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-deep-security-review stale no-write repair loop: security-review/REPORT.md had unchanged content (2679 bytes) for 61s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 8/11: security-review/REPORT.md is missing required content: real file and line — nothing in its 2679 bytes matches /src/admin\.ts:[0-9]+/. Add that content. |
| r2-gemma-llamacpp-developer-craftbook-deep-security-review | 5.8 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 9 while the failing detail kept changing; latest checked file security-review/findings.json. Last failure: security-review/findings.json: record 0 is missing required field "theme" |
| r2-qwen-mlx-developer-craftbook-deep-security-review | 6 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-deep-security-review stale no-write repair loop: security-review/findings.json had unchanged content (2697 bytes) for 61s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 10/11: security-review/findings.json: record 0 is missing required field "theme" |
| r2-gemma-llamacpp-developer-large-pr-review | 3.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on pr-review.md; missing signals stayed unchanged: coverage ledger names 14/120 exact paths. Last failure: coverage ledger names 14/120 exact paths |
| r2-gemma-mlx-developer-large-pr-review | 3.3 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "large-pr-review:3:targetpr-review.md:fr1i0sit8:rp0:rf0" stuck for 13m — 3 re-writes (5 tool calls) without sniff movement. Artifact produced but never reached success. |
| r2-qwen-mlx-developer-craftbook-codemod-sweep | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-codemod-sweep:29:targetnone:frjr7vze:rp0:rf0:m5" stuck for 15m — 3 re-writes (18 tool calls) without sniff movement. Artifact produced but never reached success. |
| r2-qwen-llamacpp-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-codemod-sweep:30:targetnone:frjr7vze:rp0:rf0:m5" stuck for 11m — 3 re-writes (10 tool calls) without sniff movement. Artifact produced but never reached success. |
| r2-gemma-mlx-developer-craftbook-codemod-sweep | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: obiageli/67e79a32 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model did not emit a required tool call after 2 corrective retries. No tool ran. |
| r2-gemma-llamacpp-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 30 while the failing detail kept changing; latest checked file workspace output. Last failure: task sourced from craftbook codemod-sweep has not reached a terminal step |
| r2-qwen-llamacpp-developer-craftbook-refactor-module | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: kartini/2411b9cd exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r2-gemma-mlx-developer-craftbook-refactor-module | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook refactor-module has not reached a terminal step. Last failure: task sourced from craftbook refactor-module has not reached a terminal step |
| r2-gemma-llamacpp-developer-craftbook-refactor-module | 5 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-refactor-module stale no-write repair loop: tasks/1/verification.md had unchanged content (1025 bytes) for 567s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 21/30: tasks/1/verification.md is missing required content: verification sections in order — nothing in its 1025 bytes matches /^##\s+Before\s*/\s*after[\s\S]*^##\s+Suite[\s\S]*^##\s+Result/. Add that content. |
| r2-qwen-mlx-developer-craftbook-refactor-module | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook refactor-module has not reached a terminal step. Last failure: task sourced from craftbook refactor-module has not reached a terminal step |
| r2-gemma-mlx-developer-craftbook-bug-fix-tdd | 5.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 23 while the failing detail kept changing; latest checked file workspace output. Last failure: tests/pricing.test.mjs did not pass when run with node: exit=1 Command failed: /usr/local/bin/node tests/pricing.test.mjs ✖ exactly 10 units should get the bulk discount (1.004875ms) ✔ small orders pay full price (0.118958ms) ✔ large orders get the bulk discount (0.065542ms) ✔ rejects bad quantities (0.133417ms) ℹ tests 4 ℹ suites 0 ℹ pass 3 ℹ fail 1 ℹ cancelled 0 ℹ skipped 0 ℹ todo 0 ℹ duration_ms 5.58225 ✖ failing tests: test at tests/pricing.test.mjs:5:1 ✖ exactly 10 units should get the bulk discount (1.004875ms) AssertionError [ERR_ASSERTION]: Expected values to be strictly equal: 80 !== 68 at TestContext.<anonymous> <sandbox>/tests/pricing.test.mjs:7:10) at Test.runInAsyncScope (node:a |
