The world this wiki

This project is an instantiation of Andrej Karpathy's LLM Wiki idea — rather than retrieve from raw documents on every query, an LLM incrementally builds and maintains a persistent, interlinked wiki, doing the cross-referencing and synthesis once and keeping it current as new sources arrive. Here the source is a year of The Economist. Each entry distils the enduring facts about a person, country, company, organization or topic — biographical background, structural details, lasting policy — and ignores the week-to-week churn of news. Every fact traces back to the article it came from.

It runs entirely from flat markdown files. Each article has a one-line summary, an Economist-style witty subhead (its "pun"), and a list of sources — the issue dates and article titles it draws on. Cross-references between entries are live: clicking "China" in any article takes you to the China page.

A few good places to start:

• Donald Trump — much-cited; a good test of how the wiki holds together under heavy linking
• China — one of the longest entries; politics, economy, AI, all the threads converge here
• Anthropic — company entries are usually tighter and more recent
• Artificial intelligence — for the topics ↔ companies ↔ people web
• Jeff Landry — small, focused entry; shows the typical shape

Or browse a whole section:

• People — 958 articles
• Countries — 149 articles
• Companies — 420 articles
• Organizations — 139 articles
• Topics — 413 articles

The pipeline behind it: a Python CLI fetches each weekly edition as an EPUB via Calibre's recipe, parses it into one markdown file per article, and then a Claude Code skill reads each section and updates the wiki with whatever enduring facts it finds — never adding anything not present in the source. A periodic lint pass rewrites stale summaries and adds the puns. The source code lives on GitHub.