JezzBall native system cursor and personal walls, 2026-11-15

Run: scratch/runs/20261115-jezzball-system-cursor-walls/. Result: partial. The qualified public-output route was published and one original ran, but zero walls were submitted. A consumed finite hover hit the strict late-ACK gate; input stayed closed without replay or rebind. Native Quit and persistent area/score/level progress were not proved. The original was shut down through the supervisor, with actual zero waits and unchanged launch pins. The impossible short-pulse range is tightened after this run; no second original is authorized or launched by this task. This task authorizes one fresh default registered original after qualification and publication. The prior zero-wall game remains closed. Rodent and QuakeIII startup diagnosis actually released before heavy execution. Their native, watchdog and supervisor waits, both EOFs and exact PID absence were checked; QuakeIII's controller exit 1 is a diagnostic failure, with Chrome/watchdog/ supervisor zero. Ancillary outer UNKNOWN remains separate. No box was created.

The coordinator task explicitly extends the earlier coordinator-authored cursor-image-only policy to personally reviewed current native SYSTEM cursor metadata. This is public renderer.canvas.style.cursor output, not an OS pointer image. Neither cursor pixels nor selected/orientation flags are manufactured. The complete raw board PNG remains unchanged; exact CSS, capture interval, native PID/start token, window and board hash are saved in a separate JSON sidecar. The preload passively observes the real host.set_cursor and host.set_cursor_image calls after forwarding their original arguments once. It records only actual public CSS, source and wall-clock timestamp. Placement requires a matching fresh system output after the finite hover's consumed receipt. A border resize CSS assignment alone cannot qualify. No native call, input event, cursor selection or game-state write is introduced by this observer.

Existing publication 9818430a4 establishes imported USER.173 system IDs 0x7f84/0x7f85 and USER.69 SetCursor reaching actual ew-resize/ns-resize host output. Original static segment 1:290b loads those two system IDs using NULL instance; the EXE and WEP4UTIL contain no cursor resources. This alone does not identify the original's current orientation or establish why the previous closed game had zero walls. Native Help and current native controls must qualify the interpretation before it is frozen for the original.

The new self-contained finite controller retains the released PID/socket/window identity checks, permanent binding, sticky UNKNOWN and no replay/rebind rules. Five-second complete-board reviews apply to hover and placement; menus allow thirty seconds. Each physical action has a fresh nonce and an independent UP prearmed at 1,750 ms, with normal gesture duration bounded to 1,500 ms. Capture and persistence follow UP. The 180-second Quit reserve is irreversible.

Every wall requires a personally reviewed post-UP finite hover capture. Only exact ew-resize or ns-resize metadata can qualify, tied to the same board, window and identity, with an explicit review and unchanged qualification proof. Missing, default, unknown and custom CSS are refused. Only ordinary supported left input and pointer-only hover are currently exposed. New/start/restart, second game, right-button bypass, Next Level and completion dismissal are absent. No private board state, ball tracking, solver or collision planner is used.

Transport ACK does not establish a wall. A separate personal consequence review ties each wall intent's actual before board to its after board, visible area, score, level, lives and native outcome. A completed-wall claim also needs two personally reviewed post-release board observations showing persistence and increased area/score. Another wall is refused until prior consequences are reviewed. A natural outcome requires native visible evidence; shutdown and native Quit remain separate.

The light checks passed 37 policy and 10 HTTP controls, including absent/custom CSS, metadata/qualification drift, orientation mismatch, missing finite hover, native-output mismatch/expiry, replay, irreversible Quit and causal consequence. Policy and HTTP controls are synthetic fixtures, never original gameplay or image reviews. All 12 actual CLI controls passed: finite click and UP, late ACK, consumed UNKNOWN, cancellation, identity replacement, child and parent death, silent-parent independent fallback, capture after UP, pointer-only hover without DOWN, and actual imported native system-cursor publication. The native fixture calls USER.173/USER.69 through production WASM and records actual host.set_cursor output in the running CLI's public canvas metadata. Its two system outputs were personally reviewed as metadata; it is not original-game orientation evidence by itself. No assignment of a test CSS value substitutes for this native control. Fixture release-child orphan waits remain explicitly unknown after deliberate parent death; actual native/watchdog/supervisor waits are retained independently.

