Sims terminal thunk and native null call, 2026-11-17

Evidence: scratch/runs/20261117-sims-terminal-thunk-transfer-installer-repair/result.json. This continues the slot41 frame investigation. No causal production repair is established. Full T-db74d863, BW2 and IS11 criteria remain unchanged and open; installer completion is not demonstrated.

Offline, the retained module f3568ba6cd1c1982cf3c79eebada41823dd9c3d6a21cc6ac9e48491af4ed6c19 and map 8e15e720bcfded5902defa7e323661b824baf53ebc25de0fdcad3e8620982bf9 load the original iKernel PE and initialize the engine COM vtables without executing installer instructions. Thunk 0x07501dc8 is index953, marker 0xCACA0010, API2105 IStream_Read. The original PE alone allocates only 256 entries; deterministic COM-vtable initialization supplies this mapping. The admitted diagnostic independently reads the same two engine metadata words.

The single new passive original identifies the actual terminal operation:

Control field Observed
Slot41 target 0x004210c4, correct native ret 8 unchanged
Earlier token1018 S_OK, exact ESP 0x074ffe88
Failing token1340 expected ESP 0x074ffe88, terminal 0x074ff4d0
Final native block 0x0040c749, entry ESP 0x074ff4d8
Terminal instruction 0x0040c75b: call dword ptr [eax+0xc]
Terminal registers EIP0, EAX0, ECX8; ESP falls eight bytes from block entry
SEH inner head 0x074ff4f8; saved head 0x074fffe8
Last API identity IStream_Read, stale across subsequent native blocks

Pinned PE instructions after 0x0040c749 load [esi+0xd8] into EAX, push the argument, set ECX to EAX+8, load [eax+8], and call [eax+0xc]. The final ECX8 implies that the first native object pointer was null. EAX0 and the eight-byte ESP decrease match the argument push and indirect CALL into zero. This is subsequent guest control flow, not the IStream_Read epilogue, an observed exception/unwind, thread exit, or host continuation restoration. The earlier last-thunk value alone did not identify the writer. The producer of the null native member remains unknown; this evidence does not establish a slot ABI, ordinal201/202, stream implementation, or marshaler repair. No synthetic success, private seed, provider bypass or guard relaxation was introduced.

Actual native controls compare observer off/on results and deliberately fail the observer output sink. Correct/error HRESULT, ret4/ret12, extra argument, callee-saved corruption, trap and retirement refusal retain parent registers, flags, FS/SEH and empty continuation frames. New authored controls distinguish an internal IStream_Read return, an early API return to zero and ExitThread yield2. A further authored native function returns from IStream_Read, then calls [eax+0xc] with EAX0/ECX8: it retains the stale thunk while reproducing the final block-to-terminal ESP decrease of eight bytes. Existing fourteen native yield12 lifetime controls and stale-token/replacement-epoch negatives are reused. One local counterexample validator initially expected a block boundary that the fixture did not contain; an explicit native JMP boundary corrected the fixture. No original retry resulted.

The inherited --trace-api-mem restriction violation remains explicit. All inherited trace/dump options were removed, including nested child arguments, before launch. An early preload rejects forbidden actual process/fork arguments; four real rejected Node spawns and ten clean actual argv records are retained. Run-local COM protocol payload logging was also disabled while preserving child exit/EOF records. The observer retains only a bounded window of CPU/ESP/SEH/block/engine-thunk control metadata and an API return-control word, with unchanged native run budgets. No inherited API-memory payload was decoded or circulated. Genuine oleaut/stdole providers remain absent from execution; prior registrar/media/module-map/provider-absence proofs are reused.

The selected, personally reviewed whole 800x600 screenshot still shows support file error 0x80004001. The installed overlay remains byte-identical. An independent watchdog was armed before launch; both actual child waits and both EOFs precede parent shutdown. The registrar exits0; Embedding needs SIGKILL and lacks its final main-thread provider audit. Its incomplete state merge is refused. Parent exits0, and watchdog/supervisor/outer actual waits finish. This is host cancellation, not native Quit or installer completion. All known owned runtime PIDs are absent and the endpoint refuses connections. No box was created: ID/expiry null, state NOT_CREATED, stop receipt NOT_APPLICABLE. No workers, archives, media copies or deployment.

Next work must identify the producer of native [ESI+0xd8] before justifying a repair. The one authorized passive diagnostic is consumed. No further original or unchanged retry was run. No production code changed, so no new canonical locked build or repaired-installer result is claimed.