Verification

Reading a parity failure

A failure names the fixture, the dialect, the worst moment and how far off it was. The size is a percentage of the property's own range, so 2px off on a 4px move shows up as the big error it is, and 2px off on a 400px move doesn't.

Start with the fixture's sentence
Each fixture says why it exists, and that sentence prints with the failure. If “Cut spring” fails, something about cutting springs short changed, before you’ve read a single number.
Check which dialects failed
One dialect failing usually points at that dialect’s export. All three failing at once points at the engine, or at a fixture whose expected values changed.
Check when it was worst
Worst at the start means a wrong starting value. Worst at the end means a wrong target or rest threshold. Worst in the middle means the curve: a wrong ease, or too few samples.
A huge percentage is usually leftover state
A number far outside the normal range usually means something was left on the element by the previous fixture, not that a curve is wrong. A 7071% error, for example, is a 0.7071 that should have been reset.

Before trusting a pass, make sure the check can fail: break the export on purpose and confirm the fixture catches it. Break something the fixture actually runs, or the test proves nothing.