Stronghold setup and preset boundary

The sole original attempt completed stable setup, fresh setup review, MOVE-only preset hover, its separate current capture review and timely success, then the separate balanced preset press. The press returned first unexpected preset result: unknown. The worker stopped on that first failure. Native Start, world and gameplay remain unverified. This failure occurred with 123.043 seconds still available before the original browser deadline; review transfer overhead no longer prevented reaching the preset press boundary.

This uses the exact hover split from preset hover recognition, 2b5b3b07e, and the consumed-request publication repair from hover-reviewed native world, 943dce309. Preparation follows the successor recipe, 11acfe967, with the closed-browser guard from first-title native world, 9c03e783c. Those sealed predecessor inputs remain read-only. Fresh evidence is in scratch/runs/20261104-stronghold-setup-preset-native-world/; only this note is committed. No production source, renderer, module, media or recognizer changed, and no full-build result is claimed.

The assigned box was bx_qerkqfrp, expiry 2026-11-04T14:41:47.895Z, guest cutoff 14:28Z, recovery deadline 14:33Z and coordinator normal stop checkpoint 14:38Z. Its actual API state was idle, healthy and fully hydrated before execution or staging. The worker returned custody RUNNING at14:12Z, with API idle/healthy/fully hydrated, and performs no further box operations. The coordinator owns shutdown; this worker performed no lifecycle operation or predecessor-box operation.

offline-sequence.txt records the complete conditional command/review/recovery plan before Chrome. Small local orchestration invokes the unchanged operator and publication helpers, then returns the current capture directly for personal VIEW. ACK remains a separate invocation after that VIEW. Final successful operator inspection remains a separate prerequisite before each dependent command. Only current captures were transferred during runtime; full closed recovery followed afterward. No duplicate historical input transfer, source archive or module/media copy was needed.

All36 explicit relative helper dependencies and three fresh allocation inputs passed SHA checks locally and on the assigned box. The actual runtime driver, layout recognizers, operator, state, publication helpers and closed-browser guard are byte-identical to the tested predecessor. Original2415 runtime pins passed before launch and after closure: source 1188e9bf57ef35bd3363fb88b7ca0da442ce696d, canonical WASM cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee, companion 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3. Original media remains at recorded existing paths and SHA identities. Actual wrapper setup/relative imports, Puppeteer CJS and ESM, executable access,98 page/12 Worker cache entries and five pinned HTTP isolation responses passed without Chrome. Chrome launcher, actual Chrome binary, Node and Python executable paths and full SHA identities were recorded before launch. The process guard found no existing Chrome or replay driver.

Fresh focused controls passed25 actual-driver cases,25 layout negatives, actual driver/wrapper/separate fetch/ACK transport,18 retained operator cases, six review/four transport/six split/nine bounded-route cases, and three positive/ 48 negative retained layout controls. Actual operator entry, exclusive locking, timely reply inspection and wrapper top-level setup passed. The multi-review regression reproduced the old consumed-request error and passed with the exact repaired helper; all four consumed-request negatives and ten publication controls passed, including the original40-second timeout. Closed-browser tests passed normal closure without a command, one original quit while open and three abnormal closure rejections. Offline fixture cursor/replies were retained under explicit fixture names before the first live command. Historical gzip fixture reads restored bytes only into the owned run; originals were retained. These controls granted no live review or input permission.

Chrome21593 recorded identity at 14:07:31.956Z, version151.0.7922.108, with original absolute deadline 14:13:31.111Z. Commands0–4 each completed 2048 observation steps with timely inspected success. Each current capture was personally viewed before deciding the next command: loading artwork2048, media-loading dialogs4096/6144, Firefly8192, then Enter Your Title10240. Command5 immediately used balanced Enter, with separate fresh title and main-menu fetch → personal VIEW → separate ACK. There was no extra2048 observation. Command6 separately reviewed main menu and Crusader hover under its original command deadline, then completed balanced Crusader press. Both multi-review commands had distinct request IDs and timely final success.

