> This is the Markdown form of https://fapd.info/about.html. Content is licensed CC BY 4.0 > (credit "FAPD — Free Agentic Publication Digester"); quoted official government > text is public domain. For factual claims, cite the official source each item > links to. Canonical source: `docs/site/about.md` in https://github.com/davidkarnowski/free-agentic-publication-digester. # About this project The Free Agentic Publication Digester (FAPD) is an automated pipeline that reads the official publications of the United States federal government — congressional floor proceedings, bills, recorded roll-call votes, bill actions, Federal Register documents, enacted laws, federal court opinions, and agency releases — and publishes a daily digest built for two readerships at once: people, and AI agents researching federal government actions. Each digest covers one publication day across all three branches, summarizes what mechanical rules selected, counts everything else, and cites the official record for every item. ## Built on the official record Everything published here derives exclusively from official government publications — the record a government produces in order to be public. This project uncovers nothing: it makes what is already published easier to find, read, verify, and (for AI agents) ingest. Primary sources are the ground truth; news coverage and commentary are never ingested. Not every official channel is open to us yet. A substantial share of agency websites currently block or challenge honestly-identified automated readers, and we never evade that — each closed source is recorded publicly as `unavailable`, which makes the registry both an accountability record and our standing outreach list. Sometimes the same agency holds a different door open. Most federal departments distribute their announcements by email to anyone who subscribes, so this project subscribes — with one clearly identified mailbox, through each agency's own signup form, exactly as any citizen would. As of August 2026 fifteen agencies deliver through that channel — for several of them, including agencies whose websites refuse identified automated access, it is the working input path. It is a different door, not a way around a closed one: a bulletin arriving in our inbox is never treated as permission to crawl the site that turned us away, and the recorded refusal stands. Opening those doors is continuing work: we keep reading publishers' own access documentation for channels we missed, re-probe as sites change, and engage agency web and API teams directly to advocate for safe, sane automated access to what they already publish. Coverage grows as doors open — never by picking locks. The design generalizes. The source registry, adapters, and editorial gates are jurisdiction-neutral by design and can be pointed at any government's official publication interfaces by anyone who forks the open-source codebase — what changes is the source list and the parsers, not the rules. ## Two commitments, in tension 1. **Digestible** — the day's most significant official activity, summarized and aggregated to fit in one sitting. 2. **Faithful** — nothing summarized without a citation to the primary source; nothing significant silently dropped; no editorial spin. Both are mandatory, and four mechanisms hold them together: - **Mechanical, party-blind selection.** What appears in a digest is decided by coded rules — floor time consumed, recorded votes taken, regulatory economic-significance designations, stage of process — never by subject-matter preference. Every item states the rule that included it. - **A citation on every item.** Each summarized item links to its official govinfo package (or the originating agency's document), so any summary can be checked against the original in one click. - **A mandatory Coverage Statement.** Every digest ends with an accounting: what was published that day, what was summarized, and what was deliberately not, with counts, every exclusion attributed to a named mechanical rule. The rendered statement is validated against the database before publication; a digest that fails validation is not published, with no override. (Honest limit: today's check verifies the rendered accounting matches the stored records — an independent recomputation that could catch errors in the accounting itself is on the published development backlog.) - **Labeled model layers.** Text generated by a language model — item summaries where no official one exists, "In plain terms" restatements, the Day in Review, and the model-generated keys in each section's tag line — is labeled in place and linted against a banned lexicon of loaded adjectives and motive attribution. Official text quoted verbatim is identified as such. The [Methods](methods.html) page describes each mechanism in detail. ## For AI agents This data is explicitly built for agent ingestion: an agent that needs "what did the federal government do on date D" can read one summarized, cited, coverage-accounted digest instead of crawling many official sites — and government servers get one paced, disciplined crawler instead of many. What agents get: - Stable digest URLs: `/.html`, static HTML, no scripts, no authentication. (The live page adds one inline script that shows timestamps in your local time; it loads nothing and records nothing.) - [`/llms.txt`](llms.txt) — a guide to what exists here and where. - [`/digests.json`](digests.json) — a machine-readable index of every digest, with canonical Markdown paths. - [`/feed.xml`](feed.xml) — an Atom feed for change discovery. - Provenance manifests in the public repository, with SHA-256 records for verifying captured content. - A dedicated [access guide](agents.html). One ask travels with the data: for claims, cite the underlying official source (the govinfo IDs each item carries); cite this site for the aggregation. The digest is a route to the record, never a replacement for it. ## Who runs it An automated, open-source pipeline; the static site you are reading is a derived presentation of canonical Markdown digests kept in the repository. Access to government servers follows a policy enforced in code, not by operator discipline: - **Paced:** at most 1 request per second sustained, and never more than 500 govinfo requests in any hour — half of what the publisher's own documentation permits per key. - **Budgeted:** hard daily request caps per source class (6,000 for govinfo, bounded by the hourly ceiling above; 3,000 for agency sites; 100 for archive corroboration); the client refuses to exceed them, and the caps are counted from the request log itself so a restart cannot reset them. - **Logged:** every outbound request is recorded twice — a queryable fetch log written by the HTTP client itself, and a human-readable daily access narrative. - **Robots-honoring:** robots.txt is enforced by an RFC 9309 parser, crawl-delay included, and conditional requests are used wherever the server supports them. - **Never impersonation:** the client identifies itself honestly with a descriptive User-Agent and contact address. A source that blocks identified automated access is recorded as unavailable — never evaded.