Wait for source resize transitions before sampling browser regression fixtures #592

Findings

Two full-CI attempts on #591 failed in browser regression fixtures before portable media materialization ran:

• Attempt 1: fluid-capture.test.ts, mixed-unit image height at width 600, delta 20px. The fixture records source heights after setViewportSize() plus a 10ms timer, although its inline custom property changes in a separate resize event handler.
• Attempt 2: reference.test.ts, fresh-navigation viewport proof, observed 33px instead of the expected resized 29px. It asserts immediately after setViewportSize() without waiting for the fixture's dataset.pose = 'resized' transition.

The mixed-unit fixture passes unchanged on both frozen main and the candidate when run in isolation. The viewport API's completion is not a guarantee that the fixture's resize event callback has run. These tests need explicit source-state readiness before sampling; changing numerical tolerances or retrying assertions would obscure that boundary.

Proposed owning-layer repair

Wait for the authored custom property to match the resized viewport in the mixed-unit fixture, and for the declared resized state in the fresh-navigation fixture. Keep all existing exact height, typography, alignment, source-page preservation and frozen-reference assertions.

Acceptance

• Deterministic delayed-resize reproduction demonstrates that old sampling precedes the authored transition.
• Both fixtures prove their expected source state before measurement, including under delayed resize delivery.
• Focused browser tests and full authoritative CI pass with unchanged assertions.
• Repair ships independently, then refactor(export): isolate portable media materialization #591 resumes through the repaired gate.

AI assistance

OpenAI openai/gpt-6.1-sol via OpenCode inspected the CI failures and test sources, diagnosed the missing readiness boundary and drafted this tracker under Chris Huber's direction. Direct execution outside Homeboy is operator-authorized.