Overview · About

About

An independent measurement layer for AI provider reliability. Cross-provider incident data, a published methodology, impact-weighted uptime, and multi-source confirmation before anything is called an outage.

What we publish

  • Per-provider impact scores and impact-weighted uptime for every confirmed incident.
  • Cross-provider comparison tables, each cited to the methodology that produced them.
  • Per-model claims only when a provider or its SDK ecosystem identifies the model.
  • The full open dataset under CC-BY-4.0 at /data.
  • Our own uptime at /status.

What we do not publish

  • Latency or output-quality claims, until independent measurement is in place.
  • Per-region uptime for providers that do not expose per-region incident data.
  • Root-cause attribution beyond what the provider has stated.
  • Single-source incidents. Anything called an outage here has been confirmed by at least one strong or two medium sources.

Disputes and errata

Dispute any number by filing a GitHub issue. Errata appear prominently in the next weekly newsletter. We do not silently fix disputed numbers.

Code MIT · Methodology and dataset CC-BY-4.0 · not affiliated with any AI provider.