Recoil hermetic native objective, 2026-11-08

Run: scratch/runs/20261108-recoil-hermetic-native-objective/.

The single original reached a new native transition/freshness boundary. It did not qualify a temporal HUD or dispatch C, M or movement. The worker then implemented and tested a generic compound-driver correction in separate retained inputs, without retrying the original. Native consumption, ramp/control access, power-control destruction and onward objective progress remain unproved.

Prerequisites and original identity

The predecessor HUD-gate repair, including its original browser bytes and artifacts, remains unchanged. This task adapts its provisioning reference under the new authorization; the completed predecessor's allocation refusal was not removed in place.

All 755 repaired pins passed locally against published Git 49874a81c5ae180c20a15c6bb169d9041f028cae, including the 23 SHA-pinned pngjs 7.0.0 files from the retained SHA-512-verified dependency closure. The hermetic negative full entry rejected missing pngjs before browser output. The good delayed-HUD full entry passed locally under wine-runtime.lock, including all 113 HTTP pins and trusted generated-fixture key events. Existing twelve timing regressions, eleven integrated cases and the native Calculator Edit → Copy control were reused by retained path/hash, not presented as fresh game evidence. The bound-input, ordinary-key, pointer and transport helpers were unchanged.

The remote launcher independently checked all 755 staged bytes before its one launch. The recovered original Worker module is 6cf120d0d3b3db1823a8105eb26ef52a9093a7f871c37a46117475e213fda1c0. The ordinary route is /index.html?debug&threads&app=recoil_demo, Chrome 151.0.7922.108, Threads, cross-origin isolation, 1024×768 browser viewport and native 640×480 Software presentation. Toolbar WebGL is not backend evidence. Published Git providers exclude peer WIP. Exact Git, relevant binary diff, owned-input patches and media path/size/SHA pins provide recovery; no source or media archive was made. No production emulator files changed or build was claimed.

Fresh briefing resolves the objective numbering

Ordinary Software → OK → Game → Start → New Game retained the visible default name user and Medium difficulty. Native Start and the natural opening followed. An independent passive recorder captured the cycling native loading briefing, including its text, objective illustrations and map. These panels are not pause.

scene-review/observed-1794128776075.png shows 1.1 Operation Short Circuit: the force field above the beach blocks access to the enemy base at the top of the hill; find and destroy the force-field power control. The illustration shows the ramp entrance and posts; the map circles its northeast beach/road area. scene-review/observed-1794128795262.png shows 1.3 Destroy Rocket Sites, with different objective imagery and map markers. Native 1.2 Missile Control and 1.4 Go to Landing Site were also captured. Thus 1.3 is a later cycling subobjective, not evidence replacing the fresh 1.1 force-field objective. briefing-review.json records this review. No actual control was identified.

Native pause and the first new boundary

disembark.png shows living HUD health 100 and timer 00:00:00. Ordinary Escape is followed by native-pause.png at 09:10:13.110Z, still showing HUD timer 00:00:03, then pause-current.png at 09:10:24.278Z, visibly showing the native Resume menu. Relative menu movement produces fresh highlighted Resume in resume-ready.png at 09:10:58.763Z. Those captures are in review/ and the independently retrieved original/.

Command 28, first-hud-gated, was the only compound attempt. It planned C then ordinary M for map review, followed by release/Escape. Resume DOWN was dispatched at 09:11:22.243Z and UP at 09:11:22.372Z. The pre-Resume baseline and subsequent probes remained the same native menu PNG. The third post-Resume probe began at 09:11:23.307Z and completed at 09:11:24.356Z: 1,049 ms old. The driver threw HUD probe stale before action, sent release/Escape at 09:11:24.357Z and Escape UP at 09:11:24.438Z. Its receipt has zero attempted gameplay keys. The prior 2,400-ms overall window also included both transition and temporal-HUD qualification, leaving no separately reserved command window.

The independent observer first captures the returned native HUD at 09:11:24.432Z, health 100, timer 00:00:03, in original/observed-1794129084432.png. That is a sampled upper bound on visible resumption, not its exact time. The enclosing HUD readings span 71.322 wall seconds with the same displayed timer; this supports excluding native menu time from that timer. It does not measure exact actor/health freeze or make loading, briefing or map panels pauses.

