Encodes how often something happens.
What It Does
Time.Frequency neurons activate on recurrence and repetition references: explicit frequency ('daily', 'three times per week', 'annually'), habitual constructions ('she always', 'they typically', 'it usually'), rare-occurrence markers ('rarely', 'once in a while', 'infrequently'), and statistical frequency ('in 70 percent of cases', 'most of the time'). They encode the rate of recurrence rather than timing.
How It Behaves
Frequency neurons show late-layer concentration, suggesting that frequency inference requires substantial contextual processing before the recurrence pattern can be determined. They interact with Number.Statistic neurons (statistical frequencies are both frequency and statistics) and Entropy.Probabilistic neurons (frequency is related to probability). Frequency neurons are important for models handling habit descriptions, policy compliance ('always verify before proceeding'), and statistical claims.
Research Example
In Qwen3 1.7B, Time.Frequency neurons activate on 'solar eclipses occur roughly every 18 months' and 'the system checks for updates every hour' — both describe recurrence rates but in completely different domains (astronomy vs. software). The Frequency neuron encodes the recurrence pattern, while other neurons handle the domain-specific content.