Measured from what the service produced, not from a separate monitor. Every figure states the window it came from.
How many city-readings each upstream failed to deliver since metrics collection began. Seven of the eleven arrive through one aggregator, so an outage there shows up across seven rows at once.
| openmeteo | 42084 |
| dwd_mosmix | 17088 |
| kweather | 11679 |
| metno | 3678 |
| metar | 3344 |
| noaa_gfs | 2505 |
| eccc_gem | 1670 |
| ukmo | 1503 |
| jma | 1336 |
| meteofrance | 1169 |
| ecmwf | 334 |
Cumulative counts since metrics collection began on 2026-08-01, which is after the verdict rule was tightened on 2026-07-31 - a 1-1 split between measurements used to count as verified and now counts as mixed. Hours judged under the old rule are not included here, and are excluded from the city grades too. verified means measurements corroborated the published value; mixed means they were split; single_source means only one measurement could be read.
| verified | 123444 |
| mixed | 73179 |
| mismatch | 30060 |
| single_source | 11115 |
| unverified | 344 |
What these numbers are not
The service has been running publicly for 66.6 days, so every percentage here comes from that window and not from a long record. A high figure over a short window is weak evidence, and we would rather say so than let the number stand alone.
This is a measurement, not a service level agreement. Nothing on this page commits us to an availability target or entitles anyone to a credit. Availability commitments for insurance or hedging use are agreed separately with KWeather.
On-time runs are derived from the gaps between anchored snapshots: collection runs hourly and a successful run leaves one snapshot, so the spacing of those rows is the record. A gap wider than 90 minutes counts as one missed slot per hour skipped.