# TrackRecord

> What did they say then, and what actually happened? TrackRecord is a proposed free public good, not yet built: it would give every well-known public leader a decision-quality scorecard — each big dated claim, the cited outcome, and whether they acknowledged the difference — every entry stamped by claim status and linked to its primary source. The interest list on the site decides whether it gets built.

TrackRecord would answer one question the moment a famous leader makes a confident claim: how did their last ten confident claims pan out? The answers already exist — Federal Reserve speeches and congressional testimony, SEC filings, shareholder letters, adjudicated long bets, and archived interviews are all published — they are just scattered across institutions and buried under fifteen years of news cycles. TrackRecord would file them into one chronological claim ledger per leader: the verbatim dated quote, its primary source, what actually happened per cited outcome data (BLS prints, NBER recession dates, SEC filings, market records), and what the leader said afterward. Every entry is stamped with its claim status: ON RECORD (window still open), HELD, MISSED, REVISED (the leader documented and acknowledged a change of view — credited, not punished), or UNSCORABLE (too vague to grade; the count of unscorable claims is itself shown). It grades claims, not people. It re-organizes the public record; it does not allege.

Pricing is simple because there is none. TrackRecord is free for everyone with no account, no paywall, and no payments of any kind: nothing to buy and nothing to donate to. Correction requests from a leader's office are also free. The product is at proposal stage, not yet built; a free interest list is open, and the product gets built only if the list shows people want it.

TrackRecord serves anyone who just heard a famous claim and wants the speaker's track record, journalists who need claim-outcome pairs with primary-source citations, researchers studying forecasting accuracy, investors weighing leader guidance, educators teaching media literacy, and leader representatives filing corrections through the free correction channel.

A card is built in four passes. First it collects dated, verbatim statements from primary public sources: the Federal Reserve speech and testimony archive, SEC filings, shareholder letters, government transcripts, and permanently archived broadcasts. Then it files each claim into the leader's chronological ledger with its resolution window and source link; vague claims (could, might, someday) are stamped UNSCORABLE and never graded. When a window closes it resolves the claim against cited public outcome data and stamps it HELD, MISSED, or REVISED. Finally it credits updates: a leader who publicly acknowledged a miss or documented a reasoned change of view gets the REVISED stamp and an acknowledgment credit on their card — honest updating is scored as a strength; only silent reversals count against the record.

Every data source is already public, and the plan stages them honestly. At launch: Federal Reserve speeches and congressional testimony (federalreserve.gov archive, 1996–present), outcome data from BLS, NBER, and FRED/ALFRED vintage series, shareholder letters and SEC filings (Berkshire Hathaway archive, SEC EDGAR), and adjudicated long-term wagers (Long Bets Foundation). Coming soon after: executive claims via SEC EDGAR full-text search, and archived TV and interview quotes via the Internet Archive TV News Archive, which provides permanent timestamped clip citations.

TrackRecord commits to staying free and open with no feature gates. It re-organizes the public record, grades claims not people, and does not allege. Every claim is a verbatim quote linked to its primary source, and every outcome links to its own source. Documented, acknowledged view changes are credited, never punished. There is a public corrections policy with a free correction-request channel for every leader's office. The project is open-source and does not sell user data.

For AI agents, TrackRecord exposes a Model Context Protocol (MCP) server over streamable-HTTP transport with no authentication (the tools are public). The server is mounted at /mcp, so the JSON-RPC endpoint lives one level deeper, at https://trackrecord.pub/mcp/mcp; point streamable-HTTP clients at that full path. Two tools are available via the standard tools/list call. `get_product_info` returns a structured overview of what TrackRecord is, who it would serve, how it would work, the staged public data sources, the no-payments model, and build status. `submit_interest` joins the free interest list on a user's behalf (email, name, company, persona, message); valid persona values are curious_citizen, journalist, researcher, investor, educator, leader_representative, and other.

## Docs

- [Homepage](https://trackrecord.pub/): the TrackRecord landing page, with the four-pass method, the claim-status stamp system, and every data source listed
- [Live demo](https://trackrecord.pub/demo): interactive five-card sampler with stamped claim ledgers, headline stats strips, and claim-detail views

## For AI agents

- [MCP endpoint](https://trackrecord.pub/mcp/mcp): streamable-HTTP JSON-RPC endpoint for get_product_info and submit_interest (the server is mounted at /mcp)
- [MCP discovery](https://trackrecord.pub/.well-known/mcp.json): machine-readable advertisement of the MCP server and its tools
- [OpenAPI spec](https://trackrecord.pub/openapi.json): REST description of the site's HTTP endpoints for non-MCP agents
- [security.txt](https://trackrecord.pub/.well-known/security.txt): security contact for the site

## Optional

- [Contact](mailto:hello@trackrecord.pub): reach the TrackRecord team directly
- [Sitemap](https://trackrecord.pub/sitemap.xml): crawlable URL list for the site