A value has several times attached to it, and they answer different questions. Freshness is knowing which one you are looking at.
Three times
| Time | Question it answers | Stored as |
|---|---|---|
| Observation time | When did the station measure this? | observedAt and localDate |
| Source update time | When did the source last revise it? | sourceUpdatedAt |
| Ingestion time | When did Snotel last store it? | Stored, not in API responses |
Observation time is a property of the world. The other two describe the pipeline, and they can be much later than the observation: a station reports on a schedule, the source publishes after that, and Snotel ingests on its own schedule after that.
How current a station looks
To judge whether a station is current, compare the latest observation time to now, not the ingestion time. A station can look stale for reasons that have nothing to do with Snotel: a failed sensor, a telemetry outage, weather that stopped a site from transmitting, or a source that has not yet published recent readings.
A station that has stopped reporting is still listed, with its period of record ending when the source says it did.
A zero is a reading, not an absence of one. A station reporting a snow depth of 0 hour after hour is current; a station whose newest observation is hours old is not. Snotel keeps the two apart: how recent a value is comes from its observation time, never from whether the value changed.
Revisions
The source can correct values after publishing them, and recent values are the ones most likely to change. When a value changes, Snotel stores the new value, increases the observation's revision, and records the source's update time so corrections apply in order. A value you read yesterday for a recent day may therefore differ from the value for that day today.
How often Snotel reads the source
| Data | When Snotel reads it | Counts as stale after |
|---|---|---|
| Hourly observations | Every hour, at 15 minutes past, reaching back six hours | 6 hours |
| Daily observations | Once a day at 07:45 UTC, reaching back three days for revisions | 3 days |
So a latest reading in the API is at most about an hour behind what NRCS has published, and a daily value the source revises within three days is picked up. A revision older than that is picked up only when that range is read again, for example by a backfill.
How freshness is reported
The latest conditions endpoint and the station_latest MCP tool judge each reading when the response is sent: fresh when a recent period has a value, no_values when periods arrive with no value, stale when no period newer than the threshold above has arrived. The verdict is about whether data is reaching Snotel, never about whether the snow changed.
The coverage page and GET /api/v1/coverage show, for each source, how many stations reported within these thresholds, when each scheduled read last delivered, and the known gaps in the history.
Snotel does not promise any particular delay between a station observing something and the source publishing it. That depends on the source's own schedule.