Commands 0–27 have exact ACKs. Command 28 has no ACK and remains sticky UNKNOWN; source ordering and zero DOWN rows do not convert it to an acknowledged zero-effect command. There was no replay, second Resume, restart, attack, guest write, private guest handler, forced scene/configuration/clock/seed or deployment. The overall Escape was independently released at the host, but a post-compound native MENU was not captured before closure. The last living sample is not a final-state claim. No native C/M consumption, fresh temporal gate, movement or objective consequence is accepted.

Tested generic integration correction

The actual original compound-v2.cjs, browser-run-v3.cjs and 755-pin inventory remain unchanged. New compound-v3.cjs / browser-run-v4.cjs use a separate 4,500-ms transition gate (validated maximum 5,000 ms) and a command window started only after a qualified HUD. The immutable overall timer and the gate-miss release/Escape timer are armed before Resume. After qualification, an independently armed action timer bounds the command; per-key releases and the 80-ms C timer remain independent of ACKs. Gameplay DOWN still requires an image no more than 250 ms old and cannot follow an absolute deadline.

Stale samples are retained and discarded while waiting inside this one bounded transition; they neither admit keys nor seed a temporal pair. A sample that ages past 250 ms during evidence saving also resets the pair. This does not resend Resume or reopen an UNKNOWN transport. Full owner/module checks remain intact; no cache or identity check was weakened to make images appear fresh. Gate failure still means release and stop, not another gameplay attempt.

boundary-baseline-qualification.json reproduces zero gameplay and the stale probe failure through the unchanged full browser entry. Its generated fixture models the observed 1,049-ms stall and a returned HUD after 2.2 seconds, with a distinct timer image at 3.3 seconds. These are explicitly generated fixtures using retained native images, not another Recoil launch.

corrected-qualification.json passes six full-browser cases: that transition replay, persistent stale images, changed owner, partial UNKNOWN, delayed C ACK and delayed arrow ACK. The transition replay admits trusted generated C/Up after the fresh temporal pair. Persistent stale images admit no gameplay. Changed ownership and partial delivery retain sticky UNKNOWN/no replay. Delayed C ACK releases C independently and rejects the subsequent stale arrow; arrow UP and overall Escape remain independently bounded. Every fixture closes Chrome with code 0; expected driver refusal is checked separately.

boundary-regression.json additionally passes seven deadline/release cases and a recorded-native-image test proving that a stale first timer cannot seed a fresh pair. Cases cover absent/late gates, synchronous deadline overrun, a separate ordinary-command window, C ACK delay, overrun between two DOWNs and an invalid gate bound. The new local closure has 757 pins. This is a substantive tested harness correction at the observed boundary, not evidence that the native objective or corrected original-game gate now works. No second original was allocated to test it.

Retrieval, shutdown and remaining work

All 558 remote files / 83,664,519 bytes, comprising 491 unique SHA-256 values, were retrieved and independently rehashed before delegated shutdown. All 250 original PNGs pass decoding/CRC checks. Verification checks the actual module, 28 ACKs/one UNKNOWN, zero gameplay DOWNs, five known runtime PIDs absent, HTTP/CDP ports closed, profile and writer gone, and last observed mouse and arrows released. Local full-entry and correction browsers, drivers, ports, profiles and writers are drained. The independent normal/watchdog children exit 0 and their supervisor ends with no children. verification.json retains these checks; the original timing/cleanup limitations above remain explicit.

The one UUID-idempotent allocation was bx_6zp7wwsr, created at 09:00:17.139Z, with initial expiry 09:26:17.008Z. An observable detached 25-minute watchdog was armed before allocation, deadline 09:25:17.150Z. The box became archived at 09:13:10.383Z, expiry null. Stop receipt stop_543466a94268adb9 completed 09:13:10.462Z, forced false, error null; final-stop-receipt.json retains the final observed state. No box was resumed or deleted. Disk remained above 5 GB.

Next is a separately authorized original using the corrected, hermetically qualified closure: qualify fresh native temporal HUD before C, demonstrate actual ordinary consumption and independent releases plus visible native MENU, then take a separately bounded movement toward a freshly reviewed ramp/control. The fresh 1.1 briefing is available, but an actionable control/target must still be identified before attack. Objective completion, full27, parity, audio and simulation FPS remain open.