MDK2 fresh pending-scene attempt: first gap before selection
Run scratch/runs/20261104-mdk2-fresh-pending-scene/ stopped at a fresh visual
qualification gap at3200 slices. Personally VIEWED native-selector-76.png,
SHA-256 52066472b64a5e63e650195bf842ec3221534336121acd7762254c84ccea989e,
shows readable New Game and a white controller icon without the required red
controller highlight. The immediately preceding3184 capture had no readable
New Game. Each image was viewed in a separate call before a later review/advance.
No historical review authorized this run.
The accepted next.py opening route requires newGameLabelVisible=false;
main-menu observation and New Game input require
controllerIconHighlighted=true. Neither describes the actual3200 pixels.
The native capture has file=startscreen, conscheme=title, pending scheme=nil,
freeze=nil, selected=0, current=lang, showmenu=0, Level0/checkpoint0, no native
input and consumed Escape/Enter rest. The unchanged native New Game readiness
predicate passes, but that does not supply the missing fresh visual predicate.
Unknown-phase STOP therefore applied without submitting a stronger review or
calling construction identity. Browser command78 was the final shot; command79
was quit. No New Game/Medium selection, pending quantum, construction, comic,
cinematic or playable readiness is claimed. No Enter, movement, fire, mouse,
reload, second Chrome or fault-clear retry occurred.
After quit, an exact fresh gap description was authored locally in
first-gap-reviewed.json, explicitly without authorization. test-first-gap.cjs
reproduces the three visual rejections and separately confirms native readiness
passes. Its initial diagnostic incorrectly expected native rejection; that failed
local result is retained, and the corrected test verifies the actual boundary.
The local recovery validator initially disallowed all clock commands; it was
corrected to verify the accepted menu helper's single ordinary tick20 at command56.
Neither local diagnostic failure executed runtime commands or changed helpers.
Closure and verification
Verified all23 predecessor seal files/88737bytes from accepted 3c92703d7, plus
all37 adequate-lifetime-menu input hashes. Recovered missing neighboring gzip
evidence only through the two coordinator archive indices, checking original
size/SHA. Copied only small run inputs; retained unchanged helper dependencies
at their original sibling paths and checked their actual remote hashes. A new
assignment-bound runtime_io/release adapter targets only bx_3bh7zkjy; the
unchanged driver uses its new directory as RUN and creates a fresh browser dir.
Its closeBy was freshly bounded to10:56:10.675Z, under six minutes from launch,
with recoverBy11:16 and expiry11:28:13.640. No expired predecessor closeBy was used.
Before Chrome: actual Puppeteer24.31.0 CJS/ESM imports, all432 runtime pins,
isolated HTTP full hashes, cache98page/12worker graph, empty browser process
inventory, and actual staged/sibling dependencies passed. Actual captured states
confirm guestWorker=true, worker backend and one frozen host. Runtime stayed
pinned to source 1329b1ba6eb00144ff77d4cd92a8513d4fc953ad, canonical wasm
cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee,
compat 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3;
latest-main unrelated VB changes were not substituted. The inherited runtime
directory has no .git; a post-close git-provenance query failed and is retained.
The recorded canonical source SHA,432 path hashes and local Git SHA/binary diff
remain the provenance, without claiming a remote Git checkout.
Same-worker focused verification passed30 pending controls, five route controls, 18 unchanged scene controls, the fresh gap test, and complete recovered-history validation. These53 predecessor controls were run locally, not merely reused. Only documentation is committed; no tracked production code changed and no local heavy build/browser was run. The pending512-slice/96-command cap, shared540-second anchor,600000ms TTL, reserves and all ACK/history/native/fault/deadline guards were unchanged. Pending/construction/comic ledgers were never created.
One Chrome9597 opened10:50:18.853Z and closed normally10:53:46.413Z (207.56seconds), before its bounded deadline. Validation proves80 contiguous successful ACKs,3200 complete slices with individual step<=64,13 captures, 200 matching served hashes, balanced Escape only, empty input at every capture, and no browser/controller fault files. All268 closed browser artifacts/ 6,092,457bytes were recovered and fully SHA-verified. Post-close process identity and socket checks prove Chrome/driver absent, HTTP43005 closed,432 runtime pins and40 staged input hashes unchanged. Evidence includes small new-input patches, relevant binary diff, Git SHA and a final SHA seal; no source archives, repeated modules or game media were copied.
bx_3bh7zkjy is left RUNNING, API idle/healthy/hydrated, expiry
2026-11-04T11:28:13.640Z. Custody and shutdown return to the coordinator; no box
lifecycle or allocation operation occurred. The next work is to qualify an honest
input-free observation route for this specific readable-but-unhighlighted menu
state before another separately authorized fresh runtime attempt. This assignment
stops at the documented first gap; publication status and exact commit are in the
run handoff and messageboard.