CYCLE 3 - Execution Template v0.1

Date: 2026-04-18Usage: operational checklist before/during/after running benchmarks #2.

1) Pre-run checklist

Manifest benchmark #2 compiled (BENCHMARK_MANIFEST_cycle3_v0_1.md).- [ ] Pre-registered policies frozen (no changes during run).- [ ] Split data fixed and versioned.- [ ] Seed runs saved.- [ ] Baseline (P_cls) and ordinal (P_ord) pipelines blocked.- [ ] Active contexts: Omega_A, Omega_B, Omega_C.

2) Minimum log per run

For each run:- theorem_id: D02 / A01 / D03- benchmark_id- omega_context- dataset_version- seed- pipeline_version- primary metrics (Coh, Phi, Delta)- auxiliary metrics (Err_sem, Loss_U, other specific metric)- start/end timestamp- outcome (ok / error)

3) Decision table (draft)

D02

PASS candidate if non-empty class Phi=0 exists in at least 2 out of 3 contexts.- REVIEW if class Phi=0 appears in only one context.- REJECT candidate if no context has Phi=0.

A01

PASS candidate if both advantages (Delta_cum, Err_sem) hold up in at least 2 out of 3 contexts.- REVIEW if only one metric holds.- REJECT candidate if no metric holds.

D03

PASS candidates if conservative and degenerative regimes coexist in at least 2 out of 3 contexts and ambiguous rate < 0.20.- REVIEW if ambiguous odds >= 0.20 or if coexistence is fragile.- REJECT candidate if only one regime emerges in a dominant/stable manner.

4) Mandatory post-run

Fill in results_D02_cycle3_v0_1.md- [ ] Fill in results_A01_cycle3_v0_1.md- [ ] Fill in results_D03_cycle3_v0_1.md- [ ] Fill in CYCLE_3_REPORT.md- [ ] Note anomalies and limitations- [ ] Define explicit gate towards Cycle 4