Delivery gate can prescribe a provably identical retry when the basis is already true (post-#549/#550 residual)

Carried over from PR #554's round-2 cross-family review (closed as superseded by #549/#550). The reviewer's live counterexample, run against the #554 branch but structurally applicable to main's gate: with the assignment basis equal to the release-weight recomputation, an ordinary band miss still gets the 're-run with --ssi-take-up-prior-weight-basis' remediation — and that retry regenerates byte-identical flags under the same seed (same priors, same draws), the same miss, and the same advice. An endless no-op ladder.

Main's delivery gate (post-#549/#550) has no conditional-advice machinery (grep finds no basis-comparison in the retry path). Required, adapted to main's gate shape rather than ported from the closed branch:

1. Retry advice only when the release-weight prior recomputation differs from the assignment prior (exact float inequality — PR #554 round 2 proved a 1e-9 tolerance wrong with a 1e-10 counterexample that flips a marginal draw and clears the gate; only float-identical priors guarantee identical selections) for at least one missing enforced band.
2. When every missing band's priors agree: say the basis is already true, that a retry would reproduce identical flags, and direct repair at support/draw realization. Name both priors on every miss line.
3. Per-band suffixes (an equal-prior band must not be told its basis differs) with one run-level remedy; run-level applicability boolean stays.
4. Nonfinite/missing prior fields in a schema-4 row = invalid diagnostics, not a silent 'priors match'.
5. Schema-3 legacy-seed loader: per-band capacity/floor equality pins (the #554 review killed a +1.0 capacity mutation only after these were restored).

Empirical context: attempts 6/7 (populace#507) validated floor-aware + delivered basis (18_64 −0.00%, 65_plus +4.53%); the no-op case becomes reachable in production precisely when the ladder has converged — which is now the expected steady state.

Start in main's delivery-gate retry path, where the payload reports no basis-comparison machinery, and trace its diagnostics and run-level applicability boolean. Check schema-4 prior validation and the schema-3 legacy-seed loader, including per-band capacity and floor equality coverage. Done means advice appears only for exact prior differences, equal-prior bands explain the no-op retry, invalid priors are rejected, and the required pins are tested.