Registered Win32 Pegged: personally reviewed default puzzle

Run: scratch/runs/20261114-pegged-reviewed-outcome/.

Honest partial: zero verified jumps, then native Game → Exit with actual exit code 0. The untouched default scene already contained six blue pegs, so eight removals could not be achieved without changing puzzles. No numeric remaining-count text was displayed. I did not start another puzzle.

I personally chose the geometrically legal vertical jump r3c4 over r2c4 into empty r1c4 from the current full scene. One source click was consumed and its independent UP acknowledged. The settled full scene had exactly the baseline SHA-256: no visible source selection and no removed peg. I therefore submitted no destination click and recorded no move, count decrease, win, solved state, or no-moves end. A prior source intent expired at 5.50 seconds and was positively NOT_SENT before DOWN; the later source used a new scene and new intent.

The native Game menu and its Exit entry were individually reviewed. Exit completed normally after 457.820 CLI seconds (458.119 seconds between supervisor launch and actual native wait), before the irreversible 180-second reserve. Native, watchdog, outer-child and outer waits all returned 0 with actual waits and both EOFs. The independent watchdog was unfired, the heavy lock released, all four waited PIDs absent, and the endpoint closed. The final independent release found the owner already exited: OWNER_EXITED, with matching actual native wait/EOF, physical UP unacknowledged. The native log also reported one held main-owner critical section at exit; this is not an all-thread-clean claim.

All 422 frozen source/module/media pins matched after the run. The registered media/import closure and original executable matched before and after. No box was created: ID/expiry null, state NOT_CREATED, stop receipt not applicable.

Validation: 23 policy tests, 12 HTTP controls, 10 actual CLI controls, and flock /tmp/wine-build.lock bash tools/build.sh passed under the heavy lock, after personally reading GL startup and StarCraft actual terminal waits/EOF and heavy release. Six selected full scenes remain, named by result.json; unselected duplicates were removed with their hashes recorded.

Next work, outside this closed single-original run: determine the native selection/drag interaction in a separately authorized task before attempting any destination. The current controller deliberately stops when a left click has no visible selection. Eight jumps also require resolving the conflict between that target and the six-peg default; no new pattern is selected here.

The registry entry pegged is unchanged. The original is test/binaries/entertainment-pack/pegged.exe, 43,792 bytes, SHA-256 bb5a4e90b71363cb7df5f818fc9a2de9841dbdfcb8a74498756b2c11438fe4c1. The current shared DLL resolver loads the original msvcrt.dll PE. It routes ABOUTWEP, NTDLL, KERNEL32, GDI32 and USER32 imports through built-in handlers; the adjacent original ABOUTWEP file is pinned too. No alternate Win16 binary, download, media copy, registry edit, clock/seed/config write, or guest memory inspection is part of this run.

tools/pegged-controller.py adds human-review policy over the released, read-only tools/barrage-guard.py transport. The controller interprets no board pixels and generates no moves. Each source and destination is independently chosen from a submitted raw full scene, with a five-second precision deadline. A destination requires a fresh visible selection and the immediately preceding source's identity-matched consumed receipt and independent UP acknowledgement. Before another source, a separate review must show the source and jumped peg empty, destination occupied, and visible remaining count reduced by one. Input acknowledgement alone cannot record a removal or natural outcome.

Binding records the truthful bootstrap or board phase and delegates permanent PID/start-token/window ownership to the guarded transport. Bootstrap permits only the native Start route; board admission requires a reviewed actual default board. Natural outcome and the 180-second Quit reserve stop gameplay irreversibly. Every native Exit step needs its own reviewed scene. The original recipe is fixed at 900 seconds with an independent watchdog at 920 seconds, real ticks, default batch size 1000, and one default puzzle.

The policy tests cover changed identity/image/geometry, expired reviews, bootstrap versus board, one native Start, pending source/destination ordering, actual removal attestation, unchanged count refusal, outcome evidence, irreversible Quit, server deadlines, finite pulses and native Exit routing. HTTP and actual CLI controls reuse the released transport regressions for identity, unique nonces, positive all-unsent cancellation, mixed/UNKNOWN terminal behavior, independent UP, parent death and actual waits. Their Solitaire fixtures and synthetic board attestations are transport tests only.