The wiki is an open project: the transcripts, the extraction pipeline, the articles, and the site all live in one public repo — github.com/frankie-eight-days/amp-hour-wiki. Anyone can improve it; every change lands by pull request and gets human review.

The one rule: every claim traces to a verbatim transcript quote. CI byte-compares every quote against the transcripts, so a PR either has real evidence or it doesn’t build.

Three ways in

Fix an article

easiest — just markdown
Awkward phrasing, a section that repeats itself, a claim that misreads its own citation. Edit the file in articles/wiki/, keep the citation markers intact, open a PR. The lint runs automatically.

Extract a new article

the real work — ~200 concepts left
Pick an unwritten concept, read its evidence bundle, and produce a packet of claims with verbatim quotes. The extraction spec is the full procedure; a packet-only PR is welcome even without the written article.

Hack on the site

quartz 5 + python tooling
The static site, the graph explorer, the landing page, infoboxes — all in the repo under site/ and tools/. Include a screenshot with site PRs.

Spotted an error?

You don’t need a PR. Highlight the offending text on any article and hit the “Report” button that appears — it opens a prefilled GitHub issue with the article, the exact text, and the page link already filled in. This is especially useful when an extraction misunderstood what a speaker meant: the quote is real but the claim built on it is wrong. Those are the hardest errors for the pipeline to catch itself and the most valuable to report.

No GitHub account? Open one — it takes a minute and the issue queue is the project’s memory. All reports are public and you can watch yours get fixed.

What review looks like

CI runs the same checks the article factory uses: verify_packet.py byte-compares quotes, lint.py checks that every paragraph is cited, every citation resolves, and nothing editorializes beyond the evidence. Green CI means review is about substance only. Full details in CONTRIBUTING.md.

Curious how the pipeline works end-to-end? Read How this wiki was built.