Disciples II fresh map-action attempt, 17 November 2026

Evidence: scratch/runs/20261117-disciples2-map-action/. The single authorized original demonstrated a missing menu button-DOWN at the public input boundary. It did not reach Single Player, NewQuest, a map, or any town/hero/turn consequence. Ordinary Quit reached the demo's closing promotional screen, but native exit was not demonstrated. Local owned-host cancellation ended the run. No further original or input retry was attempted.

Exact runtime and admission

The source was published main 2d3a248f3d1693b7f086e1a880def91c80d0b973, including mandatory sealed-launch change fe44f6f19468fb9127e55e595921d75752fcf20e. All 4,514 Git source bodies independently matched the recorded revision. The owned patch was empty and foreign shared WIP was excluded. The privately compiled module SHA-256 was 5c9f7126daff2e586685c0dd2ad46b79c5e43e8dbf569aea2919b733b075614a. The source, preload, module and manifest descriptors had seals 15. Git and patches provide source recovery; no source, media or Wasm archive was written.

The Nov16 original recipe was retained exactly except for its new holder, manifest identities and allocated loopback port. Preparation preceded the fixed 3,600-second native clock, independent 3,620-second watchdog and irreversible 300-second Quit reserve. Real ticks, actual guest Workers, the original EXE, 1024×712 scene, completion session and input policy were unchanged.

Preflight and postflight matched all 128 original media hashes and 127 support manifest rows. The EXE remained adba242a0a3563651baffa6ec7d9a6c0978b7ae844a1cccedb2a1f4e8eb7b1f8. The original exported scenario was used only as installed media; no f.sg, imported save, configuration, EXE or clock modification was supplied. The five real DLLs and existing three native OLE32 fallbacks passed the unchanged checks.

The same sealed holder passed mandatory admission negatives and actual Node Worker/child-process inheritance checks. A generated public USER fixture ran with the threaded original runner shape. Its personally viewed complete scene bound successfully; finite Down confirmed in 9.862 ms and independently released in 137.477 ms. Its authored Escape exit completed with actual zero native and watchdog waits and both EOFs. Reviewed-context units passed. This task changed documentation only; it does not claim a new canonical full build. The published sealed-launch implementation's earlier fullflock qualification remains separate.

Newly demonstrated input blocker

original/current-menu-target.png was personally viewed before the one finite Single Player click at canvas (750,44). The host receipt was CONSUMED in 78.769 ms, with physical UP acknowledged in 226.150 ms. The complete frame original/menu-consequence.png, captured 108.298 seconds after consumption, still shows the main menu and highlighted Single Player button.

The ordinary runner's existing public check_input log is more specific than the host receipt. At native client point (601,37), it records WM_NCHITTEST and WM_LBUTTONUP (0x202), but no WM_LBUTTONDOWN (0x201). The earlier mouse-move record is also retained. No further Single Player click, Enter activation, queued click or private command was sent. The endpoint stayed healthy; no UNKNOWN or rebind occurred.

Reading the retained Nov16 log establishes the same narrower observation for its ineffective NewQuest click: the old Single Player click has DOWN and UP, whereas the subsequent NewQuest click has hit-test and UP without a recorded DOWN. The later Enter DOWN/UP pair is present. These are retained log facts, not repaired provenance or a claim that the old run completed cleanly.

lib/cli-guarded-input.js explicitly defines CONSUMED as host handler consumption, and independent release purges the intent's remaining tagged queue effects. Thus the current receipt cannot establish delivery into the guest message pump. Release/pump timing is a plausible explanation for the missing DOWN, but neither its precise removal point nor guest window-procedure execution is proved. This task made no speculative helper/core repair and did not lengthen a hold or relax a bound. Core scene/window changes owned by the concurrent VB worker were excluded from this runtime and left untouched.

Quit and actual closure

After the blocker, a fresh complete review authorized one pointer move to the visible Quit target. A second fresh review of original/quit-target.png then authorized one finite Quit click at (750,394). Both intents completed with independent physical UP; the click's UP was acknowledged in 136.585 ms. Its public log has both DOWN and UP at native client point (601,332).

The resulting original/quit-promo-final.png shows the closing promotional screen 141.827 seconds after the click. The process remained live, and there was no explicit further exit control on the viewed frame. No dismissal key or click was guessed. Final-choice confirmation passed the controller's 500 ms assertion, but this final branch retains no separate numeric confirmation-ms field. Native Quit completion remains unproved.

The local cancellation file requested owned host cleanup. Native, watchdog, supervisor and outer actual waits completed zero with both EOFs; the cancellation receipt explicitly records nativeQuit:false. The holder closed normally and its independent wait completed zero with both EOFs. No deadline watchdog fired. Final PID/start-token and port checks are recorded separately from actual waits. Every original capture completed within 758 ms; the unchanged menu freshness, 500 ms confirmation, independently prearmed 1,750 ms physical UP and immutable 3,000 ms operation limits remained in force. No gameplay freshness claim applies.

The old Nov16 launcher, outer, native and holder identities with recorded start tokens were absent at this task's initial audit. No old endpoint was contacted and no old process was signalled. Missing old native/watchdog/launcher/holder joins remain UNKNOWN. The previous unsealed day-one map stays disqualified.

Only named selected full screenshots are retained, with omission/hash records for unselected observations. No box was created: ID null, state NOT_CREATED, expiry null, stop receipt not applicable. Free space on /home stayed above 1.5 GB. No workers or public deployment were launched.

Next work should reproduce the missing-DOWN boundary in a generated public fixture with a deliberately slow ordinary message pump, using the existing public input logs and unchanged finite bounds. A causal delivery repair needs that evidence before any separately authorized original; this run does not authorize replay, another endpoint or another original.