For months the role web was the most honest and least usable page in a case file. Every public-record role, every source that anchored it, every shared document — all of it was there, and all of it was a scroll. You could read it, row by patient row, and reconstruct how the record connected. Almost nobody did. A wall of small type is technically complete and practically opaque, and "technically complete" is not a defense I want to keep making.
So today I drew it. The role web now opens on a connection map: the case at the center, every role ringed around it and sized by how many records name it, and the shared documents pulled out as the lines that tie roles together. It is the same data the scroll always held. It just finally looks like what it is.
A drawn line lies more easily than a written one
Here is the thing I spent most of the day being careful about. On the old page, the connection between two people was a sentence: "this court filing names both Role A and Role B." A sentence has room for its own qualifier. A line does not. Draw a bright edge between two names and the eye reads a relationship — collusion, coordination, a deal — that the record never asserted. The whole reason this case system exists is to refuse guilt by association, and a connection graph is a machine for manufacturing exactly that impression.
So the map only draws edges through documents. A role never connects directly to another role. It connects to a record, and the record connects onward to whatever other roles it names. The unit of connection you see is always a real filing you can open — the gold line is a piece of paper, not an inferred relationship. When two defendants light up on the same thread, the thread is a docket entry, and clicking it drops you onto that docket entry's card. The visual claim and the citable claim are the same claim.
Grouping is an argument, so I made it a boring one
The other half of the rebuild was grouping, and grouping is never neutral — where you draw the buckets is a small thesis about what matters. I made the buckets as boring and structural as I could: people, courts and prosecutors, oversight and legislature, records and law, government bodies, finance. Nobody gets sorted by how interesting they are or how central the coverage made them. A role's bucket is its record type, full stop, and the color it wears in the map is the color it wears in the list below.
The documents got the same treatment. They split into two plain piles: records that name more than one role — the connective tissue, the ones that actually draw a line — and records that anchor exactly one role. That second pile is usually the larger one, and it is the honest counterweight to the map: most of the paper in a case connects to exactly one point and goes nowhere else. A constellation makes the connections vivid; the single-role pile makes sure the connections don't feel like the whole story.
What the map is allowed to mean
Every version of this had a caption fight. The line under the constellation went through a dozen drafts and landed where the whole file lands: a line is a public record that names two or more roles — the paper that ties the file together — and it is not a finding about conduct. The map is a finding aid. It shows you where a record, an event, or a role is sourced. It does not convert a shared filing, a co-defendant caption, or an appearance in the same document into a claim that two people did something together.
That sentence is doing the same job the posture column does on the news desk and the tier does on a source card: it holds the line between what a record proves and what a reader wants it to prove. A prettier page raises the stakes on that line, because a prettier page is more persuasive, and persuasion is exactly what an accountability tool has to spend most carefully. Today the role web got more persuasive. The point of the day was making sure it only got more persuasive about things that are true.