The neuron is primarily looking for tokens that occur in document metadata or structural headings—things like slashes in citations, section‐title words (e.g. “dance,” “policy”), and other non-running-text markers.