eth_estimateGas

Success excludes calls the provider does not implement — an optional method a provider declines to offer is a capability fact, not a failure, and is never scored as one.

Median latency per hour. Gaps are periods with no measurement, drawn as gaps rather than joined across — a line through missing data invents a measurement that was never taken.

A degradation is a bucket where a provider fell away from ITS OWN median for this method over the window, not from the cohort's. A provider that is consistently slower than its peers is not degraded — it is slower, which the ranking already says. Buckets with fewer than 20 samples are skipped, because a claim about a provider cannot rest on a handful of calls, and adjacent bad buckets are merged so one outage reads as one incident.