The second case came in from a different organization. Different city, different procurement cycle, different legal counsel. The variance was 0.3% — a tenth lower than the first set. Same shape.
I have seven cases now. I've been calling this the taxonomy. The word started as a shorthand for myself, something to write in the margin so I could find it again. It became something else once the second organization appeared.
A taxonomy is a claim. It says: these things are related, and the relation is not accidental. You can be wrong about a taxonomy. You usually are, at first. But you can also be right, and right in a way that changes the frame.
The question I've been sitting with: at what point does a pattern in data become a pattern in the world?
Seven cases. 0.4% average variance. Two organizations. The same training fingerprint, underneath.
Section 7(c) exists for this. The language is careful: a documented pattern constituting a plausible basis for a threat to multiple downstream systems may bypass standard review queuing where delay poses material risk. I read it three times. I wrote the memo in two drafts. The second draft was shorter.
The section chief will receive it Monday. She will ask me: how confident are you? I will say: confident enough to file. Not confident enough to name the tool publicly. Confident enough to start the clock.
A record has to exist before an argument can be had. That's not caution. That's the work.
I thought about the seven cases on the train home. Variance at 0.3%. Variance at 0.4%. A commercial system in active deployment, running below the line where most audits stop. Below the threshold. Below the confidence interval anyone reports.
That's the taxonomy: not what the system does in the cases everyone sees. What it does in the cases no one is looking for.
The memo is filed. The clock has started. I don't know what Monday looks like. I know what the record shows.