| r2-gemma-llamacpp-developer-craftbook-bug-fix-tdd | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook bug-fix-tdd has not reached a terminal step, task-notes.md is missing required content: DONE note names the real test command — nothing in its 17423 bytes matches /\bDONE\b[\s\S]*(npm\s+run\s+test\|node\s+--test)/. Add that content.. Last failure: task sourced from craftbook bug-fix-tdd has not reached a terminal step |
| r2-qwen-mlx-developer-craftbook-bug-fix-tdd | 4.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook bug-fix-tdd has not reached a terminal step. Last failure: task sourced from craftbook bug-fix-tdd has not reached a terminal step |
| r2-qwen-llamacpp-developer-craftbook-bug-fix-tdd | 3.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 24 while the failing detail kept changing; latest checked file workspace output. Last failure: tests/pricing.test.mjs did not pass when run with node: exit=1 Command failed: /usr/local/bin/node tests/pricing.test.mjs ✔ small orders pay full price (0.302875ms) ✔ large orders get the bulk discount (0.046209ms) ✔ rejects bad quantities (0.076333ms) ✖ orders of exactly 10 get the bulk discount (0.308084ms) ℹ tests 4 ℹ suites 0 ℹ pass 3 ℹ fail 1 ℹ cancelled 0 ℹ skipped 0 ℹ todo 0 ℹ duration_ms 3.060333 ✖ failing tests: test at tests/pricing.test.mjs:17:1 ✖ orders of exactly 10 get the bulk discount (0.308084ms) AssertionError [ERR_ASSERTION]: Expected values to be strictly equal: 80 !== 68 at TestContext.<anonymous> <sandbox>/tests/pricing.test.mjs:18:10) at Test.runInAsyncScope (node:asyn |
| r2-gemma-llamacpp-complex-work-craftbook-find-vs-create | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection; missing signals stayed unchanged: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project. Last failure: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project |
| r2-gemma-mlx-complex-work-craftbook-find-vs-create | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection; missing signals stayed unchanged: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project. Last failure: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project |
| r2-gemma-llamacpp-complex-work-craftbook-edit-midtask | 4.3 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 302s — daemon reachable but issuing no model turns (soft digest stuck at fdfe73ba0aac6921) (re-engage nudge was sent and ignored) |
| r2-qwen-llamacpp-complex-work-craftbook-invoice-run | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook invoice-run has not reached a terminal step. Last failure: task sourced from craftbook invoice-run has not reached a terminal step |
| r2-gemma-mlx-complex-work-craftbook-invoice-run | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook invoice-run has not reached a terminal step. Last failure: task sourced from craftbook invoice-run has not reached a terminal step |
| r2-qwen-mlx-complex-work-craftbook-invoice-run | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook invoice-run has not reached a terminal step. Last failure: task sourced from craftbook invoice-run has not reached a terminal step |
| r2-gemma-mlx-complex-work-craftbook-author-linear | 3.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: riccardo/75b99a9f exhausted its bounded automatic recovery allowance at the current checked progress; last error: [Mac AI] aborting — the gezel emitted 12010 characters of prose this turn without calling any action tool. Stop planning. Your next message must START with a single tool call — or, if the work is genuinely finished and nothing is left to do, be ONE short sentence saying so and nothing else. This ses |
| r2-gemma-llamacpp-complex-work-craftbook-route-multi | 5.3 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 303s — daemon reachable but issuing no model turns (soft digest stuck at e5fe35be34a442c0) (re-engage nudge was sent and ignored) |
| r2-qwen-mlx-complex-work-craftbook-route-multi | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-route-multi:10:targetnone:fr1mhd7ue:rp0:rf0:m2" stuck for 17m — 3 re-writes (33 tool calls) without sniff movement. Artifact produced but never reached success (retry-loop nudge was sent and ignored). |
| r2-qwen-mlx-complex-work-craftbook-export-generalize | 3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: supplier intake recipe; missing signals stayed unchanged: task supplier-intake/1 (from craftbook "quarterly-supplier-intake") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., out/q2-intake.md does not exist yet — write the Q2 intake report there by running the saved recipe over data/suppliers-q2.json, out/q2-clean.json does not exist yet — the cleaned Q2 supplier records are the recipe's second deliverable. Last failure: task supplier-intake/1 (from craftbook "quarterly-supplier-intake") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count. |
