The Record

What this is

A docket, not a dashboard. The unit is the claim with receipts; the verb is challenge it. Every screen can lose an argument: an edge that carries no relation type says untyped, a career step with no recorded date says no date, and two people the data cannot connect return no documented route rather than an invented one.

The Gap Ledger is computed at build time from the data itself, never written by hand. It is the most useful page here. If it is ever empty, the audit is broken — not the dataset perfect.

What it will not do

There is no timeline. Zero of 1,243 edges carry a date and only 931 of 1,921 career hops do. A timeline would require interpolating the rest, so the app refuses to draw one. Career order is real and is shown; dates appear only where a source recorded one.

Play tags are shown as annotations. A play is a recurring career-rotation pattern; the vocabulary is defined in PLAYS.md. It was audited on 2026-08-02, when backstop (defense official -> contractor board -> defense) was found on 23 carriers whose own role strings showed no contractor leg at all. The vocabulary was corrected and two missing plays were minted: bar, the legal revolver (senior DOJ -> large-firm partnership -> senior DOJ), and ward, the medical one (health regulator -> industry -> regulator). The audit that caught it is published in the ledger.

Co-affiliation is not coordination. Every route and every shared institution here is a documented fact about who held which post. None of it asserts that anyone directed anything.

The board

The board is the capture ledger summed per institution and ranked: who has been caught running the most of the playbook, with every record’s receipt one click down. Two boards, because the captured and the capturers are not the same list. The rank sums verified evidence from lenses with a measured index floor and nothing else; attributed material and rumor are drawn as a ghost band and never move a row, a lens that only holds a background firing rate is counted and shown and never summed, and the credit an institution earned by policing itself is already subtracted. No gap between two adjacent rows is claimed as a finding. It is an evidence-density map with the receipts stapled on, not a verdict, and the rule that produces it is printed under it.

The detectors on the collateral

The collateral page is the first measurement of the working detectors on the material this record actually holds about its own people: the verbatim public statements in the texts store, graded by the cue detector and the capture scanner against the same background rulers the instrument prints, pre-registered before the run. It did not go well, and the page says so first: one span in five that fired carried the thing the detection names, the one lens above background is above it because AI-safety people say existential risk, a cue fires inside the word haven’t, and the scanner scores a Twitter handle’s underscores as emphasis. Every board row says whether any collateral exists for it — one of thirty-six does — and every person with texts carries the reading on their dossier, beside the rate it should be read against. What a text exhibits, attributed to the text. Not a verdict on anyone.

The sibling instruments

• The Instrument — the same people graded on their prose instead of their posts. Graph and text stay on separate axes and are never averaged into one number; the MCP server’s profile_person is where the two lanes are read side by side.
• The Ratchet Atlas — this dataset drawn as a graph, for when you want the shape of a cohort rather than a query against it.
• Ratchet Clicks — the twenty-three mechanisms the dataset was assembled to test, one per chapter of The Ratchet.