> This is the Markdown form of https://fapd.info/faq.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/faq.md` in https://github.com/davidkarnowski/free-agentic-publication-digester. # Frequently Asked Questions *How the Free Agentic Publication Digester keeps time, how its collectors behave, and how to read what it publishes. This page grows as questions do; the [methods page](methods.html) carries the fuller technical story.* ## FAPD's three clocks Every document in this record has up to three dates, and they are not the same thing. We name them precisely because confusing them is how aggregation quietly misleads. - **Date of Action** — the day something happened, as described in the document's own text: the day the Senate debated, the day a judge issued an opinion, the day printed on the Federal Register's cover. This date may not be available to us directly; it depends on the metadata and text the source provides. - **Date of Publication** — the day the publisher made the document available, as stated in the publisher's own metadata. Also not always available, and not always reliable: publishing systems post late, backfill, and revise. - **Date of Observation** — the day our own collector first saw the document. This is the one timestamp we can define precisely, from our own worker records, for every document without exception. **Observation is our source of truth for filing.** A daily digest carries what our collectors observed that day (Eastern time, the publishers' own clock), and each item states its own document date beside it when the two differ. One exception: the Federal Register files under its cover date, because that is the date on which it is legally published — and it is the case where all three clocks agree by design. **Which clock the graphs are on.** The stamps we store are UTC, and they stay UTC in the page markup and the machine files. Everything we *show* on a time axis — the live page's hour headings, the source cards' 7-day activity graph and its hourly segments, the per-source day-by-day charts, the health windows — is placed in publication-clock days and hours (Eastern time, Washington, D.C.), so a bar labeled with a date covers the same day the digest for that date covers, and each of those surfaces says which clock it shows. On the two nights a year the clock shifts, one wall-clock hour holds two hours of activity or none; the graph states that day's hour count (23 or 25) rather than hiding it. The clock itself is set in one place in the code (`FAPD_PUBLICATION_TZ`), so a copy of this project run for another government can keep another clock the same way. Why observation rather than the publisher's stamp? Because the publisher's stamp can lie about availability. If a publishing system goes down and a document stamped Monday only becomes reachable Wednesday, filing by the stamp would assign it to a digest that was already frozen — and it would vanish from the record. Filing by observation, nothing we see can ever miss its digest. Congress's official Record, for example, is typically published the morning *after* the day it covers: you will find it in the digest for the day it became available, labeled with the day of proceedings it describes. ## How do FAPD's collectors actually work? At the simplest level: our collectors are polite readers. Each one visits an official government publishing site on a fixed schedule, the way you might check a library's new-arrivals shelf. It asks "what's new since I last looked?", writes down exactly what it saw and exactly when it saw it, and saves a copy of the official text. It never guesses, never rewrites what it found, and never knocks faster than a site says visitors may knock. If a site publishes rules for robots, we follow them exactly — including one site that asks for seven minutes between visits, and gets it. Everything else on this site — the digests, the summaries, the statistics — is built from those notebooks. ## How often do you look? (observation windows) Each collector runs on its own clock, so "observed on day D" has a known precision per source type. If a document appeared on a publisher's site at moment T, we typically observed it within one cycle of T: | Source type | Check cycle | Observation window | |---|---|---| | govinfo collections (Congressional Record, bills, court opinions, laws) | every ~30 minutes | within ~30 minutes of govinfo posting it | | Agency newsrooms and press feeds | every ~60 minutes per site | within ~1 hour | | Email bulletins (agency subscription lists) | every ~15 minutes | within ~15 minutes of delivery | | gao.gov | feed checks honoring its requested 420-second spacing | within ~1 hour | | The live "today" page | rebuilt within ~5 minutes of new observations | — | | Source health on the [sources page](sources.html) | refreshed every ~15 minutes | — | The daily digest freezes at midnight Eastern — Washington's own clock — and is then the canonical record for that day. Collectors keep working around the clock; anything they observe after midnight belongs to the next day's digest. ## Does a slow source mean missing coverage? No — it means labeled timing. A court that posts an opinion four days after issuing it appears in the digest for the day we observed it, carrying its own issue date in the listing. The Coverage Statement in every digest accounts for everything observed that day, and anything deliberately not summarized is counted under a named mechanical rule. Silent omission is the failure mode this project is built to prevent.