Executive summary
Claim-selective evidence upgrade
The existing canonical reproduction was retained for claims where it already had equal or stronger evidence. The following weaker claim pages were replaced with stronger current public evidence:
- Claim 4: ai-sherpa/source-screening-repro (falsified)
- Claim 5: snaykey/repro-source-screening-shared-feature-extractors (verified)
This canonical logbook independently audits all five supplied claims for OpenReview dTMrITkTr5. The main result is now supported by 1,520 seeded split-local-averaging fits, including a new high-dimensional stress grid up to d=320 and N=51,200; the empirical error stays within 1.884–2.449 × sqrt(d/(N lambda_k)). The remaining pages preserve a deterministic theorem-condition counterexample, an executable algorithm-constant audit, a release audit for the unavailable real-data pipeline, and a 500-seed screened-versus-full comparison.
Scope & cost
| Scope | Hardware | Cost |
|---|---|---|
| 1,520 rate fits + 500-seed estimator comparison + exact theorem/algorithm audits | Local CPU | under 30 CPU-min; no paid compute |
The audit now includes an explicit theorem-to-algorithm ledger and release-boundary ledger. Claim 1 remains the positive reproduction. Claims 2 and 3 are source-level wording/specification contradictions rather than failed stochastic experiments. Claim 4 remains source-verified but experimentally indeterminate, and Claim 5 remains a well-powered negative robustness check—not an exact implementation match. These distinctions prevent paper values, proxy estimators, and theorem statements from being conflated.
<!-- poster_embed.html -->
<iframe src="https://chenruishuo-posterly.hf.space" title="Posterly reproduction poster" style="width:100%;height:720px;border:0" loading="lazy"></iframe>