AoW II: native comctl demand succeeds; required package cancellation remains unresolved

Evidence: scratch/runs/20261106-aow2-comctl-terminal-control/ (ignored). Tested main 3772aa782530180d4bba8e8812a88c95dcd45c86; no production host, emulator, registry, collector or guest changes. No original launch, army command, movement-point test or gameplay claim. This does not clear the original Network666315.271 required comctl32 failure recorded in the preceding investigation.

The worker waited for Forsaken and Aranna FINAL releases, independent cleanup and the coordinator's explicit 19:00:16.343035–19:45:16.343035 UTC grant, including cleanup. Before that grant, preparation was run-local and unexecuted. The worker independently checked 110 predecessor PID identities, 25 ports and four locks. No boxes or nested workers were assigned or created.

Generated control and unchanged collection

The generated PE statically imports native comctl32.dll!InitCommonControls, calls it, creates an ordinary STATIC window, then sleeps. The ordinary shell launches its actual main Worker with the untransformed shipped Worker source. The actual slot-0 instantiated standard module is 2fa9f377dde5e477eb811baceb64032889fbe9b27a664b213477e84627e5eb94; the compatibility build is 3dd82e987c0683b45f28860748b6a19151b6623f578f10d70d201b798baf2a6a.

Both arms use the current DLL registry and ordinary shell system-data mounts, including registry and fonts. The original AoW local manifest has no per-app registry overlay. The package arm additionally uses the registered three Borland DLL/package seeds in their original order and the original manifest's five lazy DLL/package entries, preserving their VFS paths and load modes. Original media was referenced, not copied, and AoW2.exe was never launched. The generated control's refreshed input map has 1,641 pinned URLs; this is not a replacement for the preceding original's 1,661-URL/2,776-alias inventory.

The authoritative collector remains exactly 0808b400aca53cd8547a1d47d6ef2c4f3c06823a1bf723a87bbe329cf240a220. The unchanged adapter is e86f182824a7818b01acbcac43ba494518fa714588d379b8846dd8313644dd7a. Seven retained transport controls were reused after exact hashes and passing proofs were checked. No alternative HTTP-body channel, collector relaxation, invented optional 404, request retry or guest-state patch was used.

Run-local observation records owning-session Fetch commands and their results, Network events, caller body-read metadata and aborts. It never adds a caller body read or clone. A unique X-Wine-Control-Request-ID header explicitly links each server response to the browser response; server, Fetch and Network IDs remain separate namespaces. Server close/error/destroy records include the observed byte total and an incremental hash as well as the finish record.

Direct native demand

comctl-direct-mounted/comctl-terminal-proof.json passes the exact transfer checks: one request/response, verified and continued Fetch, one Network loadingFinished, no loadingFailed, and the UUID-linked server HTTP 200 finish with matching bytes/hash.

Namespace Exact direct-control identity
Page target D07C1C30E8DEDFFCBBC3A278A5CF86F2
Owning session 0FB4F7CA0DCB9955CAD9371DE32D7D6E
Fetch interception-job-163.0
Network 725777.264
Server request 163
Server trace UUID a13fb635-182a-4516-a797-d26f1a9afab5

The body is the original 577,808-byte comctl32, SHA256 28a41e6a35c4509924a32a99841a815cf49e1d3ee6912ec26a823e45472edfa4. The normal caller read all bytes and stream EOF. Native Worker logs show the EXE import patched to the loaded DLL, DllMain returning EAX=1, then the generated CreateWindow marker after InitCommonControls returned. All three actual host font loads fulfilled. I reviewed shell.png: the full shell log and generated taskbar title are visible; this is no gameplay evidence.

This arm is not a clean driver pass. A later driver assertion incorrectly expected a wine.moduleBases entry, despite the native load/return evidence. Its exit 1 and error are retained. direct-assessment.json separately hashes and assesses the successful terminal/native-demand evidence; the control was not repeated. The next arm's check uses the actual native load/import and post-call window markers instead of that map oracle.

An earlier comctl-direct/ setup failure is also retained: the working directory was not mounted, so no comctl request occurred. The only correction was a normal generated-EXE VFS file entry establishing c:\aow2demo; the driver now asserts that mount. Neither setup error was a collector fault.

Package graph: immutable first fault before comctl

The discriminating package arm fails on a different required native seed, Ml42ND50.bpl, before it reaches comctl32 demand. The first fault remains fatal and was neither cleared nor retried.

Namespace Exact package-control failing identity
Page target 82955721C24B5AA6137FB3C3EFC0BD80
Owning session EA10693C9A901BCF4BE156AD97E85983
Fetch interception-job-165.0
Network 726905.266
Server request 165
Server trace UUID de6d6d27-ddb2-4231-a27c-d5516bb2880e
UUID-linked NetLog URL_REQUEST 989

Fetch captured, verified and continued 468,992 bytes, SHA256 c503b230760bfc9f3c315f91911c3b90471368115fabf227e4e95d0e34a30993. Its continuation resolved at 19:03:30.162 UTC; Network response receipt was 19:03:30.165 and loadingFailed, canceled true / net::ERR_ABORTED, was 19:03:30.167. The UUID-linked server finished HTTP 200 at 19:03:30.148 with the same complete body and no abort. The ordinary caller separately records 468,992 bytes followed by stream EOF, with no observed controller/signal abort. These caller and Node/CDP timestamps use distinct clock bases; their precise cross-clock ordering is not asserted.

The response UUID also links the NetLog HTTP transaction. URL_REQUEST 989 records exactly 468,992 filtered bytes and a REQUEST_ALIVE end without a recorded net_error. That transport observation does not override the CDP required-request failure. package-fault-assessment.json retains the joined identities, original events, caller metadata and actual main Worker identity.

This demonstrates that native comctl demand alone can complete, and that the failure signature can occur in a generated shell without the original game executing. It does not prove that the package failure and the preceding original's comctl failure have the same cause. No cancellation cause or production host/driver/emulator mismatch is established. In particular, successful caller EOF and server completion are insufficient reasons to ignore a required Network failure.

Validation and remaining prerequisite

The locked full build passed and reproduced both module hashes above. Focused tests passed: 99 synthetic collector lifecycle cases, process-boot yield cases, and classic Worker DLL registry cases. test/test-static-server.js failed its existing repository-wide server-owner census before its HTTP tests: it additionally finds test-d3dim-palette-gpu-web.js, test-dos-recovery-safety.js, test-file-bundle.js, and test-gl-native-render-load-web.js. That failure is retained; those files were not changed. No production repair or new production regression was justified. The run-local terminal assertion distinguishes the successful direct request from the package arm's required failure.

Exact source commit, relevant binary diffs, untracked runtime-source patches, module identities, helper identities, captures and original failures are retained in the run. Git remains the source archive. All browsers closed normally; runtime holder and final cleanup evidence are retained. Box ID/state/expiry are null.

The prerequisite remains: explain the required intercepted-response completion failure, or demonstrate and repair a specific host/driver/emulator mismatch with a generated discriminating control while keeping the strict collector. The failing package arm cannot qualify the complete AoW DLL graph. Further original gameplay work remains gated; there is no army-command acceptance.