Asghan: same-plane frontal combat attempted; outgoing consequence unresolved

The fresh registered software demo reaches an orc on the same broad dirt surface, avoiding the earlier raised-edge approach. One fast attack and one distinct swipe consume the documented native chords. No attributable outgoing hit, enemy defeat or pickup is established. The new boundary is outgoing consequence attribution during moving, same-plane contact; missing input delivery and the earlier vertical separation are not sufficient explanations for this run. There is no demonstrated emulator mismatch and no emulator change.

Evidence: scratch/runs/20261108-asghan-frontal-target/. Tested main SHA: 707b0f5dab3c0602a865c6e09a9e3d65fa13a9ee, with relevant uncommitted source patches retained. Recovered executing guest Worker module: cc438f727a39e39d1016045d37ae95f60cf8f9793e6b117e14beabb80c27e07a. This differs from the preceding Asghan run's module; the current module was qualified locally before allocation. Only this note changes in Git.

Reviewed living sequence

Read CLAUDE.md, the native-attack-consumption and current-browser-action notes, and shipped readme.txt sections 3–4. Native title → Enter → visible NEW GAME → Enter starts one original sequence. Opening camera frames are not movement evidence. No death, restart or resumed original is joined into this sequence.

pier-neutral shows the grounded hero on the wooden pier. Up for 1500 ms reaches the dirt by the wall; Left for 800 ms follows the open path. A neutral path-settled capture resolves the following camera before another move. Up for 1000 ms advances along the outer portion of the broad path, with the orc and barrel ahead-right. Two separately reviewed 200 ms Right turns aim toward the advancing orc. The hero does not take the previous extra forward step onto the raised strip beside the wall.

frontal-heading shows both figures on the continuous dirt below that strip, with the hero turned toward the orc and the sword extending into their gap. This supports same-plane frontal preparation, not an exact collision-volume or strike-angle measurement. The orc moves during the live sequence.

One 1500 ms Ctrl+Up fast attack follows. Six intermediate screenshots and same-plane-fast show combat motion, changing relative heading and a standing, armed orc afterward. Transient red marks and overlapping models do not resolve whose strike caused which effect. No enemy health display or decisive isolated enemy hit reaction supplies attribution. No outgoing hit is claimed.

The fresh neutral contact-current capture shows the living orc now ahead-right on the same dirt. One distinct 1500 ms Ctrl+Left swipe-and-rotate addresses that changing frontal angle. same-plane-swipe still shows the orc standing, armed and near the hero. Attacks stop after these two different actions; there is no attack loop, ranged-ammo inference or further attack from the old pose. No clearly actionable pickup was identified, so Space was not sent blindly.

The visible hero bar measures 160 pixels at row 666 after the fast attack and in contact-current, then 120 after the swipe. No bar is displayed in frontal-heading; its zero detected green pixels are not zero health. These measurements supply no enemy damage claim. The final image shows the hero alive. Browser closure then ends the run without a claimed menu success.

The remaining distinction is a missed or interrupted attack versus damage without a uniquely visible consequence. Same-plane preparation alone does not resolve the exact heading/contact at the guest's strike evaluation. A next attempt should seek a visible actionable pickup with disappearance/inventory change, or a separately justified way to isolate an enemy reaction while both characters move; repeating these attacks is not new evidence.

Current controls and unchanged inputs

The released software native-CDP driver retains independent releases, exact ACKs, owner/window/canvas/module binding, 120-second scene freshness and sticky UNKNOWN closure. Current production Calc Worker qualification passes trusted Alt, Alt+Up and Ctrl+Up DOWN/UP consumption. Generated controls reject changed owner/window/canvas/renderer/module, wrong image, stale scene and UNKNOWN before dispatch; pending/rejected ACK controls still attempt the independent release.

The already qualified input-import observer is reused unchanged to record attack timing. It forwards actual imports exactly once, makes zero extra polls and writes no guest memory. Its transparency and actual production Worker checks pass; the original reports one instantiation, zero errors and zero dropped transition rows. This is observation with overhead, not a speed test.

Attack Native hold, UTC Actual guest down returns Released returns
Fast, Ctrl+Up 06:58:54.022–55.522 Both 0x8000 at 54.215 Both zero at 55.532
Swipe, Ctrl+Left 06:59:48.717–50.217 Both 0x8000 at 48.784 Ctrl zero at 50.300, Left zero at 50.301

attack-consumption.json contains the actual transition rows and counts. The observations corroborate consumption, not hits. DOWN/UP ACKs take 1–2 ms; the independent releases are attempted within 1 ms of their scheduled deadlines. There are 15 exact command ACKs, nine pulses, 11 balanced trusted native key pairs, zero runtime UNKNOWNs and no held keys. No replay follows any unknown.

All 120 original fixture paths, 30,448,512 bytes, match the preceding media references. All 711 runtime pins match before launch and after closure. HTTP responses independently match 69 original files, including executable, DLLs, manifest and native setup outputs; the remaining 51 have no recorded request. No installation, setup change, alternate executable, private handler, guest patch, forced scene/configuration/clock/seed or source/media archive is used.

All 15 bound captures retain process 1000, the actual guest endpoint/Worker/ instance/module, page renderer, canvas and visible process-owned Win Alis window identity. The actual route is software/DirectDraw presented by CanvasRenderingContext2D, with no owning 3D Worker/context. No GL readiness requirement is imposed. GL/full27, parity, audio and unmeasured FPS remain open.

Evidence and normal shutdown

verify.cjs passes current controls, original-media hashes, exact ACKs, balanced events, capture identities, actual attack-poll transitions, 44 CRC-decoded PNGs and retrieved artifact hashes. All 198 remote references, 14,737,528 bytes, are retrieved and independently SHA-256 checked before normal stop.

The first immediate post-quit read-only inventory fails its drain assertion; that failed receipt is retained. No gameplay command is replayed. A separate settled diagnostic inventory then passes: controller and Chrome PIDs absent, both ports closed, profile/writer lock removed, no held keys, Chrome exit 0, and all runtime pins unchanged. The earlier assertion does not identify which drain predicate was transiently false, so its cause is not inferred.

Fresh UUID-idempotent box bx_vambj3yj has an observable independent 25-minute deadline watchdog prearmed before allocation. Initial expiry: 2026-11-08T07:21:50.933Z. Normal stop receipt stop_b645588227876f39 completes at 07:00:48.859Z, forced=false, error null; final state archived, expiry null. Both lifecycle delegates exit 0 and the supervisor has no remaining children. Local qualification also closes Chrome normally. Disk remains above 5 GB. No old box is resumed/deleted, no nested worker is launched and no deployment occurs.

Git SHA plus relevant tracked/untracked binary patches, owned-input patches, recovered executing module and original-media references preserve recovery without archiving the source tree or copying media into evidence.