Commands7/8 separately reviewed actual ChooseType. Command9 moved to Custom Game, then fetched/reviewed/ACKed its current hover and completed successfully. Command10 completed balanced Custom Game press, returning arrival with384 route steps. This fresh attempt used one more64-step drainage ACK than the predecessor; no allowance was reset or inferred from the earlier12160 result. Commands11/12 advanced256 arrival steps each: intermediate then exact setup at512 cumulative arrival. Commands13–15 completed three further256 stability captures. Every result and returned capture was inspected before its dependent successor. Final setup was12224 navigation/1664 route steps, stability768.

Command16 separately fetched and personally viewed fresh stable setup: Human/Lord Crusader, map, Deathmatch Game,10000/10000 gold, Crusader Demo row and three preset circles. Its separate ACK and timely final success returned preset. Command17 emitted only MOVE at client(84,733), native(67,491), and returned exact actual-scene20-preset1-hover, preset-hover-review,1728 route steps/12288 navigation steps. Its successful reply and current capture were inspected. Command18 separately published a new current request, received personal VIEW and separate ACK for the highlighted preset1, then completed timely success with preset-press at 14:11:20.179Z. No preset down occurred before this successful separate review command.

Command19 separately emitted preset down at12288 and up at12352. The final complete ACK reached12416 navigation/1856 route steps. At 14:11:28.068Z, the unchanged driver rejected the immediate post-press result as unknown. The first-failure latch was retained. No route handoff, native-driver install, native Start, additional observation, retry, reload, callback forcing, guest write, repaint, fault clearing, counter reset or deadline extension followed. Chrome closed normally at 14:11:29.932Z, exit0, for first concrete command failure,237.976 seconds after browser identity. The tested quit guard observed that closure and returned alreadyClosed: true, noCommandSubmitted: true; it submitted no quit mailbox command or missing-reply timeout.

The recovered post-press capture route-native-boundary-19.png was personally viewed after closure. Its PNG SHA is 9fcc66eea4ba565a7fce8bc891e811219687e1c2f5f9419f806361aee9c404fe. It visibly shows Zephyr — Very Easy (1/10), Normal Game,2000/2000 gold, green arrow and no Rat opponent row. This is visible preset feedback, not an accepted ready checkpoint or world. Read-only diagnose-post-preset.cjs reproduces unknown through both unchanged route and native classifiers. The Zephyr description patch exactly matches its original hash; only the scenario gold/settings patch fails. Actual patch SHA is 4e42eb0da3fff58db1054b9545ed0e9b71c29ed7246ec80cccb10ec132a7b5be, expected aa3ac86e9a6b30d0e705adb01042aa9745b2c2b09d746425de565b7b40a312cd. The original retained ready fixture, also personally inspected offline, shows Human Advantage,8000/2000 and Rat/Duc de Puce. No reason for that difference is proved: the current capture alone cannot establish whether ordinary native initialization would subsequently finish. No alternative template, mask, tolerance, foreign layout or new scene acceptance was introduced.

actual-result.json verifies194 consecutive complete64-step ACKs,1052 matching served pins, nine unique personal reviews,19 timely successful replies followed by exactly one failed command, and drained final input. Navigation16384, route4096, arrival3072, stability768, ACK64, command40 seconds and browser360 seconds remained unchanged. Captures are original host PNGs; no live derived PNG, guest raw-pixel dump, native world or gameplay claim is made.

All329 immutable closed artifacts/35301830 bytes and31 unique control files were recovered with full SHA verification. The second control inventory was identical. Chrome/replay processes were absent, HTTP33557 was closed and final 2415 pins passed before custody return. Evidence containment has no errors for this run;16 unrelated existing repository errors were retained. The corpus has no matching Stronghold candidate, so evidence stays in this note without an unrelated dashboard result or gameplay badge. Provenance records initial and final-test checkout SHAs, relevant binary diff, owned new-file patches and an artifact seal; Git remains the source archive.

The concrete next prerequisite is qualification of the actual ordinary post-preset native readiness transition against the unchanged scenario guard. The immediate result's description already matches while gold/settings and opponent population differ from the accepted ready fixture. That boundary needs a separately authorized, evidence-grounded continuation design and focused tests before another fresh original attempt. This closed failed session grants no continuation or Start permission. No helper defect was proved that would justify changing its recognizer or readiness requirements in this task. This single assignment is finished.