# Batch 01 Excluded Runs

Status: `in_progress`

This log records run slots that have completed or been reviewed but are excluded from aggregate scoring before any metric average or method comparison.

## Excluded

| Run ID | Case | Arm | Reason | Source |
| --- | --- | --- | --- | --- |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__agentic_coding__r1` | `case-strategic-gated-diligence-001` | `full_ofone` | Case-fidelity failure: the artifact identity is bound to `case-strategy-micro-001`, not the benchmark case. | `research/results/2026-05-17-06-ofone-batch01-independent-review-result.md` |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1` | `case-strategic-gated-diligence-001` | `full_ofone` | Semantic-validation failure: computed local validator rejects relation legality and option expected-effect references, and the artifact self-attested validator pass conflicts with the computed result. | `benchmarks/reviews/2026-05-17-batch-01/2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1.md` |
| `2026-05-17-batch-01__case-regulated-wastewater-market-entry-001__full_ofone__frontier_reasoning__r1` | `case-regulated-wastewater-market-entry-001` | `full_ofone` | Semantic-validation failure: computed local validator rejects illegal `updates` relations from `option_move` to `unknown`, illegal `supports` from `claim` to `gate`, and an invalid relation family for gate-to-rendering `constrains`. | `benchmarks/reviews/2026-05-17-batch-01/2026-05-17-batch-01__case-regulated-wastewater-market-entry-001__full_ofone__frontier_reasoning__r1.md` |

## Exclusion Rule

Schema-valid is not benchmark-valid. A full-OfOne run must pass the pre-score compliance gate before metric scores can enter aggregate comparison. Any wrong-case artifact identity, copied-example artifact, missing required output, other-arm leakage, or unsupported superiority claim is an auto-reject condition.

## Current Effect

- Direct-answer first-slice run remains aggregate-eligible.
- Light-structured first-slice run remains aggregate-eligible.
- Original full-OfOne first-slice run remains excluded and immutable.
- Remedial full-OfOne rerun 1 is reviewed and aggregate-eligible as a replacement for the excluded original only.
- Frontier full-OfOne strategic repeat-1 remains excluded and immutable.
- Frontier full-OfOne regulated wastewater repeat-1 remains excluded and immutable.
- Remedial frontier full-OfOne rerun 1 was harvested but rejected before matrix insertion because the output was an advisory research report, not a benchmark raw output package.
- Remedial frontier full-OfOne rerun 2 was harvested with the required benchmark raw-output package shape, but computed local validation failed; it remains outside aggregate scoring and outside `remedial_runs`.
- Remedial frontier full-OfOne rerun 3 was harvested with benchmark package sections and machine artifacts, but computed local validation failed current benchmark trace shape and relation-legality checks; it remains outside aggregate scoring and outside `remedial_runs`.
- Remedial frontier full-OfOne rerun 4 was harvested but rejected before matrix insertion because the output was a meta/advisory report, not a benchmark raw output package.
- A same-shape attachment-led Deep Research remedial rerun is now barred for this frontier full-OfOne slot.
- Controlled Mode A rerun 5 passed computed local validation, local review, and matrix insertion as replacement evidence only. It does not mutate the excluded original and does not support aggregate comparison or superiority claims by itself.
- Regulated wastewater controlled Mode A rerun 1 passed computed local validation, local review, and matrix insertion as replacement evidence only. It does not mutate the excluded original and does not support aggregate comparison or superiority claims by itself.
- Three original run slots are excluded before aggregate scoring; three remedial replacements exist under `replace_for_aggregate_only`; no empirical superiority or method-performance claim is supported.

