What happens to each reading: stored by the sensor's own time, once, into the live layer and its alert state, into history and charts, and kept for as long as the tier says.
A reading is one value of a measurement at one time. A sensor sends readings over HTTPS, up to 500 in a call (the sensor endpoint); for each one, in order:
at is what the reading is stored, charted and judged by; when it
arrived is kept beside it. A reading stamped more than ten minutes ahead is refused — the sensor's
clock is wrong — and one older than the enterprise keeps readings is refused too.at again is already received and
changes nothing, so a sensor may safely send a batch again.What an asset's readings say now, set by the platform, never by a person — the live layer's
alert_state:
| Value | Shown as | When |
|---|---|---|
normal |
Normal | a switched-on sensor of the asset has reported |
alerting |
Alerting | an event of a rule on its readings — a limit or a zone — is open |
stale |
Stale | a switched-on sensor has readings, the newest older than the class expects |
none |
None | nothing reports on it: no sensor, none switched on, or none has sent anything |
When two apply, the order is stale, then alerting, then normal, then none: a quiet sensor wins over
an open crossing, which stays open and counted in open_events, and the asset reads Alerting again once
its sensor reports. A person's service status is a different column on a different layer.
A sensor is stale when it is switched on, has sent readings, and the newest of them — by its own time — is older than its class's Readings expected every. The platform looks every five minutes. An asset is stale while any of its switched-on sensors is; a sensor that reports again is cleared with that reading, not five minutes later. A sensor that has never sent anything is not stale: its asset reads None.
A sensor going quiet is told to people only through a rule on A sensor goes quiet. Without one it reads Stale on its page, its asset's, the live layer and the dashboards, and nobody is messaged.
The asset's page › Readings: the latest values, and a chart per measurement with a line per sensor, over a period — the last 24 hours, 7, 30, 90 or 365 days. A short period draws every reading; a longer one the average of each hour, then of each day; the note under the period says which. A yes/no measurement draws each hour's or day's last value.

| What | Kept |
|---|---|
| Each reading | the enterprise's retention — on Haiyz Cloud the tier's (Pro 30 days, Max 90 days); on a self-hosted instance its own setting (365 days unless changed) |
| Each hour's count, minimum, maximum, sum and last | 90 days |
| Each day's | until the sensor, its class or the enterprise is deleted |
| The readings an event keeps | as long as the event is kept — see Events |
A reading with a position can be personal data; it goes with the retention, with its sensor and with the enterprise.