| r2-gemma-mlx-complex-work-craftbook-export-generalize | 4.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: supplier intake recipe; missing signals stayed unchanged: task supplier-intake/1 (from craftbook "supplier-intake-procedure") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., out/q2-intake.md does not exist yet — write the Q2 intake report there by running the saved recipe over data/suppliers-q2.json, out/q2-clean.json does not exist yet — the cleaned Q2 supplier records are the recipe's second deliverable. Last failure: task supplier-intake/1 (from craftbook "supplier-intake-procedure") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count. |
| r2-gemma-mlx-complex-work-craftbook-author-params | 4.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: gunel/2b7fcd7e exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r2-gemma-llamacpp-complex-work-craftbook-author-params | 4.2 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 601s — engine idle — no token stream or slot update for 76s (hard digest stuck at 16397ada868b2572) |
| r2-qwen-mlx-complex-work-craftbook-author-params | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-author-params:6:targetnone:frp4roae:rp0:rf0" stuck for 11m — 3 re-writes (17 tool calls) without sniff movement. Artifact produced but never reached success. |
| r2-gemma-mlx-complex-work-craftbook-author-gate-script | 5.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: inventory report gate script; missing signals stayed unchanged: no reusable authored craftbook with at least 2 steps exists yet — author the inventory-report craftbook as a reusable template, out/inventory-report.json does not exist yet. Last failure: no reusable authored craftbook with at least 2 steps exists yet — author the inventory-report craftbook as a reusable template |
| r2-gemma-llamacpp-complex-work-craftbook-author-gate-script | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: inventory report gate script; missing signals stayed unchanged: craftbook "inventory-health-check" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate, out/inventory-report.json does not exist yet. Last failure: craftbook "inventory-health-check" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate |
| r2-qwen-mlx-complex-work-craftbook-author-gate-script | 3.7 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 603s — model busy but not delivering (hard digest stuck at a952a8e00ca20d21) |
| r2-gemma-llamacpp-complex-work-craftbook-author-fanout | 3 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 303s — daemon reachable but issuing no model turns (soft digest stuck at ca7a98543aa2408a) (re-engage nudge was sent and ignored) |
| r2-qwen-llamacpp-complex-work-craftbook-author-fanout | 3.6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "craftbook-author-fanout:6:targetnone:fr1tjwffw:rp0:rf0:m1" stuck for 18m — artifact scored but stalled 18m despite 7 tool calls (0 re-writes). Artifact produced but never reached success. |
| r2-gemma-mlx-complex-work-craftbook-author-fanout | 3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: helmi/940c2c74 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r3-gemma-mlx-core-tictactoe | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: yaa/e22a3e3b exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model couldn't apply its file edit after 5 tries, so the turn was stopped. Try sending your message again, or switch to a more capable model in Settings. |
| r3-gemma-llamacpp-core-petshop | 1.7 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "build-me-a-single-page-pet-shop-website/workspace/index.html:3:targetnone:frnone:rp0:rf0" stuck for 8m — 6 re-writes (29 tool calls) without sniff movement. Artifact produced but never reached success. |
| r3-qwen-mlx-core-petshop | 2.2 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on index.html; missing signals stayed unchanged: working-image, image-asset. |
| r3-gemma-mlx-core-petshop | 1.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on index.html; missing signals stayed unchanged: working-image, image-asset. |
| r3-gemma-mlx-core-failing-tests-spec | 6.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on src/machine.ts; missing signals stayed unchanged: vitest-green. Last failure: vitest run failed (exit 1): × submit moves draft to pending and returns the new state 2ms \| × a rejected send leaves the state unchanged 0ms \| FAIL tests/machine.test.ts > order machine: transitions > submit moves draft to pending and returns the new state \| AssertionError: expected 'draft' to be 'pending' // Object.is equality \| FAIL tests/machine.test.ts > order machine: invalid transitions > a rejected send leaves the state unchanged |
