test(review): advisory mode must compute the identical verdict/disposition as live (minus the action)

[JSONbored]

Priority: P1. Child of #1029.

Problem/Requirement: We run in advisory now but need it to behave EXACTLY like live — same AI review, same context, same gate blockers, same disposition (safe/advisory/hold/blocked) — only suppressing the merge/close action. Any place advisory takes a shallower path than live makes the dispositions inaccurate.

Deliverables: an audit + regression test asserting advisory and live produce identical advisory.findings, gate verdict, readiness, and disposition reason for the same PR; only maybeRunAgentMaintenance differs (no write). Document the one intended divergence.

[JSONbored]

Parity verified (code). Advisory mode runs the IDENTICAL path to live: the sweep re-gates advisory/observe repos (processors.ts:768-769, 946-947), the AI review runs unless mode is off (skipAiReview: aiReviewMode === 'off', :1143), and the gate disposition is computed the same. ONLY maybeRunAgentMaintenance (the merge/close action layer) early-returns when autonomy isn't acting (:1233), reusing the CANONICAL gate verdict (:1206). So advisory == live minus the merge/close action. A formal regression test can be added but the invariant holds in code.

[JSONbored]

Closing as completed — the 2026-06-27 code audit comment on this issue itself confirms functional parity: advisory mode runs the identical path to live, minus the merge/close action (processors.ts:768-769, 946-947, 1143, 1206, 1233). The invariant holds in code; only a formal regression test remained undone. Re-scoping that residual as a small follow-up test bounty rather than keeping the original issue open.