AoWII intercepted response terminal, 2026-11-18

Evidence: scratch/runs/20261118-aow2-intercepted-response-terminal-cause/.

The exact genuine Ml42ND50.bpl discrepancy is reproduced outside game execution. The unchanged older coherent shell and its actual WineAssembly.fetchAssetBytes streaming reader consume 468,992 bytes, SHA256 c503b230760bfc9f3c315f91911c3b90471368115fabf227e4e95d0e34a30993. The existing response-stage interceptor captures, verifies and continues that same request. Its owning Network session then reports canceled ERR_ABORTED without loadingFinished. The old adapter closes UNKNOWN on the second read. No abort, reader cancellation or earlier cleanup command initiates that failure. Historical scopes, including request 2346097.272, remain closed and failed.

The AoWII correction opts this route into the existing actual-consumer contract in browser-consumer-completion.cjs and browser-response-body-collector.cjs. It observes application-requested reads without cloning, teeing, read-ahead, replacement responses or refetch. An unpredictable invocation name in the native fetch initiator binds the exact Network ID to the default script context, unique context, document, frame and timeOrigin. Copied consumed chunks are hashed independently and must equal the immutable intercepted bytes. Binary selection excludes JSON, script and image consumers; unsupported child owners fail closed. Consumer completion remains mandatory for every selected invocation, even when Network finishes normally.

The separate terminal assessment retains every failed Network terminal. It accepts authenticated consumer completion only with one exact GET/200 request and response, verified server finish, a resolved earlier Fetch continuation, and exactly one eligible canceled ERR_ABORTED. A missing Network terminal still fails. Other transport errors, partial bytes, wrong hashes, cancellation, stale contexts, duplicate/unmatched owners and delayed completion remain failures. The whole browser is disposed on UNKNOWN without in-page restoration commands. No emulator, DLL, registry, game-state, clock or rendering behavior changes.

Real controls include the old failure, 32 corrected genuine package reads (six failed Network terminals retained), the packaged native VariantInit control and real abort, reader-cancel-before-EOF, truncated response, wrong-pin, unsupported Worker owner, timeout and replaced-document failures. The generated native control actually instantiates module 87ef39648a61ed6f54ca4ab8de1926de02c651eb2d99655015dc7e5de5407009 and reaches Native VariantInit returned vt=0; it is not original game output. The synthetic collector suite retains its 129 cases, including race ordering. Focused AoWII tests reject 22 malformed or insufficient terminal receipts.

A separate native trace of repeated genuine package reads records four ResponseBodyLoader::DidCancelLoadingBody callbacks and request-bound failed ResourceFinish events. This demonstrates a Chromium renderer body-loader cancellation boundary despite authenticated consumption. The internal C++ bookkeeping/reentrancy mechanism is not proved. No upstream browser repair is claimed. The earlier shorter startup trace did not cover the package requests and is not used to identify that boundary.

In a separate browser lifetime with no Fetch interception, the exact pinned host _readBody method consumes the same genuine package with strongly retained response/controller/bytes, an un-aborted signal and normal loadingFinished. That single comparison does not prove an interception-independent recurrence or isolate an internal cause. The real traced intercepted cases establish the boundary relevant to this correction. Independent receipt audits bind all 32 consumer invocations and the four native callback intervals to their exact failed ResourceFinish request IDs; all four consumer hash completions arrive after the failed Network event. The first corrected 32-read case similarly authenticates its six post-failure completions, with no synthetic terminal.

The source is explicitly the older coherent 0a823095c plus recorded patch, published 81ad6c802 manager and 87ef39648 module, with the same 435 media pins. It is not current main. Text recovery checks the immutable expected hash; media uses the existing files without substitution or copies. Qualified passive Device7/native raster controls and independent UP remain unchanged.

Single original and concrete remaining blocker

Correction 46f8ea1d5 was committed and immediately published with strict pull/push after focused tests and a passing locked build. Intermediate builds failed on another owner's changing callback census and region declaration; those files were left with their owner. The final build passed. Four additional production-collector synthetic races cover completion before/after failure, wrong hash and owner loss. An actual reader cancellation was repeated after the no-restoration teardown correction. The fresh independent UP control passed.

One registered original then launched through the personally reviewed whole launcher image. The actual main Worker ownership read binds slot0, endpoint1 and the exact older 87ef39648 module. The new original's required package is Network ID 2394660.272, not historical 2346097.272; its complete owning chain, server UUID and actual consumer hash are retained. The last and final transport snapshots have 759 required bodies, 465 independently authenticated binary consumers, no required failed terminals, no missing receipts and no first fault. The independent audit passes all 759. Normal Network completion also passes for these requests; no original canceled transfer is silently cleared.

The main menu and default Inioch's Legacy setup were personally reviewed. Ordinary Scenario, Single and Start clicks each retain trusted DOWN/UP evidence, prearmed independent release, normal acknowledgement and actual release-process wait/EOF. No game settings, media, registry, clock or guest state were substituted. There is no recorded application fault before termination. The control fixture's VariantInit result is not substituted for original execution.

At 03:38:48 UTC, the native supervisor received SIGTERM about 179.5 seconds after starting, before its configured 900-second deadline. The independent watchdog recorded no deadline/disk trigger and minimum /home free space 1,933,516,800 bytes. Its cancellation followed the supervisor's exit. The driver labels every incoming SIGTERM independent watchdog SIGTERM; that literal label does not identify the sender. The sender and reason remain UNKNOWN. The run ended during the natural 15-second wait after the acknowledged Start click, before world-first.png or its command completion. A command-tool lifetime limit is a hypothesis, not a demonstrated initiating cause.

No endpoint request, observation, input, replay or second original followed recognition of closure. Actual completed Device7 ground draws, effective raster coverage/last writer, terrain, army consequence and ordinary Quit are not reached. The qualified observer's retained native fixture and historical terrain images remain controls/reference only. No rendering defect or rendering repair follows from this attempt. The remaining concrete blocker is the premature native-supervisor termination; the transport prerequisite passed.

Custody and retained evidence

Every browser control and the original used native-supervise plus a prearmed independent finite disk/deadline watchdog, without a RAM gate. Actual original waits are driver124/null, Chrome null/SIGTERM and watchdog0/null, all with both pipe EOFs. Native-supervise records exit124/null, supervisor SIGTERM, no poll error and leaked:false. The external invoking command actually waited124. Profiles were removed only after retaining/hashing unique wait/EOF records. The no-interception control and native callback trace are preserved separately. Earlier reusable-adapter controls performed restoration during cleanup; the published whole-browser teardown avoids those in-page commands after UNKNOWN.

Evidence retains exact Git, owned new-file/helper patches, immutable source and media identities, one unique module, selected named whole PNGs and recovery recipes. No command-stream capture, frame series, media copy or source archive was written. There is no candidate-corpus entry for AoWII, so candidateId stays null rather than using another application's card; global historical evidence checker failures are reported separately. Final result.json is written last. No boxes: ID null, state NOT_CREATED, expiry null, stop receipt null. No workers, Python, Wine, foreign edits or deployment. Ownership is released when finalized.