Stronghold Custom Game hover recognized; original press leaves ChooseType

The demonstrated hover false negative is fixed in this run's small navigation/route helpers and passed against the actual original route. World/gameplay remains unobserved. After a personally reviewed, freshly recaptured Custom Game hover, balanced native (543,292) down/up left the same ChooseType hover visible. first unexpected route scene: choose-type-custom-hover at 08:56:51.164Z was terminal. No preset, Zephyr, route handoff, native Start or gameplay action followed. No retry or second browser occurred.

Evidence and recoverable helper/test patches: scratch/runs/20261104-stronghold-hover-recognizer-world/. Predecessor: fresh integer route, published 167179eb48450c9a9de3d7a7e4edec4698d7f83a.

Narrow image-grounded fix

I personally inspected the predecessor's exact pre/post-hover PNGs 35ae1b67… and 465c6a1a…. The latter visibly retains the original ChooseType title and left map, and adds Custom Game / Customize your own Crusader game. A tower/soldier region changes during hover; its exact animation mechanism is not established. These images justify a distinct hover state, without a production rendering repair.

Own navigation.cjs preserves the strict initial ChooseType signature and adds only choose-type-custom-hover. It requires four exact opaque RGB SHA-256 patches: full header [110,315,660,75], left map [160,480,260,120], Custom Game label [345,630,195,33], and description [240,676,400,25]. Hashes are grounded in the inspected real hover PNG. The animated tower region is excluded from this distinct hover signature. There are no fuzzy thresholds or general unknown-scene admissions. The predecessor still classifies the exact original hover as unknown; the new recognizer classifies it specifically as Custom Game hover. Unknown remains unknown elsewhere.

Own route-driver.cjs requires strict initial ChooseType before moving. After the corrected integer CSS (594,520) move has queued native (543,292) and fully ACKed/drained, it requires this distinct hover, explicit personal review, and an identical fresh full-PNG recapture before down/up. Other routes retain their expected scene and gain the same reviewed pre-press freshness check. Initial review, native coordinate plans, DOM transport, event ownership, HWND/mask/wParam/lParam checks, explicit host/renderer telemetry, complete ACK/drain, immutable queue, first-failure latch, strict Zephyr and unchanged native Start guard remain intact. The top-level driver uses own navigation for initial installation; native Start code is unchanged.

Preflight and tests

Only eight small runtime inputs were staged. The retained source/module/media remain referenced by existing paths and SHA-256, without copies. Actual full preflight.py ran through subprocess.run(..., check=True) before Chrome, passing at 08:53:24Z: all 2415 source pins at 1188e9bf57ef35bd3363fb88b7ca0da442ce696d; corrected actual helper graph including own navigation; 98 page / 12 worker cache graph; actual Puppeteer CommonJS/dynamic ESM/helper imports and syntax; and five pinned HTTP responses with isolation headers and SHA checks. The temporary HTTP preflight server closed. The new release preserves exact authorization stronghold-game-crop-freshness user task, ten-minute TTL guard, box/expiry, 09:08 hard close, 09:13 recovery, and coordinator shutdown ownership.

Focused offline results, with actual production DOM/renderer/host input code and explicitly modeled ACKs/guest scenes:

Reproduce from the checkout using node scratch/runs/20261104-stronghold-hover-recognizer-world/{hover,actual-postpress,actual-successor,route-dom,integer-route-point,native-guard}.test.cjs separately, python3 scratch/runs/20261104-stronghold-hover-recognizer-world/launch.test.py, and python3 scratch/runs/20261104-stronghold-hover-recognizer-world/check-evidence.py. The helper files are restored from owned-new-helpers.patch if needed. No production code changed, so no production full-build gate applies; the actual retained build was verified by pins rather than rebuilt or replaced.

Personally reviewed live boundary

The sole detached execution began 08:53:24.343Z; ready was checked before any command. I viewed every bounded loading capture, the original title at 10240, original menu after balanced Enter at 10432, qualified Crusader hover at 10496, and strict initial ChooseType at 10624. One loading-review description mistakenly named Firefly at 6144; an append-only correction records the actual castle loading artwork. The capture at 8192 shows Gathering of Developers.

The real corrected move reached native543292 and drained at 10688. The distinct hover PNG 8af95d757560b25af4972fc5dd76b38625174669797b686cae1f9102080afa70 passed the narrow recognizer; I viewed it live before acknowledging review. Fresh pre-press recapture matched. Actual queued down contained exactly WM_NCHITTEST/WM_LBUTTONDOWN at HWND65537, native543292, lParam=19137055, button mask1. Up contained exactly WM_LBUTTONUP with mask0. All input drained at 11008.

I personally viewed the closed final PNG 0d721dbb0a829c3ba5c9eda1645821fd49afd1c49e68ffa44fa35e514f060554: ChooseType and Custom Game hover remain; there is no preset shell. Host logs show the owned down/up were consumed. The queue/ACK evidence proves transport and drainage, not game callback execution or scene transition. A screenshot cannot distinguish delayed transition, guest button dispatch, or another guest-side cause; no additional production defect is proved and no speculative fix was applied. The original native callback/source route audit still identifies button2 as Custom Game; coordinates were not changed to probe alternatives.

Closed recovery and handoff

233 closed logical files / 16,361,192 bytes were recovered and full-SHA verified by 08:57:14.725Z, before the recovery deadline. Identical already reviewed PNGs are referenced rather than duplicated. Evidence audit passes 172 complete ACKs / 11008 steps, 1036 matching served-pin receipts, exactly one balanced native543292 route gesture, final empty queue/null active/null last/zero mask, and exactly one terminal runtime failure.

Chrome 20287 exited normally with code0 at 08:57:02.152Z. Driver 20271 is absent; private HTTP 41049 is closed; all 2415 historical pins still match. Assigned bx_v34fdr58 remains RUNNING / API-idle, healthy/hydrated, expiry 2026-11-04T09:18:46.794Z. Custody returns to the coordinator, who owns shutdown. No stop/delete/resume/allocation/sourcebox action, local browser/heavy game, subworker, successor worker, HeroesII, forced guest write/call/scene, reload or deployment occurred.

exact-git-sha.txt, relevant-uncommitted.patch, owned helper/note patches and the artifact manifest preserve provenance. Retained module hashes are cffca869… (normal) and 580f49f3… (compat), with full pathSHA entries in retained-module-media-path-sha.json. Git is the source archive; no source-tree archive, copied module or copied media was created. Only this note is committed. Next boundary: explain why the consumed balanced Custom Game press leaves ChooseType before any further original-route execution. This single assignment stops here.