The neuron fires on high-level structural markers and labels—things like case-style role names (“Plaintiff,” “Defendant,” “Respondent”), code keywords/imports (“import,” “public,” class/interface names), license section numbers and headings—i.e. tokens that denote major document sections or labels.