CYCLE 6 Execution and Reject Rules v0.1

Date: 2026-05-05
Cycle: CYCLE_6_2026-05-05
Scope: A01 only

1) Execution order (fixed)

1. Run governance compliance check.
2. Verify dataset independence proof.
3. Run P_cls on all contexts.
4. Run P_ord on all contexts.
5. Build full trajectory logs.
6. Compute pre-registered metrics.
7. Produce decision artifact.

No permutation of step order is allowed without new preregistration.

2) Hard reject triggers

Any trigger below sets theorem decision to reject_candidate:

1. Reuse of synthetic proxy formula as decisive metric source.
2. Missing context coverage (< 3 contexts executed).
3. Benchmark overlap with cycle3_inputs above allowed limit (> 0 record-level overlap by ID/hash).
4. Missing hash-chain integrity in trajectory logs.
5. Post-hoc threshold/criterion change after first failed evaluation.
6. Role-separation failure (designer == auditor).
7. Gate-entry or gate-exit signature invalid/missing.

3) Soft fail / revise triggers

If no hard reject is triggered, mark revise_needed when:

1. Fewer than 2/3 contexts satisfy process advantage criteria.
2. Matched-output subsets exist in fewer than 2/3 contexts.
3. Metrics are complete but insufficient for promotion.

4) Integrity checks

Required integrity checks:

1. All reports reference exact script-lock hashes.
2. Seal timestamp is earlier than first trajectory event.
3. Decision criteria are evaluated from pre-registered thresholds only.
4. Final report includes explicit non-claim section.

5) Output naming contract

1. results/A01_real_pipeline_metrics_v1_0.json
2. trajectory/A01_trajectory_events.jsonl
3. CYCLE_6_EXECUTION_REPORT_A01_v1_0.md
4. CYCLE_6_DECISION_GATE_v1_0.md
5. CYCLE_6_REPRO_PACK_MANIFEST_v1_0.md