Drakan menu input and finite acknowledgment — 2026-11-18

Latest result: New Game / Normal, mounted heading control, flame attack and ordinary confirmed Quit verified in scratch/runs/20261118-drakan-prepared-new-game-mouse-route/. Forward travel remains unverified. Earlier timeout scopes below remain closed.

Predecessor evidence: scratch/runs/20261118-drakan-menu-input-level-gameplay/. All predecessor UNKNOWN scopes remain closed, without endpoint requests.

The ordinary renderer keyboard route updates physical/DirectInput key state and separately queues USER input. Host dequeue is not native consumption. The guarded native route acknowledges the exact owner/thread/token only after a successful public USER return; this does not prove the guest's menu action. Historical notes also identify relative DirectInput mouse movement/clicks as the Drakan menu route. A delivered Enter alone does not establish New Game.

An authored ordinary PE creates/shows its own window, calls genuine DirectInputCreateA/CreateDevice/SetDataFormat/SetCooperativeLevel/Acquire and GetDeviceState, and polls keyboard/mouse/all USER message ranges. A two-second guest GetTickCount busy frame reproduces slow consumption without any host pause, forced poll/wake, focus assignment or clock change. Exact native USER consumption takes 1.907 seconds in calibration. GetDeviceState succeeds and returns DIK_RETURN=128 independently; the guest callback updates its caption. The original 500ms harness confirmation times out on a fresh authored scope. A predeclared 2500ms confirmation succeeds on another fresh scope with exact native receipt and independently prearmed physical UP. These controls prove an inadequate harness deadline is possible; they do not establish the exclusive cause of the predecessor original's timeout or prove D3D pressure coverage.

The new helper validates finite policy before arming: 2500ms confirmation, 3500ms fallback UP, 6500ms physical-release completion and 10000ms operation. Consumption must precede the sealed targeting/operation/fallback deadline. An acknowledged native return can only shorten the already armed release; late receipts cannot renew a pulse. UNKNOWN closes the scope with no replay. Process/window/thread/generation/queue/device/retirement predicates remain the existing native reviewed-context contract. No runtime file is changed.

The full normal production browser shell also runs this authored PE in its real Chromium Worker. Ordinary DOM key input produces native callback dispatch and successful DirectInput state, and independent CDP UP is followed by the guest's own DirectInput zero read. This is a browser positive control, not a guarded native acknowledgment or authentic game result. The fixture includes busy guest work and ordinary window painting, not a D3D render workload.

Fresh authentic outcome and publication follow validation below.

Mouse preparation additionally uses an ordinary standard DIMOUSESTATE format. The first mouse control exposed a fixture error: one message per two-second frame allowed ordinary hover/hit-test messages to delay button dispatch. The corrected fixture drains each ordinary message range and reads DirectInput from its actual input callback as well as its frame loop. A short click could otherwise end between two slow frame samples. Fresh native and full browser controls prove button=128 and independent release; browser later reads zero. The failed control is closed with actual cleanup and is not game failure. The helper accepts reviewed ordinary absolute move/down coordinates, retaining the production mapping and exact native message/actor predicates. Three wrong mouse-message receipts are rejected. Original menu selection still requires the visible difficulty/level consequence.

The fresh original used sealed 6308e7a3 source plus the three 5aa5dabd JavaScript overlays, exact a51912dd module, genuine COMCTL32/COMMCTRL and unchanged Settings100/media. Main-menu review at batch100157 was authentic. Warning Enter and UP succeeded; ordinary main focus arrived 17.087 seconds after UP (17.035 seconds after the completed-input record). Its immutable lifetime was 900 seconds, independent-watch intention920 and gameplay cutoff720, retaining180 for Quit. Mouse qualification/build publication finished at the cutoff; no New Game, level, movement or attack is claimed.

During the reserve, ordinary mouse reference/move returned exact actor-thread1, generation4/window10004 USER receipts in418ms and1148ms, with independent UP. The immediate image still showed the old cursor; a later passive whole frame showed Quit red under the moved software cursor. Ordinary Escape also returned an exact865ms receipt. A left click returned an exact723ms receipt and UP ACK, then visibly opened the genuine "Too scared to face the perils of Drakan?" YES/NO confirmation. These consequences establish this ordinary mouse route actually reaches the original menu. They do not establish ordinary exit.

