Checking systems…
Asking every component how it is doing.
How we measure
Measured, not declared.
Every five minutes each component is checked against what it actually did — not whether the server is merely switched on. A day turns amber when a component was degraded for ten minutes or more (or down briefly), and red when it was down for ten minutes or more. The checks run inside the service itself, so when it could not be reached at all, nothing was checked: that time is counted as down for Web app & API, not left out (gaps of up to 15 minutes, such as a restart, are not counted).
Web app & API
Share of real requests that fail on our side, and whether the server is keeping up. Time when the service could not be reached to check at all counts as down.
Database
A test read every few minutes: does it answer, how fast, and is there room for more connections. A check taken while the database is down is kept and recorded once it is back.
WhatsApp messaging
Whether every WhatsApp line that has been linked to a phone is connected. A short reconnect is not counted; a line waiting for its owner to scan a QR code is not an outage.
AI assistant
Whether requests to the assistant come back with an answer, how often the AI providers behind it fail, and whether its queue is full.
Email sync
When each connected mailbox was last read successfully. Mailboxes are read every five minutes; one whose access has stopped working counts as not syncing.
Meeting transcription
Whether the latest recordings were transcribed, failed, or stopped making progress.
Payments
Whether each payment method that is switched on can confirm payments. Mobile-money and bank payments are confirmed by our team.