# COVID-19 Records > A searchable, citable dataset built from 14 public document > packages released by the U.S. Senate Committee on Homeland Security and > Governmental Affairs. 5,247 pages were OCRed and segmented into > 18,094 individual communications, each with a speaker, an > audience, a date and a page citation. This site adds no source documents. Every document is public and originates from . No PDFs are hosted here; citations link to the Committee's own copy, deep-linked to the cited page. ## Structured data (OKF v0.2) The analysis is published as an Open Knowledge Format bundle — markdown with YAML frontmatter, no tooling required. - [Bundle root](https://randpaulcovid.org/okf/index.md): entry point and collection listing - [People](https://randpaulcovid.org/okf/people/index.md): activity, correspondents, statement timelines - [Topics](https://randpaulcovid.org/okf/topics/index.md): principal voices per subject - [Sources](https://randpaulcovid.org/okf/sources/index.md): the released packages, with SHA-256 - [Outlets](https://randpaulcovid.org/okf/outlets/index.md): media coverage and press direction - [Glossary](https://randpaulcovid.org/okf/glossary.md): 133 acronyms and strain names used in the records ## JSON API - [Discovery](https://randpaulcovid.org/api): every endpoint, with parameters - [OpenAPI](https://randpaulcovid.org/api/openapi.json): machine-readable description - [Guidance](https://randpaulcovid.org/api/guidance): field semantics in full, as JSON ## Agent protocols - [A2A card](https://randpaulcovid.org/.well-known/agent-card.json) — `POST https://randpaulcovid.org/a2a`, JSON-RPC 2.0: `SendMessage`, `GetTask`, `ListTasks` - [MCP card](https://randpaulcovid.org/.well-known/mcp) — `POST https://randpaulcovid.org/mcp`, Streamable HTTP, tools `search_records` and `get_record` Both return cited records. Retrieval only: no language model runs here and no prose is generated, because a summary of these records would carry none of the citations that make them checkable. An MCP client already has a model; the useful division of labour is that this end returns records and the model reasons over them. Search accepts natural questions. `mode=hybrid` merges keyword and semantic retrieval and is the right default for a question; `mode=keyword` is better for exact identifiers such as Bates numbers or grant IDs. ## What the fields mean `grade`, `citation`, chat `audience`, `topics` and the limits of coverage all carry specific meanings that a reader taking the numbers at face value will get wrong. They are defined once, in machine-readable form, at https://randpaulcovid.org/api/guidance — not restated here, because an agent that reads this file, the bundle root and that endpoint would otherwise meet the same paragraph three times before reaching a record. This is an independent technical demonstration, unaffiliated with the Committee and with anyone named in these records: https://randpaulcovid.org/about