Run independent adversarial demo-integrity review and close blockers

Outcome

An independent model and cold reviewer attempt to expose scripting, overclaiming, mandate bypass, confusing tools, or brittle failure handling before a high-stakes demonstration.

Scope

• Review the deployed build, merged spec/rulings, rehearsal evidence, setup skill, console, run banners, private reliability artifact, and fallback behavior.
• Try unsupported inputs, no matching inventory, prohibited candidates, ambiguous phrasing, stale search/selection evidence, provider outages, monitoring failures, revocation races, and hostile questions.
• Attempt to make Farewright rank/recommend and to make the model bypass execution guards with supplied arithmetic or a partial over-cap plan.
• Compare every spoken/UI claim with an observable event, provider mode, or M0-F evidence.

Acceptance

• Reviewer did not implement the reviewed behavior or author its fixes.
• No facilitator steering, preselected money values, hidden relaxed constraint, Farewright ranking, or invented provider outcome remains.
• Complete/no-match/unavailable search, autonomous/approval/prohibited candidate, execution-refusal, and labeled-fallback branches are accurately represented.
• Search evaluation and CandidateSelected are visibly non-authorizing; current execution guards prevail after drift.
• The reviewer cannot use #34 to derive or suggest user inputs; any such path is blocking.
• Every blocking finding is fixed and independently re-reviewed or explicitly accepted by the product owner.
• Product owner records the final go/no-go decision and residual demo risks.

Dependencies

• #22 security/effects gate, #31 live/outage rehearsals, #34 private reliability artifact, and #32 cold-user rehearsals.

Review

This issue is the mandatory stage 6 demo-integrity gate artifact.

Start with the deployed build, merged spec and rulings, rehearsal evidence, setup skill, console, run banners, private reliability artifact, and fallback behavior. Exercise the listed adversarial and outage branches, then confirm every blocking finding is fixed and independently re-reviewed, with the product owner recording the final go/no-go decision and residual risks.