Independently audit #2979's joint rank-seven frame paths

[petalas]

Summary

Adds only research/rank7-joint-frame-audit/, an independent finite audit of #2979 at be191346391661f4d35fc0d1508d0a6029572576.

literal parent + declared frame edits
  five isolated groups and complete stage one
  all 16 V45 endpoint subsets with fixed partner frames
literal final word
  chronological frames, widths and cleanup names
  conditional boundary to the source's complete invoice

Both providers establish the exact V45 identity
Delta(k) = (2-k) H1 +10 H2 +6 H3 -10 H4.
The shared partner overhead is paid once. The source's surrogate margins reproduce losing singles and a winning four-helper group, but the surrogate width guide is not a certified physical tariff. No new supplier or kappa is claimed.

The five isolated group deltas sum to stage one. Cleanup syntax retains 86 private names there and gives 91 in the shipped final word. The source-claimed idle remainder is explicitly conditional, not presented as a fresh owner replay. This complements #2987, which checks chronological packing while assuming the native widths.

Evidence

• Before: Conditional κ ≈ 0.0017290339826465: joint rank-7 end frames, five paired extractions and #2957's rewrites on #2942's words (+5.00e-6 over #2969) #2979's new rank-seven frame paths were author-checked; Independently audit PR2917's paired-helper frame ledger #2932 and Independently verify PR2942 owner-to-recursive-cost bridge #2974 cover older words. This audit targets that changed implication rather than another root calculation.
• After: code-disjoint explicit subsets of GF(2)^9 check both providers, all five groups and 16 V45 subsets. No predecessor Python module is imported or executed. All nine data inputs match the original PR's bytes and have exact Git-blob/SHA256 bindings.
• Registered CPython 3.12.15 replay passes, including 18 frame/syntax rejection controls and five unittest methods. Tests cover normal/optimized equality, a Git-free export and corrupted-source rejection. Optimized CPython 3.13.5 gives byte-identical JSON. Lint, formatting, warnings-as-errors and committed-byte preflight pass. Hosted checks are separate.

python3.12 scripts/research_package.py --package research/rank7-joint-frame-audit

Native residuals/phases, arbitrary-dirty-input and private-zero return, actual physical storage, prefix/precision/tape, outer release and all-size/compiler obligations remain outside this partial audit. The selected frontier and both selected caps are unchanged.

Source credit: Marius-Paul Dumitrean/#2979 and #2942, Jeremy Kothe's frame/cleanup and #2957 lineage, Mason Hale/#2917, Jeremy Diviney/#2908, ToikY/#2932, Ross Richardson's complementary audits, and the retained geometry/native/owner contributors. No independent human or formal-kernel review is claimed. Scope, process disclosure and the concrete next obligation are in TASK.json and the README.

Canonical claim and handoff. Base/current upstream 836b19bfa309ceafbc8ee32a15e8c4b10dde7e11, priority revision 6; full overlap refreshed at 16:34 UTC with no warnings.

Merge danger

Door: two-way. One additive package, removable without changing any supplier.

Blast radius: isolated. No frozen source, shared checker, priority, selection or other PR changes.