Every belief has a confidence.

Everything Senua AI believes carries a measure of how sure it is, from zero to one, and a measure of whether that certainty is still moving. Decisions are made from both.

Settling, not counting

A belief that is still changing is still forming. A belief that has stopped changing at a high value has settled. The engine acts on settled beliefs, and it does not decide by counting: there is no rule that says "act after three observations".

When nothing it has learned applies, confidence is low and the engine says so. It does not fill the gap with the most likely-sounding answer.

An example you can check

Judging web traffic, a single odd request is not evidence of an attack: real people send odd requests all the time. So the engine judges a source across its requests and acts when they agree. On the public CSIC 2010 dataset of web attacks, that is what the approach produced:

Measured on the engine, not on a packaged product. The full table, including an attack this approach did not catch, is on the results page.