Aranna movement: retained rejection is capture expiry

Causal player movement remains unverified. The retained world reacquisition did not fail an ownership or input-mode check: command 49 was refused because its reviewed capture was 17.931 seconds old. The unchanged harness requires capture age at dispatch to be at most 15 seconds. No emulator or harness implementation was changed, and no new box or original was launched.

Evidence: scratch/runs/20261107-aranna-world-movement/ (R). This is a read-only diagnosis of the released point-transform run, not another gameplay attempt. Its original response49 was personally inspected again: the hero and green selection ring stand outside the snowy house, with portrait, compass, and action HUD. It supplies no causal before/after player-action pair.

Exact rejection and resolved helpers

R/diagnosis.json records execution of the retained contract against the exact saved tickets and live snapshots. All 52 relevant source/helper pins match the prior runtime manifest, including production browser/renderer input, the probe, passive snapshot, mapper, adapter, and wrapper chains. This verifies these diagnostic inputs, not complete runtime/media coverage for a new launch.

Original command Input Capture age at dispatch Recorded and replayed result
46 Canvas acquisition click 15.890 s Freshness refusal, zero effects
47 Canvas acquisition click 11.406 s Completed
48 Absolute move to (490,315) 16.700 s Freshness refusal, zero effects
49 Canvas acquisition click 17.931 s Freshness refusal, zero effects

Command 49 reviewed response48, captured at 07:56:12.805Z. Its review was written at 07:56:29.998Z; live validation occurred at 07:56:30.736Z. The capture had therefore expired before the review was written. The error is BeforeInputRefusal: capture freshness exceeds 15 seconds, from the current run's contract wrapper before the inherited contract executes. The adapter persists effectsAttempted: 0; command completion is rejected-before-input.

The actual adapter chain is point-transform → relative-move → atomic-observation → native-point → setup-input. Each retained.cjs compiles the inherited text with the current run's filename. Its require('./contract.cjs') therefore resolves the point-transform freshness wrapper. That wrapper invokes relative-move → atomic-observation → native-point → setup-input validation. The driver follows the same inheritance chain; native-point replaces toPage with the current run's point-mapping.cjs. No wrapper silently selects another run's mapping or freshness policy.

Command 49's ticket and live snapshot have the same exact Worker target, session, module, endpoint, app, geometry, protocol epoch 13, movement epoch 12, and mode epoch 25. Both have null lock and positive relative intent. A pure diagnostic call to the inherited contract without the outer freshness check accepts those saved values. This is not permission to remove freshness; the actual full contract correctly rejects them.

Live recentring and real automatic lock expiry

Response49's lifecycle now is 1192487.840 ms. Its latest warp is at 1192305.780 ms, only 182.060 ms earlier; recentringUntil is 1194305.780 ms, still 1817.940 ms ahead. The warp run is 65. The last automatic poll was false at 1152228.260 ms, 40259.580 ms earlier. That old false poll does not determine the result: the passive snapshot ORs it with current recentringActive. The saved response44–49 snapshots all agree with that formula. The active recentering signal is not demonstrated stale lifecycle state.

The existing passive trace establishes the earlier real lock transition. Command 47 requested pointer lock with user activation at 07:55:50.621Z and received a trusted lock-change event. At 07:55:52.982Z the host called exitPointerLock from releaseExpiredAutomaticLock. At that moment recentring had expired and the automatic capture hint was false. Response47 correctly shows unlocked/stable absolute input. Before command48, repeated guest warps resumed and its live state became unlocked/active-relative again. Even without freshness, command48 fails actual stable unlocked input required.

This is a real runtime transition, not evidence of a changed owner or module. The last attempt to review response49 subsequently failed in runtime-gate.cjs with kill ESRCH: the supervisor had ended. It did not enqueue a fresh action.

Checks, boundary, and lifecycle

node scratch/runs/20261107-aranna-world-movement/diagnose.cjs passes exact receipt replay, 52 retained pin checks, current-recentring calculations, trace transition assertions, and nine pure-contract negative checks: retired owner, wrong module, expired capture, partial lock transition, changed protocol, intervening movement, lost relative intent, changed exclusive mapping, and changed app endpoint. UNKNOWN/partially dispatched receipts remain nonrecoverable. Counterfactual timestamps were used only in cloned pure function arguments; retained evidence was not edited or dispatched.

These are not real-browser transition tests. No corrected helper exists to qualify, and no safe follow-on player action was demonstrated from the retained expired captures and transient lock. Under the task's boundary/stop rule, this worker stops at the exact demonstrated refusal rather than launching another original on an unproved repair. A future attempt would need a personally reviewed capture delivered inside 15 seconds, a completed ordinary acquisition if still needed, then another freshly reviewed action under the resulting real mode. Lock duration and resulting player displacement would still need testing; neither is established by the pure-contract acceptance or command47 ACK.

No journal was reset. Global serials 8257/8359 already exceed the target-proof limit of 8192; the retained target-click path remains unavailable under its unchanged quota. The acquisition path does not consume that proof, and its pure validation acceptance must not be read as fresh target-proof coverage. The prior resource broker overflow remains: no complete post-overflow resource coverage, FPS, audio, or full27 claim is made. There is no demonstrated CPU, Win32, or memory mismatch and no production repair/build claim.

Inherited actual runtime identity remains module SHA-256 88ff2708a2dec89af430fac2ba778f39857c2716e851299c3ecbfa62fd5e7d1f, 2,221,881 bytes, slot 0 / endpoint 1, legacy API / requested WebGL / WebGL 2 / ANGLE Mesa Intel UHD Graphics 620. Its runtime manifest names inherited commit d14a8e882f9cee83408ba67587591f90bd921957; current main is not substituted for that runtime identity.

The prior worker explicitly released ownership after delegated normal shutdown. Its retained terminal census has all 13 recorded PIDs absent, three ports closed, temporary profile absent, and all four locks free. A new read-only boat --no-update --json info bx_c9bc8zsn reports stopped, completed stop, forced:false, stopped at 07:59:13.590Z, and archiveAfter:null. The prior API called this state archived; the current CLI spelling is retained separately. The box was neither resumed nor stopped again. This worker owns no test box, runtime process, browser profile, port, or runtime lock to hand back.

Git SHA, relevant working-tree diff, owned diagnostic patch, exact evidence hashes, validation, publication, free disk, and final boundary are recorded in R. No source tree or media archive was created. result.json is written last.