Native IDispatch transport remains incomplete

No production transport correction is published by this task. Sims installation, native Quit, FullT-db74d863, and BW2/IS11 remain open. The authorized fresh Sims original was not used. Evidence: scratch/runs/20261117-sims-native-idispatch-returned-interface-transport/result.json.

The prior 11d04acfc6c9065739415c24a4abe7e03d69182e finding remains valid: the genuine ISetupMedia.OpenCAB succeeds, and encoding the returned ISetupCABFile fails eager validation of inherited IDispatch slots 4–6. Neither OpenCAB nor its WORD argument handling was changed here.

Additional native boundary

A small authored x86 fixture calls the actual WAT ITypeInfo AddRef and returns that actual interface from its GetTypeInfo-shaped slot 4. The fixture executes through ComGuestContinuation, with its real stdcall frame checked and restored. It does not execute IKernel or implement an application or Microsoft provider. The returned native ITypeInfo cannot currently marshal:

The native standard IDispatch metadata still passes slot 3 and rejects slots 4–6. Eager validation, the standard marshaler, proxy factories, WAT metadata, and WAT Invoke behavior were left unchanged. No private registry seeds, synthetic success, or cross-owner guest pointer copies were introduced.

Microsoft's IDispatch operation ordering and GetTypeInfo contract require handling the returned type-description interface. Adding canonical inherited signatures alone would leave that positive return path unsupported.

Unintegrated codec candidate

lib/com-automation-types.js is uncommitted WIP, with an exact new-file patch and hash in the run. Nothing in the production runtime imports it. It contains typed-value helpers for counted BSTRs, UTF-16 names, scalar and BYREF VARIANTs, DISPPARAMS, EXCEPINFO and descriptor fields. It is a candidate, not an accepted transport implementation.

The focused native controls passed seven groups and eleven malformed-value checks. They cover embedded-NUL BSTRs, scalar/DECIMAL VARIANT bytes, a real x86 write through aliased BYREF storage, actual IUnknown reference marshaling and retirement, a real x86 deferred EXCEPINFO fill callback, and reconstruction of an actual native TYPEATTR into separately owned storage. Tracked candidate allocations, authority exports/requests and native continuation frames finish at zero. A retirement guard rejects new candidate writes and permits cleanup in the existing owner's cleanup context.

These are local codec controls and a negative native marshaling boundary. They do not prove a returned IDispatch-derived proxy/stub, inherited names/LCID/ DISPID/named-argument behavior, complete descriptor graphs, SAFEARRAYs, atomic caller output publication, or transport cancellation/disconnection/nesting for the candidate. The FUNCDESC/VARDESC helpers have no complete positive lifetime qualification. Do not import the candidate or relax validation on its strength.

The canonical flock /tmp/wine-build.lock bash tools/build.sh passed. The native controls used its compatibility module dcaefe74632b53f344649dfa67ae4bb937950d0c669576aa54f004629a865438. All 26 selected earlier proof hashes and 22 media/provider hashes were checked and reused; reader/783 continuation/ordinal ABI/BW2 parser tests were not blanket-replayed. Independent deadline watchdogs were armed before the build and native controls. Their actual child waits and both stream EOFs are retained.

The concrete remaining blocker is a real owned standard type-description transport, including descriptor allocation/release and returned-interface lifetime, plus canonical inherited IDispatch contracts and their complete proxy/stub tests. This task does not deliver that implementation. No fresh original should be spent on this candidate. No boxes, workers, deployments, source archives, media copies or downloads were created; no image or ordinary input claim is made.