What

Makes the tracker that spots WebGL / WebGPU canvases see the cases that slipped past it, and counts the frames that fell back to plain DOM.

• The accelerated-canvas tracker (pageMotionBlur.ts) is now a self-contained function. Besides wrapping getContext, it wraps transferControlToOffscreen: a canvas handed to an OffscreenCanvas creates its WebGL context out of reach, so a transferred canvas counts as accelerated. Its registry lives on window, so independent installs agree.
• It runs before any author script. The engine evaluates it on every new document (installPageMotionBlur); insertRuntimeTag, used by the bundler and Studio's preview, puts it at the very start of <head> for pages that opt into blur; the runtime's own install is idempotent with both. Before, it was installed when the runtime evaluated, so a context made by an earlier author script was missed. Either miss left a frozen (Chrome 152) or black (Chrome 157) capture with blur still on.
• report() carries failedFrames, the cumulative count of frames that rendered plain DOM because their capture failed or timed out. failures keeps its streak meaning (it resets after a success), so an "on" report can no longer hide a render where captures keep failing.
• Core's unpacked size budget goes from 3,903,211 to 3,930,000 bytes. With the lifecycle, the compositor and the tracker, core measures 3,903,549 here, just over the old budget; the rest of the stack needs about 3,925,300 at its tip, so the ceiling is raised once, with about 4.7 KB of headroom, rather than in several PRs.

Why

Part 5 of 9 in the capture motion blur stack: #5875 → #5967 → #5968 → #5969 → #5970 → #5971 → #5990 → #5991 → #5992. It builds on #5969, which added installPageMotionBlur in the engine.

Stack goal: opt-in capture motion blur for compositions, in Studio preview and in renders, drawn with Chrome's html-in-canvas. A frozen or black canvas inside a "blurred" frame is a wrong frame, so the runtime has to know about every accelerated canvas, whenever it was created.

Notes

• This PR only makes the runtime see those canvases; what it does about them (turn blur off, and say so) is the existing behaviour, extended to the full defect list in feat(core): block capture motion blur on html-in-canvas defects #5971.
• The tracker is a function whose source is injected, so it must stay free of imports and closures (its toString() is evaluated as a script).
• The engine's log line that prints failedFrames for a session arrives with the runtime's reason codes in feat(core,engine): report why blur is off, and what each render session did #5991.

Testing

• Core pageMotionBlur.test.ts: the tracker records getContext WebGL / WebGPU canvases and transferred canvases, is idempotent across installs, and its source runs standalone.
• Core htmlDocument.test.ts: the tracker goes at the start of <head> for a page that opts into blur, once, and not for a page that does not.
• Core captureMotionBlur/index.test.ts: blur stays off for a canvas handed to an OffscreenCanvas; failedFrames counts every fallback frame though the failure streak resets; engine pageMotionBlur.test.ts: the tracker is the first script installed.
• bun run build, tsc --noEmit in core and engine, and their tests pass; comment ratchet against feat(producer,studio): wire capture motion blur into renders and Studio #5969's branch passes (the tracker's doc comment is two lines, to stay at the file's comment share); every package is within its packed and unpacked size budget.

🤖 Generated with Claude Code