Skip to content
Measurement7 min read

Measured, not asserted: how a number earns its place

A false-positive rate without a denominator is a marketing figure wearing a lab coat. Here is how ours is produced.

Every vendor in on-chain security reports a false-positive rate close to zero. Almost none report the denominator, and a ratio without one tells you nothing.

How ours is produced

The harness runs in continuous integration on every merge, against corpora of real captured benign traffic rather than synthetic fixtures. A false-positive regression fails the build, and so does a recall regression. The number is a condition the codebase keeps satisfying in order to compile, not a measurement somebody took once.

Every detector carries its own evidence

Detectors ship with golden cases, positive and negative, all exercised on every run. Detection quality and coverage of input are tracked as separate axes: a detector that never fires can be precise or unreachable, and conflating the two reports a property of the data as a property of the detection.

What you can check yourself

Every alert ships as a threshold-signed, expiring attestation that you can verify without trusting us. Every screening verdict names the source it consulted and the time it consulted it, and where no live source was reachable the verdict is insufficient data, never a pass.

That last one costs a cleaner-looking dashboard and ships anyway, because a customer who believes something was screened when it was not has been misled at the moment it matters most.

The next breach is already being dug.

Every engagement is scoped to your protocol by the team, and pricing follows that scope.