LMarbles distinct targeting and completion deadlines, 2026-11-16

New finite intents keep three deadline domains separate. The reviewed capture and personal review must still be fresh at DOWN admission and actual native consumption (5 seconds in game, 30 seconds in menus). The native consumption guard retains the minimum of targeting expiry, operation expiry, independently prearmed release cap and gameplay cutoff. Its existing hold cleanup is unchanged.

The operation starts once on entering pulse, before policy and lease work, and ends 3000 ms later. It no longer ends at targeting-image expiry. The existing 2.15-second admission allowance still applies against the operation deadline; it does not require 2.15 seconds of targeting-image life. NEW strict intents reserve the existing 500 ms confirmation interval against targeting expiry, including the native integer-millisecond deadline floor. This derived reserve adds no numeric limit. Admission is checked before prearm, after prearm and immediately before dispatch. Expiry while arming or checking policy disarms an unsent child safely. DOWN confirmation remains 500 ms, independent physical-UP acknowledgement remains 1750 ms from dispatch, and release confirmation is also bounded to 500 ms. Release is not a new targeting decision and does not require an unexpired targeting image.

Postrelease progress and full-frame capture consume the remaining original operation time. Their request timeouts cannot start a fresh three-second completion window. No numeric contract, capture/review timestamp, running deadline, native paint binding or permanent-context rule was enlarged or reset. The prearmed release child still runs after failed/lost DOWN confirmation; UNKNOWN remains closed to capture, binding and new input.

The fresh original exposed one additional boundary: native validation can pass before calling its effect handler, while its published consumedAt timestamp falls after the deadline. NEW receipts now must also satisfy floor(dispatchAt * 1000) <= consumedAt < guard.deadline. A violation permanently closes the controller; the independent UP is still awaited. This is a receipt validation fence, not a claim that it can undo an effect already made by native code. An actual native fixture crosses expiry inside the effect callback and proves rejection, independent release and zero subsequent HTTP/rebind/replay.

Evidence: scratch/runs/20261116-lmarbles-distinct-admission-completion-deadlines/result.json. The published411626622 original and earlier sg1 outcome are historical and unchanged; neither endpoint is contacted or restarted by this work.

Verification

Fifteen native cases pass (the final fourteen-case suite plus one additional native confirmation-reserve refusal). Nine budget tests cover the separate exact freshness and operation admission boundaries, unchanged limits and actual prelaunch 3600/3620/300 watchdog/reserve records. Eleven reserve/lifecycle tests pass. Native tests use authored Solitaire fixtures with actual threaded USER windows, the production HTTP transport and an independent release process, never an original game as a fault fixture. They exercise near-expiry consumed DOWN with UP after image expiry and timely completion, stale admission/consumption with spare operation time, spent prearm and dispatch budgets, slow confirmation/UP/completion, focus/owner cancellation, new capture/nonce after proven NOT_SENT, independent fallback and permanent closure with zero later HTTP requests.

The initial, pre-reserve positive native boundary had 465.8 ms remaining at dispatch and consumed DOWN 442.9 ms before targeting expiry. Independent UP was acknowledged 410.5 ms after expiry (876.3 ms from dispatch); the complete operation took 2485.7 ms. The physical-UP fault additionally stalls the actual native event loop, retaining the independent child and unchanged deadlines, and triggers the 1750 ms failure.

The first diagnostic suite retained one startup bind refusal: Solitaire was still painting its initial deal. Fixture-only settling precedes its single bind in the final suite. Production paint binding is unchanged. An earlier 180 ms positive fixture was safely NOT_SENT when prearm exhausted its targeting life; this is retained as a diagnostic test expectation failure. The final positive leaves 900 ms before starting, without changing production limits. Its DOWN dispatch has 799.9 ms remaining, consumption precedes expiry by 788.1 ms, and UP is acknowledged 39.6 ms after image expiry. The native stale-consumption and crossing-effect fault fixtures bypass only the derived admission reserve to exercise the independent native and receipt fences; they retain the actual native expiry and immutable operation start. The required canonical flock /tmp/wine-build.lock bash tools/build.sh passed and shared outputs were released. Independent runtime sources come from published Git358a20bae58e01713036332cb443426d98cd7b48 plus recorded owned patches, with canonical sealed module83703369cc49c3e16077a4abd43def27262487f83dd52583d1b433adf7095f5f. No foreign uncommitted source is adopted. The 86-file original-media verifier passes.

Single fresh original: consumption-timestamp blocker

The one original launched after material publication85f7dd61a and successful native proof, with 3600 native seconds, the independently prearmed 3620-second watchdog and irreversible 300-second Quit reserve recorded before launch. Personally reviewed menus confirmed Normal/Original, then the ordinary level-8 confirmation entered Tier1 Puzzle8, Moves96. The final complete frame shows Moves94: two slides are verified. There is no clear, tally or Puzzle9.

Full-frame PNG display intermittently appeared blank despite byte-identical captures. Whole-frame JPEG presentations, including smaller complete displays, were retained alongside their original PNGs. No crop, OCR, board-memory read, solver, route/progress import, queued input or clock/config/EXE change was used. Expired reviews were cleanly NOT_SENT, with new captures/nonces for subsequent intentions. One review description overclaimed a blank duplicate presentation; its attempt was already NOT_SENT without effects and an append-only correction is retained. No historical record was rewritten.

Audit found down3 reports consumption 3.349 ms after targeting expiry and 4 ms after its integer guard deadline. Its UP took 214.7 ms from dispatch and operation completion 2113.0 ms, both timely in their separate domains. No visible slide followed that DOWN. The native pre-effect guard and post-effect timestamp straddle expiry; this does not prove exactly when within the callback the physical effect occurred. It does demonstrate absent receipt-time validation. The controller had allowed later intentions before the audit discovered this gap; those records remain unchanged. Ten submitted finite intents have UP ACKs within 178.5–281.3 ms and completion within 1787.1–2511.0 ms.

The audit permanently closed this new endpoint and prohibited further HTTP, capture, rebind and input. No original was restarted. Host cancellation produced actual native exit0 and both EOFs, plus actual watchdog/supervisor waits. It was not ordinary native Quit. The subsequent receipt fence is qualified only with native fixtures; there is no second original attempt or post-fix completion claim. The new demonstrated blocker is the original's late consumption receipt and its resulting permanent closure. Further original work requires new authority.

The first sealed holder's invoking tool ended143 without a normal-close receipt; its PID/descriptors are absent. Each original/fixture had already verified and resealed its own immutable inputs. A later qualification refused the vanished holder before launching native processes; those failures were retained and stopped. A fresh owned holder recompiles the same pinned module for final fixtures. One concurrent canonical compiler was killed137; it is retained as a failed build; sequential canonical builds after the receipt fence and final derived reserve both passed with actual tool waits. Synthetic fixture partial-JSON startup/transport failures and an earlier final-suite failure are retained; corrected controls and the complete final fourteen-case suite pass. Final holder source/module hashes were verified before normal shutdown; its actual tool wait is exit0 and its descriptors are closed. The first holder's abnormal tool termination remains a distinct limitation.

No box has been created; box ID null, state NOT_CREATED, expiry null, stop N/A.