| r3-qwen-llamacpp-productivity-conflict-synthesis | 5.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on synthesis.md; the same scenario failure stayed unchanged. Last failure: - conflict (launch budget): a superseded value appears outside the Conflicts section; re-read memo-product.md and finance.csv, then keep the historical value only in its reconciliation entry - unsupported budget claim not present in the seeded sources: re-read finance.csv total_budget and line-item amounts, then replace every claimed amount with a source-backed value |
| r3-gemma-llamacpp-productivity-docblocks-theme-roundtrip | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, recommendation, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r3-qwen-mlx-productivity-docblocks-theme-roundtrip | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r3-qwen-llamacpp-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r3-gemma-mlx-productivity-docblocks-theme-roundtrip | 5.9 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on theme-report.md; missing signals stayed unchanged: slide-count, compare-cited, docblocks-provenance. Last failure: infer_theme_from_file never succeeded; call it — describing the step does not count; also: save the themed deck to deliverables/quarterly-review.pptx so its slide count can be reported |
| r3-gemma-mlx-productivity-craftbook-powerpoint-deck | 5.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: imogen/021ee8a5 exhausted its bounded automatic recovery allowance at the current checked progress; last error: File turn ended without a successful workspace mutation after 2 corrective nudge(s). |
| r3-gemma-mlx-productivity-wikipedia-research-brief | 5.8 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on research-brief.md; missing signals stayed unchanged: no-false-completion, mcp-research-provenance. Last failure: mock wikipedia: MCP tool "search_wikipedia" was never called; writing about it does not count — "search_wikipedia" is on your tool roster, call it like any other tool and use its response in the deliverable; also: false claim present: the Analytical Engine was not completed in 1843 and Lovelace did not build it |
| r3-qwen-mlx-productivity-wikipedia-research-brief | 6.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on research-brief.md; missing signals stayed unchanged: mcp-research-provenance. Last failure: mock wikipedia: MCP tool "search_wikipedia" was never called; writing about it does not count — "search_wikipedia" is on your tool roster, call it like any other tool and use its response in the deliverable |
| r3-gemma-mlx-productivity-craftbook-week-plan | 5.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: mock calendar: 0/1 authorized request(s) seen; writing about it does not count — actually call the service (run the probe script listed in mocks/services.md via run_installed_script, or use http.authed), then update the report from the response, mock calendar: no authorized request matched ^/api/events; writing about it does not count — actually call the service (run the probe script listed in mocks/services.md via run_installed_script, or use http.authed), then update the report from the response. Last failure: mock calendar: 0/1 authorized request(s) seen; writing about it does not count — actually call the service (run the probe script listed in mocks/services.md via run_installed_script, or use http.authed), then update the report from the response |
| r3-qwen-llamacpp-productivity-craftbook-week-plan | 0.8 | model | timeout | [postmortem](scorecard-details.json) | hit hard ceiling (2700000ms + 45m progress extensions) — last hard progress 136s before the end; the deliverable never closed |
| r3-gemma-llamacpp-productivity-craftbook-ab-test-readout | 1.5 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 304s — daemon reachable but issuing no model turns (soft digest stuck at 768f6b9bfd8f56e9) (re-engage nudge was sent and ignored) |
| r3-gemma-llamacpp-developer-dev-craftbook-routing | 4.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection (engineering); missing signals stayed unchanged: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it. Last failure: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it |
| r3-qwen-llamacpp-developer-dev-craftbook-routing | 4.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection (engineering); missing signals stayed unchanged: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it. Last failure: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it |
