Method

Entities. The list covers every town, village, city, county legislature, school district, BOCES and IDA in the three regional registries used by the meeting-video project, plus the state's school-district list. School districts added only to file investigation documents (Buffalo, Syracuse, Olean and Binghamton areas) appear in the library but are not counted here. IDAs outside the Capital Region are still being added.

Venue discovery. We read each entity's website and agenda/minutes pages, along with existing civic-source registries, to find the document platform it uses: CivicPlus AgendaCenter, IQM2/Granicus, Legistar, BoardDocs, eScribe, Laserfiche, Municode Meetings, Diligent/iCompass, BoardBook, or a plain page that links PDFs. Fetchers cover CivicPlus AgendaCenter, IQM2, Legistar, plain agenda/minutes pages, and BoardDocs public portals (when enabled).

Fetching. A Cloudflare Worker cron lists each venue about every 6 hours and fetches agendas and minutes. For non-BoardDocs hosts, robots.txt is honored (including Crawl-delay), with at most one request per host every 2 seconds and a project user agent. BoardDocs public pages are collected when enabled, using a normal browser user agent, one in-flight request to the shared host, at least 3 seconds between requests, a per-site daily cap, and a circuit breaker. Attachment PDFs are downloaded; scanned PDFs become OCR jobs for team Macs. No meeting video is downloaded.

Text. When a PDF has a usable text layer, its text is extracted on the Worker. Scanned PDFs, under 200 characters per page, become OCR jobs that team Macs claim and OCR locally. Per-page text and confidence are stored.

Extraction and scoring. Minutes are parsed with conservative rules for motions, resolutions, tallies, roll calls, per-member votes, attendance, adjournment and executive-session motions. Each indicator follows the rubric. Evidence snippets are verbatim, at most 25 words, with a page number, and personal names are masked.

Limits.

Backfilled posting dates are mostly unknown, because the documents were first seen after the meeting.
Notice by newspaper or physical posting can't be seen online.
Absence from a venue isn't absence from the public record.
The rule-based extraction can miss or misread formats, so every flag links its source document.

Collection identity. User agent: Mozilla/5.0 (compatible; meeting-docs/1.0; +/meetings/oml/method).