- Where we measure from
-
Outside our own infrastructure — no probe runs on the servers it
measures.
- What we measure
-
The same endpoints customers use, once per minute — never an
internal health check. Three are public and anyone can reproduce
them. The fourth signs in with a read-only API key, because
"the API answers" and "the API accepts your credentials and
returns your account" are different promises and only the second
one is what a customer is actually buying.
- How a minute is counted
-
Up if any vantage reached us, down only if every vantage that
reported failed, and unmeasured if none could take a reading — or
if a published incident says the reading we did take was our own
instrument failing. Unmeasured minutes leave both sides of the
fraction: they are never counted as uptime and never as downtime.
- What the percentages mean
-
Measured uptime: up minutes divided by measured minutes, with no
maintenance windows removed. The denominator counts only time we
could have measured, so it begins when monitoring began and ends at
the current minute — we are not credited for time before the prober
existed, and not charged for the rest of today. These figures are an
operating indicator, not a contractual SLA.
- When minutes are excluded
-
A published incident can declare that its own window was
unmeasurable — that our instrument failed rather than the platform.
Those minutes then count as unmeasured, exactly as a minute nobody
probed does. They are never counted as uptime, the samples behind
them are never edited or deleted, and the exclusion covers only the
incident's own span and only the components it names. Every row it
touches prints how many minutes went, next to the percentage, and
links to the entry that took them. If you would rather judge us on
the raw record, the API publishes both numbers.
- Where the incidents come from
-
A person writes them. Nothing on this page opens an incident
automatically from a run of failed probes, because a run of failed
probes is not yet a claim about the platform — it can equally be
our own measurement breaking, and telling those apart is a
judgement rather than a threshold. The consequence is that an
empty list means nobody filed anything, which is not the same as
nothing having happened.
- Where the incidents come from
-
A person writes them. Nothing on this page opens an incident
automatically from a run of failed probes, because a run of failed
probes is not yet a claim about the platform — it can equally be
our own measurement breaking, and telling those apart is a
judgement rather than a threshold. The consequence is that an
empty list means nobody filed anything, which is not the same as
nothing having happened.
- When this page is wrong
-
When the prober stops, this page says so and keeps showing the
last measurement with its timestamp, rather than the most recent
state it can guess at. A stale reading labelled stale is useful; a
confident reading that is quietly ten minutes old is not.