NLT143 Research · Downwatch

How Downwatch is made

This page lists the sources behind the site, how the data moves, and how the map and globe place each point. Counts come from the files themselves.

Data dates are filled from the files.

Where the numbers come from

Every chart and map on this site reads public files or public feeds. We do not have a hidden database.

File What it provides Count As of
Loading…

Live status feeds

We keep a catalog of public status pages and feeds, checked from this project.

The live map reads a shorter working set through /api/status. That proxy caches results for about 60 seconds. The map itself checks again every 3 minutes.

Country outage signals come from IODA at Georgia Tech, through /api/ioda. That proxy asks for the last 24 hours and caches the result for about 2 minutes. Those signals are labeled REPORTED. A signal is not a confirmed outage.

Cloudflare Radar has a public outage center. Its live outage API needs a private token, so this site links out instead of calling it.

Firm Feed HTTP at check Checked Note
Loading…

Link check

Source URLs in the datasets were fetched from this project and recorded in link_audit.csv.

Report lists

The Report page compares early-access firms with a comparison group.

How the data moves through the site

  1. Public status pages and AWS Health A Cloudflare worker reads them at /api/status, caches the result for about 60 seconds, and the live map and infrastructure page use it. Data center status uses /api/infra. Hub operator feeds use /api/hubs. Both of those cache for about 2 minutes.
  2. IODA country signals The same worker reads IODA at /api/ioda, caches the result for about 2 minutes, and the live map draws country rings.
  3. The live wire Public posts on X are searched, then each hit is checked against official statements or credible press. Accepted items are written to wire.json with a CONFIRMED, REPORTED, or CLAIM label.
  4. Most reported A catalog in services.json is scored from official feeds at /api/reported and from public X post counts in crowd.json. The ranked list is on Reported. Crowd counts are CLAIM. They are never a confirmation by themselves.
  5. Sourced history files Outages, hacks, breaches, and cascade links live in JSON files on the site. The Report, History, and Connections pages all read those files.
  6. Publish The site is static files plus API routes, served on Cloudflare. Custom domains point at that same site. This page does not name internal project IDs.

The live wire scanner

The wire file records how each scan was done. Counts and run times are filled from wire.json when this page loads.

New items are meant to go through a small checker script so duplicates and empty sources cannot sneak in. If a run is still in progress, this page shows the latest file we can read.

API health

We fetch every local API route on both live domains, every URL in status_feeds.csv, and the data JSON files the site reads. Results are recorded in api_health.csv.

The table fills from the latest check file.

Name Kind HTTP Latency Freshness Schema Note
Loading…

Pages and the files they read

How the map and globe place a point

Libraries and base map

Cables

Submarine cables and landing points come from TeleGeography's Submarine Cable Map, used under CC BY-SA 4.0, and were fetched on Nov 17, 2026.

Open the TeleGeography Submarine Cable Map.

Firm dots on the live map

Events on the history map and globe

Each event in history.json has a latitude, a longitude, and a placement note that says why it sits there.

Cloud regions and points of presence

Locations come from each provider's own public list. Airport codes are placed with OurAirports coordinates (public domain). City names are placed with GeoNames cities15000 (CC BY 4.0).

Provider Sites on our map Location list
Loading…

Limits of the maps

What the labels mean

The Report page also has a strict and a loose rule for what counts as a major 2026 incident. Those rules are explained on the Report page.

How we grade a consumer service

The Reported list uses three public levels: No problems, Possible problems, and Problems. The rule is meant to be stricter than a raw complaint count.

Crowd counts

Each service has a public X search such as "Steam down" or "Steam outage", excluding retweets. We count posts by the hour for the last 24 hours, and by the day for the last seven days. The last two hours are compared with that service's own typical hour over the prior six days. A two-hour spike needs at least eight posts, at least three times the typical two-hour baseline, and at least ten posts above that baseline. A day-level spike, which also stays a CLAIM, needs at least 20 posts in 24 hours, at least three times the median prior day, and at least 25 posts above that median.

This is a CLAIM signal. Queries can overcount jokes, ads, and short or ambiguous names such as X, EA, Teams, Google, or A1. The count is not a unique-user report, and it is not a Downdetector figure.

Downdetector

We do not scrape Downdetector report counts. Those figures are proprietary. Each service card links out to the matching public Downdetector page so you can read their view. A 403 from an automated fetch is treated as a bot block, not as proof that a page is missing.

Country filter

The country selector shows services we catalog for that country, plus global services. The connectivity signal for a selected country comes from IODA, the Internet Outage Detection and Analysis project at Georgia Tech. An IODA signal is labeled REPORTED. It is not a confirmation, and it is not a grade for one consumer app.

What this does not claim

Plain English

Action steps to review