Stronghold hover-reviewed native replay: stopped at a second review publication
The sole original native attempt reached the main menu, then stopped at a concrete publication prerequisite before Crusader, Custom Game, preset, native Start or world. The second fetch within title-enter selected the already acknowledged title request; publication of the new main-menu request during transfer correctly triggered immutable current request replaced. The first failure stayed terminal. No world or visible gameplay is claimed.
Evidence and the tested next helper are in scratch/runs/20261104-stronghold-hover-reviewed-native-world/. Only this note is committed. The exact predecessors are preset hover recognition, commit 2b5b3b07e, and current-request native replay, aabe0e753.
Before Chrome, the assigned bx_246a6dv2 was confirmed healthy, idle and fully hydrated. The nine replacement runtime files, operate.py, unchanged operator-state.py and three unchanged publication helpers matched their tested identities. complete-dynamic-input-closure.json explicitly resolves all36 predecessor closure entries, mapping the replacement files into the fresh run while retaining their sibling dependencies; three additional entries cover authorized runtime_io.py, release JSON and the cache checker. hover-layout.json SHA-256 is 0632e31b8c0a62be221de882e019aa1dedcca512a983436db548e719f60ab318. All39 staged identities passed remotely. Two initial staging-path mistakes were caught and corrected by preflight, before any browser; the replay helpers were unchanged.
All2415 original runtime pins passed before and after the attempt. Runtime source remained 1188e9bf57ef35bd3363fb88b7ca0da442ce696d, canonical WASM cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee, companion 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3. Module and media identities reference existing paths; none were copied. Actual wrapper top-level setup, lazy imports, Puppeteer CJS/ESM imports, executable access,98 page/12 Worker cache entries and five SHA-pinned HTTP isolation responses passed without calling main or creating a browser.
Fresh controls ran against the staged helper identities:25 actual-driver cases,25 focused layout negatives, actual wrapper/driver/separate fetch/ACK transport integration,18 operator controls, six personal-review controls, four transport controls, six split controls, nine bounded-route controls, three positive/48 negative retained layout controls, actual operator entry, exclusive lock, timely reply inspection and actual wrapper setup. Ten unchanged publication controls passed, including a40.058-second actual timeout. Full RGBA and independent column recognition remain exact; there is no mask, tolerance, synthetic positive or substituted layout.
Chrome14266 recorded identity at 13:08:38.732Z, with the original absolute browser deadline 13:14:35.625Z. Startup observation commands0–6 each completed and were separately inspected. Their personally viewed captures show streamed loading at2048/4096, the original loading artwork at6144/8192/10240, Gathering of Developers at12288, and the original Enter Your Title dialog at14336. These observations consumed the unchanged aggregate navigation budget; no counter or deadline was reset. This startup was slower than the predecessor's10240-step title arrival.
Command7, title-enter, published a fresh title request at14336. Its current capture was personally viewed and separately acknowledged. The unchanged driver sent balanced Enter and naturally reached the main menu at14528. The immediately following fetch still selected the previous request from command7. At13:11:43.614Z the wrapper published its new main-menu request under that same command index; transfer revalidation rejected the replaced request. The operator latch stopped the wrapper, and reply7 reported operator terminal latch: quit only. The final main-menu screenshot was personally viewed after recovery; that later inspection granted no input permission. Command8 was submitted quit NEXT after the lock released. Chrome had already closed normally, so quit's missing reply timed out; its later cursor text is retained without treating it as a new permission.
Chrome closed at 13:11:45.012Z, exit0,186.280 seconds after identity. actual-result.json verifies227 complete64-step ACKs/14528 steps,1032 matching served pins, six Worker preloads, one completed personal review, drained final input and the rejected command7. No physical route instance was installed: the attempt stopped before its prerequisite. There were no retries, reloads, forced callbacks, guest writes, repaint, fault clearing or second Chrome. Captures are original host PNGs; no guest raw-pixel dump or derived PNG is claimed. Historical preset RGBA identities were read-only verified from their retained adjacent gzip originals; compressed evidence remains intact.
All275 immutable closed artifacts/8420813 bytes and12 unique control files were recovered with full SHA verification. A second control inventory was identical. Chrome and the replay process were absent, HTTP36015 was closed, and the final2415 pins passed. Custody returned at 13:13:40Z with bx_246a6dv2 RUNNING (API idle, healthy and hydrated), expiry 2026-11-04T13:43:13.381Z. The coordinator owns normal shutdown at13:40Z; this worker performed no lifecycle action and makes no subsequent stop claim.
The same worker made a narrow offline successor repair in this run's review_publication.py. Fetch now reads its own successful ACK history for the pending command and waits past a request already personally acknowledged in that command. It still strictly checks command binding and shape before skipping, retains the original absolute deadline, never ACKs while fetching, and still rejects replacement during transfer. The live attempt used the unchanged predecessor helper; the repair was neither uploaded nor tried in a second browser. narrow-publication-fix.patch records the delta.
multi-review.test.py executes the actual staged wrapper's personal-review function twice under one command using original recovered title/main-menu captures. The original helper reproduces the consumed-request return. The repaired helper waits through retained old publication, fetches the distinct new request without acknowledging it, requires a separate second ACK and permits timely final reply inspection. Four further controls cover consumed request never republished, future binding with consumed ID, malformed consumed ID and history from another command. All ten retained publication controls pass against the repaired helper, including an actual40.133-second timeout. These are offline fixtures and grant no live review or gameplay permission.
The next separately allocated recipe must stage this repaired review_publication.py, unchanged review-request.py/ack-review.py, the exact hover helpers and a newly authorized allocation-bound runtime_io.py; regenerate all dependency/staging identities and rerun the controls before its sole Chrome. Do not reuse this run's launch guard or old identity manifest after the offline repair. Within a multi-review command, each next fetch now waits for a request absent from that command's ACK history; personally VIEW its exact capture, invoke ACK separately, and inspect the command's timely final success before any dependent command. Preserve the original16384 navigation,4096 route,3072 arrival,768 stability,64 ACK,40-second command and360-second browser bounds. Startup consumed14336 steps here, so remaining navigation capacity is an additional constraint; no larger allowance is authorized.
Only after original ChooseType, separate Custom Game hover review/press, bounded setup arrival and three stability captures may setup review permit MOVE. Preset continuation remains route-next MOVE-only → separate route-review-preset-hover current fetch/personal VIEW/separate ACK → timely inspected success → separate route-press-preset. Fresh Zephyr handoff and the unchanged native-Start guards remain mandatory. On the first failure, quit NEXT only. The next native execution remains unperformed.
Git SHA, relevant binary diff, owned small helper/note patches and an artifact seal preserve provenance without source archives. No production code changed and no full-build result is claimed.