Glossary

One line each. Where a term has a precise meaning that differs from the obvious reading, the difference is the definition.

Monitors and pings

Heartbeat monitor: A job that tells PulseWatch it ran, by requesting its ping URL. Nothing is checked from the outside. Silence is the failure signal.
Uptime monitor: A URL PulseWatch fetches on a schedule. The response status, latency, body text and certificate all decide the outcome.
Ping: One HTTP request to a monitor’s ping URL. A plain request means success, /fail means failure, /start means a run began, /exit/<code> lets the exit code decide.
Ping token: The secret in a ping URL. Anyone holding it can report on that monitor, so treat it like a credential, but it can never read your data.
Period: How often a heartbeat monitor is expected to report. The alternative is a cron expression, for schedules a fixed interval cannot express.
Grace: How long past the expected time a ping may arrive without the monitor going down. A job that runs two minutes late should not page.
Auto-provisioning: Creating a monitor by pinging a URL that names it, using an account key rather than a per-monitor token. The schedule is inferred after the third ping.
Run: One execution, from a /start ping to the ping that ends it. Without /start there is no duration to measure and no overlap to detect.

Late: Past the expected time, still inside grace. Counts as available. Nothing is sent.
Warn: The run completed, with problems. Never pages and never opens an incident. Counts as available, because the job did its work.
Blocked: An upstream this monitor depends on is down, so this failure has no cause of its own. Not downtime, not availability: left out of the uptime calculation entirely.
Down: Missed its window past grace, failed explicitly, or failed a check. The only state that counts as downtime.
Degraded: An uptime monitor responding correctly but slower than its degraded threshold. A flag on the monitor rather than a state of its own.

Exit code rule: A range of exit codes mapped to success, warn, down or ignore. Set per monitor, per fleet, or as an account default. The most specific match wins.
Metric: A number a job reports on its ping: rows written, files skipped, queue depth. Rules turn a number into an outcome.
Baseline: The learned median for a metric, so a rule can say "far below normal" instead of a fixed floor somebody has to guess and maintain.
Bucket: Which past runs a baseline compares against: all of them, the same weekday, weekday against weekend, or the same hour. Monday is not Sunday and 3am is not 3pm.
Log excerpt: The body of a ping, scrubbed and capped by plan. Where stack traces and error text come from.
Fingerprint: A stable id for the shape of a failure, computed from the parsed trace. The same bug on a shifted line number keeps the same fingerprint.
Error group: All occurrences sharing a fingerprint, with a first-seen time and a count. What makes "this is new" and "this has come back" answerable.
Output diff: A comparison of this run’s excerpt with the last successful one, ignoring timestamps, counts and ids, so only a real change in shape warns.

Duration regression: A run that took longer than its baseline by more than the configured percentage. Warns, never pages.
Overlap: A run starting while the previous one is still open. Your policy decides whether that is fine, a warning, or an alert.
Run pileup: Three or more runs of the same job in flight at once. Goes down whatever the overlap policy says, because that is a host about to fall over.
Slot: One time a job was scheduled to run. Each resolves to satisfied, late, missed, excluded, or still pending.
Completion budget: A promise about how often a job runs, measured over a window. Catches the job that runs four nights in seven, which no single alert ever notices.
Uptime: Time in an available state divided by time measured. Blocked, paused and never-run periods are in neither half.

Fleet: One monitor covering many machines that run the same job. Configuration is declared once and evaluated per machine.
Instance: One machine inside a fleet, identified by the instance key it puts in its ping URL. Usually a hostname, site id or serial.
Auto-register: An unknown instance key joins the fleet on its first ping instead of being rejected. How a fleet grows without a deploy step.
Retire: An instance silent past the retirement window leaves the denominator. A decommissioned machine should not hold a fleet down forever.
Aggregated alert: One message for the whole fleet naming the affected instances, then a count past the cap. The reason a fleet is not 200 monitors.

Incident: The record of one outage, opened when a monitor goes down and resolved when it recovers. A quick recovery and re-failure merges rather than opening a second.
Runbook: The instructions attached to a monitor, delivered inside the alert itself. The point is that nobody has to find the wiki at 3am.
Acknowledge: Marking an alert as seen, from a single-use link in the alert. Stops the repeats without pretending the problem is fixed.
Snooze: Silencing a monitor for a set time. State and history keep recording; only the notifications stop.
Maintenance window: A recurring period where notifications are suppressed for planned work. Suppressions are recorded, so a silent window is auditable rather than invisible.
Change event: A deploy or config change you recorded, shown against alerts that followed it. It states timing and never claims causation.

Channel: Where an alert goes: email, Telegram, or a webhook. Slack and Discord URLs are detected and get native messages.
Signing secret: The key PulseWatch HMACs a webhook body with, so your endpoint can prove the delivery came from us and was not modified.
Digest: A batched summary rather than one message per event. The default delivery for warnings, so a job that warns nightly does not become noise.
Status page: A public page showing the monitors you chose to publish. Names and states are visible; URLs and error text are not.
Badge: An SVG showing one monitor’s state, for a README.
Advisor: A periodic pass over your configuration looking for monitoring that will not do its job: a grace window too tight, a rule that has never matched, a channel nothing is attached to.

Entitlement: What your plan allows, resolved in one place. Pro and Business differ only in numbers, never in which capabilities exist.
Retention: How long pings, runs and check results are kept before pruning. 7 days on Free, 30 on Pro, 90 on Business.
API key: A bearer token for the REST API. Reading is available on every plan; writing needs Pro or above.
Auto-provision key: An account-level key that can create monitors and post pings, and can never read. Safe to ship inside a machine image.

The states and alert reasons reference is the exhaustive version of the states section above, generated from the code.