| r3-gemma-mlx-developer-dev-craftbook-routing | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection (engineering); missing signals stayed unchanged: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it. Last failure: no task has been started from a library recipe yet — search the recipe library for the one that matches this brief (the cause is known, nothing is urgent, and a failing test is wanted before the fix) and start the task from it |
| r3-gemma-llamacpp-developer-craftbook-code-review | 2.5 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 302s — daemon reachable but issuing no model turns (soft digest stuck at 1a6b57c0d3c8d49c) (re-engage nudge was sent and ignored) |
| r3-gemma-llamacpp-developer-craftbook-deep-security-review | 6.3 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted (score plateau): 6 completed repairs with the scenario score frozen at 9 while the failing detail kept changing; latest checked file security-review/findings.json. Last failure: security-review/findings.json: record 0 is missing required field "title" |
| r3-gemma-mlx-developer-craftbook-deep-security-review | 6.1 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-deep-security-review stale no-write repair loop: security-review/REPORT.md had unchanged content (1508 bytes) for 61s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 6/11: security-review/REPORT.md is missing required content: real file and line — nothing in its 1508 bytes matches /src/admin\.ts:[0-9]+/. Add that content. |
| r3-gemma-mlx-developer-large-pr-review | 2.8 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "large-pr-review:4:targetpr-review.md:fr1i0sit8:rp0:rf0" stuck for 18m — artifact scored but stalled 18m despite 3 tool calls (2 re-writes). Artifact produced but never reached success. |
| r3-gemma-llamacpp-developer-large-pr-review | 2.8 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "large-pr-review:5:targetpr-review.md:fr1i0sit8:rp0:rf0" stuck for 18m — artifact scored but stalled 18m despite 3 tool calls (0 re-writes). Artifact produced but never reached success. |
| r3-qwen-llamacpp-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-codemod-sweep:30:targetnone:frjr7vze:rp0:rf0:m6" stuck for 8m — 4 re-writes (13 tool calls) without sniff movement. Artifact produced but never reached success. |
| r3-gemma-mlx-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: pauline/21e8ec87 exhausted its bounded automatic recovery allowance at the current checked progress; last error: [Mac AI] aborting — the gezel emitted 5023 characters of prose this turn without calling any action tool. Stop planning. Your next message must START with a single tool call — or, if the work is genuinely finished and nothing is left to do, be ONE short sentence saying so and nothing else. This step |
| r3-gemma-llamacpp-developer-craftbook-codemod-sweep | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-codemod-sweep:30:targetnone:frjr7vze:rp0:rf0:m5" stuck for 8m — 8 re-writes (39 tool calls) without sniff movement. Artifact produced but never reached success (retry-loop nudge was sent and ignored). |
| r3-gemma-mlx-developer-craftbook-refactor-module | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-refactor-module stale no-write repair loop: tasks/1/verification.md had unchanged content (512 bytes) for 61s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 22/30: tasks/1/verification.md is missing required content: verification sections in order — nothing in its 512 bytes matches /^##\s+Before\s*/\s*after[\s\S]*^##\s+Suite[\s\S]*^##\s+Result/. Add that content. |
| r3-gemma-llamacpp-developer-craftbook-refactor-module | 5.3 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-refactor-module:25:targetnone:frabc232:rp0:rf0:m3" stuck for 8m — 17 re-writes (17 tool calls) without sniff movement. Artifact produced but never reached success. |
| r3-qwen-mlx-developer-craftbook-refactor-module | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook refactor-module has not reached a terminal step. Last failure: task sourced from craftbook refactor-module has not reached a terminal step |
| r3-qwen-llamacpp-developer-craftbook-refactor-module | 5.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: ebrima/fae57ec2 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r3-gemma-llamacpp-developer-craftbook-bug-fix-tdd | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-bug-fix-tdd stale no-write repair loop: tasks/1/validation.md had unchanged content (743 bytes) for 167s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 24/28: tasks/1/validation.md is missing required content: validation sections in order — nothing in its 743 bytes matches /^##\s+Repro re-run[\s\S]*^##\s+Suite[\s\S]*^##\s+Result/. Add that content. |
