Status
Every vendor in this category advertises 99.9% and none will show the measurements. These are the measurements, including the intervals where there is nothing to show.
Last 90 days
How this is counted
An interval with no traffic is not counted as up. It is counted as nothing. A service nobody called is a service nobody has tested, and quietly filing those as successes is how a product with four customers advertises five nines. The denominator is printed above so you can judge the figure rather than take it.
An inconclusive verdict is not an error. It is the product working. Counting it as failure would make honesty look like unreliability, and would create an incentive to guess.
Reduced depth is recorded separately from failure. When the monthly upstream budget is spent, lookups still answer from the free tiers and are not billed. The service is up and shallower — a real distinction, and one worth seeing rather than having averaged away.
This is self-reported, and it runs on the same platform as the API. An outage bad enough to stop the recorder leaves a gap rather than a red mark, and one bad enough to take down this page means you cannot read it at all — in which case assume the answer is no. A status page hosted by the thing it monitors is worth less than an independent one, and saying so is worth more than pretending otherwise.
The same argument applied to accuracy is the published benchmark, whose dataset you can download and rerun. What shipped and when is in the changelog.