Sources, excerpts and coverage

Every fact, quote and briefing in the platform points back to a source document: a page, PDF or file that was fetched or uploaded, stored exactly as received, and kept with a record of where it came from. This page explains what that record holds, how to upload your own documents, how excerpts work, and what the coverage banner tells you.

Two kinds of source

• Public records are fetched by connectors. A connector is a small program that reads one official source (a county's notices page, a city's agenda list) on a schedule. Every organization can read public records.
• Uploads are files your organization adds. They are private to your organization, whatever their terms, and no other organization can see that they exist.

Open Sources to browse both. The list follows the location selected in the header: with a location selected, it shows documents about that place and the larger places it belongs to (for example its county, state and country), plus documents not tied to any place. You can filter by visibility (public or your uploads), by kind (meetings, agendas, public notices, …), by connector, and by place, and search titles, publishers and file names. Superseded versions are hidden unless you ask for them. Filters stay in the page's address, so a filtered list can be shared as a link.

Provenance

Each document's page shows its provenance panel:

The stored bytes never change. When a connector finds that a source has changed, it stores a new version and marks the old one superseded; both stay readable, so anything built on the old version still shows what it was built on. Download original returns the exact bytes, always as a download, never rendered in the browser.

Terms labels

The terms label decides what published work may do with a document's text.

A connector's label comes from a person's review of the source's published terms; a connector cannot fetch on a schedule until that review is recorded. An upload's label is declared by the uploader, who knows where the file came from. Declaring an upload "public domain" does not make it public: uploads always stay private to your organization.

Uploading a document

Researchers, editors, project leads and finance stewards can upload (owners can too). Open Sources → Upload a document, then:

1. Choose a file: PDF, HTML, plain text, CSV, JSON or XML, up to 25 MB. A scanned PDF without a text layer is stored, but has no text to excerpt until OCR arrives.
2. Choose its terms. This is required.
3. Optionally give a title (otherwise the title found in the file is used), its kind, the date it was published, the place it is about (the places of the selected location are offered), and where it came from, for example a URL or "handed out at the October hearing".

The platform extracts the text, records the provenance above and adds an entry to the audit log. If your organization already uploaded the exact same bytes, nothing new is stored: you are taken to the existing document and told so.

If an upload is refused, the page says why: the file type cannot be read, the text cannot be extracted, the place does not exist, or the file is over the size limit.

Excerpts

An excerpt (a span) is an exact passage of a document's extracted text, identified by its character range and page. Excerpts are what claims cite: AI proposals cite them, and published facts that rest on a document's wording must point to one.

• To save one, select text on the document's page and choose Save excerpt. Researchers, editors and project leads can save excerpts; everyone can read them.
• The platform checks the selection against the stored text, so an excerpt is always verbatim. Saving the same range twice returns the existing excerpt.
• An excerpt is at most 5,000 characters. Leading and trailing spaces in a selection are dropped.
• Excerpts you save are private to your organization. Public excerpts on public records come from the platform's own pipelines.
• Each excerpt has its own link. It opens the document at the excerpt, with the passage highlighted.

Coverage

"Nothing found" means something only if someone was looking. The coverage banner above lists of sources (and, as other features adopt it, above meeting lists, briefings and alerts) says how well the selected places are monitored, per place and per kind of source:

The banner starts with the overall status, for example "Coverage: Partly monitored.", then says which kinds of source are watched and which are not. Open Coverage by place and source kind for the detail. A place and kind with no record at all always reads as Not monitored: the platform never assumes coverage it has not verified.

Which places are covered

Today the platform monitors public sources at three levels:

• Federal, for every US location: proposed rules and their comment deadlines, and advisory-committee meeting notices (Federal Register), and rules open for comment on Regulations.gov.
• Washington State, for every Washington location: the Legislature's committees, hearings and bills, legislators (Open States), and the Governor's boards and appointment openings.
• Kitsap County: the Board of Commissioners' meetings and agendas, the county's advisory groups and their vacancies, and the Planning Commission's agendas and hearing notices.

A location's coverage includes its nation and state, so a profile anywhere in Washington gets federal and state coverage even though only Kitsap's county sources are monitored.

City and county councils elsewhere come in coverage packs. A coverage pack is the set of connectors for one place, built for the agenda website that place uses (Legistar, CivicClerk, CivicPlus AgendaCenter, AgendaSuite, and more). Packs exist for 25 US cities and counties, among them Seattle, King County, Bremerton, Port Orchard and Bainbridge Island, and for the UK Parliament's committees and the Irish Oireachtas's bills. A pack counts as coverage only after a person has reviewed every source's terms, a 7-day run has passed and someone has activated it. Until then its place reads Not monitored yet, with the reason "we are still checking a source for this place before we follow it, so anything shown from it may be incomplete or out of date". Coverage requests explains the steps.

Where to look when nothing is monitored

Where a place is Not monitored, the banner lists official sources to check directly: the place's own website, its legislature, official gazette or law database, and its government portal. Where coverage is partial or stale, the same list is in the banner's details. Each site says when its link was last checked; one the platform could not reach automatically says link not checked yet, with the reason.

Where a place is Not monitored, the banner also offers Request coverage for <place> (the city or county first, then districts). The request goes to the people who run the platform. They rank requests and check the place's official sources; see Coverage requests.

Outside the United States nothing is monitored yet. A location in another country reads Not monitored, with that country's official sites (the United Kingdom, Canada, France, Germany, Mexico and Australia so far).

How connectors fetch

Connectors are polite by design. Each one:

• identifies itself with a User-Agent that names the platform and a contact address;
• reads the site's robots.txt file first, and records a refusal instead of fetching a page the site does not allow;
• keeps to a limit on how often it asks each site for pages, however many fetches run at the same time;
• asks whether a page changed since the last fetch (conditional requests) and skips identical bytes, so unchanged documents cost nothing;
• watches each page's structure and marks coverage Partly monitored when a layout change means it may be missing items.

Connectors that read an API with a key (Open States, Regulations.gov) send the key in a request header, never in the address stored with the document.

A connector does not run until a person has reviewed its source's terms and recorded the review. Each source was surveyed when its connector was first built (access, robots.txt, terms, rate limits, licensing). That source survey is the starting point for the review, not the review itself.

Connector health

Open Sources → Connector health to see every connector your organization can see, public ones and your own uploads:

Filter by visibility or health. Open a connector for its 30-day counts, the pages that changed structure, the terms review (who, when, the terms page), the source survey, the coverage it provides for its place, its recent runs with their first error, and a link to the documents it recorded.

Not yet available

• Feeds your organization configures itself (a private connector on a URL you choose).
• Full-text search inside documents; search matches titles, publishers and file names.
• Reading scanned PDFs (OCR).
• City and county agendas from a coverage pack that has not been activated (most of them: each waits for a person's terms review and a 7-day run). PrimeGov, Granicus and BoardDocs places are not monitored: PrimeGov's robots.txt refuses automated clients, Granicus could not be reached for recording, and BoardDocs blocks automated requests.
• The Kitsap Planning Commission's PDFs themselves: the county's web firewall blocks automated downloads, so the platform records their links and titles only.
• Monitoring anything outside the United States (the UK Parliament and Oireachtas packs await activation).