API reference

Version number, increasing with each change.

First minute classified under this version, UTC, RFC 3339. null for a version with no start, which covers every minute before the next version.

Summary of the rules this version applies.

The version classifying minutes now: the highest version whose effective_from has passed. null only before any version has taken effect.

Every version, ascending. Always a JSON array.

Human-readable error message.

Slug-style identifier from the registry.

Human-readable service name.

Provider region for regional services. Omitted for global services (CDNs, DNS resolvers, payment APIs).

Brand or parent company behind the service. Editorial rather than derived from id or name: every aws-* service reports "Amazon Web Services" whatever the product is called, and linode reports "Akamai / Linode" for the parent company. Omitted when the registry row carries no provider.

Whether the service appears on public ranking surfaces. true, the default, lists it on the leaderboard and the services index. false keeps it probed and queryable through /v1/services/{id} and /v1/services/{id}/uptime but off those rankings, which is how a newly onboarded service is held back until it has enough history to rank fairly.

Always present, never omitted: an absent field would read as "the API does not know" rather than "hidden".

Public landing page for the service. Always present.

The provider's own status page, where it publishes one. Omitted rather than empty when it does not, so a client never has to special-case "".

Carried on the list endpoint as well as the detail one, so rendering a directory of services needs a single request.

Number of active checks measuring this service. Checks registered but not yet active are excluded, so the count matches what the fleet is really executing. 0 for a newly registered service whose checks have not landed yet.

Availability percent over the rolling trailing 24 hours, computed from per-minute consensus. null when the window has no measurements yet.

Availability percent over the trailing 7 days, computed from the per-day rollup, which is anchored to UTC midnight. null when the window has no measurements yet.

Same as availability_7d, over 30 days.

Same as availability_7d, over 90 days.

Slug-style identifier from the registry.

Human-readable service name.

Provider region for regional services. Omitted for global services (CDNs, DNS resolvers, payment APIs).

Brand or parent company behind the service. Editorial rather than derived from id or name: every aws-* service reports "Amazon Web Services" whatever the product is called, and linode reports "Akamai / Linode" for the parent company. Omitted when the registry row carries no provider.

Whether the service appears on public ranking surfaces. true, the default, lists it on the leaderboard and the services index. false keeps it probed and queryable through /v1/services/{id} and /v1/services/{id}/uptime but off those rankings, which is how a newly onboarded service is held back until it has enough history to rank fairly.

Always present, never omitted: an absent field would read as "the API does not know" rather than "hidden".

Public landing page for the service. Always present.

The provider's own status page, where it publishes one. Omitted rather than empty when it does not, so a client never has to special-case "".

Carried on the list endpoint as well as the detail one, so rendering a directory of services needs a single request.

Number of active checks measuring this service. Checks registered but not yet active are excluded, so the count matches what the fleet is really executing. 0 for a newly registered service whose checks have not landed yet.

Availability percent over the rolling trailing 24 hours, computed from per-minute consensus. null when the window has no measurements yet.

Availability percent over the trailing 7 days, computed from the per-day rollup, which is anchored to UTC midnight. null when the window has no measurements yet.

Same as availability_7d, over 30 days.

Same as availability_7d, over 90 days.

Active checks for this service. Always an array, never null, and empty for a newly registered service whose checks have not landed yet.

Each vantage point's current view of the service, taken from the most-recent minute that probe reported on, within a 5-minute lookback. A probe that has not reported inside that window is absent from the array rather than listed as unknown. Always an array, never null, and empty for a service with no recent measurements.

Stable check identifier, formed from the service id, the check type and the target it covers.

Probe type. Currently one of http, tls, storage, dns. Typed as a string rather than an enum so adding a probe type stays a non-breaking change.

What the probe hits: a URL, an IP and port, a hostname. The format depends on check_type.

Where the real target embeds a private identifier, such as a storage bucket name, the API substitutes a redacted placeholder of the same shape (for example https://redacted.s3.eu-west-1.amazonaws.com/probe.bin). The check is still described, down to the operation and the region, without publishing the identifier. Only the public wire shape is substituted; the probe hits the real target.

