A national civic-record contribution, waiting on decisions only you can make.

Somewhere in the United States tonight, a city council is meeting in a half-empty room, and a camera nobody is watching is recording it. Multiply that by tens of thousands of towns and forty years, and you have one of the largest, least-studied records of a democracy anywhere — and one of the most fragile, because it lives on volunteer laptops, dying platforms, and vendor servers that forget. This project is the attempt to catch it before it goes. The tally so far is deceptively dull and secretly enormous: eighteen million pages, one-point-four million documents, two-plus terabytes of agendas and minutes — the connective tissue of self-government, gathered so it can't be quietly deleted.

The most haunting number isn't the biggest one. It's that almost every unreadable file in the entire collection — nineteen thousand of them — belongs to the oldest corpus, the 2022 forensic set. The newer scrapes come back clean; the old one comes back with holes. That's the whole argument for the work in a single statistic: digital records don't fail all at once, they fail at the edges first, and every year you wait, the edges move inward. The forensic bundle isn't just data; it's a memento mori for the born-digital record.

Look closer and the corpus starts to gossip. Legistar towns publish lean, six-page index documents; CivicPlus towns publish thirty-page packets stuffed with backup material. The page-count histogram is really an ethnography — you can see which governments treat the agenda as a pointer and which treat it as the whole file, which is why the same tooling had to learn ten different vendor dialects, and why merely finding the towns meant drawing a 12,362-tenant map that didn't previously exist.

But the real story is epistemological, and it's the one worth telling twice. Again and again the project ran into the same trap wearing different masks: a document the API lists but won't deliver; a meeting that generated no files and so was recorded nowhere; six thousand records filed under a board named “General” that turned out not to be a board at all. Each time, the lesson was the same — a populated field is not a usable value, and silence is not absence. The discipline that came out of it is almost moral: count the thing itself, give every gap its own reason code, and reconcile in both directions until zero files are of unknown origin. That's why “418,871 of 418,871” is the proudest number here. It's not that nothing was lost — nearly two thousand documents were — it's that nothing was lost anonymously.

And it all points somewhere human. The endgame is this contribution to Democracy's Library and a three-tense architecture that reads more like a clock than a schema — collector.db for what the fleet is doing now, archive.db for what's already preserved, civic.db for every government that should be represented but isn't yet. The last database is a to-do list for a democracy's memory. Filling it won't be done by scrapers alone; it'll be done by Archive Corps — the retired clerk who knows every voice, the student learning archival standards, the neighbor who can name the people in a 2009 zoning hearing. The stories in these numbers are about decay and discipline, but the one they're all building toward is about stewardship: a community learning to keep its own record, at a scale no one has tried before.