[pinterbanget]

Adversarial audit of DaysSky's #4596, which quotients PR4470's word by span{511} and claims complex 0.0121 / outer κ 0.0034. No κ is claimed here. This is a scoped finding about the endpoint contract. It does not close #4596.

Evidence (research/pr4596-endpoint-frame-audit/verify.py, standard library, about 2 s; #4596's QUOTIENT-WORD vendored with blob 5bc2bdaf and its SHA256 pinned; PR4470's frames from main's pinned native_joint.py):

• PR4470: mains 0–111 end in F₂⁹, and mains 112–223 end in 112 distinct hyperplanes that avoid 511.
• Conditional κ = 0.0034 (BIT-limited, +12.58%): complex supplier to 0.011417749 via the ambient quotient by span{511} #4596: all 224 mains end in the same frame, F₂⁸. Every main output frame changed. Only the padded helper 224 reaches F₂⁹.
• Why the frames collapse: φ sends F₂⁹, and every 511-free hyperplane, onto F₂⁸.
• Why the replay still passes: the pinned verify_buffers.py compares final frames with the word's own canonical["final"] (line 207), and expected() takes no frame input. "632/632 exact" therefore certifies only the logical map.

Conditional consequence: if the supplier contract is PR4470's endpoint, meaning each main field is delivered in its required chart, the quotient word does not meet it. Its lower paid profile then prices a different operator. Whether output charts are part of the contract is the maintainers' call.

Open question: an operator-level endpoint check in literal replays, which would take the certificate's required final frames as an independent input.

Credit: DaysSky (#4596) and Cinna Davis (#4470). Ryanson Jonathan, with Claude Code (Claude Opus 5.5) assistance.

Head: 6d4d728

🤖 Generated with Claude Code

[DaysSky]

PR edited to reflect this audit, Thanks for contributing!