Go Figure! (Win16 WEP4)

App: wep16_gofigure, test/binaries/wep16/WEP4/GOFIGURE.EXE, VBRUN100.

2026-09-20: black label fields after parent exposure

The gameplay gate reached a positive puzzle target but found zero white pixels in the bordered fields. Visual inspection confirmed black rectangles in the puzzle row, score, timer and target, not a screenshot-coordinate error. The fields are guest ThunderLabel windows, not native edit controls. Example: wide HWND 0x10019, parent 0x10003, style 0x50810000, 30x30 client at (134,203).

The API trace shows VB correctly requesting white with SetBkColor(0xffffff). The erase trace shows the child-exposure path using class brush 6 (COLOR_WINDOW+1). win16_rearm_visible_child_erases directly filled those children and cleared their erase bit without entering their window procedures. That shortcut bypassed VB's own background work. Class-brush availability is not evidence that the guest has handled WM_ERASEBKGND.

An isolated full-source WASM with only that direct-fill branch disabled restored white fields and passed the unmodified Go Figure gameplay gate. Its screenshot was inspected: white bordered fields, legible score/timer/ target, and working puzzle buttons. Rodent/Rattler gameplay and WEP1 8/8 also passed this artifact. This matters because the shortcut was originally introduced for VB paint ordering; see the historical status-panel section of the Rodent note. The later callback/validation fixes mean that historical workaround must not be treated as today's contract.

The main-source change removes the direct fill and retains the exposure erase request for BeginPaint's real far callback. It does not force a white brush, special-case VB, or change the expected screenshot colors. The far regression asserts that exposure retains the request even with a class brush.

Evidence in /private/tmp/:

• wa-gofigure-win16.log: API calls, including white background requests.
• wa-gofigure-erase.log: direct class-brush exposure fills.
• wa-gofigure-trace.png: black-field baseline, visually inspected.
• wa-gofigure-no-class-erase.wasm: isolated experiment.
• wa-gofigure-no-class-erase.log: Go Figure gameplay PASS.
• wa-gofigure-no-class-erase.png: restored fields, visually inspected.
• wa-gofigure-no-class-erase-vb-gameplay.log: Rodent/Rattler PASS.
• wa-gofigure-no-class-erase-wep1.log: WEP1 8/8 PASS.

Reproduction after a completed build:

WINE_ASSEMBLY_WASM=build/wine-assembly.wasm node test/test-win16-wep4-gameplay.js gofigure

This is functional paint evidence, not a native Win98 pixel-equivalence or performance claim. Fuji Golf's separate player-name-control failure remains open. Parent exposure message timing and ShowWindow's other documented gaps still require separate work.

Final verification: normal and compatibility builds pass (/private/tmp/wa-gofigure-exposure-build.log), as does the current-source far regression (wa-gofigure-exposure-contract.log). After build completion, the shipping artifact passes Go Figure and Rodent/Rattler gameplay again.