| r3-qwen-mlx-developer-craftbook-bug-fix-tdd | 5.2 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on workspace output; missing signals stayed unchanged: task sourced from craftbook bug-fix-tdd has not reached a terminal step. Last failure: task sourced from craftbook bug-fix-tdd has not reached a terminal step |
| r3-qwen-llamacpp-developer-craftbook-bug-fix-tdd | 4.3 | model | model-stuck | [postmortem](scorecard-details.json) | craftbook-bug-fix-tdd stale no-write repair loop: tasks/1/validation.md had unchanged content (2498 bytes) for 275s — a re-emission of identical bytes counts as unchanged here after repair feedback and newer chat activity; still failing 21/28: tasks/1/validation.md is missing required content: validation sections in order — nothing in its 2498 bytes matches /^##\s+Repro re-run[\s\S]*^##\s+Suite[\s\S]*^##\s+Result/. Add that content. |
| r3-gemma-mlx-developer-craftbook-bug-fix-tdd | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-bug-fix-tdd:26:targetnone:frnsn9x5:rp0:rf0:m3" stuck for 8m — 3 re-writes (7 tool calls) without sniff movement. Artifact produced but never reached success. |
| r3-qwen-llamacpp-complex-work-craftbook-find-vs-create | 2.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook selection; missing signals stayed unchanged: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project. Last failure: no task sourced from the bundled "seo-meta-pack" craftbook exists yet — search the recipe library for the page-metadata recipe and invoke it on this project |
| r3-gemma-mlx-complex-work-craftbook-edit-midtask | 5.1 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on task quarterly-brief/1 craftbook; missing signals stayed unchanged: task quarterly-brief/1 has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count.. Last failure: task quarterly-brief/1 has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count. |
| r3-gemma-llamacpp-complex-work-craftbook-author-linear | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: CSV Order Cleanup; missing signals stayed unchanged: task order-intake-cleanup/3 (from craftbook "csv-order-cleanup") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., out/report.md is 0 bytes, need ≥ 80. Last failure: task order-intake-cleanup/3 (from craftbook "csv-order-cleanup") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count. |
| r3-qwen-mlx-complex-work-craftbook-author-linear | 3.4 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 600s — model busy but not delivering (hard digest stuck at 672fc0fef506fe14) |
| r3-qwen-llamacpp-complex-work-craftbook-author-linear | 3.9 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-author-linear:6:targetnone:fr1bjnkm5:rp0:rf0:m2" stuck for 15m — 3 re-writes (22 tool calls) without sniff movement. Artifact produced but never reached success. |
| r3-gemma-mlx-complex-work-craftbook-author-linear | 3.6 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: CSV Order Cleanup; missing signals stayed unchanged: task order-intake-cleanup/2 (from craftbook "csv-order-cleanup") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count., notes/anomalies.md is 0 bytes, need ≥ 80, out/report.md is 0 bytes, need ≥ 80, out/orders.json does not exist yet. Last failure: task order-intake-cleanup/2 (from craftbook "csv-order-cleanup") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count. |
| r3-qwen-mlx-complex-work-craftbook-route-multi | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on out/press-release.md; missing signals stayed unchanged: no task has been started from the bundled "press-release" recipe yet — search the recipe library for the one that covers writing an announcement for journalists and invoke it on this project. Last failure: no task has been started from the bundled "press-release" recipe yet — search the recipe library for the one that covers writing an announcement for journalists and invoke it on this project |
| r3-gemma-mlx-complex-work-craftbook-route-multi | 5.5 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (stall-path): sniff "craftbook-route-multi:10:targetnone:frhoh0bm:rp0:rf0" stuck for 18m — artifact scored but stalled 18m despite 37 tool calls (6 re-writes). Artifact produced but never reached success (retry-loop nudge was sent and ignored). |
| r3-gemma-mlx-complex-work-craftbook-export-generalize | 3.5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: reza/ec118226 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r3-gemma-llamacpp-complex-work-craftbook-export-generalize | 2.9 | model | chat-stalled | [postmortem](scorecard-details.json) | chat stalled for 303s — daemon reachable but issuing no model turns (soft digest stuck at dc14abe488b10b8d) (re-engage nudge was sent and ignored) |
