Method
Outage Sphere reads the public status page of each of 90 security, IT and AI vendors once a minute and repeats what the page says, with the time it was read.
At , 86 of 90 pages could be read.
One source for each vendor
A vendor's state on this site comes from one place: the vendor's own public status page, or the feed that page publishes. Nothing else can change it. Crowd reports from other sites are not used.
The reader is called OutageSphereBot. It says who it is in its User-Agent, with the contact research@example.com. It asks each vendor's host at most once a minute, never signs in, and never goes around a refusal. If a page refuses the reader or does not answer, the vendor is shown as unknown until the next read works. A page that shows its status only after running scripts is also unknown: the reader does not run them.
What the words mean
The wording quoted here is the wording of the vendor's status feed, as written (for example "Partially Degraded Service"). A few vendors word their page for people a little differently from their feed; the feed is what is quoted. Where a vendor's page for people does not open for the reader but its feed does, the page says the state was read from the feed. Beside the vendor's wording the site uses its own words so that vendors can be compared:
| Here | On a Statuspage-style page | Elsewhere |
|---|---|---|
| Operational | indicator "none" | Slack "ok"; Google, AWS and Azure with nothing open |
| Maintenance | indicator "maintenance" | |
| Degraded | indicator "minor" | Slack "incident"; Google "service disruption"; AWS "degradation"; any open Azure item |
| Partial outage | indicator "major" | Google "service outage"; AWS "disruption" |
| Major outage | indicator "critical" | Slack "outage" |
| Unknown | the page could not be read at the last try | |
| Status page requires sign-in | the page only opens for signed-in customers; no state is guessed | |
An incident's impact (minor, major, critical) and its state (investigating, identified, monitoring, resolved) are the vendor's own ratings. The site does not re-rate them. A notice the vendor posts with no stated impact is called a notice and is not counted as a problem. The answer word on a vendor's page ("No", "No, but", "Partly", "Yes") follows the vendor's top-level indicator, never the state of single components.
Some products are read from a larger page: GitHub Copilot is the part GitHub's page calls "Copilot", Gemini is the product of that name on the Google Workspace dashboard, and ChatGPT is read from OpenAI's page. Each of those pages says so.
The one word for the stack
The big word on the home screen comes from a fixed scale. It counts incidents that are open now and that the vendor rates minor or worse. It does not count maintenance, notices, or long-running incidents (below).
| Word | When |
|---|---|
| All clear | no vendor has a counted incident open |
| Minor issues | the worst counted incident is rated minor, at fewer than a quarter of the vendors |
| Degraded | one or two vendors have an incident rated major, or a quarter or more of the vendors have a minor one at the same time |
| Major outage | any incident is rated critical, or three or more vendors have a major one at the same time |
A vendor whose page could not be read counts as neither up nor down.
Long-running incidents
An incident open for more than 7 days is long-running. It stays on its vendor's page and is listed in its own short block on the home screen and on Today, but it does not set the word for the stack and is not lit on the sphere. The vendor's own state is unchanged by this: if its page still says "Partial System Outage", so does this site.
Regions on the sphere
A region is lit only when the vendor names it, in the incident's title or in the list of affected components (a city, a country, a cloud region such as me-south-1, or an area such as Europe). When a vendor names no region, nothing is lit for that incident; it still appears in the list and on the ribbons. A maintenance window is a single lamp at the place named. An incident resolved in the last 24 hours leaves a fading glow.
Outage pages
An incident can get a page of its own when the vendor rates it major or critical, or when it stays open for hours; the most recent and the largest come first, no more than 150 at launch and a few a day after that. Every other incident is a row in the vendor's history. An outage page carries the vendor's updates in order, the duration, the regions and components the vendor named, a cause only where the vendor wrote one, and the post-mortem when the vendor publishes it. The page's address is fixed the first time the incident is seen and does not change if the vendor edits the title.
Two clocks
Every page shows when it was last checked. Separately it records when a fact on it last changed (a state, an incident, an update). Only the second time is given to search engines as the page's date; a check that finds nothing new does not make a page look newer.
How long, and history
A duration on this site is how long the entry stayed open on the vendor's status page, from the start the vendor gave to the time it marked the entry resolved; it is not the impact window a vendor may describe in its own text. An entry a vendor posts after the fact has no such span and is shown as "not stated". The typical figure on a vendor's page is the median of those durations across the incidents the vendor resolved in the last 90 days, shown only when there are at least three. History comes from the incidents the vendor's page still lists; a busy vendor's page may reach back only a few weeks, and the days before that are drawn as "before the feed begins" rather than as quiet days. AWS and Azure publish only what is open now, so their pages here carry no history.
Chatter is a separate signal
Where the site shows how many public posts mention an incident, that number sits under its own label and only appears at ten or more posts in an hour. It never changes a vendor's state and never opens or closes an incident.
Freshness
The live file behind the home screen is rewritten every minute. The home screen says "Live" only when the file on screen was read in the last five minutes; otherwise it says "Last read" with that time.
Visitors
The site sets no cookies and runs no tracking script. Every file a page loads comes from the site itself. Its servers keep a log of requests: the address a request came from and any address a proxy passed along, the time, the page asked for with anything after a question mark, the page that linked to it, what the browser says about itself, and technical details of the answer such as its size. A log is deleted after 90 days.