# Results-to-claim audit report (WP31, 2026-10-30)

* `scripts/verify_manuscript_claims.py`: **AUDIT PASS** — 25 registered claims; every artefact path exists; every run id has a
  manifest (states `completed`, or `stopped-run-used` for the two deliberately stopped runs cited as failures); every numeric
  macro used in the manuscript is generated by `scripts/make_tables.py`; no promotional phrase; no placeholder section.
* Tests: `python -m pytest tests -q` → 23 passed (15 mandated invariants, controller target-freedom with and without the
  residual/entropy features, dense-write ablation, 5 cache-equivalence tests).
* Manuscript: `paper/main.pdf`, 48 pages, latexmk with 0 undefined references/citations (one harmless font-shape warning).
* Pre-registration: `protocol/preregistration.md` frozen v1.0 at 2026-10-30T04:04-0400; `protocol/locked_test_manifest.json`
  unsealed at the same time; test splits were read only afterwards (synthetic test regimes: 5 nodes; FineWeb test: 4 systems).
* Deviations: `protocol/deviations.md` (bootstrap init/scale, controller features, label stream, dense 103M infeasible,
  backbone budget 98.3M instead of 196.6M tokens).
* Not executed: Phase 11 (MLX port), Phase 12 training (459M/1.18B only instantiated on meta), GSM8K/EvalPlus, quantisation, distillation, joint fine-tuning (one degraded attempt), dense editor at 103M
  (collapsed twice), MLX/Metal profiling.
* Checksums of the key artefacts: `reproducibility/checksums.sha256`.