Static evidence can prove a lot.

It can prove imports resolve, types match, declarations stay stable, and source spans still point to the same symbols.

It cannot prove every interface a user experiences. HTML, CSS, JSX, hydration, focus, browser layout, and accessibility all have runtime meaning. Some merge claims need runtime proof.

Runtime Is A Surface

A UI change can preserve source shape and still break the interface.

Examples:

button still exists, but moved outside the viewport
label text remains, but accessible name changed
class compiles, but container query flips the layout
handler exists, but click target is covered
hydration succeeds, but focus order changed

Those are runtime claims. They need runtime evidence.

Probes Should Be Small

The system should not always need the whole application.

A useful runtime proof can often run against a micro-slice:

component fixture
route shell
CSS module fixture
DOM state snapshot
event sequence
accessibility tree query

The real environment still matters for final gates, but micro-slices make evidence cheaper and more focused.

Browser Evidence Needs Shape

A screenshot alone is not enough.

Runtime proof should produce structured observations:

bounding boxes
computed styles
accessible names
focus order
event results
layout shift windows
console errors
hydration status

The screenshot can help humans. The structured evidence helps the merge system route.

Scope Matters

Runtime proof should not claim more than it measured.

If a probe covers the happy path, it should say that. If it covers desktop but not mobile, it should say that. If it runs without login state, it should say that too.

Honest limits make the evidence usable.

The Mental Model

Runtime proof is how semantic merge talks to interfaces that only exist after code runs.