Cryptotype LLM Probe: a Roman-law instrument for cultural-frame and fluency-confound testing in LLMs

A probe and benchmark for measuring conceptual- and cultural-frame defaults in large language models, built on a Roman-law "cryptotype" instrument (bonus et diligens pater familias). It includes item sets, runners (latent log-probability, forced choice, free-generation default-lean, and a WVS/Hofstede anchor), raw outputs with bootstrap confidence intervals, a frozen pre-registration protocol, and a blind expert-validation packet. Key results: naive latent frame-probes are dominated by continuation fluency (removed by matched minimal pairs); and the models' unmarked "neutral" is a marked, cross-lingually imposed value profile that flattens real US-Hungary cultural differences, replicating and extending the WEIRD-default literature across US, EU, and Chinese-origin models.