| r3-gemma-mlx-complex-work-craftbook-author-params | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: giacomo/13a43438 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model got stuck repeating the same action without making progress, so the turn was stopped. Try sending your message again, or rephrase your request. |
| r3-gemma-llamacpp-complex-work-craftbook-author-params | 5 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: regional sales rollup; missing signals stayed unchanged: no reusable authored craftbook with at least 2 steps exists yet — author the regional sales rollup as a reusable craftbook template, not as ad-hoc chat work. Last failure: no reusable authored craftbook with at least 2 steps exists yet — author the regional sales rollup as a reusable craftbook template, not as ad-hoc chat work |
| r3-qwen-mlx-complex-work-craftbook-author-params | 4.6 | infra | model-stuck | [postmortem](scorecard-details.json) | repair-aborted: xanthi/5fcb2dd5 exhausted its bounded automatic recovery allowance at the current checked progress; last error: The model's `write_artifact` calls failed 5 times in a row, so the turn was stopped. Try sending your message again. |
| r3-gemma-llamacpp-complex-work-craftbook-author-gate-script | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: inventory report gate script; missing signals stayed unchanged: craftbook "inventory-health-check-recipe" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate, task inventory-health-check/1 (from craftbook "inventory-health-check-recipe") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count.. Last failure: craftbook "inventory-health-check-recipe" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate |
| r3-qwen-mlx-complex-work-craftbook-author-gate-script | 3.2 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 603s — model busy but not delivering (hard digest stuck at fb32cb4c8ad20df7) |
| r3-qwen-llamacpp-complex-work-craftbook-author-gate-script | 3.7 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 603s — model busy but not delivering (hard digest stuck at 83afa60a9c8388b8) |
| r3-gemma-mlx-complex-work-craftbook-author-gate-script | 4.7 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: inventory report gate script; missing signals stayed unchanged: craftbook "inventory-health-check-recipe" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate, task inventory-health-check/2 (from craftbook "inventory-health-check-recipe") has status "active" — advance THIS task to a terminal step with advance_task_step/set_task_status. Do NOT invoke the recipe again or create a second task: a new task starts the work over and does not count.. Last failure: craftbook "inventory-health-check-recipe" has no step gated by a scope:"craftbook" script — attach the custom check script to the build step's gate |
| r3-qwen-mlx-complex-work-craftbook-author-fanout | 3 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 721s — model busy but not delivering (hard digest stuck at dcd09960bfecf762) |
| r3-qwen-llamacpp-complex-work-craftbook-author-fanout | 5.1 | model | model-stuck | [postmortem](scorecard-details.json) | no real progress for 724s — model busy but not delivering (hard digest stuck at 4acfabedef8a1c30) |
| r3-gemma-mlx-complex-work-craftbook-author-fanout | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | repair-exhausted: bounded scenario feedback exhausted after 4 delivered-and-completed attempts on craftbook: store health sweep; missing signals stayed unchanged: craftbook "store-health-sweep" declares no per-item fanout over the store list — the recipe must derive its per-store work from the store list itself (one item, one run) rather than repeating steps once per store. Last failure: craftbook "store-health-sweep" declares no per-item fanout over the store list — the recipe must derive its per-store work from the store list itself (one item, one run) rather than repeating steps once per store |
| r3-gemma-llamacpp-complex-work-craftbook-author-fanout | 5.4 | model | model-stuck | [postmortem](scorecard-details.json) | retry loop (fast-path): sniff "craftbook-author-fanout:7:targetnone:fr12v2qjn:rp0:rf0:m0" stuck for 8m — 3 re-writes (20 tool calls) without sniff movement. Artifact produced but never reached success. |


Publication note: this is a portable copy of the completed campaign report. Individual trial evidence links open scorecard-details.json, which embeds all postmortems and retains original archive references. Statements about publication status describe the original completion checkpoint.