Methodology

The homepage button records a report only when a visitor actively clicks it. The site does not directly test Claude service health, ping Anthropic infrastructure, or confirm outages on Anthropic's behalf.

Each browser gets a persistent local client ID. On the server, that browser ID is combined with the request IP address and a server-side salt, then hashed into a reporter key. Raw browser IDs are not used as the public count identity.

Accepted reports are rate-limited in two ways. First, one browser or connection can count only once every 10 minutes. Second, the same connection is burst-limited if too many reports arrive from it inside a short window. These controls reduce obvious manipulation, but they do not eliminate all false or misleading reports.

Reports are stored in Redis in production so the count is shared across Vercel instances. The homepage chart is built from accepted report timestamps. Day view uses hourly buckets. Week and month views use daily buckets.

Status labels such as low reports, elevated reports, or crowd spike are internal crowd-signal thresholds based on recent unique reporters. They are not official provider incident declarations.

All public time labels on the tracker use New York time and explicitly show the timezone abbreviation so the chart can be read consistently during incidents.

Visitors should compare this tracker with the official Anthropic status page before drawing conclusions: https://status.claude.com/.