TIM schema preflight and fresh visible progress, 2026-11-15
Evidence: scratch/runs/20261115-tim-schema-checked-belt-progress/result.json.
Published route: a05fb264d08157c39e72d76ae69325a6982679d7 on main.
The one authorized original is closed with exact partial progress. Both
predecessors explicitly released source/build/runtime; their direct waits and
process closure were personally checked. The coordinator repaired exactly two
stale shared census hashes before this task began heavy work.
The preceding 7d27c1d28 route / 5d5f8b21c report remains closed UNKNOWN.
Its malformed from/to drag was an operator schema error, not an engine
defect. It is neither reclassified nor retried. Its actual retained final
complete scene was personally viewed: three Puzzle 1 motors, no belt.
The exact CLI accepts one JSON action after --gesture, with click fields
kind,x,y,button,holdMs, and drag fields
kind,x,y,toX,toY,button,holdMs. The strict constructor checks the
exact field set and types before reading a live scene, creating a review or
intent, or calling policy. It supplies no defaults, aliases, coordinates,
intent, sequence or retry. Caller-selected holds are integers 80–1400 ms;
the existing independent UP remains at most 1750 ms. Frame bounds are checked
against actual capture dimensions before review construction. Existing policy
and native ambiguity/UNKNOWN and positive UNSENT recovery rules stay intact.
The original scanned manual was read, including personally viewed printed
pages 12 and 13. It describes drag, a separate placement click, then a handle
click. Belts require a click at the first attachment point and another at the
second, with the red line between them. The complete retained first-belt and
settled pre-Run scenes from 20261114-tim-advance-puzzle2 were personally
viewed. The former shows an attached lower belt; the latter shows three belts
and a ramp. The retained Puzzle 2 dropped-cage scene was also personally viewed;
handles alone do not prove a direction change. These are historical images,
not progress in this new run. Every new action still needs its own personally
reviewed current complete scene, new intent and nonce.
Native dummy controls exercised the actual CLI command path for seven individual click, placement drag, placement click, belt pickup, two endpoint clicks and handle-click shapes. Sixteen malformed/missing-field commands produced zero adapter calls, queued events, buttons and native callbacks, including 200 subsequent native steps. The fixture contains no TIM model and cannot prove a game mechanism. Only a changed actual original scene can do that. Native Quit was not reached; host shutdown is reported separately. No boxes or workers have been created.
Fresh qualification passed: 26 policy tests, 12 schema tests, native mouse
context, all 13 existing native controls, the exact CLI controls above, and
flock /tmp/wine-build.lock bash tools/build.sh. The 78 named wait records
include 16 direct native and 16 watchdog waits; all actual waits have both
EOFs, and all 78 recorded PIDs and matching process groups are absent. All
442 qualification pins are unchanged. Unexposed historical outer transport
EOFs remain UNKNOWN; they are not substituted for direct process evidence.
The route was published before the fresh original started. All 13 installed
files and all 15 original media/manual pins matched the preceding closed run.
The current source, host, build and actual native toy module were pinned:
442 input pins remained unchanged through direct closure. The native module
was 346006 bytes, SHA256
44dbe86c807add8ad8c1f8010ed455ab8b011d282f9387dfaf42481109710fa1.
No source/media/module bundles were made; Git SHA and binary patches retain
relevant uncommitted hunks. Foreign WIP stayed untouched.
The original had a 900-second native deadline, independently prearmed 920-second watchdog and irreversible 180-second native Quit reserve. Ordinary title/manual-page-one bootstrap was separate from board review. The actual printed page one was personally viewed before its three icon selections. Every action had its own personally reviewed complete scene, manual choice, nonce and intent. Ten stale reviews received native positive-UNSENT receipts and were followed by independently reviewed new scenes/intents. No aliases, action queues, automatic retries or game-memory/config/physics writes were used. Caller-selected mouse/key holds were 100 ms; no capture or thinking occurred while held.
The visible new result is stronger than the preceding run: three separately placed Puzzle1 motors, three individually attached two-endpoint belts, an anchored ramp, runner Run, airborne basketball, then native PUZZLE 1 COMPLETED! with total bonus 626. Continue opened the Replay/Advance dialog; Advance opened Puzzle 2: Tutorial: Mirror Images. The ordinary Play control opened its full editable board.
In Puzzle2 the right mouse cage was separately dragged and anchored. An ordinary hover exposed its smart handles without picking it up. Clicking the visible red circular handle at the lower left actually flipped the cage: the pulley moved from its right side to its left, while the brick enclosure and cage location stayed fixed. This changed actual scene proves this handle and direction result; the dummy controls do not. The basketball was then carried too far right and visibly crossed out because it overlapped the wall. The final complete scene retains that invalid carried placement and the flipped cage. No Puzzle2 belt, Run, completion or native Quit is claimed.
The next explicitly chosen hover to move the carried ball inward was written
after policy admission, then the native boundary raised review expired.
For that scene, capture time was 1794729851.022; submission was
1794729856.014871 (age 4.992871 seconds), and native rejection was
1794729856.030 (age 5.008 seconds). The permanent UNKNOWN is preserved; this
written command is not reclassified as positive UNSENT. There are 36
written commands, 35 consumed commands and 35 actual post-UP consequences.
All input stopped immediately; no cleanup input, rebind, replay, unchanged
retry, recovery, or second original followed. This is not the earlier
malformed-schema error and does not establish a TIM mechanism defect.
The run ended after 336069 ms. Direct native exit was 1, watchdog exit 0,
supervisor-child exit 1 and launcher exit 1; each exact child has an actual
wait and both EOFs. All recorded original PIDs and matching groups were
personally absent, and source/build/runtime were explicitly released for the
next worker. nativeExited:false with default nativeExitCode:0 is host
shutdown, not native Quit. The in-memory native-created tim.cfg overlay
was hashed; installed media stayed unchanged and operator game writes were
false. Historical/unexposed outer transport EOFs remain UNKNOWN.
Ten selected complete actual result-named PNGs retain the first belt, three belts, ramp, Run, completion, Advance, Puzzle2 instructions, handles, changed direction and exact final partial scene. Other new frame PNGs and temporary manual rendering were removed with hash records. No owner-sent images were omitted (none were sent). No boxes were created: ID/expiry null, state NOT_CREATED, stop receipt not applicable. This task stops on the exact partial; any further original requires fresh owner authorization.