Stronghold setup transition onward, 2026-11-04
Evidence: scratch/runs/20261104-stronghold-setup-transition-onward/.
Predecessor: quiet RPC original,
commit 3384ac75d69d2ef55f660553ff125b86e4fce186.
The single authorized original attempt stopped at its first fault: another unrecognized partial setup frame. The causal recognition repair passes offline; there was no actual retry. Complete setup, preset selection, Start, world and ordinary gameplay remain unproved.
Exact entry and preflight
The local contract binds stronghold-setup-transition-onward-2248 and its exact
stronghold-quiet-rpc-original-gameplay-2230 dependency: OWN 22:30:00Z, LOCAL
HEAVY RELEASE 22:39:17Z and file RELEASE 22:45:12Z. Foreign, missing, quoted,
stale, future and assignment-suffix records reject. Browser/bootstrap/command/
cleanup bounds remain 720/60/40/180 seconds, with task cutoffs 23:35Z/23:38Z.
The fresh composed entry passed all 2,429 filesystem pins, actual process/nonce binding, private ancestor flock, real filesystem initialization and HTTP server creation. The plain browser mock threw at launch; an independent child-spawn prohibition recorded zero spawn attempts. The server closed. The real dependency gate was used both before launch and in the later offline repair check. Caller actions/fields and six unique startup capture names passed preflight; the actual composed startup replay covered 8,448 steps and 132 ACKs. All 1,016 HTTP content, cache and isolation checks passed.
The actual launch Git SHA is 69713b13c8415407f4603e8297532a738c1dde5d.
The launch pins and relevant binary diff are retained separately from the later
repair pins. Both existing modules and all 1,856 original media paths were
hashed in place; no media/module copies or source snapshots were made. Normal
module SHA is 78a0b1d61b23428a1ae909d739440ae0de059bbbd84a69a2ad3e4c37471ad664;
compatibility is 99b579cd17b57f94145df92ae6d28110f24fc284d13a1b2c0ebe3596bd10a77f.
No guest/configuration writes occurred. No archived inputs needed restoration.
Observed rendering and native progress
I personally viewed the full fresh selector and startup captures, title at 8,448, main menu at 8,576, Crusader hover at 8,640, ChooseType at 8,768 and Custom Game hover at 8,832 before their ordinary controls. Loading artwork was still visible at 6,400 in this run. An inaccurate initial description of the 2,304 image was explicitly corrected in the append-only personal inspection log.
The balanced Custom Game gesture completed in 1,536 steps at 10,304, with mouse
release drained. Continuous native history recorded setup return at boundary
22,783,070, callback return and Custom Game path completion at 22,789,897.
Arrival observations at 256/512/768 still displayed the previous hover screen.
Command 15's 1,024-step arrival sample captured partial setup at total step
11,328 and failed at 22:54:22.448Z: first unexpected natural scene: unknown.
I personally inspected the entire fault image. It shows translucent Human/Lord Crusader panels, numbered 1/2/3 presets, Deathmatch Game, the Crusader Demo row and “Not available in the demo” over the sword/shield background. This is a partial render, not complete layout or readiness. Different sample timing and pixels do not establish that it is further along than the predecessor.
There are 177 complete 64-step ACKs at the original 16ms. The last continuous native report has 39,098,384 boundaries, 60 records, 51,439 bytes, no observer fault and no pending frames. Input queue is empty, active/last input null and mouse mask zero. The remaining 10,901,616 native boundaries are headroom only, not proof that complete rendering or gameplay fits the unchanged limits.
Offline causal repair and limits
The predecessor recognized one exact partial-frame identity. This newly
observed full-game RGBA identity is
5cf84f22b5e0e11ba4b5a945b1bd031febd8be322cffa382f9c02670c6df237f.
The helper now retains both exact identities in a frozen list. Geometry remains
the opaque 860×668 game region at (10,194) in a 1200×900 capture. Recognition
grants only setup-transition-observation in bounded arrival, never complete
setup or input readiness; other unseen frames still reject. It is not a general
fade classifier.
Regression replay runs the composed capture and arrival functions against the actual fault image/state: OLD returns unknown and rejects; NEW permits arrival observation. Seven pixel mutations, geometry/alpha failures, known-scene precedence, arrival cap, terminal mailbox and scene20 no-readiness guards pass. The predecessor frame regression also passes. One continuous finite 50M/360s native history, byte/record/owner/nesting/input/ACK/command/arrival guards and original clock remain unchanged. No history reset, cap increase or readiness permission was introduced. The existing diagnostic route still ends before preset/Start; reaching another partial image cannot justify removing that guard.
Focused local-contract tests, both-module original initializer tests, title and
route tests pass. flock /tmp/wine-build.lock bash tools/build.sh passes.
Post-repair composed entry again passes 2,429 pins and the real dependency gate,
plain mock plus independent spawn prohibition with zero Chrome. Its separately
named offline pins never replace actual launch provenance. Released large logs
are gzip compressed only after SHA-256 roundtrip verification, with mappings.
Closure
The first fault requested normal Browser.close; Chrome 1325427/start ticks
187975499 exited code 0/signal null at 22:54:22.843Z. Node 1325328/start 187975379,
external flock 1324072 and wrapper 1324080/start 187974208 are absent; wrapper
exit is 0. HTTP 41707, DevTools 39209 and prelaunch mock HTTP 33243 refuse
connections. The exact profile was removed after checking process ownership;
external flock was free at 22:54:32.879Z. Assignment-bound LOCAL HEAVY RELEASE
was posted at 22:54:44Z. No additional actual command or Chrome followed.
Box ID/state/expiry: NA. Coordinator retains shutdown ownership. After publishing this repair, stop. A future separately authorized attempt needs its own exact entry/dependency preflight and later rendering/native-budget evidence before any preset/Start/world readiness claim.