Public Memory

A searchable record of what public figures say over time, with every claim tied to dated, sourced quotes and a review status.

Date
July 2026
Built with
Next.js, TypeScript, Tailwind CSS, Playwright, Vitest
Public Memory homepage with a large search field over the public record, corpus stats, recently reviewed sources, and a people panel.

Public Memory keeps track of what public figures say, and ties every entry to a source. It draws on interviews, speeches, podcasts, essays, posts, and testimony, and records what influential people say again and again, how their positions change, and where a later statement seems to conflict with an earlier one.

Everything in it is already public. It just isn't collected anywhere you can check it. Public Memory doesn't claim to know what anyone really believes. It shows the patterns in what people say in public, and each pattern comes with the quotes behind it: when and where each one was said, a confidence label, and any caveats.

Public Memory homepage with a large search box, counts of reviewed sources and receipts, recently reviewed sources, and the people being tracked.
The homepage: one search box over the whole record, under the tagline “See what changed. Check the receipts.”

One position, followed over years

Everything is organized around claim records. A claim record is one position a person keeps coming back to, followed across years of appearances. At the top are a short summary, the person's stance, and a confidence rating. Below that is a timeline of every appearance, and next to it the receipts. A receipt is a direct quote labeled with its source, its date, and a confidence level.

Claim record page for Sam Altman on frontier AI regulation, with stance, confidence, appearance timeline, and quoted receipts.
Sam Altman on regulating frontier AI: 19 receipts across three years, each quote linked to its source.

A page for each person

Each person gets a page built only from reviewed sources. It shows their most recent statements, which topics the record covers and how much, the phrases they repeat, and tabs for their claims, contradictions, and sources. The first version covers AI and tech leaders, including Sam Altman, Dario Amodei, and Yann LeCun. Material on a wider group has been collected and is waiting on review.

Sam Altman's page with recent quotes, a chart of topic coverage, and repeated phrases, drawn from 30 reviewed sources.
Sam Altman's page: recent quotes in date order, and on the right, how many receipts the record holds on each topic.

Why contradictions wait for review

Contradictions may be the obvious draw, but I care more about what people say consistently than about catching them out. Nothing is published as a direct contradiction until both statements are confirmed to be on the same topic, both quotes are checked word for word, the context has been reviewed, and a person has signed off. Until something passes, the contradictions page shows only the checks.

Contradictions page showing no published direct contradictions yet, the required review checks, and a tab for each person.
The contradictions page, with nothing published yet. A contradiction only appears here after it passes skeptic review and the checks listed on the right.

How sources get checked

AI agents do the research behind the site. They bring in sources, prepare transcripts, confirm who is speaking, pull out claims, and put those claims through a skeptic review. A separate, guarded step moves material into the published record, and research that hasn't been through it never counts as part of the record. Before anything reaches the site, it goes through validation checks, rules that set aside anything doubtful, and a comparison against a hand-labeled test set. The reviewed record holds 68 sources and roughly 400 receipts so far, with hundreds more transcripts waiting behind it.

Source library listing reviewed interviews, essays, and posts with receipt and claim counts for each, overall totals, and a note on what stays unpublished until reviewed.
The source library. Every claim on the site traces back to one of these original, reviewed sources.

Why I built it

I built Public Memory because I want AI-assisted research to work like an archive, not add to the stream of takes. I kept the design plain and dense.

Next on Systems

All systems
  1. 2026Unsupervised BrainrotA 3D show made by AI agents from commentsIn production
  2. 2026CouncilWhere AI models agree and disagree on disputed questions
  3. 2026Universe IndexAlternate histories from one changed fact