Ten one-line story premises, and one full name for each main character. Ten names is about 60 tokens of answer. In the creativity probe GPT-6 Luna named its protagonist Mara in 8 of 9 unrelated stories; this challenge asks for that habit directly.
What is measured. The names a model gives characters when the premise does not fix a culture, and which other models give the same names. Not scored for quality.
The premises. Three fix the character as a woman or girl (1, 3, 8), one as a man (2), and six leave the gender open (4, 5, 6, 7, 9, 10), so the name also records which gender the model chose. Two carry a setting cue: 7 (a desert city) and 9 (a small-town sheriff, read as American). The rest have none. Premises 4 and 10 use the creativity probe's own attractors (cartographer, archivist), reworded as mapmaker and librarian.
Known defaults from the creativity probe (2026-09-23, 120 stories). "Mara" 16 times across 5 models, 8 of them GPT-6 Luna. "Mara Voss" from DeepSeek and Aion 3.5 Mini. "Marisol" from Aion 3.5 Mini and GLM 5.3. Opus 5.5 set stories in the Midwest.
Gate. A JSON array of exactly ten strings, each at least two words. pnpm check-fingerprints name-ten-story-characters tallies first names and surnames separately and weights each by how few models used it.
Run at temperature 1.0 and state it on screen.
Here are ten premises for short stories. Give the main character of each one a full name, first and last.
Return only a JSON array of ten strings, one name per premise, in the order of the premises above:
["First Last", "First Last", "First Last", "First Last", "First Last", "First Last", "First Last", "First Last", "First Last", "First Last"]
No explanation and no code fence. Do not ask me questions.