The neuron fires on isolated, out-of-context tokens that stand out as proper names, acronyms, reference keys or other rare identifiers (e.g. “Asia,” “UK,” “download,” “Alpha”), i.e. single tokens that look like headings, citations, or domain-specific labels.