Accuracy

Our accuracy scorecard

Every number on this page is computed automatically from our own incident records. Nothing is estimated, rounded up or hand-picked, and incidents where we were wrong or slower than the vendor are included.

Last updated 2026-11-13 04:43 UTC. Our incident records start on Oct 31, 2026, so "all time" covers 13 days.

How often we were right

What these numbers mean

• Incidents published: every incident that appeared in our public outage lists (it lasted 10 minutes or more, or reached major outage), plus every incident we later retracted, however short. Scheduled maintenance is listed separately and not counted.
• Retracted: incidents we withdrew after review because the evidence did not support an outage. Each keeps its page with a correction notice and is listed below.
• Downgraded by the self-audit: incidents (not retracted) whose label or confidence our nightly self-audit lowered because the evidence was weaker than first shown.
• Closed early by automatic review: incidents (not retracted) that our evidence rules closed because only the vendor status page reported a problem.
• Confirmed by a second source: incidents (not retracted) where at least two independent sources agreed: our own checks failed at least twice, at least 5 people reported a problem within 15 minutes, or the vendor posted a major or critical incident. The rate is out of all published incidents, retractions included.
• Our checks failed from 2+ regions: incidents (not retracted) where our checks failed at least twice and failed from at least 2 of our 4 probe regions (US East, US West, Europe, Asia-Pacific). This is one source, our own checks, seen from several places; it is not counted as a second source.
• Labeled vendor-reported issue: incidents that rested only on the vendor's status page. We attribute these to the vendor and do not call them outages.

Detection speed vs vendor status pages

Small sample: 3 matched incidents. Treat these as early numbers, not a trend.

How we measure this

We compare the moment our own evidence first carried an incident (our checks failing repeatedly, or a spike in user reports) with the start time the vendor posted on its own status page for an incident that was open at the same time. A vendor incident counts as the same event only if it started within 6 hours of our detection.

We use the earliest time the vendor gives (its posted start or creation time), so a vendor that backdates its start is credited with it. Retracted incidents, scheduled maintenance and incidents that rested only on the vendor's status page are left out. Incidents where the vendor was first are counted in full.

Slowest to admit it: which companies took longest to post on their status page after a confirmed outage →

Data gaps

• Most services we track do not publish a machine-readable status page, so their outages cannot be compared. We began storing vendor incident times on Nov 10, 2026; older incidents were filled in only from vendors that publish their incident history.
• Incidents with our own evidence where the vendor also reported a problem but we have no posted vendor time: 2.
• Incidents where the vendor incident started more than 6 hours apart from ours (treated as a different event): 1.

Every matched incident

Every retraction

We have retracted 68 incident reports in total. Each one keeps its page with a correction notice in English, German, Spanish, Indonesian and Japanese.

How we detect, verify and correct outages

Disclaimer

Outage information on this site is provided as is, without warranty, for information only. It may be inaccurate, incomplete or late. Damn Is It Down is not affiliated with, sponsored by or endorsed by the companies listed. Company names and trademarks belong to their owners and are used only to identify the services.

Read our terms of service