Status
Whether each measured endpoint is serving right now, and how it has held up against an availability objective this site publishes and applies identically to everyone. That objective is ours — not an SLA, and not a promise any provider made.
1 provider is refusing most of what we ask
90 days ago46 of 90 days measured · 292,759 requests refused rather than answered, excluded from the figuretoday
Error budget at 99.9%: 533× over — 19,105 failures against the 36 failures this many requests earn
too few recent requests to say how fast it is burning
90 days ago46 of 90 days measured · 76,560 requests refused rather than answered, excluded from the figuretoday
Error budget at 99.9%: 2.1× over — 3,402 failures against the 1,651 failures this many requests earn
burning 0.0× over the last 490 requests
90 days ago46 of 90 days measured · 34,389 requests refused rather than answered, excluded from the figuretoday
Error budget at 99.9%: 29% spent — 993 failures against the 3,482 failures this many requests earn
burning 0.0× over the last 1,021 requests
90 days ago46 of 90 days measuredtoday
Error budget at 99.9%: 27% spent — 471 failures against the 1,716 failures this many requests earn
burning 0.0× over the last 507 requests
90 days ago46 of 90 days measured · 51,579 requests refused rather than answered, excluded from the figuretoday
Error budget at 99.9%: 69% spent — 1,183 failures against the 1,707 failures this many requests earn
burning 0.0× over the last 505 requests
An SLA is a contract: a promise a provider made, with a remedy attached when it is broken. Almost none of the endpoints measured here carry one, so the bar below is ours. A provider that misses it has not broken a promise — it has failed to clear a threshold we set.
A single threshold says little: at 99.9% almost every free endpoint fails, and at 99% almost everything passes. Each provider is reported at the highest rung it actually cleared, and reported as not enough data — never as failing — where fewer than 500 requests were eligible. Not knowing and failing are different findings.
The pill beside each endpoint and the uptime percentage beside it are judged at different bars, deliberately. The recent window holds a few dozen requests, which can tell “mostly working” from “mostly broken” and cannot tell 99.9% from 100% — so the pill uses 90% and the percentage uses the objective. A green pill beside a 99.5% figure is those two statements agreeing, not disagreeing.
eth_blockNumber2026-10-08 10:10 · 6m measuredDurations are floors: only minutes with evidence inside them are counted, so an incident is never reported as longer than what was measured. The full record
This is an objective we publish and apply identically to every provider. It is not a service level agreement: nothing here is a promise any provider made, and a provider that misses it has not broken one. What a provider does commit to, where it commits to anything, is quoted on its own page with the source and the date it was read. Availability is taken over requests the provider could reasonably have answered — rate limits, methods it does not implement and methods it sells rather than gives all leave the denominator, and the counts that left are published beside the percentage so it can be judged. A day nothing was measured is drawn as a day nothing was measured, never as a day everything worked, and it is excluded from the uptime figure rather than counted as perfect. Every measurement comes from a single vantage point, so a component shown as operational was reachable from there.
The full record, including the persistent defects deliberately kept out of the incident list, is on the incidents page; what each provider says it permits is on its own page, and how every figure here is produced is on the methodology page.