How we measure pixel fidelity (and why nobody else does)

Most exporters say 'pixel-perfect'. We measure it. Every FrameFlow export is re-rendered and pixel-diffed against the original at each breakpoint, and the score ships with the result.

We report three numbers. Static fidelity is the headline: the percentage of deterministic pixels that match, excluding inherently-animating regions. Raw fidelity counts every pixel — the honest floor. Dynamic percent is the share we mask as animating.

Masking isn't cheating, it's accuracy. A continuously-moving marquee can never pixel-match between two single-frame captures — the clone reproduces it faithfully, it's just in a different phase. We detect those regions two ways and report exactly what was masked, with red-vs-blue diff heatmaps.

The same verify loop also drives repair: real differences (red) are located, attributed to the DOM element under them, and optionally pinned to guarantee fidelity. It's the discipline that turns 'looks right' into a number you can trust.