The numbers are not consistent with the documentation. The documentation from https://cs.chromium.org/chromium/src/third_party/catapult/telemetry/telemetry/web_perf/metrics/smoothness.py says:

mean_pixels_approximated looks at tiles which are missing or of low or non-ideal resolution.

mean_pixels_checkerboarded looks at tiles which are only missing. It does not take into consideration tiles which are of low or non-ideal resolution.

Which means mean_pixels_approximated >= mean_pixels_checkerboarded. But the dashboard (below) shows that in some cases the former is close to 0% while the latter is around 50%. Either we are computing the metrics wrong, or they don't mean what they claim they mean.

https://chromeperf.appspot.com/report?sid=0f54afcaeeee97171ce29a93c9a63d10da009001177f1ff9d46342d178571e33