Verification record · AUDIT-DS-STABLE-MARGINS-COMPILE

Publication-grade audit of DS16 stable margins and the compile verdict

Verification recordpublication: internalno direct precedent found

Statement

Independent audit verdict: DS-STABLE-MARGINS-PRICE, DS-PROFILED-COMPILE-CERTIFICATE, and DS-STABLE-STATE-SELECTION are all hardened. PRICE/FUNNEL/FLOOR survive after replacing a pointwise SLLN at a data-dependent centroid limit with a compact tilt-codebook uniform strong law, making the all-labelings event pathwise and measurable, distinguishing v*+(kappa) from v*(kappa), and fixing the raw-label/exact-centering convention. The compile verdict is restricted to the only currently established unconditional registry path; DS14 applies to certified sequences, not one finite diagnostic, and certificate-branch inhabitation remains open. Independent Fraction arithmetic reproduced both fixtures and four exact census instances; an N=100 public-library run reproduced the centered nuisance collapse and non-centered retention. The census range is corrected to 5-944, and 0.004-0.046 is qualified as an aggregate researcher summary rather than a per-run bound. Primary-source search found codebook-rigidity prior art but no direct combined theorem, so search_gap remains a claim-specific conclusion, not novelty.

Assumptions

the audit session did not produce DS16 and froze source commit 1b58518
the exact audit harness is pure stdlib Fraction code and does not import py/ds_stable_margins.py or reuse its classification logic
the theorem verdict is restricted to d_psi = d_lambda = 1, K >= 3, equal weights, exact scores, class (L)+(S), raw fixed-rule labels and centered empirical moments
the measured verdict does not assert an asymptotic basin-selection theorem

Closest prior work

Why each source is not this result: the search records

Nearest sources, checked 30 August 2026

The full search record

Relationships

Evidence references