What 500 public-records requests told us in one window

Read together, the requests people file with city governments form a map of matters that have not reached a court yet. Here is what the first read shows, and where it stops short.

Published July 4, 2026 · PreDocket · Sample: 500 requests, 5 city portals

Most people treat a public-records request as an errand. Someone needs a fire report, a dispatch log, a permit history, so they ask the city for it. On its own that is administrative. Read a few hundred of them at once, though, and the errands start to line up into something with a shape. A records request names parties and a theory in plain language, and it lands in public the day it is filed. That is weeks or months before a complaint, a claim file, or a deal announcement makes any of it official.

This is the first edition of what we plan to publish on a recurring basis: an aggregate read of the request activity we see, with names stripped out and only the patterns left in. It is drawn from a starting sample of 500 requests across five city portals, of which 140 carried a real commercial or legal signal. That is a narrow window, and we say so plainly below. Treat this as an early read, not a national census.

The mix is heavily pre-litigation

We sort each qualifying request into one of six types. Across the 140 signals in this window, the distribution was lopsided in a way that surprised us less the longer we looked at it.

Pre-litigation is half of everything. These are the requests where someone is pulling a property's incident history, a set of inspection records, or a dispatch log against an address, and the plain reading is that a liability or premises theory is being built. Regulatory activity is the next largest block at a quarter of the total. Subrogation, the type most people expect to dominate a records feed, came in at 8 percent.

We want to be careful about that last number, because it is easy to over-read. A low subrogation count in one window does not mean carriers and recovery vendors are not pulling loss reports. It means that in these five cities, over these months, the recoverable-loss pulls we could confidently identify were a small slice. Subrogation activity also hides better than pre-litigation: an insurer's investigator often files a short, procedural-looking request that reads like routine records retrieval unless you already know the loss it points at. The honest read is that subrogation is present and identifiable, not that it is rare.

Where the requests came from

The five portals in this sample are all city governments, and the volume was not evenly split. Two cities produced the bulk of the qualifying signals.

This is mostly a story about how each city runs its records office, not about how litigious each city is. A portal that publishes the full request text produces far more usable signal than one that publishes only a case number and a status. So the ranking above is best read as a coverage ranking. It tells you where the window is currently wide, not where the underlying activity is highest. As more cities come into view, we expect this ordering to move.

The commercial requesters leave a fingerprint

The most useful thing in the data is not any single request. It is who is doing the asking. Where a requester identifies themselves, or signs the request text in a way that gives them away, a recognizable set of professional categories shows up. We can count those categories without ever naming an individual filer.

Of the 140 signals, we could tie a clear commercial-requester fingerprint to 37 of them. That is a floor, not a ceiling, because many cities blank the requester field and the identity only leaks through the text. The categories, ranked:

The point of this table is not the precise counts, which are small and will move. It is that the buyers of a feed like this are already sitting inside the feed as its most active users. Forensic and origin-and-cause firms, recovery vendors, and law firms are pulling these records one portal at a time, on their own matters, and none of them can see what the others are asking. The demand for reading this data at scale is visible in who is reading it by hand.

Clustering is rare, and that is the interesting part

Most requests stand alone. In this window, of 130 distinct matters, only 6 drew more than one request, and the largest single event pulled five separate requests within days. A cluster like that is the strongest signal the data produces: several unrelated parties converging on one address or one event, each from a different angle, none aware of the others. Those are worth far more than their share of the volume suggests. You can see a live example of one on the sample report.

What this is, and what it is not

This first edition is a narrow read. Five city portals over a few months is enough to show that the patterns are real and countable. It is not enough to make national claims, and we are not making any. A request also proves only that someone asked, not that a claim or a deal is real. The mix will shift as coverage widens, and we would rather publish the small honest number now and let it move than round it up.

If you work in subrogation, litigation finance, or claims, the guides below turn these patterns into something you can act on. And if you want to see the underlying signal feed, sourcing and coverage are shared on a call.