saey-log-likelihood-ratio-proxy-formula
IN premise — summaries/2026/08/24/bricken-2023-monosemanticity-chunk-3.md
Created 2026-08-25T02:57:54+00:00
The scoring proxy for feature specificity/sensitivity is the log-likelihood ratio log(P(s|context) / P(s)).
Summary
The system judges whether a pattern is truly characteristic of a given situation by measuring how much more likely it is to appear there than it is to appear anywhere by random chance. This sets the mathematical yardstick that every specificity and sensitivity score in the system rests on, deciding what counts as a meaningful signal versus background noise.