UT GOTY: CLI memory-only handoff prerequisite fails

Evidence: scratch/runs/20261118-ut-goty-memory-store-install-gameplay/. This is explicitly a memory-only route. Durable installation remains an OPEN, separate outcome. Genuine UT Setup and the original game were not launched: the required authored installer/application handoff failed first. Neither genuine partial disk overlay nor the unique CD1 ISO was written or copied. No full VFS/media bundle was exported.

The clean published graph is Git 5dc1e3a489e47bba7a924a19f56efcfe9c26583e, including payload-reuse commit f55de39b8. Compiling its Git objects in memory produced module SHA-256 a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8, 2,418,710 bytes. The CLI and runtime modules were loaded from those exact Git objects without adopting held worktree changes. Its local test support modules were separately checked byte-for-byte against that revision. Shared foreign runtime diffs, including the held timer work, remained unchanged.

The authored ordinary PE installer initially confirmed that its installation path did not exist. Through CreateDirectoryA/CreateFileA/WriteFile it installed an embedded ordinary child PE at C:\MemoryInstall\child.exe (6,433 bytes) and C:\MemoryInstall\payload.dat (4 bytes). It reopened both through ReadFile and compared every byte with the embedded originals. That check passed. The retained manifest records the matching sizes and SHA-256 hashes; only the small authored inputs/executables and logs are retained, not a VFS snapshot. The child program is authored to read the installed payload and create a read receipt.

The same installer then used ordinary CreateProcessA, WaitForSingleObject, GetExitCodeProcess and ShellExecuteA against the installed child. On the public CLI default, CreateProcessA returned TRUE with handle 0x000e3001, its wait returned zero, and ShellExecuteA returned 33. Neither call actually started the child: its debug/read receipt was absent, the VFS contained no child receipt, and no emulator child fork occurred. The installer deliberately exited 42 to report this failed handoff. The enclosing Node process exited zero; that zero is not a successful application-handoff result. There was no crash or runtime UNKNOWN and no later endpoint/replay/extension.

The missing boundary is concrete. Although lib/overlay-store.js:10 describes memoryStore() as the default CLI backend, test/run.js has no call to that function. Its overlay attachment at line 6446 is guarded by OVERLAY_DIR and constructs only nodeDirStore. Without that option, writes live in the ordinary volatile VirtualFS, with no attached memory-store journal. The executed process did not load lib/overlay-store.js at all.

Actual ordinary CLI child execution is opt-in through --spawn-processes (or an app's equivalent setting). That transport unconditionally calls saveVfsToHost(vfs, inDir) at test/run.js:4335, then starts the child with --vfs-tree at line 4367 and --save-vfs at line 4372 before merging its output back. Those are native disk snapshots, not an in-memory store/session handoff. Readonly media source-context forwarding does not remove the writable snapshot. No public option selects a memory-only alternative. This path was audited, not executed: taking the full snapshot would violate this task's restriction. No fake filesystem, guest injection, forced callback, extraction, configuration, registry/save patch, extra memory setting or RAM gate was used.

Thus the bounded cause is a missing supported CLI memory-only application handoff, before UT admission. This is not a new UT Setup failure or evidence about its game compatibility. No runtime repair was attempted or published; native/browser old-fail/new-pass repair claims and a locked code build do not apply to this documentation-only change.

Install 1200 / total 2400 / final Quit-cleanup reserve 300 seconds were declared before the authored launch. No UT clock started. The authored control had its own fixed 30-second CLI cap, 50-second supervisor cap and 60-second cleanup cap. Its real runtime was 13 batches / 0.031 seconds after loading. The independent 100 ms disk watchdog remained active through preparation, native exit, supervisor wait, both output EOFs and board-watcher cleanup; its observed minimum was 1,863,434,240 bytes, above the 1,500,000,000-byte floor. Its independent finite deadline was not reached. Native supervision recorded no survivors; all recorded owned PIDs were subsequently absent. Actual prepare, supervisor, guard and board close/EOF receipts are retained. The terminal tool returned exit zero for the controller; it does not expose separate stream EOF flags for that outer boundary, which are not claimed.

Whole-file verification after closure confirmed both genuine disk partials' indexes and every referenced blob unchanged, including their inodes. The unique 649,633,792-byte ISO remained SHA-256 e184984ca88f001c5ddd52035d76cd64e266e26c74975161b5ed72366c74704f. Recovery/source-mount recipes, exact Git and relevant binary diffs, module and artifact hashes live in the run. No UT installed manifest or completion, startup/gameplay images exist because UT was never admitted. The authored volatile installed state was lost when its CLI process exited; the retained authored inputs can regenerate it, but it is not a persistent installation.

No browser, agent, worker, test box, Python, Wine, deployment or held-title work was invoked. Box ID, expiry and stop receipt are null; state is NOT_CREATED. After publishing this note and sealing result.json last, ownership is released and this task stops.