The following move toward YES timed out at the fixed2500ms HTTP deadline. Independent release still returned ACK. This is sticky UNKNOWN; no later endpoint request, renewed review or replay occurred. The retained trace shows host preparation approximately88ms after dispatch, handler return138ms later, the actual WM_MOUSEMOVE queued for10004, and no subsequent observed public USER return for that ticket. The last observed actor consumer is sequence3219/thread1, PeekMessage(all,remove), result0. Its actor diagnostic says same memory/window and not retired, but targetOwnerThread0/generation0/visiblefalse are not a complete target fence and do not prove the intended window's consumption. That incomplete bounded trace cannot distinguish later actor work, missed native return or transport completion. It provides no actual late consumption latency with which to justify another longer deadline. The normal browser controls are positives; no authentic browser failure is claimed.

Containment terminated at the actual900-second deadline and reports no leaks. The reused deadline helper had a separate demonstrated defect: its containment deadline killed the receipt-writing wrapper and its watcher before it retained native/watchdog pipe-close receipts. owner-complete therefore explicitly says retainedActualWaits=false. Those missing original receipts cannot be recovered or invented. stdout/stderr are streamed prefixes, not proven complete logs.

The targeted drakan-menu-custody helper repairs this ordering. Native termination remains at the sealed deadline, with finite escalation and actual pipe close; the independently prearmed920 watcher lives outside that containment group. The outer wrapper cleanup boundary is935. A real TERM-ignoring child reproduces the old missing waits with the retired deadline constants; a fresh corrected control retains both actual waits and both EOFs, with no leaks. This custody repair does not resolve the original input UNKNOWN, so it does not authorize another unchanged original input retry. The concrete remaining blocker is an unresolved original native return/ACK after genuine Quit confirmation, and the gameplay goal remains unverified.

Prepared New Game route

Evidence: scratch/runs/20261118-drakan-prepared-new-game-mouse-route/. All preparation and custody qualification finished before the fresh original's 900/920/180 lifetime. Exact retained source/module/media reached the main menu. The first absolute move established DirectInput's reference; the next ordinary relative movement visibly placed the software cursor on New Game, then the jewel turned red. A stationary click opened the genuine difficulty dialog. Normal was hovered, visibly red, and clicked. Loading advanced through Wartok Canyons assets and reached a rendered close-up intro cinematic. This proves New Game and the difficulty consequence, not playable movement or attack.

The subsequent reviewed Escape had only873ms remaining on its original 15-second capture lifetime when DOWN was sent. Its native result was CANCELLED for immutable deadline expiry, with no native consumption and independent UP ACK. This is a known cancellation, not a measured late ACK or HTTP UNKNOWN. The helper nevertheless closed the scope; no later endpoint request followed. Filesystem cancellation retained both actual native/watch waits and EOFs.

The targeted helper now reserves the unchanged2500ms confirmation budget before arming and immediately before DOWN. Insufficient remaining image time returns NOT_SENT without effects. Its regression reproduces the873ms admission error, tests the strict equality boundary, and a fresh real native control with a13-second-old capture refuses before DOWN without closing the scope. For a separately prepared fresh retry, the helper also accepts predeclared finite review limits up to60 seconds for game/90 for menu. This uses Alice's already published98e56ec00 cli-native-scene review calibration, pinned before launch; the original capture clocks, current actor/window/generation/context checks and immutable in-flight deadlines remain authoritative. A20-second-old authored native capture succeeds with the declared60-second policy, exact USER receipt and independent UP. Normal browser USER/DirectInput controls, receipt negatives and scene identity/staleness negatives pass. No ACK deadline was increased based on the original cancellation. Locked build passed before publication and affected original retry.

The affected retry passes the previously cancelled Escape with a660ms native receipt and UP. The first close-up was misidentified: its later zoom reveals illustrated dragon story artwork, not Rynn or level geometry. A correction is retained without rewriting the input reviews. Escape opens/closes the real menu over that story; an ArrowUp pulse there is not movement evidence. A separately reviewed ordinary Enter reaches the actual mounted rainy outdoor level, with ridge/conifers, river, bridge and health/HUD visible.