Probe cadence in seconds.

Stable probe identifier from the fleet registry, e.g. hzr-fsn1. Use it as the key when joining probes against probe metadata such as location and provider.

Per-probe consensus across the service's checks within the most-recent reported minute. unknown is not valid here: a probe with no recent report is absent from the array entirely.

p50 latency this probe observed in the most-recent minute, from the service's http checks only, taken as the maximum across them.

null when that minute has no successful http measurement, which is a different case from the probe not reporting at all: that probe is absent from the array.

Timestamp at the start of the bucket, UTC, RFC 3339.

Percent of measured minutes in the bucket that were classified up. null when the bucket has no measured minutes, in which case status is unknown.

Worst minute in the bucket: down if any minute was down, else degraded if any was degraded, else up. unknown when no minute was measured.

Count of minutes in the bucket with consensus down.

Count of minutes in the bucket with consensus degraded.

Count of minutes in the bucket with consensus unknown.

Count of minutes with consensus up, down or degraded: the denominator of availability_pct.

Success-count-weighted p50 (median) response time, in milliseconds, across every http check and probe pair inside the bucket. null when the bucket contained no successful http measurement, so a chart should break the latency line there rather than plot a zero.

Consensus methodology version that classified the bucket's minutes (see /v1/methodology). A bucket spanning a change of version reports the newer one.

100 * total_up / total_measured_minutes across every measured minute in the period. null when no minute had measured data.

Sum of down_minutes across every point.

Sum of degraded_minutes across every point.

Sum of unknown_minutes across every point.

Sum of measured_minutes across every point.

Success-count-weighted p50 response time, in milliseconds, across every successful http measurement in the period. null when the period contained none. Useful as the headline "Average response time" for a service detail page.

Every consensus methodology version the period's points were classified under, ascending. More than one entry means the period spans a change of methodology. Empty when the period has no points.

Echoed back from the path parameter.

Buckets in ascending bucket-time order. Always renders as a JSON array (never null); empty [] is valid for a service with no measurements yet.

100 * up minutes / measured_minutes, truncated to 2 decimal places. null when measured_minutes is 0.

Minutes with consensus down.

Minutes with consensus degraded.

Minutes with consensus unknown (fewer than 3 probes reporting). Not part of measured_minutes.

Minutes with consensus up, down or degraded.

Cross-references /v1/services/{id}.

Human-readable service name.

Start of the first bad minute, UTC, RFC 3339.

End of the last bad minute (exclusive), UTC, RFC 3339. null while the incident is ongoing.

ended_at - started_at in minutes, or the span so far while ongoing. Includes merged gaps, so it can exceed down_minutes + degraded_minutes.

Minutes in the incident with consensus down.

Minutes in the incident with consensus degraded.

down if any minute was down, else degraded.

Effective lower bound after applying the 90-day limit.

Newest first by started_at. Always a JSON array, never null.

Server wall-clock at response build time, UTC, RFC 3339.

1-based standard competition rank on the published 2-decimal availability_pct. Equal values share a rank and the next rank skips (1, 1, 3). Services with availability_pct: null share the last rank. Clients should not recompute rank.

Slug-style identifier; cross-references /v1/services/{id}.

Human-readable service name.

Minute-weighted availability percent over the requested period, truncated to 2 decimal places. null when the service has no measurements in the period.

Minutes with consensus down in the period.

Minutes with consensus degraded in the period.

Minutes with consensus unknown in the period.

Minutes with consensus up, down or degraded: the denominator of availability_pct.

Rank movement against the previous publication, positive for an improvement and negative for a decline. Always absent in v1, because historical leaderboard state is not persisted yet. The field appears once it is.

Echoed when a ?category= filter was applied; absent on the unfiltered call.

Ranked rows, ordered by rank ASC then name ASC. Always renders as a JSON array (never null); empty [] is valid after a category filter that matches no services.

Server wall-clock at response build time, UTC, RFC 3339. Useful for cache debugging, and for rendering an "as of" line alongside a ranking.