Stronghold: sequential replies fixed; later lobby observed, review timed out

The run-owned operator now blocks every dependent command until the preceding final reply has been inspected. A timeout, rejected reply or mismatched reply persists a terminal latch that admits quit NEXT only. Late success cannot clear it. The single authorized original runtime reached the complete later lobby layout and three further input-free captures, but its required personal review missed the unchanged 40-second reply deadline. No preset, native Start, world or gameplay was authorized or reached.

Input: intermediate native onward, commit 1b200969ff0ad08cb2ce3d6266a60dc12d8f196a. All321 sealed files /8060368 bytes were verified before use. Implementation, tests, closed captures, helper hashes and provenance patches are in scratch/runs/20261104-stronghold-review-reply-sequencing/. Only this note is committed; production code and WAT were unchanged. No full build was run or claimed.

The operator holds an exclusive command lock and persists the pending index before submission. A successful, matching final reply must be parsed and recorded before another command is admitted. Timeout/failure writes a separate remote terminal latch, which the live wrapper checks at command admission and during personal-review waits. Reviews remain separate fetch, personal VIEW and acknowledgement operations. Each acknowledgement binds the pending command index, unique request ID, capture SHA, scene, frozen step count and original request time. Review files use unique request IDs, so identical pixels cannot reuse an older acknowledgement. Review fetch/acknowledgement fails after the deadline or terminal latch. No dependent actions are prequeued.

The preflight used only dependency/source/hash/import, HTTP/isolation, cache and pin checks. The copied actual-worker.test.cjs staging/execution path was removed from the new workflow entirely: no synthetic callback fixture, forced ESP/EIP/argument writes or forced guest callback execution occurred. The actual renderer and passive natural observer supplied runtime observations. The exact accepted navigation/model, whole-RGBA and independent-column recognizers, presentation/freshness helpers and observer bytes were reused with SHA verification.

Offline checks passed:18 operator controls, five controls exercising the actual wrapper review function, five launch-order controls,48 layout negatives,28 controller sequences,12 route cases,27 observer controls,20 freshness cases,20 retained history reports with16 rejecting mutations and33 ACK/budget/wrapper guards. The copied textual guard expectation was corrected to require the stricter quit-only latch; the final suite passes. Model fixtures establish scheduling and guard behavior, not fresh runtime permission. Actual remote preflight passed all2415 pins, CJS/ESM/Puppeteer imports and transitive helper hashes,98 page/12 Worker cache entries and five isolated HTTP responses. It did not run guest fixtures.

The sole Chrome19326 started at11:22:44.712Z in the assigned existing runtime /home/user/stronghold-corrected-native-start-runtime, box bx_9xp85xy3. Source remains 1188e9bf57ef35bd3363fb88b7ca0da442ce696d; actual canonical WASM is cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee. No latest-main module, reload, retry or second Chrome was used. The browser bound360000ms, minimumTTL600000ms, expiry reserve300000ms, arrival3072/stability768/aggregate4096 steps,64-step ACKs and16384 navigation cap were unchanged; hardclose11:33, recovery11:38 and coordinator shutdown11:41 remained unchanged.

Five separately completed and inspected input-free observations reached10240 steps. Command5 title-enter received separate personally viewed title and main-menu reviews, then returned success at11:24:08.987Z before its40-second deadline. Only after inspecting that reply was command6 menu-crusader submitted. Separate fresh menu and hover reviews preceded its balanced original mouse gesture; its successful reply was inspected before command7 installed the route. The original ChooseType at10560 steps was personally viewed and separately acknowledged. Command8 route-next then received separate fresh ChooseType and Custom Game hover reviews before its balanced original gesture.

Natural input-free continuation produced these whole-layout observations:

Frozen steps Full RGBA classification Independent classification
11200 accepted intermediate waiting layout intermediate
11456 later lobby candidate later lobby
11712 later lobby,256 further steps later lobby
11968 later lobby,256 further steps later lobby
12224 later lobby,256 further steps later lobby

Thus candidate plus three further successive256-step later captures were actually recorded. The final fresh review request was published at11:26:08.626Z, with SHA d82cfa207bc5145bdf3628b6433800f019204ebf3e6968fb437d00cceb18ba6b. Command8's absolute reply deadline was11:26:09.539Z. Its personal review was not completed before that deadline. The operator latched timeout; the wrapper rejected the pending review with operator terminal latch: quit only at11:26:09.828Z and closed Chrome normally at11:26:09.899Z, exit0, elapsed205187ms. No further ACK or gesture followed the terminal failure.

After inspecting the completed timeout result, the very next submitted command was quit9. Chrome had already closed, so quit9 remained queued and was not processed; its absence of a final reply did not clear the failure latch. There was no menu command after timeout. Review fetches attempted after command completion/terminal failure were rejected, and no late positive review receipt was uploaded. The final lobby capture was personally viewed after closure solely for documentation: Human Lord Crusader, complete map, Deathmatch Game,10000 gold, buttons1/2/3 and Crusader Demo list. That view grants no runtime input permission.

actual-result.json verifies191 complete64-step ACKs /12224 steps,1052 served-pin matches, six paused-before-start Worker preloads, seven separate positive personal reviews and drained final input. Mouse evidence contains only the launcher and the balanced Crusader/Custom Game gestures; no preset input occurred. Raw natural observer rows record one slot0 ID2 ChooseType callback, its custom branch and subsequent scene20 request, followed by currentScene20 snapshots. natural-observer-summary.json preserves those rows and the observer's explicit unqualified coverage label; no callback-absence or broader coverage claim is made.

All294 immutable closed artifacts /26688734 bytes were recovered and fully SHA verified, with unique driver log, executed wrapper, terminal latch and queued quit recovered separately. Executed wrapper bytes match the tested local wrapper. Post-close checks confirm no Chrome/driver and HTTP38143 closed; all2415 retained pins remain unchanged. Git SHA, relevant binary diff and small owned-source patches record provenance without source archives or media/module copies.

Custody returned at11:28:19.124660Z: bx_9xp85xy3 remains RUNNING (API idle, health ok, hydrated), expiry 2026-11-04T11:47:28.508Z. Coordinator owns shutdown; no lifecycle operation was taken. The remaining blocker is completing the separate fresh later-lobby personal review within the existing reply/wall bounds before any preset. This assignment stops after its own note commit and one strict pull/push attempt; no successor or additional runtime is launched.