# kemek-monitoring status page Public operational status for components monitored by kemek-monitoring. ## What this surface exposes - /status HTML status page (components, overall status, active incidents) - /status.json Same data as JSON: { version, overall_status, components[], active_incidents[], updated_at, monitor_impaired_since } Both lists name the same unit: active_incidents[].component matches a components[].name. This page reports COMPONENTS, never the individual checks behind them, so an incident tells you which component is affected and not which target. - /status/faq How the uptime figure is calculated, in prose - /llms.txt This file ## What "status" means - operational — all probes for the component succeeded recently - degraded — at least one probe is slow or failing assertions - outage — at least one probe is failing entirely - maintenance — planned work declared in advance; probing continues and results are recorded, but a failure here is expected and does not open an incident. A component under maintenance whose outage STARTED before the window still reports "outage". - unknown — no probe data yet ## Uptime, and why there are two numbers Each component carries: - uptime_90d 0..1 or null. THE PUBLISHED FIGURE. Two kinds of probe are removed from both the numerator and the denominator — not counted as up, not counted as down: those inside a declared maintenance window, and those taken while this monitoring service had itself lost its uplink. - uptime_90d_raw 0..1 or null. The same period counting everything, maintenance included. - coverage_90d 0..1 or null. The share of the window we were actually watching. Counted in HOURS, not probes — a probe-count denominator can be inflated by the same malfunction it is meant to reveal. null means unknown, never zero. - uptime_90d_verified 0..1 or null. uptime_90d x coverage_90d: the share of the whole window we can vouch for. Unobserved time counts as neither up nor down, only unverified. Lower bound of the honest interval; the upper is uptime_90d_verified + (1 - coverage_90d). - maintenance_runs_90d integer. How many probe results uptime_90d excluded as declared maintenance. A count of results, not a duration. - uplink_runs_90d integer. How many probe results uptime_90d excluded because THIS monitoring service could not reach the internet at the time. Those probes failed, and the failure was ours rather than the component's. Reported apart from maintenance_runs_90d because the two are different admissions. - days array, oldest first, one entry per UTC day of the window: { day, uptime, sample_runs, maintenance_runs, uplink_runs }. Empty if no daily readings were taken. "uptime" is 0..1, the same units as uptime_90d — not a percentage. THREE STATES, and they are three different claims: uptime is a number when we watched that day; null with maintenance_runs above zero means we watched and every probe was declared work; null with nothing at all means UNKNOWN — no record, or a day before the check existed. Unknown is not zero. Do not collapse the last two: one is planned downtime and the other is our ignorance. null means there was nothing to divide — no checks in the window, which includes a period spent entirely under maintenance or entirely inside an outage of our own. It is not 100%. ## Incident timeline Each entry in active_incidents carries: - updates array, oldest first: { at, kind, body }. Operator-written notes. "at" is a UNIX millisecond timestamp; "kind" is one of investigating, identified, monitoring, note. THERE IS NO "resolved" KIND. Whether an incident is over is decided by probe data, not by a note — an incident disappears from active_incidents when it closes. A note can never assert a recovery the measurements did not show. Empty means nobody has posted. It does NOT mean the incident is unattended, and there is no synthesised "we are investigating" entry. ## When WE are the thing that broke - monitor_impaired_since epoch ms, or null. Non-null means this monitoring service had lost its own network connection as of this response, and had not yet recovered. READ THIS BEFORE TRUSTING components[].status. Every component status is the verdict of that component's most recent probe. When monitor_impaired_since is non-null, those probes were taken while we could not reach the internet, so a component reporting "outage" may have been serving normally the whole time — the failure is ours, not the component's. The statuses are deliberately NOT rewritten while impaired. Doing so would hide a component that is genuinely broken: we suppress opening new incidents during one of our outages, so any incident listed in active_incidents while monitor_impaired_since is non-null started BEFORE we went blind and is real. Relabelling would also blank components whose probes never needed the uplink at all, and every probe that succeeded as the connection came back. So: treat a non-null monitor_impaired_since as "component statuses are unverified from this vantage point", and treat active_incidents as reliable regardless. The uptime figures already exclude the affected probes; see uplink_runs_90d. Uptime is a ratio of CHECKS, not of minutes. A degraded result counts as available. Prose version at /status/faq. ## What is NOT exposed publicly Targets, probe intervals, response times, error details, and operator notes are operator-only and intentionally absent from this surface. ## Agent-friendly notes - /status.json is the contract. Prefer it over scraping HTML. - Use the User-Agent identifying your agent; outbound probes from this service identify themselves as: KeMeKMonitor/0.115.4 (+https://monitor.kemek.net/status/faq) - updated_at is a UNIX millisecond timestamp.