Checking the National Weather Service…
A sensor network you join from your phone. Report what you see — a rain gauge, hail on the deck, a creek level, a corn stage, smoke — in a short common language. Neighbors and official stations check every reading; corroborated readings become events, raise your trust, and roll into a daily digest.
The network lives inside a free messaging app. No typing, nothing to pay, no account to create here.
Telegram is a free messaging app, like WhatsApp or iMessage. Install it and sign up with your phone number.
Already have Telegram? Go to step 2.
Press Open the chat above, then the Start button at the bottom of the chat. That's you joined.
The chat is @intelligence_network_bot, a program that answers straight away, day or night.
Tap 📍 Share my location, then tap what you'd like: warnings for your area, a morning brief, or both. Done.
On a computer, tap 🗺 Pick my county instead. To report rain, hail or flooding later, tap its button in the chat. Others see a handle like s-4f2a1 and your ZIP, never your name or number.
For other topics or places: answer three questions, tap Open in Telegram, then Start. You're joined and subscribed in one go.
/subscribe weather.warnings cookOpen in Telegram
Later on, send /subs to see what you signed up for, or /unsubscribe all to stop everything. You can change your mind at any time.
One reading per clause. Metrics, units and named sizes come from the topic packs — the same files the network runs on.
<metric> <value>[unit] [@<location>] [at <time> | <N>m ago] [#tag] [-- note]
Who reports what, where it corroborates, and how the fortnight moved.
County lines from the Census Bureau. Darker = more human readings; dots = official stations reporting.
score = severity × corroboration × trust × official agreement. Only verified events (two sensors, an official source, or a trusted sensor) are pushed.
| Score | Topic | Event | Peak | Sensors | Official | Verified | Updated |
|---|
Every morning the network writes a Google Doc: events ranked by score, the official alert recap, station extremes, a triaged reading list, and the network's own vitals.
Three ways — pick whichever suits you:
/subscribe digest. A link arrives every morning at 8am./digest you@example.com and Google shares the folder with that address.Every day also gets its own folder in Drive holding the same digest as a PDF, a Word file and an HTML page, plus CSV tables of the events, official alerts, county activity, station extremes, contributors and the reading list.
The same document that lands in Drive every morning — readable here, no account needed.
The network is the people in it. Trust is earned by agreement, not granted.
Coordinate a spotting session, ask what a reading means, compare notes after a storm.
Have a rain-gauge log, a soil test series, a stream survey, a PurpleAir sensor? Automated sensors post JSON to the bot; datasets get a home in the catalog.
Every reading is compared with readings of the same metric nearby, inside the metric's distance and time window: other people, weather stations, NWS storm reports and warnings, and NOAA's radar estimates of rain and hail (MRMS). Agreement corroborates both readings. A sensor's trust, kept separately for each topic, is
trust = (K·prior + agreements) / (K + agreements + disagreements) K = 4 · prior = 0.5 · trusted ≥ 0.8 · half-life 180 days
Old outcomes fade. Radar counts half as much as an official report. Two people under one roof, or two who always agree, count as one witness. New sensors show "new" instead of a number that looks exact; after that, the board shades an 80% range. Nobody starts trusted; nobody is condemned by one odd reading.
| Sensor | County | Readings | Corroborated | Trust |
|---|
Sensors appear by handle only. The public snapshot never carries names, ZIP+4s or coordinates of people.
What the network already ingests, what you can pull yourself, and the papers behind the idea.
The page is built from fourteen JSON files that anyone may fetch and reuse: data/network.json, map.json (the day's notable readings and the news tied to them), counties.json, events.json, alerts.json, storms.json, reports.json, gauges.json, activity.json, graph.json, leaderboard.json, digests.json, topics.json, sources.json. Same anonymisation as the page.
For other systems: the week's verified events as GeoJSON and as CAP 1.2 in Atom (the format emergency managers' tools read), and storms as GeoJSON. For research: explore the database in your browser with Datasette Lite (people's readings, storm reports, events, storms and alerts for 90 days; people as handles with ZIP and county only), or download network.sqlite.