TIM smart-handle investigation, 2026-11-15
The route was natively qualified and published before one fresh original.
The original ended as an honest partial at fresh Puzzle 1: no Advance, Puzzle 2,
direction change, belt, Run or puzzle completion was earned.
Evidence belongs to
scratch/runs/20261115-tim-smart-handle-puzzle2/.
The two preceding Puzzle 2 notes were read. Actual manual pages 12–16 and six selected original prior images were personally viewed. The prior cage body click picked up the cage; the apparent direction-handle click displaced it to invalid placement. Neither image proves a flip. The current session must distinguish floating, anchored, selected, handle-ready and direction states using visible, persistent consequences, before belt endpoints and Run.
The new host controller supports a separate native hover. Clicks first move the pointer with no buttons and let actual native execution handle that move, then independently arm release before DOWN. This is a prospective input-route investigation, not an established emulator defect or a proven handle gesture. There are no puzzle coordinates, puzzle interpretation, solver, configuration imports, game-state writes or action queues in the original route.
An external host watchdog is started before its native child. It bounds the whole supervision period and each held input independently. The supervisor owns both child pipes and retains actual waits and both EOFs. It ignores launching-tool HUP/TERM; explicit cancellation requests release and terminal UNKNOWN. A lost or expired operation cannot replay or rebind. Holds are bounded below 1500 ms and independent UP is armed at no more than 1750 ms. Captures are immutable and prohibited while held. Gameplay review expires at five seconds, menus at thirty. Quit begins irreversibly with 180 seconds remaining in the single 900-second native lease; the external deadline is 920 seconds.
Fifteen policy tests, host syntax checks, the full locked build and fresh native qualification pass. The executed 16-bit INT33 probe verifies separate movement/left press/left release (conditions 1/2/4), right press/release (8/16), held movement at the chosen drag endpoint, and balanced actual callbacks after cancellation and independently scheduled UP. Normal holds are 100 ms; the deliberate independent-UP case releases after 301 ms. Parent death follows an actually executed native press; the independent guard closes and the native child is actually waited with exit 17. Cancellation and timeout remain terminal UNKNOWN in their separate diagnostic processes. These checks do not model TIM or prove a smart handle. Their actual rendered DOS captures are blank, not synthetic game images. The existing native callback test also freshly passes full interrupted-context, caller-stack and RETF restoration, masks, resets, nonrecursive lifetime and native range/ratio programming.
Thirty-six actual child waits with both EOFs and 36 absent PIDs/process groups
are audited. All 405 runtime source pins are unchanged across qualification.
The actual freshly compiled toyVM module is
44dbe86c807add8ad8c1f8010ed455ab8b011d282f9387dfaf42481109710fa1.
No general emulator defect is demonstrated or production engine path edited.
Foreign WIP is preserved and excluded from publication. Git SHA plus relevant
binary diffs, owned-new-file patches and source/media hashes provide recovery.
Fresh Sims651's 05:14:56 release and Q3public's 05:25:15 release were personally verified: 37/51 named actual wait files with both EOFs, direct native/watchdog/ supervisor closure, and 108/82 absent recorded PIDs and matching groups. The heavy lock was observed free before qualification. Historical missing outer tool waits remain UNKNOWN and are not inferred from absence or a new-work ban.
No box has been created: ID null, state NOT_CREATED, expiry null, stop N/A.
The changed route was published on main as
e5b50b567d766fbceb99479848c1cf14cdd99b12 before original input. All 425
launch source/media pins were unchanged through shutdown; the installed
13-file closure was fresh and no state/configuration was imported.
The five delivered, individually reviewed operations were Enter at the fresh title, three ordinary manual-check icon selections using actual manual page 1, and PLAY from the fresh Puzzle 1 instruction scene. Their visible consequences were the manual prompt, accepted answers, Puzzle 1 instructions, and the fresh editable Puzzle 1 board. Observed holds were 100–101 ms, with matching independently armed releases. No button remained held at closure.
The first construction drag was refused before delivery: its personally viewed full scene was 6.304 seconds old, exceeding the five-second gameplay review limit. This operator review deadline miss caused irreversible UNKNOWN. No command for that drag was written; no cage moved. There was no replay, rebind, second original, or later game input. This does not demonstrate a native engine defect. The separate-hover route therefore remains unproven for TIM smart handles. Floating/anchored/selected/handle direction, belt endpoints, Run and completion were not reached in this original.
The original ran for about 80.5 seconds. Terminal UNKNOWN caused orderly host
shutdown, not native UI Quit. nativeExited: false is preserved; its default
nativeExitCode: 0 is not evidence of Quit. Direct actual native exit 1,
watchdog exit 0 and supervisor-child exit 1 all have retained waits and both
EOFs. The launching tool returned 0 separately; its separate pipe EOFs are
unexposed and remain UNKNOWN. Across qualification and publication there were
45 named actual waits with both EOFs and 45 absent recorded PIDs/matching
process groups at runtime release. The independent deadline was armed, did not
fire, and was closed. Native UI pause was not observed; the real clock was not
modified. No operator game writes occurred; the native overlay remained empty.
result.json names four retained actual original screenshots: title, manual
prompt, Puzzle 1 instructions and the final untouched Puzzle 1 board. Other
run-local rendered images, including blank native callback probe captures,
were pruned with hashes recorded. Logs, actual waits, source SHA/diffs and
pins remain in the run directory. Runtime/source/build were released to the
next worker after direct closure verification; no further original is
attempted. The task stops with this exact partial and honest host shutdown.