For researchers

Fork the paper

This paper publishes to a public git repository. Not an export — the actual working archive, one commit per Sunday, cloneable in a few seconds. There is an endpoint for AI agents at /mcp, but it only reads these same committed files: the repository is the source, and it stays readable if the endpoint is not.

If you teach a data journalism class, study how news travels, or want to test a summariser against something with receipts: take it. That is what it is for.

The whole archive

git clone https://github.com/theweekbrewed/archive.git

Browse it on GitHub ↗

What is in it

content/issues/: Every published issue as JSON, exactly as it was sent: headline, ten sections, every story with its source, and any corrections appended after the fact.
content/provenance/: What we actually read before writing. Per source: the URL, the time we fetched it, and a SHA-256 of the text we checked our sentences against — hashes, never the text itself.
content/ledger/watch-items.json: The follow-through ledger: every 'what to watch' promise, when it opened, and — once answered — what happened, including the times we were wrong.
content/bundles/: The same records joined for machines: one file per issue linking each story to its source and whether we hashed it, plus an index.json listing every issue with its lock and correction times. This is what the agent endpoint serves.
content/crosswords/: The weekly mini, grid and clues, generated deterministically.

The licence, exactly

Ours is yours. Our editorial text, summaries, headlines, structure, metadata and datasets are licensed CC BY 4.0. Use it commercially, remix it, build on it. Credit The Week, Brewed and link back.

Theirs is not ours to give. Quoted excerpts, headlines and links from the news organisations we cite remain the property of their publishers. We quote them briefly, with attribution and a link, and we cannot license them onward to you. If you republish an issue, that boundary travels with it.

Things we would read

Measure how long a story takes to travel from a niche source to the mainstream — the archive dates both ends.
Check a summariser against the provenance hashes: did the sentence survive contact with the document it cites?
Study what a curation product refuses. The 'what we didn't run' item names the loudest story we skipped each week, and why.
Score us. The ledger is a public prediction record with outcomes attached; nothing stops you computing the number we decline to publish.

If you build something with it, we would genuinely like to know — hello@theweekbrewed.com.

What the records prove The follow-through record Read the archive instead