phrases indicating a response or reaction (e.g., "responded", "response", "in response", often with the following "by").

programming-language code tokens and structural keywords (i.e., parts of source code).

passages that signal a judge's stance or authorship in a legal opinion — e.g., where a justice announces concurrence, dissent, or speaks in their judicial capacity.

The neuron detects the start-of-sequence (beginning-of-document) token.

the neuron fires on numeric/math content — tokens that are numbers, math operators/terms, or words introducing calculations (e.g., "Solve", "Convert", numeric values, units).

It detects apostrophe characters (the single-quote/curly quote) used in contractions and possessives.

This neuron detects numeric tokens — numbers, years, measurements, zip/postal codes and other digit sequences (including decimals).

Questions asking for advice or instructions framed from the speaker's perspective (tokens like "I", "should", "do", "how should I", "what should I do").

the neuron detects numerical tokens and numeric/measurement references (e.g., measurements, years, table/figure/table-number labels).

the neuron detects document-structure/formatting tokens (e.g., end-of-turn markers and other markup or boundary tokens).

It detects negation fragments—roots of negative contractions like aren, wasn, haven, isn, don, didn, hasn (the "n't" parts).

the neuron detects requests or expressions of need/obligation — tokens used when asking for help, stating what should/needs/would be done, or posing a question.

Instances of first-person self-reference — tokens that mark the speaker referring to themselves (personal perspective/possession).

mathematical notation and formula tokens (LaTeX/math expressions and other math-specific symbols/terms).

the presence of third-person personal pronouns (especially singular forms like "he"/"she"/"he's") referring to people.

sentences or phrases expressing strong emotion or subjective opinion (highly charged, evaluative language).

tokens that are proper nouns—names of people, authors, taxa, or other titled entities (e.g., surnames, journal/species names).

It detects the start of a quoted utterance (opening quotation marks / the beginning of a dialogue turn).

finds technical, domain-specific tokens and named entities typical of scientific/academic writing.

It detects instructional/advice-oriented content — passages offering tips, how-to guidance, or suggestions.