Method
Each incident page is built from individual claims. Every claim has a label and at least one source. The explanations summarize those claims. Our own interpretation appears only in a separate, marked “Independent analysis” box.
The archive covers outages that affected North America (the United States, Canada and Mexico), including global outages that reached it.
Each incident has three separate labels: whether the service is working, what is known about the cause, and what is known about malicious activity. One never implies another.
Operational status
- InvestigatingThe provider is investigating; service is affected.
- IdentifiedThe provider has identified the problem and is working on a fix.
- MonitoringA fix is in place and the provider is watching for recurrence.
- RecoveringService is returning but effects continue (backlogs, cancellations).
- ResolvedService is restored.
- Status not confirmedThe sources read do not establish the current status. The last documented state is shown with its date.
For an unresolved incident the page gives the date of the latest source that confirms the status and the time its sources were last checked (“Last confirmed Nov 15 · sources checked Nov 15, 15:45 UTC”). Unresolved incidents are re-checked at least every two days. If a check is overdue, the badge changes to “Last confirmed: Recovering” and the incident is no longer counted as active, even if the page itself has not been rebuilt. Resolved incidents show when they ended.
The end time of an incident says what it marks, such as “Ground stop lifted” or “Faulty update remediated”. That is often not the same as full recovery, which is shown separately as complete, not yet complete, or not documented. An incident can have an end time and still be recovering. When a provider gives a separate end of the main impact, both durations are shown, and summaries are checked against them.
Timeline milestones
Shown when a source documents them, with a time or, when the sources give none, a date: first observed, acknowledged, mitigation, recovery, resolved, incident report. Missing milestones are shown as not documented rather than estimated.
Providers and incidents
Provider pages separate failures in the provider's own systems from disruptions to its services caused by a failure elsewhere, such as Windows computers crashed by a CrowdStrike update. A provider whose systems played a documented part in another provider's outage is listed as contributing.
Failure diagrams
“How the failure happened” draws only the steps a source states. Under the diagram, each step lists its evidence and whether it comes from a final incident report or only from preliminary statements. Each step must rest on the same sources as the stated cause or a published claim. When the sources do not describe a sequence, the page says so instead of drawing one. Dependencies are never inferred.
Root cause
- UnknownNo cause has been published, or officials say it is not established.
- PreliminaryA cause has been stated, but the final report is not out.
- ConfirmedThe provider or agency has published the root cause in its own report.
Security evidence
A missing statement is never read as a finding. “Explicitly ruled out” is used only when a provider report or completed investigation says so.
- No determinationSources do not address malicious activity either way.
- No evidence reportedThe provider or officials say they found no evidence of malicious activity. This is not a final finding.
- Under investigationOfficials say possible malicious activity is being investigated.
- Confirmed malicious activityThe provider or officials confirm malicious activity.
- Explicitly ruled outA provider report or completed investigation states it was not malicious.
Claim labels
- ConfirmedStated by the provider or agency in a final report or official record.
- PreliminaryStated officially before a final report. May change.
- ReportedReported by news outlets. Not confirmed by the provider or agency.
- UnverifiedA single or anonymous source. Not corroborated.
- DisputedSources give conflicting accounts.
- DeniedDenied or contradicted by the provider or agency.
Connections between incidents
- Confirmed linkA provider or agency says one incident caused or affected the other.
- Possible linkA documented shared dependency and overlapping times. Not confirmed.
- Unverified linkSomeone has claimed a link, with no evidence yet.
- Not connectedProviders or officials say the incidents are unrelated.
- Similar failure (comparison)The same kind of failure. No causal link.
Types of evidence
Every source is placed in one group, and each claim shows which groups support it. A claim is labeled confirmed only when an official source or a copy of an official record supports it. Our own interpretation is shown separately as Independent analysis.
- Primary: provider and official sources. Published by the provider or agency itself: incident reports, status records, advisories, statements.
- Copies of official records. Third-party copies of an official record that we cannot read directly, such as a provider's admin-only status page.
- Independent reporting. News organizations reporting statements, impact and their own findings.
- Reference and third-party analysis. Encyclopedias, aggregators and outside analysis or measurement. Used for context, not as the only support for a confirmed claim.
Sources
Provider reports and official statements come first. News reports are used for impact and for statements made to the press, and are labeled as reported. When sources disagree, the claim is labeled disputed and both are cited.
Each source shows its publication date where one is recorded, and the date it was last accessed. Each incident shows the date it was last verified against its sources. Links are checked automatically; sources that block automated checks are reviewed by hand.
A statement we cannot trace to a readable source is held for editorial review and not published, even if other outlets repeat it.
Automatic collection
Official status pages are checked every 15 minutes. When a provider reports a major incident, a page is created from its own updates: title, times, status and the updates themselves, each cited to the status page. Pages created this way are marked as such until reviewed.
The provider's statements are quoted word for word. Facts are extracted with an AI model only when the model can point to the exact passage that states them; anything it cannot quote is discarded. Explanations written by the model are labeled, use only those facts, and are rejected if they add numbers, causes or claims of an attack.
Causes, statements about attacks and failure sequences are never published automatically. They are held until reviewed. Collected updates are never merged into an existing page on timing alone: only a matching official identifier links them.
Updates
Pages for recent incidents are updated as new statements and reports are published. When a fact changes after publication, the page lists the change with its date rather than replacing it silently.
Unresolved incidents must be re-verified at least every two days. If the status has not been re-confirmed, the page says so and the incident is no longer counted as active.