Date: 2026-07-27
Source commit: 316f4b7e38ea685b4bd466caa5ebd7d717a65bbe
Outcome#
V9-005 now has one fail-closed admission contract that composes its five external flagship claims. The retained package is blocked on the policy, candidate, cohort, journey, outcome, intervention, beta-gate, and freshness requirements. No fixture or earlier component implementation is treated as a learner study or efficacy result.
Required evidence#
scripts/v9/verify-flagship-evidence.mjs requires:
- an approved and preregistered powered protocol, a fully consented real learner/teacher cohort, an exercised withdrawal path, privacy review, and a signed appointment for an evaluator independent of the optimization team;
- exact-candidate event evidence showing every enrolled learner completed Ask a Wonder, an adaptive lesson, an explorable, and an evidence-grounded revised path;
- at least four weeks of randomized, learner-blinded and teacher-review-blinded controlled measurement against a distinct baseline, with non-compensatory gates for immediate learning, misconception repair, transfer, delayed mastery, week-four retention, usability, trust, and teacher/learner preference;
- actual control-plane teacher/operator inspection, correction, takeover, withdrawal, and rollback, including stale-evidence invalidation, processing cessation and purge, and previous-revision restoration; and
- executed closed-beta, mobile/offline, localization, cost, safety, privacy, scale, canary, and incident gates, including canary rollback and incident resolution before efficacy claims are enabled.
Every artifact is nonempty and SHA-256-bound. The final package must name a full source commit and remain within its approved freshness window.
Retained audit#
The acquisition registry has no V9 learner, teacher, or evaluator cohort. There is no approved powered protocol from which to set participant minima, so the canonical policy deliberately retains null sample thresholds rather than inventing them. The candidate digest is a zero placeholder and all five external evidence groups are absent. Audit mode records every flagship check as false; strict mode exits nonzero.
All five V9-005 ledger boxes remain open. Closing them requires research approval, external recruitment and consent, independent evaluation, deployed learner and operator surfaces, a four-week study, and actual release-control drills.
This boundary changes no user-facing UI, so Playwright, mobile, and UI accessibility automation are not applicable to this repository-only slice.