WinHelp qualified-path native/fullVM requalification — 2026-10-20
Historical source last updated, 2026-10-20. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
PASS: all seven unchanged callback scenarios, with 70 successful same-VM
siblings in each (490 total). The prior three WinHelp failures before DLL
entry no longer reproduce on canonical source
1262cbfcf8b5881543f2fb18c0f80b4bd9429306. No production change, fixture repair, oracle relaxation, or rerun was needed.
| Scenario | Original oracle | Successful siblings | DLL Probe entries |
|---|---|---|---|
| edit-ordinary | PASS | 70 | Not applicable |
| edit-nested | PASS | 70 | Not applicable |
| edit-error | PASS | 70 | Not applicable |
| edit-timeout | PASS | 70 | Not applicable |
| help-ordinary | PASS | 70 | 72 |
| help-nested | PASS | 70 | 73 |
| help-timeout | PASS | 70 | 72 |
Actual callback observations
WinHelp ran through the public WinHelpA route, RegisterRoutine and the qualified
c:\ew-native.dll loader path. No direct callback bridge was used. All three
cases returned public acceptance 1, rejected an unknown macro with 0, and
completed the caller. The final actual DLL argument was 71, with guest-written
computation 0x1047. Ordinary and nested cases retained outer argument 0x1234
and computation 0x2234 before siblings. The nested case recorded the extra
Probe entry and passed the inner saved-register/ESP assertions.
The timeout case recorded the actual spin marker 0x71ae0001, with zero outer
computation as expected, then recovered and completed all 70 siblings. This
is an internal callback timeout, not an external process timeout. All child
processes exited zero; no wall-clock guard fired.
The original oracle checked exact outer stdcall ESP restoration:
122683384 before and after in every scenario, plus ECX/EDX/EBX/EBP/ESI/EDI
restoration, stack canary 0x51ac0ffe, and caller completion 0xc001c0de. The fixture's DLL uses two stack arguments and ret 8; RichEdit uses four
arguments and ret 16. Generated code and disassembly are retained. Actual continuation, argument/computation writes and caller snapshots establish
return-path behavior. The DLL code loads EAX 0x76543210 before return, but the
public API deliberately reports acceptance 1: this fixture does not
independently sample raw DLL EAX at the return instruction. No such observation
is claimed.
RichEdit ordinary/nested text results, error 123, timeout error 1, callback
entry counts, actual buffer bounds, nonvolatile clobber isolation and nested
restoration passed the original oracle. Error and timeout outer API returns
were correctly 0; subsequent siblings still returned successfully.
Build, source identity and controls
The prior successful MadBomber recipe was reused in a new isolated remote
directory /home/user/winhelp-qualified-fullvm. Before building, all 6,441
source blobs were verified, the complete predecessor closure and existing
dependencies were available, and the include/glob gate passed. Only four
tracked blobs differed from ee139a094: the integrated production fix, its test,
and two notes. No npm install or speculative closure-staging loop occurred.
bash tools/build.sh passed, including the unchanged 399-entry callback census,
vendored compiler provenance, manifest and layout gates, WATX compilation and
compiled data overlap checks. The exported source directory has no .git;
as recorded in the build log, the API-table gate checked IDs/index structure
but could not compare append-only history against HEAD. Exact Git blob
verification provides source identity separately.
Unchanged fixture commit: 2c902c79f362560112ca30d41e4332688046d55a. The fixture is an explicit test overlay, not a production source overlay:
test/test-edit-winhelp-callback-native.js: Git blobc88a5de8d7db19f9e4726466d47e6cf09a8a3509.test/fixtures/edit-winhelp-native/validate.js: Git blob9a6cfcb1d872f4cd0b1d2d73458a75c85f46243e.- Pinned
freecell.hlp: Git blob5418925c6c66c548daadd61531ea4c2083288e7a.
Existing offline controls passed: seven generated/disassembled cases, seven argument negatives, 14 ABI negatives, 14 missing-evidence negatives, seven synthetic oracle positives and 84 synthetic oracle negatives. Incorrect callback pop and bad IAT controls are offline validation, not actual malformed-x86 execution. Synthetic logs were never counted as runtime evidence. Local replay of the unchanged oracle against all seven recovered actual logs passed. The prior source-only review/replay and its validation were inspected separately.
Evidence, preservation and release
Self-contained evidence:
scratch/runs/20261020-winhelp-native-qualified-path/. verification.json holds decoded observations; recovered/native/ holds the
original logs, runtime receipts, executable bytes and overlays. recovered/wine-assembly.wasm, source/fixture manifests, canonical build log,
offline controls, preservation inventory, scripts and RECOVERY.md retain the
module and recovery recipe independently of runtime expiry.
All 168 recovered artifact hashes and the 2,174,111-byte evidence archive hash
were verified. Before/after checks passed for 6,441 source blobs and 22,999
preexisting files, including Army, ICM, callback, JS COM, dependencies and the
MadBomber exact-ee139 source/module. The original module remains
5e66c75862052b13b73f80bdcbcd67a1a65fd99d3e9e79d33f96548d00cc2a21. The actual Node executable hash is recorded in postflight.json.
This report-only task does not modify shared HEAD/index, production, laptop Heroes II/Warcraft III, or merge main. No screenshot or gameplay qualification is claimed. No remaining failure was observed within these seven scenarios. Task commit and verified remote branch are recorded in the final evidence receipt and append-only repository messageboard.