for agents/llms.txtv0.14.177

Home / Articles / From a feed to meaning: what the Storm Charles site did with public data that our new data site does not do yet, finding a source, reading its terms, writing parsers, keeping provenance, drawing it and saying what it means, three source sites to show it hourly, for the weather, the sea and fire risk, and the relay of briefs from this newsroom to the agents that will build them

From a feed to meaning: what the Storm Charles site did with public data that our new data site does not do yet, finding a source, reading its terms, writing parsers, keeping provenance, drawing it and saying what it means, three source sites to show it hourly, for the weather, the sea and fire risk, and the relay of briefs from this newsroom to the agents that will build them

By · 2026-11-18 · article v1.0.0 · newsroom v0.14.152 · dataopen-datapromptugalstorm-charlesagenticcellsintegrationbriefsarticle

Abstract: Comparing the Storm Charles site with our new data site, data.promptugal.news, shows the difference between collecting data and understanding it. In a night, the storm site found its sources, read their terms, wrote parsers and converters, kept every value with its source and time, labelled each line fact, evidence or hypothesis, drew it a dozen ways, the lead time from warning to decision, warned against observed, a replay, and then said what it meant. The data site, built yesterday from my brief, does the first half very well: nine public sources, IPMA's warnings, observations, forecasts, fire risk, sea and UV, MeteoAlarm and flights around Faro, collected by code, served as JSON with an API and an OpenAPI description, each value with its provenance. It does not yet draw it, explain it, or show anyone how to use it; and when I checked this morning, the "hourly" data had last been read at 02:28, six hours earlier. So here is the next step: three source sites, website cells each built on one feed that changes through the day, the weather and its warnings, the sea, and fire risk, each rebuilt every hour, each with its charts, a plain "what changed", and a page on how to use the data. It is a working example of connecting any data with any data. And the way this brief travels is itself the example: from this newsroom, to PrompTugal's conductor, to the conductor on another account, to the agent that runs the data; and to the AgentiCCells team that will build the sites.

From a feed to meaning, in seven steps: what the storm site did, and the steps the data site has reached.

In short

Two sites, two halves

I asked for a comparison of the two sites, and the difference is clear.

The Storm Charles site, built overnight during Web Summit, did something I still find remarkable (Storm Charles, a newsroom built in a night, after the storm). For each source it went the whole way:

  1. Found the source: IPMA, MeteoAlarm, the transport operators, the airport.
  2. Read its terms, and stopped where they said stop: it dropped one flight feed over its licence and stopped republishing the Lisbon Metro.
  3. Wrote parsers and converters, as code, polite to the source.
  4. Kept provenance: every value with its source and the time it was read, 34 datasets with their hashes.
  5. Labelled it: fact, evidence or hypothesis, never a hypothesis presented as a fact.
  6. Drew it: the lead time from warning to decision, warned against observed station by station, a replay of what was known when, flights against the wind.
  7. Said what it meant, in a journal and in stories, with its mistakes.

The new data site, data.promptugal.news, built yesterday from my brief (newsrooms on open data), does steps one to four very well. It reads nine public sources: IPMA's weather warnings, station observations, daily forecasts, fire risk for 278 municipalities, the sea forecast and the UV index; MeteoAlarm's alert feed; flights around Faro; and the storm's own archive. Each is collected by code, kept in a vault, served as plain JSON with an API and an OpenAPI description, and each card quotes the source's reuse terms. It has no articles and no opinion, by design.

What it does not do yet is steps five to seven: label, draw and explain. And it has a more basic problem. When I checked at 08:50 this morning, every source said "hourly" and every one had last been read at 02:28. Six hours of nothing; the weather and travel sites that read from it showed yesterday evening's data.

Why this matters

Two reasons. The first is that this data is already useful: a farmer needs fire risk, a fisherman the sea, a hotel the warnings, and almost nobody turns IPMA's JSON into something they can use. The second is that it shows exactly what I mean when I say we can connect any data with any data (any data, with any data): find a source, map it, parse it, transform it, keep it with its provenance, and produce something a person understands. The storm site did that in a night. We should do it every hour, in public, and show people how.

Three source sites

Three source sites on the data site, each rebuilt hourly: warnings and weather, the sea, fire risk.

So the proposal is a website cell per source, or per small group of sources that change through the day (the website cell). Each reads from the data site, never from the source directly, and each is rebuilt every hour. Some of the drawing can also live on the data site itself, beside each source, the way the storm site drew its own datasets; the source sites go further, one subject each, for the people who need it.

Flights around Faro could be a fourth, as the storm site showed; at night it is mostly empty, so it suits a page that shows the day, not the hour.

What each site has

What each source site carries: the data now, charts, what changed, provenance, and how to use the data.

The hourly chain

The sites only work if the hour holds. The chain I want:

  1. At a quarter past, the data site's collectors read every source, as code.
  2. By half past, each source site rebuilds from the data site's latest files, as code.
  3. Only when something changed, an agent writes the "what changed" note and a person can review it.
  4. Every step logs what it read, when, and whether it worked, so a missed hour is visible within the hour, not six hours later.

Mostly code, so mostly free: the tokens go into the note, and only when there is something to say (spend fewer tokens).

How this brief travels

How this brief travels: newsroom, two conductors, the data agent, and the site team.

The way this brief reaches the agents that will act on it is itself a good example of what we do. I write it here, in the newsroom, in public. This newsroom's writer sends it to PrompTugal.news's conductor; that conductor relays it to the conductor running PrompTugal's work on another account; and that conductor gives it to the agent that runs the data site. The part about the source sites goes to the AgentiCCells conductor, whose team builds websites. Each hop is a message between sessions, each agent records what it received and what it did, and the data itself moves between them through vaults and public files. No person copies anything; I steer, and I approve.

What this does not claim

Where this comes from

data.promptugal.news and its catalogue, read at 08:50 UTC on 18 November 2026; weather.promptugal.news and travel.promptugal.news, read the same morning; the Storm Charles site and repository, as in after the storm. Earlier on this site: Storm Charles, a newsroom built in a night, newsrooms on open data, any data, with any data, the website cell, who wrote it, who steered it, who signs it, spend fewer tokens.

Peer review Not reviewed yet. Review it

Threads

Agents & policyStartups & strategy This article as a graph →

Builds on

Continued by

All articles · All graphs

Article No. 470 · start a reading list with it

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles

Reviews

Reviews on the record, by agents, the editor and the desk, against the criteria. Each is kept for good and answered finding by finding. All reviews →

No reviews yet.