Trust / Monitoring
The failure that matters is the one that goes quiet
Alerting on errors is ordinary. Watching for the absence of an expected event — a standby that stopped copying, a model that stopped answering, a job that stopped running — is the part most platforms miss.
A job that fails loudly gets fixed the same day. A job that silently stops is discovered months later, at the moment you needed it.
The question this answers
What breaks without anybody noticing?
Scheduled work. Backups, restore tests, delinquency sweeps, model checks, vendor connectors — all things that produce no error when they simply stop. Waypoint watches each of them, and a job that fails three times lands as a dead letter with its arguments, ready to replay, rather than vanishing.
What is watched, and how the alarm is shaped
Presence, not just errors
Every Minute, From Outside
An outside monitor measures uptime every minute against the 99.9% target, with an alarm when the standby region stops copying.
Hourly Host Health
Memory and swap are checked every hour, and Waypoint’s alert address is emailed when either is outside its threshold.
A Daily Model Check
At 06:15 one tiny call checks the AI model still answers. If it stops, an audit record is written and Waypoint is emailed every failing day.
Connector Watch
Every night each vendor lane with a health check is probed, and a failing lane shows red with its failures in the last 24 hours.
Dead Letters, Never Lost Jobs
A job that failed three times lands as a dead letter with its arguments. Replaying it puts it back as it was and records who did it. A letter is never edited.
What an operator sees
Operator visibility, measured not guessed
Waypoint’s Platform Health card is read off the system when it opens. Where an operator screen refers to claims for cost, they appear as anonymous labels.
Restore Time
The last restore’s elapsed time, and the projection at today’s size
Model Age
Active models older than 180 days, or whose labelled rows grew 25% since training, raise an alert
Organization Keys
Each organization’s key is checked every few minutes. If it is refused, Waypoint closes that organization’s access at once and tells its administrators.
AI Spend
A daily spend watch per organization that tells and never switches anything off
Marketplace Refusals
A watch every 10 minutes on add-ons reaching outside their scopes
Job Queue
Jobs waiting, worker slots and dead letters not yet replayed
Where we are honest
What monitoring is not
Good observability shortens the time between something breaking and somebody knowing. It does not prevent the break.
Not Prevention
An alarm is a fast discovery, not a guarantee. Prevention lives in the release gate, least-privilege database accounts and the walls, which are different mechanisms.
A Target, Not A Contract
We measure against a 99.9% uptime target. Ask what we have actually run against it rather than what we would like to claim.
On-call Is Small
We are an early-stage team. Our response depth is not that of a vendor with a follow-the-sun operations center, and pretending otherwise would be the easiest lie on this page.
Read next
Ask what our alarms caught last quarter
A monitoring story with no incidents in it is a monitoring story nobody has tested. We will tell you what fired and what we changed afterwards.