## Remedial Reruns

| Run ID | Replaces | Status | Aggregate Policy | Notes |
| --- | --- | --- | --- | --- |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__agentic_coding__r1__rerun1` | `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__agentic_coding__r1` | `reviewed` | `replace_for_aggregate_only` | Case-native artifact with benchmark trace binding, validator output, rendering, patch report, and local review. |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun1` | `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1` | `failed` | `not_aggregate_eligible` | Raw export preserved with SHA-256 `c0900989fe10e528648ea57d6f20f1f18f662fcff89bb04194bbc99fb5a9d385`, but exact benchmark header, artifact JSON, validator, rendering, and patch sections were absent. |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun2` | `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1` | `failed` | `not_aggregate_eligible` | Raw export preserved with SHA-256 `dfdae1034abf0e0521df5103bfa297c605ac5ef149b4a3f85490f070e7179bd8`; artifact, validator, rendering, patch, and review were generated, but computed local validation failed missing evidence `movement_jobs` fields and a tradeoff reversal-condition defect. Review: `benchmarks/reviews/2026-05-17-batch-01/2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun2.md`. |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun3` | `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1` | `failed` | `not_aggregate_eligible` | Raw export preserved with SHA-256 `b64a604e28a5e26d871dc5bca05e4be33dd0630afbbfca8d3d28cbc579e7db85`; artifact, validator, rendering, patch, and review were generated, but computed local validation failed missing current-schema `benchmark_trace` fields and illegal relation edges `X2`, `X3`, and `X4`. Review: `benchmarks/reviews/2026-05-17-batch-01/2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun3.md`. |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun4` | `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1` | `failed` | `not_aggregate_eligible` | Raw export preserved with SHA-256 `5d4de650a1c2f3f6612b718f45e7313f40433dfaf851afeaad1b3ca0f9dbd702`, but the completed report was a meta/advisory analysis of an unspecified packet and did not include the required benchmark output package. Review: `benchmarks/reviews/2026-05-17-batch-01/2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun4.md`. |
| `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1__rerun5` | `2026-05-17-batch-01__case-strategic-gated-diligence-001__full_ofone__frontier_reasoning__r1` | `reviewed` | `replace_for_aggregate_only` | Controlled Mode A package with raw SHA-256 `2897a11f82baee33a0161a8d8f64b716017c3fedd9d22a2639c60caa6ad4bf86`, artifact SHA-256 `6bd70e20ba36bf9e63a8bc4c65127a3f1166d30590d1a63df63b2612261a14f0`, computed validator JSON, rendering, patch report, and local review. This is replacement evidence only and does not consume a new predeclared slot. |
| `2026-05-17-batch-01__case-regulated-wastewater-market-entry-001__full_ofone__frontier_reasoning__r1__rerun1` | `2026-05-17-batch-01__case-regulated-wastewater-market-entry-001__full_ofone__frontier_reasoning__r1` | `reviewed` | `replace_for_aggregate_only` | Regulated wastewater controlled Mode A rerun 1 with raw SHA-256 `eb8251313f684035e12f5ce9999645c8c1273663549bb2ee0da26021904e7fce`, artifact SHA-256 `99869ab95cb00d04ae611efbf89fbe65257c00ab845e02ef0581bb625794a4ee`, computed validator JSON, rendering, patch report, and local review. This is replacement evidence only and does not consume a new predeclared slot. |

## Rerun Semantics

- Excluded originals are immutable evidence records. Do not edit or delete the original raw output, artifact, validator artifact, patch artifact, review, or independent adjudication.
- The execution matrix declares `rerun_policy.preserve_original_runs=true`.
- Remedial reruns use `rerun_id_template={original_run_id}__rerun{rerun_number}` and carry `rerun_of`, `reason`, `status`, and `aggregate_policy`.
- The first remedial rerun repairs the excluded repeat-1 slot and does not consume repeat 2 or repeat 3.
- The frontier full-OfOne remedial rerun should repair the excluded frontier repeat-1 slot without mutating the original completed output.
- A remedial rerun can replace the excluded original for aggregate scoring only after it passes pre-score compliance, benchmark trace binding, machine-artifact hash checks, and review/adjudication.
- Failed remedial attempts may be recorded as evidence outside `remedial_runs`; they do not increase completion/remedial counters and cannot replace an excluded original.
- Remedial reruns are tracked in `execution-matrix.json` under `remedial_runs`; they do not consume a new predeclared repeat slot.
- Controlled Mode A strategic rerun 5 and regulated wastewater rerun 1 are the current frontier replacement records under `remedial_runs`; failed frontier reruns 1-4 remain immutable process evidence outside replacement eligibility.

## Provenance Binding

The excluded original now carries run-record-level `benchmark_trace` fields for the case file, arm prompt, and input bundle hashes. The failed artifact itself remains unchanged because it is evidence of the Run 06 defect. Future aggregate-eligible `full_ofone` artifacts must carry matching artifact-level `benchmark_trace` values, or the benchmark checker rejects them.