Original Sims installer after native FT — 2026-11-15

The authorized target remains the owner's ordinary four-disc Nova release, D:\Setup\Setup.exe, 168448 bytes, SHA-256 fe95a9fc8b2cdb5add76fbd326b1a11801eaa43c7d908f20cbdf413fd4d8dfde. Media identities and recovery remain in the original report. No directly unpacked game executable is an installation substitute.

The published native FT result, feature fc63bbcbac4d835258b56b1642cfc52f13c98460, proves synthetic original x86 FT execution in local, cooperative and Worker owners. Its one untouched genuine COMCTL32/COMMCTRL ordinary attach returned TRUE and published a native LIVE DLL. That genuine attach made zero FT calls. It established neither installer execution nor UI nor genuine FT invocation.

Static startup and necessary bootstrap route

Setup and the statically inspected original iKernel both import COMCTL32 ordinal 17 only. The selected DLL exports InitCommonControls there at original VA 0xbfc077e9; its actual instruction is C3 (RET). There is no static evidence that this startup import needs QT or another unsupported thunk direction. DllMain at original VA 0xbfc1a1cd calls its ordinary six-argument ThunkConnect32 wrapper at 0xbfc2c5d2.

The generated FT bodies also exist: the indirect relay pointer is at 0xbfc3503a, FT_Exit24/20 at 0xbfc3511b/0xbfc351a8, and the FT_Thunk import wrapper at 0xbfc64b42. They reference the separate SMapLS/SUnMapLS helpers. Their presence does not establish execution during original startup; the previously successful genuine attach executed none of these FT bodies. Unsupported helpers and directions must retain their guards until a concrete ordinary call demonstrates a prerequisite.

The former CLI's explicit --vfs-mount pass ran after PE DLL initialization. Consequently that option could not supply a system companion requested by COMCTL32 DllMain. The change makes those explicit inputs available before ordinary initialization, while retaining the final override pass for app manifests. The normal VFS DLL resolver then sees the unchanged original companion under its documented installed name:

--vfs-mount=test/binaries/sims-native-common-controls/kommctrl.dll=C:\Windows\System\COMMCTRL.DLL

This is the original SUBASE.INF commctrl.dll, kommctrl.dll, kommctrl.dll installed-name mapping, documented in the common-controls source note. The selected COMCTL32 stays version 5.80.2614.3600/SHA 28a41e6a...; the companion stays version 4.10.0.1998/SHA 39bbfafa.... No header, version, export or executable bytes are changed, and no installation registry is fabricated.

Qualification and original-attempt status

Evidence is in scratch/runs/20261115-sims-original-installer-after-ft/. The published controller caps holds at 1400 ms, retains independent UP at 1750 ms, and treats ordinary installer cancellation as a final native choice without requesting an automatic post-choice capture. Actual exit still needs the matching native wait; an input ACK never proves native Quit.

Fresh qualification passed 24 policy, 12 HTTP transport, 13 actual CLI UI-control and three ordinary native bootstrap controls. The bootstrap positive performs an actual DllMain file read; absent and wrong bytes keep initialization failed. The UI controls cover native pointer/menu delivery, timely independent release, late/lost ACK, identity change, cancellation, parent/child death and no capture while held. They execute public fixtures, not the original installer.

All 20 fresh FT/LS32/NE/lifecycle/GDI/native-Quit regression suites passed. Both the unmodified shared-tree full locked build and the complete committed HEAD plus whole owned feature full locked build passed. Qualification uses the latter module/source view; it reads unrelated tracked code and metadata from Git without editing or adopting the released operation-5/data-span/census WIP. The qualification closure names 180 actual waits with both EOF and verifies 188 PIDs/groups absent. One deliberately orphaned release wait and historical outer/ancillary EOF remain UNKNOWN. An initial UI qualification invocation used a relative evidence path that broke Node preload resolution; it was stopped, and the corrected absolute-path invocation passed all 13 controls.

One original attempt: pre-guest host failure

Feature fe7e6219f288174b5f0755bb2f1a30ae107cfa91 was published to main before launch. Fresh launch pins name 3159 existing inputs, including the four archives/BINs, exact setup, selected DLLs, TLBs, fonts, PATH, modules and host libraries. All 3159 remained unchanged at original closure. Nonempty actual /proc/self/maps startup/exit receipts each name and hash eight mapped host files; there is no claim that a guest DLL was mapped from an empty census.

The single authorized ordinary D:\Setup\Setup.exe attempt used real ticks, default scheduling, 900 native seconds, an independently prearmed 920-second watchdog and the final 180-second Quit reserve. It stopped before PE initializer execution and before its HTTP UI endpoint with:

Error: Process desktop: ETIMEDOUT
    at receive (lib/process-desktop.js:97)
    at new ProcessDesktop (lib/process-desktop.js:270)
    at test/run.js:3026

The native harness actually exited 1; watchdog, supervisor child and outer launcher exited 0. Each has a matching actual wait and both EOF. No input was sent. There is no current installer screenshot, UI, genuine installation, original FT invocation, native Quit or household gameplay. The native-write installed overlay remains empty. This attempt never reached the COMCTL32 route and is not another successful or failed genuine DLL attach measurement.

A host-only public Worker probe reproduced a fault in this run's capture preload: host-maps.cjs used a PID-only output name with exclusive creation. Workers share their process PID, so its second startup write raised EEXIST. The original main thread was synchronously waiting for authority-ready; its Worker exited 1 before loading process-desktop.js (zero Worker source pins). These observations identify a capture-bootstrap failure rather than a proved unsupported installer thunk. The initial qualification omitted this additional capture preload; the ordinary bootstrap controls alone did not catch it.

The evidence hook was repaired after the original pin check by including worker_threads.threadId in both startup and exit names. The exact old hook is retained as host-maps-original-used.cjs, with before/after hashes and host-maps-worker-fix.patch. A public Worker now exits 0. All three native DllMain file-read controls also pass with the repaired exact source/capture preloads and --spawn-processes. These are generated fixtures; they are not a second original run. Recover the naming rule with:

const {threadId} = require('worker_threads');
const name = `${label}-host-${phase}-${process.pid}-thread${threadId}.json`;

The next original task must use that qualified Worker-safe capture route and a fresh bounded authorization. The actual ordinary installer entry/UI boundary remains unobserved. It should not implement every guarded thunk direction in advance, infer FT invocation from attach, or replay this consumed attempt.

Runtime release 392f01e4e0238d5f0aae897396c9f1180adffe96651d47230d3ea61fcf8d0067 names 205 actual waits with both EOF and verifies 216 PIDs/groups absent, including the stopped passive board watcher. Both build/runtime locks were free. Deliberately orphaned and historical outer/ancillary UNKNOWN remain UNKNOWN. Foreign operation-5/data-span/census bytes were preserved throughout; no foreign WIP was committed. No box, worker or deployment was created.