Sims original installer after native coercion

Correction, 2026-11-17: the ordinal-17/202 identifications and “passed copy” progress claims below are invalid. Static reference DLL baeb2f7c1b8be56738d34e1d1ddf8e0eebd3a633215dc1575e14656be38b939d and Wine’s export specification identify 17 (0x11) as SafeArrayGetDim, 27 (0x1b) as SafeArrayCopy, 148 as SafeArrayPtrOfIndex, and 202 as CreateErrorInfo. No genuine Automation provider was loaded or executed to obtain this static evidence. Prior incorrect publication and tests are retained as history, not acceptance.

The engine 985eb55ecb750da812876b8569d5f1999a30a24bcc54f9bab4d3fc44dfedb931 contains raw ordinal lookup 0x80000011 at IAT 0x46c290. The actual caller pushes one argument at 0x40e921, calls that IAT, and compares EAX with 1 at 0x40e92a. The corrective run 20261117-sims-correct-ordinal17-getdim-installer-progress adds real UINT dimension return and ret4, corrects both resolver maps, and moves Copy ownership/error tests to ordinal 27. The exact x86 caller replica fails the old module with EAX 0, ESP four bytes high and a clobbered canary; corrected dimensions 1/2/3 preserve exact ESP and canaries. PtrOfIndex’s native bounds/borrowed-owner implementation belongs on 148; 202 remains unresolved until its real API is implemented. The old writer had acknowledged this correction and left corrective WIP before interruption; that WIP was reviewed and adopted.

Evidence: scratch/runs/20261117-sims-native-coercion-installer-file-progress/result.json (written last).

The authorized initial original used the retained published compatibility module 75dd393b0c0144e342a371f42852fa5ff0f2ed2dc5ce2f121cdddb3d3112c317 (2,403,800 bytes), production module 832621ca4212c8872cc91d1eb331c57b6382113ee50375232ddafb4beb51e63d, and matching region map 8e15e720bcfded5902defa7e323661b824baf53ebc25de0fdcad3e8620982bf9. Normal bootstrap, registrar and Embedding ran with genuine OLEAUT32/OLEPRO32 and stdole files absent from execution. Prior reader, metadata, Invoke, continuation, WORD, ordinal and BW2 proofs were hash-verified and reused. No installer code or private registry seed was substituted.

It passed native collection coercion and subsequent component methods, then trapped in SafeArrayCreate. A diagnostic names-only compilation, stripped and compared byte-for-byte with that exact module, identifies the native handler. The retained genuine engine's entry at 0x4082d8 reaches its first call at 0x4082f8, pushing VT_BSTR, one dimension and the native bounds. The old handler only supports scalar elements and traps on BSTR. The whole error image shows support-file installation failure 0x80004001; this is not installed-file progress.

Commit ead475c4ec50623d15c1d0d78559ff8d8075c3ab implements native BSTR SAFEARRAY allocation, FADF_BSTR ownership and element destruction. VariantClear propagates a locked-array error without discarding the original owner. Native public API tests cover multidimensional descriptors, negative lower bounds, empty arrays, embedded-NUL strings, locked destruction/clear, borrowed BYREF wrappers, exact frame restoration and 128 repeated ownership cycles with stable heap usage. The old published module fails the new regression. Canonical locked build and the existing native coercion regression pass on compatibility module 83b3b6d82f66ca2271f47a68d2c27d6f2bef61034830f5aca58b419d85c800d0.

The distinct published correction's fresh original passed array creation and reached OLEAUT32 ordinal 25 through IAT 0x46c2e0, at native call 0x407cdf. This is SafeArrayGetElement, which was absent. It again shows 0x80004001, with no physical input or native Quit claimed. The follow-on implementation copies BSTR owners independently, validates all indices, preserves existing lock counts and returns HRESULT failures without publishing partial outputs. Public named and ordinal retrieval, non-square multidimensional indexing, negative lower bounds, scalar values, independent string destruction, bad indices, invalid descriptors, lock overflow and exact frames pass native tests. Repeated retrieval/destruction keeps the warmed heap stable. The retained BSTR-only module fails the new public lookup regression. Canonical locked build and native coercion regression pass on compatibility module 010cdbf9c8df0bfbdb6bd8c82a380064fd0d7ef025928ad6a1cd5c8a992919f9.

