AoW II: package completion discriminator remains inconclusive
Evidence: scratch/runs/20261106-aow2-package-cause-fresh/ (ignored).
Tested main edf9b29c96c5d9933cb541752bf296d6d5d8c823, with relevant
binary diff and new runtime-source patches retained. No production host,
driver, collector, emulator, registry, guest or media change. No original
launch, terrain acceptance or army command occurred.
The required-package completion failure remains unresolved. The predeclared stream-versus-existing-arrayBuffer control did not qualify a causal consumer difference: the stream cell reproduced the selected package failure, while the arrayBuffer cell's immutable first fault was a different, earlier seed. Both faults remain fatal. No further launch or speculative fix followed.
This follows the generated comctl control and the registry investigation. It establishes no common cause with the preceding original's comctl32 failure.
Authorization, inputs and observation
Preparation stayed read-only/run-local while queued behind fresh Aranna menu-phase2021 and legacy POINT/NONE2021. Their FINAL releases and independent cleanup preceded the coordinator's explicit 21:02:20.187445–21:47:20.187445 UTC runtime/source/build grant, including cleanup. The worker rechecked predecessor cleanup and acquired the runtime locks before tests or browser execution. There were no nested workers or boxes.
The predecessor's unexecuted control-plan.md and observation-draft.cjs were
read without editing their originals. All 12 retained evidence hashes matched.
Raw owning-session Fetch/Network events, ordinary caller metadata and
UUID-linked server/NetLog evidence were inspected first. The historical
observer logged command resolution but discarded result objects; exact old
Fetch result envelopes cannot be recovered from it. Its cancellation coverage
also omitted stream/reader cancellation and reader release.
The single hypothesis was conditional: progressive host consumption might cause the generated package completion cancellation while the host's existing arrayBuffer branch completes it. Direct comctl had already succeeded through the stream path, so stream use alone was never accepted as the cause.
Both fresh cells used the retained generated native InitCommonControls PE,
ordinary full shell and unmodified shipped Worker. Current inventory refresh
retained 2,775 alias records and 31 native PE modules, original registry/system
mounts, seed order vcl50.bpl, vclx50.bpl, Ml42ND50.bpl, and the original
five lazy DLL/package entries. Media was hashed in place, not copied.
The collector and adapter retained their exact prior hashes:
- Collector:
0808b400aca53cd8547a1d47d6ef2c4f3c06823a1bf723a87bbe329cf240a220. - Adapter:
e86f182824a7818b01acbcac43ba494518fa714588d379b8846dd8313644dd7a.
Seven retained transport proofs were hash-checked. No new optional-file policy,
alternate body transport, retry, extra body read or clone was introduced.
The sole treatment omitted onTransfer for the selected Ml42ND50 request,
using the existing arrayBuffer branch with the same signal and retention map.
All other package requests kept ordinary consumption.
Observation attached before collector setup and child resume, recording Fetch command begin/result/error envelopes. Body results retain keys, encoding, encoded length/hash; the collector retains decoded bytes/hash. Small results are retained exactly. Normal caller reads, stream/reader cancel, reader releaseLock, controller/signal abort and lifecycle operations were observed without obtaining another body. Node receive order and page clock metadata are separate; cross-clock timestamp proximity is not causal proof.
Both actual slot-0 Workers instantiated standard module
280014368b95a768bd31397971715f5d95089c3a22d018758b5d58f3d4a1650e
(2,180,788 bytes). The recorded compatibility build is
7d476f1d52fdc4538de94a436df69164eb02df26285c3a50a2b0516ec4996eca
(2,183,434 bytes); it was not instantiated in these browser cells.
Immutable first faults
| Identity | Stream cell | ArrayBuffer cell |
|---|---|---|
| First required failure | Ml42ND50.bpl | vclx50.bpl |
| Page target | 27B0D94974514835CCE0C1C9A13F52DF | 5E308F640DC1E0CCD14ED9D263550439 |
| Owning session | 0F930A4284F56E0C5747D325539CFFDA | 4CDE714D140E456A736AB8718CDEFF9B |
| Fetch | interception-job-165.0 | interception-job-164.0 |
| Network | 871990.266 | 872763.265 |
| Server request | 165 | 164 |
| Response UUID | 5ad96a40-d7da-46a6-96a8-4cd1c47952b6 | 0831b492-05cf-40e0-8933-dda687d6b7f1 |
| UUID-linked NetLog URL_REQUEST | 995 | 980 |
| Complete bytes | 468,992 | 248,832 |
The stream cell captured the expected Ml42ND50 body, SHA256
c503b230760bfc9f3c315f91911c3b90471368115fabf227e4e95d0e34a30993.
Fetch.getResponseBody resolved at 21:04:30.937 UTC with base64Encoded=true,
encoded length 625,324 and encoded SHA256
98f496091100e358601826ad688b72b37067c36c0bab834329dba179fb747a56.
Fetch.continueResponse began at .938 and resolved with {} at .940.
Network response receipt followed at .946 and fatal canceled ERR_ABORTED
at .950. The UUID-linked server finished HTTP200 at .930 with matching
bytes/hash. The ordinary stream returned 468,992 bytes and then EOF.
The arrayBuffer cell's first fatal response was vclx50, which kept ordinary
stream consumption. Its decoded SHA256 is
0bca4d6adf70b4a01f0b6fe0e3d4d1ff08d97bdebb6ae2e836dde9d3a89cc148.
Fetch.getResponseBody resolved at 21:05:01.472 UTC with base64Encoded=true,
encoded length 331,776 and encoded SHA256
26663fcb53118766921f83c104aba6e2d7a56c75ad8209faa2c753e37d2d81c7.
Fetch.continueResponse began at .474 and resolved with {} at .477.
Network response receipt was .496 and fatal canceled ERR_ABORTED at .520.
The UUID-linked server finished HTTP200 at .459 with matching bytes/hash;
the ordinary vclx50 caller read all 248,832 bytes and EOF. The selected
Ml42ND50 treatment was invoked once, but cannot qualify a cell already
faulted on another required seed.
Each UUID-linked NetLog records the matching filtered byte total and one
REQUEST_ALIVE end, without a recorded net_error. Neither this nor caller EOF
overrides Network.loadingFailed. The instrumented page cancellation/release/
lifecycle paths recorded no such operation in the retained observation window.
That is limited to those JavaScript paths; the actual browser cancellation
mechanism remains unidentified. Each cell retains its exact first-fault
assessment, command/Network/server records, caller receive-order ledger and
UUID-selected NetLog events.
Both screenshots were personally reviewed: ordinary generated-shell library download dialogs, not a completed native-return control or gameplay. The cells each exited 1 on the retained required fault; both browsers closed normally with exit 0. No first fault was cleared or retried.
Validation and decision
The locked full build passed. Focused suites passed: 99 synthetic collector lifecycle cases, process-boot yield checks and classic Worker DLL registry checks. Fourteen run-local assessment controls passed, including rejection of complete-body-plus-Network-failure, missing/rejected command results, wrong UUID/hash/session, missing/duplicate terminals and incomplete collection. These validate assessment behavior, not successful native package completion.
The predeclared different-first-failing-URL stop rule applies. The consumer hypothesis is unqualified; no specific host/driver/emulator operation explaining cancellation was demonstrated. There is no warranted production fix or production regression to add, and no qualified original launch prerequisite. Further work needs a separately assigned concrete discriminator and runtime grant. The original, terrain and two reviewed ordinary army commands remain gated. Final cleanup and publication identities are retained in the run.