[Pentexa]

Post-Merge Independent Audit v0.6

Independent post-merge scientific audit of the integrated repository (Research v0.13 + Academy Memory v0.6 + Quality v0.5 + Artifact Gate).

Headline results

• Suites: Python 210/210 · Academy 234/234 · Audit 149/149 (incl. new verifier regression tests) · Rust 32+9 · hygiene OK.
• Manifests (13): 2927 PASS / 0 FAIL / 19 INCONCLUSIVE (memory-limit metadata + generic-format structural items only).
• Artifact gate: PASS for both results/v13 and results/academy_v01/memory_v06.

Research v0.13 — independent recomputation (33/33 PASS)

• Stable exact D1: 16.88% → 39.38% (per-seed [0,0,0.675,0] → [0.775,0,0.8,0]); novel: 23.12% → 33.13% ([0,0.25,0.675,0] → [0.275,0.25,0.8,0]).
• Δ CIs recomputed: stable +39.38 [0.00, 78.75] (includes 0 → directional pilot evidence, correctly worded); novel +33.13 [6.88, 66.25] (excludes 0).
• -gen ablation equals v0.12.1 rows exactly → rolling generator is the sole driver.
• F/H/R preservation: all cells == v0.12.1; -nodc 71.88 reproduced; metargon12 in final_l reproduces the v0.12 study exactly.

Academy Memory v0.6 — independent recomputation (34/34 PASS)

• Fresh 0.3125; original/memory_wiped 0.28125 (bit-identical vectors); controls random_walk 0.0547 / novocab 0.4609; ctx_full 0.5078; atomtx 0.25.
• All paired deltas + bootstrap CIs (seed 20261108, 2000 rounds) reproduce exactly.
• EVREN pool: 24/24 solvable & unique; original frozen 0.40625, Δ vs fresh +0.0944 CI [0.0, 0.1875] (directionally positive, not significant).
• Storage failure confirmed: verified_then_contradicted 56.5–198 on classic variants; mean contradicted ≈67 vs verified ≈3 → zero-shot transfer unproven (absent in measured cells).

Verifier fix (Audit-owned)

• v12_temporal_minus_notp_exact_D1 / temporal_minus_notp_exact_D1 key ambiguity resolved data-driven (unique delta-reproducing pair, family hint, or value-identical candidates); duplicate-roster entries deduped. 5 regression tests added.
• No Research/Academy implementation files touched; no results changed.

Deliverables

• generalization_audit/docs/POST_MERGE_AUDIT_V06.md
• results/verification_v13/post_merge_audit_v06.json (machine-readable)
• results/verification_v13/post_merge_b_recompute.json, post_merge_c_recompute.json
• generalization_audit/tests/test_comparison_key_resolution.py
• Recompute scripts under generalization_audit/tools/post_merge_*.py

No merge performed; please review.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.