API contracts: SafeArrayCreate, SAFEARRAY flags, and SafeArrayGetElement. These establish declarations and ownership behavior; no provider implementation was copied.

The GetElement correction's fresh original passed retrieval and trapped at OLEAUT32 ordinal 26 through IAT 0x46c2a8, return address 0x407be2. SafeArrayPutElement now clones the incoming BSTR value directly, publishes the staged replacement and frees the prior element. Self-aliasing, caller/array independent ownership, null assignment, scalar writes, bounds errors, caller locks, overflow errors and exact frames pass native tests, including 128 repeated replacement cycles with a stable warmed heap. The previous module fails the new public lookup regression. Its API contract specifies direct pointer-valued BSTR input, rather than a pointer to that pointer. Canonical locked build and native coercion regression pass on module a1f3337502bfcfef337e4ceabd97fa821b221c1aeca5617a13d7b6e0ee432c7e.

The third original's PNG previews appeared teal while native records reported the support-file error. This discrepancy is retained explicitly; no ordinary input or native Quit is claimed, and that endpoint was closed. Both actual child exit/EOF receipts are complete, including the forced Embedding child. Native parent exit is -15 with both EOFs. The supervisor and watchdog have complete waits; the outer helper wait is missing and is not reconstructed from PID absence.

Lifecycle limits remain explicit. The initial run's trapped Embedding child did not complete orderly shutdown; the parent was killed with actual exit -9 and both EOFs, and the child wait/EOF receipt remains incomplete. Absence of its PID is not a wait receipt. The second run's independent watchdog forced only the trapped child while its parent remained alive: the actual parent recorded SIGKILL plus both child EOFs and then exited 0 with both EOFs. Missing forced-child provider exit audits remain missing. No old UNKNOWN endpoint was reopened. No box, worker or public deployment was created.

The PutElement correction was published as fb9ea07a98f807c74f18a0fa617ff9c01102ae8c. Its fresh original passed assignment and reached missing OLEAUT32 ordinal 17 through IAT 0x46c290, return 0x40e92a: SafeArrayCopy. The personally reviewed whole image shows the same support-file error. The parent, registrar, Embedding, watchdog and outer helper have actual wait and both EOF receipts; Embedding required SIGKILL and has no fabricated orderly exit or provider audit.

The SafeArrayCopy contract requires independent strings and ownership. The native implementation stages descriptor/data/string owners, preserves source locks, publishes only a completed array, and destroys partial copies on allocation failure. Scalar and BSTR arrays are supported; interface and VARIANT element kinds remain fail-closed until callback-aware support is implemented. Native tests exercise named and ordinal 17 lookup, multidimensional and empty arrays, scalar byte preservation, counted strings, independently destroyed copies, caller locks, invalid descriptors, null arguments, allocation refusal and partial string-copy rollback. Repeated success/failure cycles pass with a stable warmed heap and exact caller-frame restoration. The previous published module fails the new lookup regression. Canonical locked build and native coercion regression pass on compatibility module fdba623468d84860ff8a11b7788d239e6c620124d0834c52bfa4e1644688a9c8. These synchronous APIs create no pending callback or cancellation state; prior native continuation/cancellation proof is reused by exact hash.

The Copy correction was published as 5a7bb47a1168c3e4df4026c7af03501f5b69825e. Its fresh original passed the copy and reached missing OLEAUT32 ordinal 202 through IAT 0x46c2b8, return 0x402ca1: SafeArrayPtrOfIndex. The selected whole preview is teal although native records report the support-file error; the discrepancy remains UNKNOWN and no ordinary input was sent. Actual parent, registrar, Embedding, watchdog and outer wait/EOF receipts are complete, with forced Embedding termination explicitly recorded.

The SafeArrayPtrOfIndex contract returns a borrowed element address under the caller's lock. The native implementation uses all signed indices and stored bounds, checks offset arithmetic and preserves the lock count and element owners. Pointer lookup also supports descriptors without a VARTYPE prefix, including caller-owned record data. Native tests cover three non-square dimensions, negative lower bounds, byte-exact pointer identity, in-place data access, independently retrieved strings, extreme invalid indices, null pointers, zero element sizes and empty dimensions. The previous module fails its new named/ordinal lookup regression. Canonical locked build and native pointer/ownership plus coercion tests pass on compatibility module bc4b86d8ad7c878a6b017b89c92a6f314f65db451978b9bb811875456595f127.

