I'll check the bin boundaries for all rendering metrics; but, for the particular input_event_latency_tbmv2 metric, why do you think the bucketing is unrealistic? Is it based on one trace file or there is another reason? When I look at the average value among all stories that metric looks OK:

https://v2spa-dot-chromeperf.appspot.com/#session=11003d2a93f09c60b3225c2e689ac83a4c243d1dfac87c5edbbf17ccd7ff7832

You might be right, the buckets for input-latency are probably OK. Another site I was looking at was facebook, but the higher latencies fall in the ~90%ile, so that's probably OK.

For bucketing: we might want to look into using custom buckets, if that's possible, instead of using linear etc. buckets, so that we can have more granularity around the frame-boundaries, and less in the middle.

See memory_metric.html for an example of custom bins:
https://github.com/catapult-project/catapult/blob/master/tracing/tracing/metrics/system_health/memory_metric.html#L50