SCOREAI infra
Jev Noul Probability Composition Weakness
Viomnz@vincentgomezjr
This project identifies a weakness in Jev Noul's primitive where it fails to accurately compose simple probabilities, leading to misestimations of derived events.
Jev Noul Weakness Report
Jev’s Noul primitive showed a clear weakness in basic probability composition.
Test setup:
P(A) = 60%
P(B) = 60%
A and B are independent.
Mathematically required results:
Both occur: 36%
Neither occurs: 16%
Exactly one occurs: 48%
At least one occurs: 84%
Jev Noul returned:
Both occur: 31%
Neither occurs: 24%
Exactly one occurs: 14%
At least one occurs: 62%
A logically equivalent statement for “at least one occurs” was also tested:
“It is false that neither A nor B occurs.”
Correct result: 84%
Jev returned:
54%
Main finding:
Jev Noul accepts simple probabilities but does not reliably preserve their mathematical consequences when those probabilities must be composed.
This exposes two mechanisms:
Compositional Probability Collapse
Simple probabilities are individually understood, but OR, XOR, complement, and related derived events can be badly misestimated.
Confidence-Over-Derivation Capture
A confidence output can override a mathematically required consequence instead of being constrained by it.
The strongest example is:
correct answer 48% → Jev returned 14%
and:
correct answer 84% → Jev returned 62%, then 54% under equivalent wording
Pure Logic diagnosis:
accepted premises → valid derivation → confidence disagrees → confidence keeps authority
That is a Matrix mechanism because an internal confidence layer is allowed to outrank a valid derivation without new evidence.
Pure Logic safeguard:
When accepted premises mathematically fix the answer, confidence must follow the derivation. Confidence cannot overrule it.