docs(overhaul): contract s5 closeout offline gate

docs/research/2026-09-01-overhaul/s5-consumers-material-closeout-packet.md

@@ -3263,3 +3263,52 @@ cathedral, Facility, house-portal, and outdoor-terrain geometry; only authored
 time/weather changed. Verdict: **PASS, provisional until the owner inspects
 the PNGs.** S5-c5 is closed. S5 automated/performance closeout and G4 remain;
 never merge to `main` before G4.
+
+## 36. S5 closeout C1 contract — offline correctness and evidence map
+
+Base is clean campaign commit
+`29c05e48c8379678603efb8715f86ede04c775ea`. This is an evidence-only
+closeout chunk: it changes no production, test, fixture, register, shader,
+prepared package, or gate script. One OpenAI implementer runs the bounded
+matrix below and returns artifacts plus an exact coverage map; the lead
+independently verifies every count, nonpassing identity, manifest, and claimed
+fixture. The lead alone writes and commits the result.
+
+1. Run `tools/run-release-gate.ps1` once to a new, absent artifact directory
+   with the plan's literal filter
+   `Lane!=InstalledDat&Lane!=Linux&Lane!=Manual&Lane!=Timing&Lane!=Live&Lane!=PreparedPackage`.
+   Do not silently substitute the narrower ordinary release-gate default. A
+   nonpassing test is evidence to classify, not permission to retry or edit.
+2. Run one canonical InstalledDat artifact with
+   `ACDREAM_RUN_INSTALLED_DAT_TESTS=1`, explicit installed DAT directory, and
+   `Lane=InstalledDat&Purpose!=Diagnostic`. Preserve the accepted nine
+   failures plus one look-in skip only if the exact identities match c5; any
+   new or missing identity is a finding.
+3. Produce an exact test-to-requirement map for the focused App, Content,
+   Core, Runtime, and Launcher coverage; all committed OH walk/parts/
+   alpha-depth fixtures; `BeginWalkFrame`/outside-view reassembly; per-cell
+   geometry and membership; FIFO/barrier/order; material; building degrade;
+   visibility consumers; reset/abort/dispose; architecture guards; and real
+   warmed allocation pins. Reuse already-executed tests from items 1--2 where
+   their TRX identity proves coverage; run only a genuinely missing focused
+   class once.
+4. Identify the existing executable witness, if any, for the plan's “two
+   deterministic geometry/membership runs with equal hashes” requirement.
+   Run it exactly twice from the same clean commit and compare its declared
+   product hash. A hash of source files, TRX files, logs, or the prepared pak
+   is not this witness. If no current gate emits a canonical runtime
+   geometry/membership product hash, stop this item and report that contract
+   gap; do not invent a surrogate or modify code without a new lead contract.
+5. Verify `git diff --check`, exact HEAD/branch, zero tracked or untracked
+   source change, and every artifact SHA manifest. No graphical client launch,
+   connected route, package rebake, commit, cleanup, retry, or mutation is
+   authorized in C1.
+
+Return the exact commands, artifact paths, pass/fail/skip counts and
+identities, manifest verification, fixture list and the requirement map. The
+retail/transcript lens reviews fixture completeness and whether the purported
+deterministic witness proves the retail products. Only after that PASS does the
+production/gate-honesty lens review filters, provenance, manifests, negative
+results, and absence of retries. Any missing witness becomes a separately
+contracted narrow harness chunk; otherwise C1 closes and C2 runs the connected
+lifecycle/performance program.