Recoil force-field objective attempt, 2026-11-08

Run: scratch/runs/20261108-recoil-forcefield-objective/.

Objective incomplete; the first new gameplay limitation is loss of the tank while locating the beach exit. One fresh original reached Mission 1 through Software → OK → Game → Start → New Game, default user, Medium, with the opening playing naturally. There was no restart, combat firing, starting-biped destruction, objective-counter increment, demonstrated force-field removal or onward access. No emulator mismatch was established and no production code or game media was changed. The earlier movement and score-1550 enemy destruction acceptances remain in the linked prior notes, not this run's deliverable: current gameplay and native objective.

The installed game/docs/index.html manual was read, including arrows, relative aim, R/O mission panes, M wireframe map, C turret-to-chassis centering, V camera perspective and Backspace chassis-to-turret centering. The fresh native original/mission-brief.png repeats “1.1 Operation Short Circuit”: the force field above the beach blocks the hilltop enemy base; find and destroy its power control. Its map marks the start on the island's east side. The manual describes green reticle triangles as damageable targets, not proof of mission relevance.

Personally reviewed route evidence, relative to the run:

In-hold frames original/held-1794122247983.png and original/held-1794122271668.png were also reviewed. They show cliff approach and incoming cyan fire near the tank. No outgoing attack or power-control damage is attributed to these effects. The last readable objective counter remains the initial 0/4; no post-death numeric objective reading is available.

The initial locked shared-tree build failed before compilation at the producer census for peer-owned host.js, lib/guest-thread-host.js and lib/thread-manager.js WIP. build.log preserves that failure. Following the coordinator's messageboard clarification, the tested runtime instead uses published Git blobs from 59abb98bf4bd1fdb51e4aacb30038256042885ad, individually staged without the active peer WIP. The published WAT closure was compiled in memory with unchanged, hash-checked published compiler tools; no second checkout or source archive was created. published-compile.json pins every WAT input, and compiler-provenance.json verifies the compiler files against that commit. The actual CDP-recovered original and local Calc modules match the compiled pin: 7ec49cbbae4cb177909fd8fd9d3599713ee5f2a5ffb7f7e8c890848a540052bf. This does not turn the failed shared-tree build into a passing build.

runtime-pins.json records all 726 staged closure/input files and their Git/artifact providers. The existing untracked lib/wininit-boot.js is retained as its explicit patch. The tested published-source patch is empty; excluded-peer-wip.patch records the untested shared changes separately. All 107 original files plus the existing manifest match the supplied pins; all 113 local HTTP fixture/dependency checks pass. Game files are referenced by path and SHA-256, not archived into evidence. The executable remains 82632b1b46153014af0caf1e0c9dfdebbb0d8482ba65b2735064f10972afa076. The first fixture-identity.json mistakenly used empty app.files; its zero-file success was vacuous and is withdrawn, preserved separately as fixture-identity-invalid-zero-files.json. The corrected report follows the actual localFileManifest: 106 manifest entries + executable = 107 originals, plus the manifest pin, 76,019,820 bytes altogether. Every supplied path, size and SHA matches a staged runtime pin. fixture-regression.json rejects empty, missing and duplicate registrations and a wrong staged executable hash. The preparation helper was corrected for future use; the original was not replayed.

The unchanged qualified capture/admission chain retains exact owner/module identity, screenshot hashes, original capture timestamps, the 120-second review bound, independent release and sticky UNKNOWN closure. Real-Chromium generated controls pass owner replacements, stale/slow capture, slow sampler, partial delivery, UNKNOWN/late ACK rejection, relative motion, trusted mouse down/up, and trusted R/O/M/C/V/Backspace down/up with released key states. R/O/M semantics were observed in the original; C/V/Backspace were qualified for native delivery locally, not exercised in gameplay. All 21 original numbered commands have exact successful ACKs; no original UNKNOWN or input replay occurred. The fire-named menu operations only select New Game and Start.

Two local staging failures are retained: a compile helper initially read the wrong compiler-result property, and the first local Calc server supplied Git bytes with working-tree content lengths, truncating changed JavaScript. Both were corrected before box allocation. The second Calc run confirmed the actual published module and live owner chain. All local browsers/PIDs/ports/profiles were drained before the original. These are helper failures, not guest fixes.

The environment is headless desktop Chrome 151.0.7922.108, V8 15.1.206.16, 1024×768, cross-origin isolated, Threads enabled, registered unchanged route /index.html?debug&threads&app=recoil_demo. The original selects Software. Observed rendering is software page Canvas2D, with no owning D3D render-worker endpoint; the toolbar's WebGL selection is not backend proof. Original process 1000 / endpoint 1 / slot 0 and document epoch 3b6f7b42-7007-4f56-8768-5905d2b51eac retain software renderer object :5 in the passive monitor. Physical hold intervals and passive guest/canvas changes do not measure simulation steps or FPS.

All 307 remote artifacts / 26,957,532 bytes were retrieved and independently SHA-256 verified before normal stop. The first post-quit inventory asserted before all PIDs disappeared; that exact failure is retained. A separate later read-only observation and settled inventory confirm all 15 recorded PIDs absent, HTTP/CDP ports refused, profile/writer lock absent and inputs released. No browser or input was relaunched to obtain that observation. The independently prearmed 25-minute watchdog and normal-stop child both exited 0; the supervisor finished with no children. Box bx_8547pftx is archived, stopped at 07:19:52.506Z, final expiry/archive-after null (initial expiry 07:37:54.570Z). Stop stop_b32ed826a009544a completed at 07:19:52.526Z, forced=false, error null. final-stop-receipt.json retains the final GET. Sampled disk stayed above 5 GB. No resume/delete, RAM gate or public deployment was used.

evidence-check.json passes published-provider pins, the real nonempty fixture closure, 68 original PNG CRC/IDAT/hash checks, exact ACKs, bounded releases, actual module equality, retrieval hashes and complete runtime/lifecycle drain. Owned result containment/schema and note whitespace checks pass. The repository evidence checker retains seven unrelated containment errors.

Next: a separately authorized fresh attempt should take the now-visible uphill ramp earlier, with a shorter observation/decision interval before incoming fire exhausts health. Identify the actual power control from fresh terrain/reticle evidence before attacking, then require objective text/counter change or proven force-field/onward-access consequence. No source fix is justified by this run. Full27, parity, audio and unmeasured FPS remain open.