"It's a public record" is supposed to mean you can find it. In local government, it mostly doesn't. The single most important infrastructure, land-use, and spending decisions in the country clear city councils and planning commissions every week, fully in the open - and remain effectively invisible to anyone outside the room. Not because anyone is hiding them. Because the record is built in a way that defeats search.

We see this most starkly in the data-center boom. A billion-dollar campus clears a council under the agenda line "Request to Enter into a Development Agreement". A 20-megawatt power deal for a hyperscaler reads as a routine "power-purchase agreement". A hyperscaler's water study is filed as "a professional services supplemental agreement". No keyword search of agenda titles would ever surface any of it. Multiply that across thousands of city halls and millions of agenda items, and a national building boom is hiding in plain sight. The same is true of fiscal stress, zoning shifts, procurement, and every other decision that matters.

Four reasons the public record resists search

  • The title is auto-generated; the substance is in the attachment. Agenda software names an item from a resolution number and a stock phrase. The acreage, the megawatts, the dollar figure, the operator - the things you'd actually search for - live in the attached staff report or ordinance that almost nobody opens. Title-based search sees none of it.
  • Every platform is different. Cities run on dozens of different agenda systems - Legistar, Granicus, CivicPlus, BoardDocs, eScribe, PrimeGov, and more - each structuring (or not structuring) data its own way, with uneven archives and inconsistent exports. There is no common format to query against.
  • Nothing is normalized. The same company is "Project Jade" in one county and "confidential applicant" in the next. The same decision is an annexation, then a rezoning, then a development agreement, then a plat - each on a different agenda under its own forgettable title. Without normalization, you can't connect them, count them, or compare them across jurisdictions.
  • There are 11,610 of them. Even if you solved the first three for one city, the decision you care about is just as likely to be in any of thousands of others. No human team monitors the public record at that scale.

What it actually takes to read it

Closing this gap isn't a scraping project; it's a data-modeling problem that has only recently become tractable. It requires three things done together:

  • Read the documents, not the titles. Extract and index the full text of agenda items and their attachments - the staff reports, ordinances, and development agreements where the substance lives.
  • Normalize across platforms, jurisdictions, and decades. Turn meetings, items, motions, and votes into comparable, time-stamped objects that can be linked - so a motion in Phoenix can be related to a vote in Pittsburgh.
  • Verify, don't just match. Raw text-matching is noisy: the documents that mention "data center" the most are often long IT contracts that have nothing to do with the land rush. Surfacing the real signal means a second, stricter pass that separates the genuine decision from the passing mention - the difference between a list and an intelligence product.

That combination - deep domain modeling plus modern AI, applied to the full text of the national record - is what turns "technically public" into "actually findable". It's also why the leading indicator lives here: by the time a decision reaches a permit feed, the news, or a state dataset, it's already resolved. The record is where it happened first.