Methodology
How we verify incidents
Automatic sources
Feeds are checked every 15 minutes. Provider and network incidents use a 14-day window; SEC filings, Canada OPC announcements and NCSC news use a 30-day window; news articles use a 7-day window. Editors can disable automatic publishing; those items wait for review. Manual edits and review decisions are preserved.
- Cloudflare, GitHub, Zoom and Atlassian: only incidents the provider rates major or critical are imported, and they publish automatically. Minor incidents are not tracked.
- Google Cloud: high-severity incidents or service outages publish automatically. Medium-severity disruptions wait for review; low-severity items are ignored. Location reports determine the region.
- Microsoft Azure and Microsoft 365: broad incident notices publish automatically. Availability notices are not incidents and are ignored.
- US SEC EDGAR: Form 8-K filings explicitly listing Item 1.05 publish automatically as data breaches and are marked Official. Amendments are attached to the existing company incident. Discovery dates are not inferred.
- Office of the Privacy Commissioner of Canada: relevant breach and privacy investigation announcements wait for editorial review.
- UK banks — Monzo, Starling Bank, Atom Bank, Kroo, Monese, Wise and Tide: only incidents the bank rates major or critical are imported, and they publish automatically as Banking & payments outages in the UK unless the incident names another place.
- UK National Cyber Security Centre (NCSC): news items mentioning incidents, attacks, breaches, ransomware, outages, compromises or disruption are checked by AI; general warnings are dropped, real incidents get an organisation name and a summary in our own words, and drafts are published only after editorial review.
- BBC News Technology, The Register (security) and Computer Weekly (IT security): articles about outages, cyberattacks, breaches, ransomware, hacks or data leaks from the last 7 days are imported. AI identifies the organisation, type and likely region and writes a one-sentence summary in our own words. When an existing incident for the same organisation is dated within two days, the article is attached as a corroborating source instead of creating a new item. News items are only published after editorial review.
- Cloudflare Radar: independently monitored internet disruptions in the UK, US, Canada and European coverage regions publish automatically. Reports link to Radar and are not provider statements.
Each provider incident ID is a separate event. Nearby reports only merge across different feeds or with hand-curated incidents. Statuspage incidents clearly limited to places outside our main regions remain drafts in Other. Outage Watch is focused on the UK, Europe, US and Canada, with major incidents elsewhere included. News drafts record the main affected country, and incidents elsewhere are kept with region Other for review.
Provider notices are linked directly and labelled Auto-imported. News-based items require editorial review before publication. Automatic publishing does not change the confidence rules below.
Confidence levels
Confidence is calculated by our database from the sources attached to each incident — not by hand. Rules are checked in order:
- OfficialA data breach backed by at least one regulatory filing or regulator action.
- High3 or more independent sources, including a provider statement or an official source.
- Medium2 or more independent sources, or 1 source plus a provider statement.
- LowAnything else.
Source types and reliability
Every incident lists the sources behind it. Here is what each kind of source means and how much it counts.
Regulatory filing
OfficialA legally required disclosure, e.g. an SEC Form 8-K or a state attorney-general breach notice. Confirms a breach on its own.
Regulator action
OfficialA regulator publicly investigating, fining, reprimanding or signing a compliance agreement (ICO, OPC, national DPAs).
Provider statement
StrongThe affected company or operator speaking for itself: status pages, post-incident reports, customer notices, official spokespeople.
Network telemetry
StrongMeasurement from internet monitoring firms (e.g. ThousandEyes). Shows that something broke, rarely why.
News
SupportingOriginal reporting by a news organisation. Wire copies republished elsewhere count once.
Trade press
SupportingSpecialist industry publications covering a sector in depth.
Analysis
SupportingBlogs and research write-ups. Often relay a provider report — we record that as the origin.
Reference
ContextEncyclopaedic or background pages. Useful context, not confirmation.
Crowd reports
ContextUser-submitted outage reports. Signal volume, not cause.
Independence rule
We count distinct original sources. When an article relays another outlet’s story (for example a Reuters wire copy republished by several sites), we record the original as its origin and count it once. A source with no origin is treated as original reporting.
Date tags
- Started — when an outage began.
- Discovered — when the organisation says it found a breach.
- Disclosed — when the breach was first made public by the organisation.
- Regulator action — when a regulator publicly acted (investigation, letter, fine).
Dates carry a precision (day, month or year). If a date has not been published we show Not stated — we never guess. Discovery-to-disclosure time is shown only when both dates are known to the day.
Records affected
An undisclosed record count is shown as “Not disclosed”, never as zero.
Summaries and links
Summaries are written in our own words and kept short. Every incident links to the original reporting and filings so you can check them yourself.
Corrections policy
Anyone can report an error by emailing prodfanatics@gmail.com. Editors review every report; confirmed errors are fixed and the incident’s updated date changes. Incidents found to be wrong are marked retracted rather than silently deleted.