The CSS Regression a Boot Check May Not See

Scope: Engineering or methodology note based on a September 17, 2026 snapshot. Source-level behavior is not a public availability or universal compatibility claim. Any scenarios are illustrative, not customer outcomes.

A page can preserve its text, landmarks, and basic document structure while looking visibly wrong. A class name changes, a stylesheet no longer applies, or spacing collapses. A semantic DOM comparison may still pass.

That is not a paradox. It is a reminder that every observation has a scope. KodeBack’s current boot comparison should not be described as complete visual verification.

Make the blind spot concrete

Imagine a synthetic button with unchanged accessible text and role. An edit changes its CSS class from the intended class to a nonexistent one. The button may remain in the document and still satisfy the observed landmark checks while losing its appearance.

Now consider an event handler that only runs after a second interaction. A boot-only check does not exercise that path. The absence of a detected difference says nothing about the behavior that was never observed.

A passing result can still be useful

Boot parity answers a narrower question: whether the current candidate satisfies the declared boot observations under the selected conditions. That can catch meaningful regressions in visible content, replay health, console behavior, and declared network subjects where measurement is supported.

The mistake is not using a narrow check. It is allowing the word “parity” to erase the qualifier. A useful report and a useful demo should name the scope every time the result is promoted into a product claim.

Add evidence according to the maintenance risk

A style-sensitive change needs appropriate visual review or measurement. An interaction-sensitive change needs an authored journey and observable consequences. A canvas application needs evidence beyond the presence of the canvas element. These are proposed evaluation requirements, not a claim that the ordinary boot command currently supplies every adapter.

The current source includes separate evidence foundations, but their existence does not mean they automatically run in the default product path. Product integration and qualification must be stated independently.

For an engineering handoff, keep a brief “not observed” section beside the passing result. It should identify appearance, unexecuted actions, and other relevant gaps rather than leave the next maintainer to infer them. That does not weaken the boot check. It lets the check remain useful without carrying a promise it was never designed to fulfill.