How camera liveness is measured

Most camera directories assume a camera works because an agency lists it. The numbers in uptime-summary.json come from checking instead: every camera in the directory is probed on a rolling schedule with real HTTP requests, around the clock.

What a check does

• Each check reads the image feed's last-updated time.
• A camera is live only when its image is genuinely current.
• A camera serving an old frame is stale: the feed answers, but the picture is not current. Stale cameras look fine until you check the timestamp.
• A camera whose feed stops answering is dead.

Timestamps in the data are real check results, never a fixed label. Nothing is set by hand.

What the summary fields mean

• live_now / stale_now / dead_now: the fleet's state at the moment the report was generated.
• checkLivePct: the share of all checks over the last 14 days that found a live image; a stability measure rather than a snapshot.

The time series

data/uptime/history.csv records these fields once a week per state, from 2026-08-31 onward. Each row is the report as generated that day: the "now" columns are that moment's fleet state, and the 14-day columns cover the two weeks before it. Consecutive weekly rows therefore overlap by a week in their 14-day window; treat the 14-day rate as a smoothed reliability measure, and the "now" columns as the weekly point sample. Rows are never edited after the fact. A state appears only from the week its feed became measurable.

Fairness rules

The published report excludes states whose feeds cannot be measured and does not rank fleets under 50 cameras, so small networks are not shamed by small samples. The full ruleset and the always-current numbers are on the camera uptime report.