A subsequent requested1000ms movement was rejected by a helper assertion before DOWN: its advertised80..1000ms range conflicted with2500ms confirmation and3500ms fallback requiring strict spare time. No movement input was sent; this was a deterministic preparation error, not native UNKNOWN. That scope was closed and retained actual native/watch EOFs. Duration admission now returns NOT_SENT before arming for an incompatible pulse. Unit strict-boundary negatives and a fresh authored native1000ms refusal followed by a700ms consumed/independently released pulse qualify the correction before another affected original. The confirmation/fallback deadlines are unchanged.

The third original reaches the mounted outdoor level directly with reviewed Enter. A700ms ArrowUp pulse returns native consumption and independent UP; whole frames do not establish forward travel. The next reviewed leftward pointer movement hits the sealed2500ms HTTP timeout, with independent UP ACK. The lifetime closes immediately; a subsequent capture command is rejected locally by the sticky failure before any endpoint request. Both native/watch actual waits and EOFs are retained following filesystem cancellation.

Offline inspection of the already-streamed trace now supplies genuinely new latency evidence: nonce e224a202-01d2-4cf3-b82e-5ba20680d65d settles as consumed 3034.772ms after dispatch, before its sealed native target deadline. This does not revise the HTTP UNKNOWN, reopen that lifetime, or establish visual motion. Unlike the predecessor's unmeasured timeout, it supports independently qualifying a fresh policy:4000ms confirmation,5500ms fallback,10000ms UP and 13000ms total operation, all fixed before a new lifetime. Existing native identity/review/receipt checks and actual900/920/180 custody remain unchanged.

Qualification reproduces old2500ms HTTP failure with a3500ms authored guest frame; a fresh4000ms control consumes at approximately3350ms and independently releases. Receipt, duration, review-boundary and scene negatives pass. A normal production-browser USER/DirectInput keyboard+mouse control passes after the locked build. Its earlier concurrent-build mouse USER-dispatch assertion failure is retained, with releases and actual browser/watch EOFs; no cause or authentic browser failure is inferred from that authored control. No original starts until these preparations, build and strict publication finish.

Final affected original: verified heading, attack and ordinary Quit

All tests, locked build and strict publication edf62a65d finished before the fourth original's fixed900/920/180 lifetime. Exact6308+5aa5 sources, published98e scene calibration, unchanged a519 module, genuine COMCTL32 and COMMCTRL, original64 media pins and unchanged92-byte Settings100 were sealed and checked before launch. Actual main/Worker source/module reads match the sealed inventory. This is the qualified native route, not an authentic-browser run; no game/runtime logic or identity override was introduced.

Personally reviewed ordinary pointer movement shows New Game red before its stationary click, then Normal red before its click. The level loads genuine Wartok Canyons assets. Reviewed Enter advances the illustrated story to the mounted outdoor level: dragon/rider, health HUD, ridge, bridge and river. The left120-pixel pointer action has exact native USER consumption and independent UP; the later whole frame shows the peak shift from about250 to360 pixels, the central conifer shift right and a new tree entering the left edge. This verifies player heading control. A stationary1000ms ordinary left click has exact native consumption and independent UP, then a whole frame visibly shows bright orange flame at the dragon's mouth. This verifies attack effect, not enemy damage. A1000ms ArrowUp also consumes with UP; its slight landmark differences do not establish forward travel confidently.

Reviewed Escape opens the genuine menu over the level. Quit is hovered and visibly red before clicking. The genuine confirmation says there is a game in progress and asks whether to quit Drakan. YES is hovered and visibly red before the final stationary click. Both final native consumption and independent UP ACK succeed. Without any later endpoint request, filesystem inspection retains guest Exit0, actual native process exit0/no signal, both stdout/stderr EOFs, watchdog exit0 with both EOFs, and no leaks. No cancellation request exists for this successful scope. Native stats report664.838 seconds; normal exit occurs before gameplay cutoff720 and preserves the entire Quit reserve. The fixed watchdog/containment deadlines were never extended.

Selected key images include original-retry/story-result.png (the earlier whole mounted level specifically requested for retention), and original- measured/mounted-before.png, heading-settled.png, attack-settled.png, quit-highlight.png, quit-confirmation.png and hover-confirm-yes.png. Four originals in this task all retain both actual native/watch waits and EOFs. The run audit checks87 actual recorded child waits with both EOFs and nonlive PIDs, including successful and failed authored controls. The deliberate old-custody negative retains its expected missing-inner-wait failure separately. Box ID/expiry/stop receipt are null, state NOT_CREATED. All old UNKNOWN scopes remain closed. Ownership is released after result.json is written last.