back to table · edit · history · where entries came from · files · download
1810 bytes, as of the version from 2026-10-29 14:26 (current). Recorded here, not run.
# T610 build audit disposition
The API audit was run after filling the draft. Its sole finding asks whether
normalisation=global,total names two quantities and recommends either splitting
or explaining the one-number-in-two-conventions exception in the definition.
The table follows the explicit family boundary in issue 244 and the archived
canonical skill (see skill-evidence.json and the run's skill manifest). The
same positive category invariant is held as G and D=sqrt(G), with G=D^2,
identical category indices and coverage, compact values in both forms, and a
constant-cost conversion. The definition now explicitly says "two invertible
normalisations"; the normalisations comment and argument-selection.md explain
why this is one table. The value header names both G and D, and each row's
normalisation is labelled by the appropriate symbol.
The snapshot's audit_table._one_table_or_several is deliberately structural:
it does not parse this explanation and still emits the question for any small
non-numeric orthogonal index on a draft. Consequently clean=false remains,
with this single explained finding. It is retained for independent critique;
no parameter renaming, numerical relabelling, suppression or extra draft was
used to evade it. There are no other API audit findings. No database-backed
--links audit was available; the linked companion pages and primary references
were read directly.
Numerical validation: verify(sample=None) matched 234/234 entries. At 600 bits,
all full-alcove sums and exact cyclotomic evaluations pass containment in the
written intervals. Live readback checks pass the D^2=G, level–rank, level-one
and SU(2) identities, and all 37 sums from the companion T609. The independent
critique is still required; these are build checks, not an acceptance sidecar.