UT GOTY progress-control observation and installation

Evidence: scratch/runs/20261118-ut-goty-progress-control-install-completion/. The previous installation scope is closed and its unique partial overlay is preserved. No request or input is sent to that scope.

An authored real modal dialog reproduces control.progressUnavailable's qualification predicate in direct native, normal CPUWorker and browser modes. The live class-17 child has style 0x50000001, guest procedure 0x401000, generation 3 and initialized native values min 7, max 913, position 318, step 10. The strict native getter returns unavailable solely because the child is subclassed. The guest callback chains ordinary updates and painting through CallWindowProc; its own ordinary UpdateWindow consumes creation/update damage. This establishes a causal observer defect for this supported storage shape; it does not retrospectively prove the old original's unsaved state.

The strict native_progress_get contract stays unchanged. A separate read-only native_progress_storage_get validates live class, supported procedure, non-null initialized native state and its mapped span. JS fences child and parent generations, actor, thread, procedure, style and parent on both sides. The storage contract identifies the procedure and paint owner; it does not claim virtual PBM results or authorize painting over a guest callback. Capture continues to include all controls and checks range/position. Refused captures retain bounded numeric discriminants (including missing-state status and exact out-of-range values), without titles or arbitrary memory spans.

Matching authored old refusal/new qualification passes in native, CPUWorker and real browser routes. The old module is b9b3ba8e195212293e955e778292aef27a0b73817179b12fc1620e272a3ecef9; the new module is 204a3300f6706bd0da127fc0f55b1a0335f4b61e5c2543a6919bb976bcb50e24. The whole browser PNG shows the real modal dialog and partially filled bar. Focused tests cover absent state, unsupported procedures, generation changes, identity changes, retired owner, destruction, invalid range and position, read-only memory equality and the existing reviewed-input metadata/target restrictions. The first authored attempt lacked a naturally serviced paint; the final fixture uses its ordinary UpdateWindow. An initial scene test command named a nonexistent file; the correct context suite was subsequently run.

Original installation/startup outcome will be appended after the affected original run. No installer completion or game startup is claimed here.

The affected original used published code 81b092da3, module 204a3300..., unchanged CD1 SHA-256 e184984ca88f001c5ddd52035d76cd64e266e26c74975161b5ed72366c74704f, genuine D:\System\Setup.exe, inherited D:\ root, default C:\UnrealTournament and default choices in a new normal writable overlay. Launch was 09:17:02.750Z. Ordinary Install consumed token 5/invocation 116 at 09:19:22.016Z; independent UP acknowledged at 09:19:22.333Z with actual release-child wait 0 and both EOFs. The admission free space was 2371952640 bytes; subtracting the conservative 702545920-byte expansion left 1669406720 bytes, above the 1.5 GB floor.

The original's former observation blocker is repaired. Its first post-Install capture retains BOTH progress controls, including the initially hidden rows: HWNDs 0x10043/0x10044, generations 67/68, style 0x50000000, parent 0x1003d generation 61 and guest procedure 0x10579235. Both have initialized native 0..100 ranges, position 0, step 10. A subsequent read-only live diagnostic confirms strict getter availability 0, storage status 1 and actual positions 50/19. The saved installation PNG shows Engine.dll and a partially filled bar. An offline comparison on that SAME retained original snapshot rejects the old predicate and accepts the new one with no context/control failures. This is stronger evidence than assigning a cause to the old scope's unsaved state.

Installation did not complete in the fixed window. It reached compressed Maps\AS-Frigate.unr expansion, with the genuine file progress advancing 1 -> 6 -> 14 -> 25 -> 40 -> 45 -> 60 percent and total progress 89 percent. The last personally reviewed whole capture was 09:25:56.631Z; all runtime observations ended there. Execution samples were active (yield 0), including Setup progress callback 0x104145fd and byte-copy loop 0x1041d5d3. The latter increments a byte index and calls the array-add import for each byte; retained disassembly documents the sample, not a proved throughput defect. No crash, progress-state failure or completed installation was observed. The precise bounded result is unfinished normal map expansion before the installation cap; the underlying throughput cause is not established. Original game startup and ordinary game Quit were not attempted or proved. There is no timer extension, retry, reopened scope, guest/media/configuration/registry patch, or performance repair claim.

The fixed install deadline was 09:26:15.547Z. The auxiliary install guard's initial marker collided with the consumed Install action's completion file, so it exited immediately; a replacement retained the ORIGINAL deadline. Its enclosing tool subsequently returned 143 and the guard disappeared without a child-wait/EOF receipt. Filesystem-only cancellation was issued at 09:26:29.904Z (14.357 seconds after the install cap, inside the predeclared 40-second cleanup window). No later runtime endpoint was requested. Native, supervisor and disk watchdog have actual exit-0 waits and both EOFs; the owner-complete record retains the latter two child waits. The enclosing launch tool returned 143 and the enclosing-owner wait is missing. These are explicit custody gaps, not successful waits or a game Quit result. The original scope is closed UNKNOWN and was never replayed.

The independent 100 ms disk guard recorded a 2017161216-byte minimum through runtime closure. A later filesystem check fell to 1132089344 bytes; the board was alerted, no new runtime/build was launched, and no foreign files were removed. The new unique partial overlay and the predecessor's 16 hashed overlay artifacts are preserved. The original's new overlay remains partial; manifest and blob hashes, media verification, Git patches, module identities and recovery instructions live in the run. An initial unretained board watcher also lacks actual wait/EOF proof; its replacement is separately retained. No box was created: ID, expiry and stop receipt are null. Foreign host/render/NE/AoW2 and Hexen helper work, and held BloodII work, were preserved.

Publication correction: commit 81b092da3 accidentally included one pre-existing uncommitted timer observer call in $run, whose definition remained in held foreign WIP. The shared-tree build passed, but clean Git could not compile. The corrective commit removes that unrelated call; after publication the call and its exact untracked review bytes are restored as uncommitted held work. No BloodII run or timer investigation occurred. The corrected clean Git module is a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8 (2418710 bytes), compiled from Git with only that single-line removal and passed through the native/normal CPUWorker fixture and negatives. This differs from the original-tested 204a module, which contained the unarmed foreign timer observer; both exact identities and patches are retained. The original was not rerun against the corrected clean module.