Test run · July 2026 · not a client deliverable

CADi’s first end-to-end run against a public synthetic benchmark — 6,362,620 PaySim rows on SCM 3.0.0. It demonstrates the mechanism: a decision made at the authorization moment, with a named reason, and the fraud label never in scope. It is not a client deliverable and not a fraud-accuracy claim. It reports recall without the operating point that decides whether the screen could actually be run — the flag alerts on 23.9% of all transactions, roughly 190 alerts for every fraud found — states no baseline comparison, and recorded no provenance stamp while it ran. That re-instrumented run has since landed (SCM 3.2.0, 1 August 2026). It reproduced this run’s confusion matrix exactly - the same 8,012 catches and 201 misses - and added the transaction amount this run never recorded. On that measure recall is 87.5% by value against the 97.6% by count quoted below, because the 201 missed frauds average 5.7× larger than the 8,012 caught. Every recall figure in this document is by count unless it says otherwise.

CADi × PaySim · benchmark test run · full dataset

6,362,620 decisions, zero labelsSYNTHETIC BENCHMARK

Every transaction decided at authorization time. The fraud label was never in scope — it was used only afterwards, to score the decisions.

97.6%
of fraud caught, by count
87.5%
of fraud caught, by value
8,012
of 8,213 fraud flagged
0.53%
precision @ 0.13% base rate
0
fraud labels seen by the engine

The catch

1,520,581 flagged for review (23.9%)containing 8,012 of the 8,213 fraud201 slipped through, carrying 12.5% of the fraud value
8,012
true positive — fraud, flagged
1,512,569
false positive — clean, flagged
201
false negative — fraud, missed
4,841,838
true negative — clean, passed

Where the flag fires — by channel

TRANSFER
42.9% of 532,909
CASH OUT
42.9% of 2,237,500
PAYMENT
15.2% of 2,151,495
DEBIT
13.6% of 41,432
CASH IN
0.0% of 1,399,284

 = channel where fraud occurs in this dataset. The detector concentrates in exactly those channels — and is silent on CASH IN — without ever being told where fraud lives.

What a normal report can’t show you

22% of review load, zero yield
A tuning insight before deployment

332,507 flags fired in channels that carry no fraud at all. Scope the screen to the two fraud-carrying channels and the review queue shrinks by 22%, precision rises 0.53% → 0.67%, and recall is unchanged — measured on the full stream before anyone commits to the control.

201 misses, one mechanism
The escape route, named

Every missed fraud is a partial drain — the account was not emptied to zero. That is not a loss statistic; it is the specification for the next detector.

0 flags on 1,399,284 deposits
Negative evidence

Perfect silence on the one channel where the mechanism cannot occur. A statistical score sprays; a mechanism tracks real money movement.

Honesty rail. PaySim is a public synthetic benchmark; its fraud is generated by a known drain-to-zero rule, and “fraud occurs only in TRANSFER and CASH OUT” is a property of that generator. This run demonstrates the mechanism — label-free, decision-time flagging with a per-decision reason — not a production fraud-accuracy claim. At 0.53% precision the signal is a first-line screen for review, never an auto-decline.