Offline readback found that all five selected error PNGs are byte-identical (71210f7fae135bf42e04f1bef1f98686fad3499c9a96d7cbf344d86976f01e47), including those whose tool previews appeared teal and those personally shown as the complete Sims error. The preview discrepancy is therefore not evidence of a different native scene. Its limits remain explicit and no input was derived from the discrepant previews.

Sims installation/native Quit, FullT-db74d863 and BW2/IS11 remain open until their actual outcomes are established.

The corrective publication 0e524a7d431be93fa0c81fa599a0f57f0af784b7 passed its own fresh genuine installer test. The actual Embedding trace records GetDim returning 1 at both 0x40e92a and 0x40e9ab, followed by real ordinal-27 Copy returning S_OK at 0x40e954 and 0x40e9d5. This is the first valid original GetDim/Copy progress proof. It then traps at the actual three-argument OLEAUT32 ordinal 20 caller, IAT 0x46c29c, return 0x40e5f1: SafeArrayGetLBound. The full image personally shows support-file error 0x80004001; no input, installation or native Quit is claimed. The pinned module’s names-only recompilation strips byte-for-byte to the tested 20fdb2dee0792da846438ffcad257587677f9b6be67a4782d624e85726eb50ce and identifies the fallback dispatch trap.

The follow-on native bound queries use one-based dimension numbers and the descriptor’s reversed bound order. GetLBound returns the signed lower bound; GetUBound also checks signed arithmetic overflow. Static export evidence assigns them ordinals 20 and 19 respectively. Queries leave the descriptor, data and locks unchanged and publish no output on error. Native named/ordinal tests cover non-square dimensions, negative/extreme lower bounds, empty bounds, bad dimension numbers, null arguments, descriptors without VARTYPE/data, exact frames and overflow output atomicity. The preceding corrected module fails the new lookup test.

The bounds correction was published as cfa5c043c5e5618a8c75c16eadf5f3e1dd8e0273 after the canonical locked build and focused native SAFEARRAY/coercion tests passed on module 75beef304a4137417f9b2d4dbf3caa02f662fe26a46de66e1d189d5df6602770. Its own distinct affected original executed GetLBound20 at 0x40e5f1 and GetUBound19 at 0x45786d/0x40ffad, all returning S_OK, and continued through further engine/file activity. The new concrete blocker is VariantCopyInd, ordinal 11, via IAT 0x46c2c4, return 0x4017a6, with destination 0x074ffa40 and source 0x074ffa84. Static exports independently identify the same two-argument API. No third unchanged original was attempted; implementing its actual indirect variant and ownership contract is the next work.

Both new originals were closed without UI input. Their actual registrar exits are 0; the trapped Embedding children required SIGKILL, with actual parent-recorded exit and both EOF receipts before parent termination. Parent, supervisor/outer helper and independent deadline watchdog waits and both EOFs are complete. Provider audits record zero genuine Automation provider reads. Forced Embedding main-thread provider exit audits are missing and remain explicit; a worker audit with exit 0 is not a substitute. The first watchdog’s conservative child-receipt matcher used the wrong basename and waited its full child deadline; independent audit of the real parent-PID/child-PID receipts proves they arrived before the parent wait. The matcher was corrected before the second launch, whose pre-parent audit marks both child exits/EOFs complete.

The first selected whole image personally shows the support-file error. The second tool preview appeared teal, but both retained PNGs are byte-identical (71210f7fae135bf42e04f1bef1f98686fad3499c9a96d7cbf344d86976f01e47), and native records again report 0x80004001. The preview discrepancy is retained; no UI input, completed installation, installed game files, native Quit, FullT-db74d863 or BW2 completion is claimed. No boxes, other workers or deployment were created. All old UNKNOWN runtimes remain closed. Final evidence is scratch/runs/20261117-sims-correct-ordinal17-getdim-installer-progress/result.json, written last.