The current native Help decoder read all 12 original topics, including the wallbuilder's up/down versus left/right meaning, finite lives, anchoring, partial-wall collisions, the 75% threshold, scoring and native Exit. The EXE, HLP and WEP4UTIL hashes match the earlier static evidence. The full locked build passed in 70.826 seconds; the native far-Pascal cursor suite passed on that module. Native Help was decoded again after build with the module hash recorded. Publication, frozen qualification and complete launch pins preceded the one original.

One original and its terminal failure

Route 3a59384d86cee21a401282982116cc39efb7d85e was committed and strictly pulled/pushed before launch; origin/main matched. The interpretation was frozen only after personally reviewing native Help, the static original load sites, actual native far-API output and the current CLI controls. The tested module SHA-256 is f866c8a9247f51fabd296b012d5b963c51afb97fc29f786b22798c39d4e2cff0.

All 557 launch pins matched afterward. They cover WEP4 media and sounds, WEP4UTIL, known autoimport DLL candidates, all TLBs including stdole2, fonts, the runtime/source closure and recursive CLI helpers, compiler/PNG packages, the generated passive observer, and resolved PATH Node/Python plus their shared libraries. All eight actually mapped native-process files were pre-pinned, including /home/user/.nvm/versions/node/v24.18.1/bin/node; /usr/bin/node was not substituted for the actual executable. Git plus the 101,086-byte relevant binary patch records the launch source; no source, media or WASM bundle/copy was staged. Small run-specific review/submission inputs are retained separately with final artifact hashes; the short submission wrapper was created during the run and was not a prelaunch pin or a guest runtime module.

One registered default game ran with real ticks, default batch 1,000, limit 1,200 seconds and an independently prearmed 1,220-second watchdog. The complete initial board already had two red atoms, Lives 2, Score 0 and Area Cleared 0%. No New/Start input occurred. A truthful active-board review permanently bound PID 93490/start token 275411891/window 0x10001 once. The selected bootstrap and last actual pre-hover PNGs were personally viewed and retained unchanged. They contain the complete chamber and counters. Initial/current pre-hover CSS was default, with no observed native cursor output; it was not accepted as orientation evidence.

Two scene reviews expired and one had insufficient remaining freshness. All three were local refusals before intent/prearm/physical input. A fourth current complete-board review admitted one pointer-only center hover, nonce 28bd5090-d20a-4fca-af7b-b734b45e2778, requested for 40 ms. Its receipt is CONSUMED, and native stdout records WM_MOUSEMOVE. No mouse DOWN or wall intent was sent. The strict controller requires more than 40 ms of ACK headroom; therefore every positive transport delay makes a 40 ms pulse fail that check. This is a controller admission defect, not proof of slow transport or an engine cursor failure. The independent child released promptly on parent EOF, with actual physical-release ACK about 25.34 ms after consumption and child exit 0 with both EOFs. That is not a normal consumed-pulse completion. The sticky input failure remains recorded, with no retry/rebind or wall admission.

The post-hover capture never ran because of the closed gate. Thus the original's post-hover CSS/orientation, final counters and persistent-wall outcome remain unproved. The log later announced a native Game Over question, but there was no personally reviewed final outcome image; a visible native loss is not claimed. No restart answer, second game or native Quit input followed. An orderly supervisor cancellation is explicitly host shutdown, not UI Quit. Native 93490, watchdog 93487 and supervisor 93414 have actual zero waits, both EOFs, no drain errors, heavy release, absent PIDs and closed HTTP port 8667. The invoking tool also reported exit 0; predecessor ancillary outer UNKNOWN is kept separate.

The causal tightening raises the admitted minimum pulse from 40 to 80 ms, before intent reservation, prearm or HTTP. The >40 ms ACK margin, 1,500 ms maximum, 1,750 ms fallback and all sticky failure/identity/replay gates remain unchanged. Negative policy/HTTP/current-CLI controls reject 40 ms without input; the current CLI also exercises an 80 ms pointer-only pulse and normal independent UP. The final fix is tested and published after this original, without replaying it. No change is made to the closed prior zero-wall game or the engine source.

Final validation: 38 policy, 11 HTTP and 13 actual CLI controls passed, together with the full locked build in 51.806 seconds. The prior native far-API proof still describes the identical module hash. The old published source was read from Git into memory for a pretransport negative control: it reaches policy for 40 ms, whereas the fixed source refuses before policy or any transport. No old-source file copy or original-game rerun was used for that control.