# BrokenAI

> Live, head-to-head reliability scoreboard for the major AI providers (Claude, OpenAI, Mistral, Grok, DeepSeek, Kimi). Composite reliability score across uptime, incident frequency, severity, and time-to-fix, sourced from each provider's official Statuspage / RSS feed and refreshed on every visitor request.

BrokenAI is an open monitoring site. All numbers come from each vendor's official status feed; we do not run synthetic API probes, so the data reflects what each provider has publicly acknowledged. Scores update at most once every 60 seconds.

## Pages

- [Home](https://brokenai.live/): Live reliability scoreboard with a per-provider score table, current live component status, and recent incident history across every tracked provider.
- [30-day reliability report](https://brokenai.live/monthly-report): Side-by-side 30-day metric breakdown — uptime, incidents, severity, MTTR percentiles, busiest day/hour, longest clean streak — across every tracked provider with the leader cell highlighted per row.
- [13-week SLA compliance](https://brokenai.live/sla): How often each provider meets its stated 99.9% SLA target across rolling 7-day windows for the past 13 weeks.
- [Scoring methodology](https://brokenai.live/about): How the 0-100 composite reliability score is calculated. Weights: 50% uptime, 20% incident frequency, 20% severity, 10% MTTR.
- [Embed badges](https://brokenai.live/embed): README-ready Markdown and HTML snippets to embed live reliability badges on a third-party site.

## Per-provider pages

- [Claude reliability profile](https://brokenai.live/p/claude): Live uptime, recent incidents, MTTR, component breakdown, failure-type mix and worst incidents for Anthropic's Claude, sourced from https://status.claude.com.
- [OpenAI reliability profile](https://brokenai.live/p/openai): Live uptime, recent incidents, MTTR, component breakdown, failure-type mix and worst incidents for OpenAI, sourced from https://status.openai.com.
- [Mistral reliability profile](https://brokenai.live/p/mistral): Live uptime, recent incidents, MTTR, component breakdown, failure-type mix and worst incidents for Mistral AI's Mistral, sourced from https://status.mistral.ai.
- [Grok reliability profile](https://brokenai.live/p/xai): Live uptime, recent incidents, MTTR, component breakdown, failure-type mix and worst incidents for xAI's Grok, sourced from https://status.x.ai.
- [DeepSeek reliability profile](https://brokenai.live/p/deepseek): Live uptime, recent incidents, MTTR, component breakdown, failure-type mix and worst incidents for DeepSeek, sourced from https://status.deepseek.com.
- [Kimi reliability profile](https://brokenai.live/p/moonshot): Live uptime, recent incidents, MTTR, component breakdown, failure-type mix and worst incidents for Moonshot AI's Kimi, sourced from https://status.moonshot.cn.

## How the data is collected

- Most providers expose Atlassian Statuspage v2 JSON (`/api/v2/summary.json` for live state + components, `/api/v2/incidents.json` for last ~50 resolved). Mistral publishes via Checkly SSR (`/_payload.json`); xAI publishes via Instatus RSS (`/feed.xml`). BrokenAI fetches each, merges them so resolved-transitions land on the same row, and stores a normalized incident table.
- Visitor-triggered fetching (60-second TTL) keeps live data fresh during the day; a daily cron at 06:00 UTC pulls the deep-history feed to backfill incidents that opened and closed between visitor polls.
- Composite scores are recomputed for every supported window (live, 1h, 12h, 24h, 7d, 14d, 30d, 60d, 90d) on every successful fetch.

## Optional

- Source feeds: Claude → https://status.claude.com, OpenAI → https://status.openai.com, Mistral → https://status.mistral.ai, Grok → https://status.x.ai, DeepSeek → https://status.deepseek.com, Kimi → https://status.moonshot.cn.
- Maintainer: threatner_ on X (https://x.com/threatner_).
- Tagline: Is AI broken right now?