An API can be available while its observations are obsolete
A public dashboard displays a temperature value, but the reading is several hours old. Stale-data alerting identifies when observation age exceeds the requirements of the intended service.
Define freshness thresholds
Different sensors and datasets have different expected reporting intervals and acceptable delays.
Use observation time
A newly received historical reading should not be mistaken for a current measurement.
Distinguish missing and delayed data
Communication failures, batch uploads and sensor faults can create different patterns.
Avoid false certainty
A repeated value may reflect stable conditions or a frozen device. Combine relevant checks.
Make alerts actionable
Assign responsibility for investigation, recovery and communication with downstream users.
Expose quality status
Clients should be able to distinguish current, stale, missing and quality-flagged observations.
Break-test: a gateway reconnects and sends old readings
Can the service avoid clearing the stale-data warning until genuinely current data resume?
Connected knowledge
See Data Freshness Monitoring, Sensor Failure Detection and NEA