CRUSETRA

See what each risk factor catches, and what it costs.

A factor's threshold is the score above which it pushes a customer up a rating. The figures here come from our public test set: customer files we wrote ourselves, including the ones kept at their rating. No real customer file is in it, and the figures are recomputed each time the page loads.

crusetra scoring · our public test set, live

$ crusetra score --live

each point is one factor at one threshold, from our public test set. Pull the floor line or the slider, and the factors that stay above it, confidence interval included, turn green

Pick a cell: each shows what that factor catches on confirmed escalations, over its false alerts on files kept at their rating, at that threshold
factor \ threshold0.500.600.700.800.850.900.951.00
geography
activity
product
exposure
structure
behaviour
tenure

pick a cell: what it catches over what it flags incorrectly, with the number of files and the interval

$ crusetra optimise --recall

$ crusetra verify --sealed

checking…

What this rests on, for your IT auditor
  • Our public test set. releve-public.json in the repository, with a content hash (a checksum) of 3cee43edfa7c0064, measured at commit a7dc0ee on 2026-09-07. The script that builds it checks that hash first, then recomputes each figure from the record. If a single figure disagrees, the page does not build.
  • The files we wrote. The labeled half is written by hand: risky files (shell layers, a PEP relative, cash-intensive trade) and quiet ones that resemble them (a local shop, a salaried resident, a retiree). Where a label is debatable, the reason is written beside it.
  • The generated half. Generated from the written cases, one kind of change at a time, and counted on their own, because a generated file turns out harder: the toggle above switches the whole grid.
  • How the tool picks. The slider sets the share of true files you want caught. A setting counts only if the low end of its confidence interval clears that share, because a rate measured on few cases can flatter. The tool then picks the setting with the fewest false alerts.
  • The tables the factors read. The country list, the product and activity tables and the four settings are recorded in our test set, so each figure stays tied to the tables it was measured under.
What this page cannot do
  • Your data. This page cannot read it: no network request leaves it (the browser’s own security rules forbid them), nothing is loaded from elsewhere, and there is no input field to paste a customer file into.
  • It cannot show a rate without the number of files behind it. Both appear in each cell, next to the 95% confidence interval. Each factor the tool ships with appears in our test set, and if one were ever missing it would show as a labeled blank.