Item 26: Admissibility red team, rerun under skeleton v2

The admissibility work bound to skeleton v1 and v2 changes what it is about. A restriction on the conduct set is now nameable as a capability assignment, so the red team's candidate families must be re-asked one question: **are they κ-statements in disguise?** A family that is really a capability restriction is not an admissibility criterion, and treating it as one hides an architectural assumption inside what looks like a legality condition.

Carry the standing constraints unchanged: a candidate must exclude the quote-responsive diagonal, retain ordinary realized conduct and a meaningful fully-updated comparator, permit intended advisory influence, resist laundering through semantically equivalent intermediates, and leave the trust-forcing proof machinery itself admissible. The last is the one that bites. Wave 1's two-sortedness finding — settlements restricted differently from weights, selections and schedules — is the substantive content and should be re-tested against the execution layer rather than assumed to carry over.

Deliverable shape: for each candidate family, a verdict on whether it is expressible as a capability assignment, with a witness either way; witness-checked or enumeration-verified. Acceptance check: the house checker accepts the submitted instances at the declared class. Context: projects/deference/notes/FINITE_MODEL_SKELETON.md §4a and §8.2; prompts/2026-08-11-phase-ii-authority/REPORT.md §9.2; the wave-1 red-team round. A solution ships: the per-family verdicts, and an explicit statement of which previously proposed admissibility conditions are now reclassified as architectural.