Which risk factor suffices, at which threshold?

Seven risk factors, from a country list to the deviation from the declared profile, swept across fifty‑one thresholds.
Recall and false alerts carry their intervals, and every figure can be verified by you.

git clone https://github.com/ArslaneSempai-ui/cascade-scoring npm ci --ignore-scripts npm run measure:yours -- --customers=your-customers.csv --reviews=your-reviews.csv Your review outcomes, measured on your machine. Nothing of yours goes up.

Cascade · five instruments, one method Routing · extraction

Where should the next dollar go?Seven tiers, from a regular expression to a human, measured on sealed records.Open Routing Screening · name matching

Which matcher suffices, at which threshold?Seven matchers, from strict equality to a multilingual embedding, measured on your own alert history.Open Screening Monitoring · transactions

Which scenario suffices, at which threshold?Seven scenarios, from a bare amount to a peer profile, measured on your own dispositioned alerts.Open Monitoring

Scoring · risk rating

Which risk factor suffices, at which threshold?Seven risk factors, from a country list to the deviation from the declared profile, measured on your own periodic-review outcomes.You are here · Scoring

Dossier · the regulator’s piece

Is the whole chain measured, sealed and fresh?Four sealed answers, five controls, one dossier a reviewer verifies without us.Open Dossier

The shelving of weights, state 01: Geography sees nothing here. On purpose.

the empty bay: geography sees nothing in this bookevery dossier lives in ordinary countries: no list

finding 01blind by design, said

Every written dossier resides in an ordinary Union country, because no official country list was copied into the data. The geography factor therefore scores zero on both sides, and that blindness is the finding: risk lives in declared facts, not in passports.

0%recall of the geography factor on the 42 written escalated dossiers, interval [0–8]

0%false alerts on the 42 maintained twins, interval [0–8]: nothing to alarm on either

The shelving of weights, state 02: Products carry the strongest signal. And its cost.

the tallest pile: products carry the signaland its ivory shadow: one twin in four reviewed

finding 02one twin in four alarms

Score nothing but the products a client holds and most escalated dossiers trip it: correspondent accounts, private banking, crypto rails. So does a quarter of the honest book, and every one of those reviews is analyst time.

92.9%recall of the product factor at 0.50, interval [81–98], 42 written escalated dossiers

23.8%false alerts on the 42 maintained twins, interval [13–39]: the price of the signal

The shelving of weights, state 03: No single factor holds the floor.

the brass shelf at ninety: no pile reaches itbehaviour: clean, not one ivory box, half seen

finding 03the floor stays unmet

The tool's rule asks the lower bound of measured recall to reach 90. On 42 written dossiers no factor gets there alone: the best bound stops at 81, and the cleanest one, behaviour without a single false alert, sees only half. That is the case for scoring on more than one factor, measured.

81%the highest lower bound any factor reaches: product at 0.50, recall interval [81–98]

52%the cleanest factor's bound: behaviour at 0.50, zero false alerts, interval [52–79]

The shelving of weights, state 04: Written and generated, kept apart. Both ways.

written dossiers: the bound stops under the shelfthe variants do worse, and still stay apart

finding 04a rule, not a convenience

The same cell, two provenances, and here the generated variants are the harsher half: the best factor drops ten points of recall on them. Separation is not a way to hide a flattering number. It is a rule, and it holds whichever side it favours.

92.9%recall at product 0.50 on the 42 written escalated dossiers, interval [81–98]

79.4%the same cell on 126 generated variants, interval [71–86]: harsher, and still kept apart

The shelving of weights, state 05: Your book, your factors.

an ivory archive box: your book goes heretwo files replay the grid; nothing leaves

finding 05yours replace both

Every dossier behind these findings is ours: written twin by twin, or generated under a seed. Two files from your own systems, reviews and customers, replay the whole grid on your machine. Nothing leaves it, and the report lands next to your files.

42written escalated dossiers behind the measured half, seven per declared nature

126generated escalated variants behind the synthetic half, natures preserved by construction

Pick any cell, read what your threshold costs.

Recall over false alerts of each risk factor at each threshold, on the written files
factor	0.50	0.60	0.70	0.80	0.90	1.00
geography	0% 0% fa	0% 0% fa	0% 0% fa	0% 0% fa	0% 0% fa	0% 0% fa
activity	38% 0% fa	0% 0% fa	0% 0% fa	0% 0% fa	0% 0% fa	0% 0% fa
product	93% 24% fa	93% 0% fa	43% 0% fa	17% 0% fa	0% 0% fa	0% 0% fa
exposure	26% 0% fa	17% 0% fa	7% 0% fa	7% 0% fa	0% 0% fa	0% 0% fa
structure	33% 0% fa	33% 0% fa	12% 0% fa	7% 0% fa	2% 0% fa	0% 0% fa
behaviour	67% 0% fa	52% 0% fa	43% 0% fa	40% 0% fa	31% 0% fa	0% 0% fa
tenure	17% 12% fa	12% 10% fa	5% 5% fa	2% 2% fa	0% 0% fa	0% 0% fa

Measured on the 42 written escalated files and 42 maintained look‑alikes: recall on top, false alerts below. The marked cell, product at 0.50, carries the strongest recall lower bound; under its default rule the tool retains NO cell at the recall floor on these cases: the record says so, and the instrument shows it live. The synthetic variants stay apart, on the instrument.

Cascade · ScoringOpen the live instrument Every risk factor at every threshold, recall and false alerts with their intervals, live from the sealed record, and the tool's own selection rule under your recall floor.

Cascade Scoring, on film.

The five findings, explained

The amethyst Cascade robot, palms up, projecting the two lower bounds of recall the fiche names: product on the left, behaviour on the right, both under the 0.90 floor.

hosted on YouTube · link pending upload

The appendices your reviewers will ask for.

Method & what is measured What the frontier reads, and what it refuses.

Security & data handling The declared tables, the seal, and what never leaves.

Terms of engagement What the grant allows, for how long, and what a client buys.

Privacy No data is collected. Written down, and verifiable.

Accessibility Usable by keyboard, by screen reader, and with motion turned off.

Pricing, in figures Contact The fine print The repository, public