Native DialogBox invocation identity for the VB5 CCE license

Run: scratch/runs/20261117-vb5-native-license-dialog-identity/.

The retained original installer imports DialogBoxIndirectParamA; its license window is HWND 0x10005, owner zero, native control class zero. Those properties alone do not establish dialog provenance. The indirect handler delegates to the real DialogBoxParamA handler, which already owns the executing actor’s modal pump, callback and return continuation.

The repair records the actual creation tuple before callbacks: HWND, generation, process, thread, owner, initial DLGPROC and return site. The read-only export requires that tuple to match the live window, actual actor-local and shared modal pumps, callback table and continuation, with neither pump ended. A suspended outer invocation cannot qualify while its nested invocation is active. Ordinary CreateWindow and modeless CreateDialog windows cannot acquire this identity.

Reviewed context accepts this exact identity for a real DialogBox, including ownerless pre-main dialogs. The existing strict native MessageBox identity path remains intact. Public captions and child control state remain in the context; password text and private application state are excluded.

Real native BUTTON controls dispatch synchronously in WAT. A new actor-local sequence advances only after that exact native WM_LBUTTONDOWN dispatch returns. The host requires one new sequence with the reviewed HWND, generation, process, thread, parent and message. Generic handler return is insufficient. Queued input still requires its existing actual check_input return; independent UP and all UNKNOWN/deadline fences remain mandatory.

qualification-final/ passed the sealed Git-plus-owned-source native suite: direct and indirect pre-main DialogBox creation, nested invocation and restored outer lifetime, real main creation, main-owned and ownerless dialogs, five real BUTTON transitions in each positive mode, and focus/caption/child-caption, disabled, foreign process/thread, retired HWND, changed generation/callback, real stale wait, ordinary class-zero impersonation and modeless refusal. Faults close permanently with no later requests. Existing native13 context tests and unit/transport guards passed. build-final.log records the passing canonical flock build; guarded-current.log records the current guarded-input regression.

The original installer pipeline is separately authorized once this repair is published. Its outcome and exact immutable Git/module/source/media identities, actual process waits and selected personally reviewed frames are recorded in result.json. No source archive or copied original media is needed for recovery.

One published original attempt

Repair 66e2124e3 passed strict pull/push. Its runtime used 5,067 exact published source bodies (no foreign WIP), sealed module bc9ce60ee3c95a396f3d60f849b267d6fde84a890454afc6d4956cc5338f33e2, and actual sealed Worker/process inheritance. All native qualification and three MessageBox question shapes passed again before the original deadline began.

The genuine original question and ownerless license dialog both completed via personally reviewed real Yes buttons. The license identity was HWND 0x10005, generation 5, PID 1000, TID 1, owner zero, DLGPROC 0x1002e5b, return site 0x1005e78. Confirmation/UP/completion were 55.2/210.9/255.5 ms for license acceptance (106.8/266.8/461.6 ms for the initial install question).

A new ownerless initialization/extraction dialog then had a valid independent DialogBox identity: HWND 0x1000b, generation 11, DLGPROC 0x1003241 and the same native return site. The retained post-acceptance public snapshot includes visible progress control 0x10010, generation 16, class 17, ID 2106, title Generic1, plus the Initializing... Please wait... static. The next capture was refused with empty context failures and control.unsupportedClass. Class 17 is the native ProgressBar (msctls_progress32); the reviewed context currently admits only classes 0, 1, 2, 3 and 16 and does not yet export/qualify progress range, position, step and state. The custom class-zero child User1 is not evidence of a second trusted dialog.

Permanent UNKNOWN closed this endpoint after ten HTTP requests, one binding, and two finite inputs. There were zero later requests, rebinds, replays or second originals. The worker requested orderly host SIGTERM after the latch; actual native wait was 0, the independent watchdog was joined with actual wait 0 and both EOFs, sealed descriptors closed, PID disappeared and port closed. This is host shutdown, not native UI Quit. Partial temporary extraction is not installation. Installed IDE, shipped AXListPicker project and native Quit remain unproved. No box was created.

Next work requires native progress semantic state and lifetime qualification, not a wider class whitelist. The consumed original authorization does not permit another attempt in this task. result.json and terminal-audit.json retain the exact blocker and process evidence.