A new company publishing a 99.9% uptime figure has either been running for years or is making it up. We haven't been running for years. So instead of a number we can't stand behind, here's the commitment and the measurement plan — and this page fills in with real figures once there are real figures.
Current rolling figures
NO NUMBER HERE IS BETTER THAN A NUMBER WE INVENTED
What we can tell you today
- The health endpoint is live and public: /api/health returns JSON with region and environment. Check it yourself any time.
- The widget on /status polls that endpoint from your browser while the page is open, so what you see there is a real reading, not a hard-coded row.
- We have not yet wired a third-party synthetic monitor. Until we do, we have no independent record of availability — and we'd rather say that than imply one exists.
How we intend to measure it
This is the method we're committing to, so you can hold the eventual numbers to it:
- An external synthetic monitor — not us — pinging /api/health on a fixed interval from multiple geographies.
- A ping counts as failed if it returns a non-2xx status, times out, or returns malformed JSON.
- Uptime = (successful pings) / (total pings) × 100, published to four significant figures rather than rounded up.
- Scheduled-maintenance windows will NOT be excluded from the calculation. If we take the app down to deploy, it counts against us.
The target we're holding ourselves to
99.5% monthly, measured over a calendar month. That's a goal we've set, not a contractual guarantee, and we're deliberately not attaching automatic service credits to it until we have enough operating history to know what we can actually honour. If an outage materially affects your month, email us and we'll make it right case by case — that's a smaller promise than an SLA, and it's one we can keep today.
TARGET · NOT A CONTRACTUAL SLA · HELLO@LADLE.CHAT
What won't be counted, when counting starts
- Anthropic API downtime, unless it breaks LADLE end-to-end. If inference degrades but the app still serves, that isn't a LADLE outage.
- Individual-user network issues — their ISP, a VPN blocking our domain. Once monitoring is live, if it succeeds from multiple geographies we will call the service up.
- Feature-specific bugs that don't affect the health endpoint. Those belong on /status/incidents and won't move the uptime figure.
When this page fills in
The first 30-day figure publishes one full month after external monitoring goes live, the 90-day figure three months after that, and the 365-day figure a year in. Each one lands here with the month-by-month breakdown underneath it. Until then this page stays empty on purpose — which also means that if you are reading it and it is still empty, monitoring is still not wired up.