Stronghold ordinary post-preset readiness: supported boundary, 2026-11-04
An input-free settling repair is not proved. The original executable has a pending-preset application path gated by a native left-press edge. The recovered host state does not establish whether that path was entered, remained pending, or completed. Keep the first unexpected-result rejection. No driver, recognizer, patch, template, tolerance, accepted scene, or Start behavior changed.
This is the offline successor to
the c13849a61 failed attempt,
not a continuation of its closed session. Evidence is in
scratch/runs/20261104-stronghold-post-preset-readiness/; original artifacts remain
at their original paths. No browser, emulator/game, remote API, runtime, box,
public deployment, or subworker was used.
What the recovered capture establishes
The original seal verifies recovered/closed/route-native-boundary-19.png and its
state JSON under 20261104-stronghold-setup-preset-native-world. PNG SHA-256 is
9fcc66eea4ba565a7fce8bc891e811219687e1c2f5f9419f806361aee9c404fe.
Personal inspection shows Zephyr description, Normal Game, 2000/2000 gold and
empty opponents. The unchanged native classifier returns unknown: the Zephyr
description patch matches 9aad37f9…4547386d, while settings/gold is
4e42eb0d…2a7b5be, not accepted aa3ac86e…312cd. The retained ready image
20261104-stronghold-browser-village/review/zephyr-ready.png, SHA-256
e0dda59a268679e21f6bc31f15e80ea2420c51544267d4c789edd38fbd07ef93, remains a
positive scenario control (Human Advantage, 8000/2000, Rat).
Boundary19 passes the existing host readiness check: frozen 16ms, one host, owning Worker, explicit drained input and zero button mask. Navigation used 12416 steps, route 1856. Command19 was the first failure; the original deadline still had 123.043 seconds left. None of these host observations proves native preset application. There is no later sample from that failed session to use.
Original native transition
The existing original PE is referenced in place, never copied:
test/binaries/candidates/stronghold-crusader-demo/installed/App_Executables/Stronghold Crusader multiplayer demo.exe
(2699264 bytes, SHA-256
9024c533095f2227b0105b7e8a9e4cf28ec4fb206b165629787b2c97cf99f13f).
The focused test derives VA mapping from PE sections and pins the decisive
instructions; bounded disassembly is retained in this run. These are static
control-flow findings, not a trace of command19.
- Scene20 constructor
0x4259d0installs descriptor0x5ba528. Its numbered preset parent0x5ba668references activation0x42d880and a distinct hover callback0x42d850. First child0x5ba6b8carries index 1. This avoids confusing the separate generic button parent at0x5ba7a8with the numbered preset. - Hover callback, conditional on
[0x1681318]==99and[0x995674]!=0, writes its argument to description index[0x69332c]at0x42d866. Setup drawing reads that index after0x429b1bcalls updater0x42db40. Thus description alone cannot establish activation or applied settings. - Activation
0x42d880also requires[0x1681318]==99and no pending selection[0x69165c]==0. Choice 1 branches to0x42d9de, calls reset0x42d7d0, then writes pending selection 1 at0x42da1cand saved choice 1 at0x42da26. It clears[0x1bd26a4]at0x42da71with ESI zero. - Updater
0x42db40checks pending selection. At0x42db53it compares native[0x1bd26a4]to zero and branches past application when zero. With a nonzero edge it calls0x49e720, rechecks pending selection, calls application handler0x42e490at0x42db72, then clears pending and sets[0x692404]=1. - Main-loop call
0x576a83passes mouse object0x1bd2670to poll0x468930. Poll compares current held+0x40with prior held+0x4c, clears+0x34, and sets it to 1 only for a newly held transition at0x468b50.0x1bd2670+0x34 == 0x1bd26a4: the updater gate is the native press edge. Choice-1 application0x42e577calls reset, sets configuration fields and calls opponent addition0x4a4780with argument 1.
This rules out treating the pending path as an unconditional next-frame apply. It does not prove that command19 took that path or that an extra click is correct. Do not add input, repeat preset activation, or replace the guard based on this static finding. Other native paths or a render lag remain possible.
Existing driver and tested boundary
The unchanged run-owned 20261104-stronghold-setup-preset-native-world/route-driver.cjs
SHA-256 is bb38b5787c277c94c2913a9ddba72a8663b9fd2fb706a88e320bcb83419633de.
route-press-preset requires the separately reviewed exact preset1 hover,
performs balanced DOWN/UP through explicit per-event drainage, captures, and
immediately rejects anything other than the existing expected zephyr-ready.
There is no post-press settling phase. This identifies an implementation
boundary; it does not establish that adding a wait repairs the native result.
Run node test/test-stronghold-post-preset-readiness.js. It verifies original
sealed capture/state (adjacent gzip decoded in memory only after matching its
uncompressed seal), exact-ready positive, original PE gates, and executes the
unchanged driver's state machine with an explicitly synthetic IO/base adapter.
The route controls accept exact ready, reject unknown/world/unchanged setup,
preserve terminal first-failure without extra input/steps, reject an expired
live deadline and exhausted aggregate budget, enforce the 4096 maximum and
64-step chunks, and reject foreign input telemetry. The synthetic adapter does
not establish native delivery, execution, or eventual convergence. Existing
navigation 16384, route 4096, arrival 3072, stability 768, per-ACK 64, command
40s and original browser 360s bounds remain unchanged.
Exact missing causal evidence and next dependency
Before authorizing a driver wait, a separately scoped fresh original attempt
needs a bounded read-only guest trace spanning preset DOWN consumption, the
activation callback, UP consumption and setup update/presentation. At minimum
capture native PC/callback argument plus [0x1681318], [0x69165c],
[0x690f7c], [0x69332c], mouse +0x34/+0x40/+0x4c, and entry/return of
0x42e490 with the associated displayed settings. The current host snapshot
contains none of that causal state. Specifically distinguish:
- Activation never entered or was rejected by its native gate.
- Activation queued choice 1 and is waiting on a press edge.
- Application completed but a later input-free native update/present is pending.
Only case 3 with observed convergence and a demonstrated bound supports the requested settling phase. Cases 1/2 require diagnosis of native input/order, not an unconditional delay. Any future observation must fit the same aggregate, ACK and original deadline caps and reject genuine first failures; it must not resume this closed failed session or execute Start to test readiness.
Outcome: tested, evidence-grounded supported boundary; readiness gap remains unrepaired pending that native trace. No driver ownership was claimed. No box was assigned; box ID/state/expiry are not applicable and lifecycle remains with the coordinator. Serialized full build and focused test results are retained in the run. Git is the source archive: exact base SHA and owned new-file binary patches are retained, without source/media/module archives.