Datumline methods

Idea Factory Council

decorrelated multi-model arms, red team, blind judge and a no-stake governor

The problem

Asking three chatbots the same question and picking the answer you like is not a council. The answers are correlated (same training, same vendor, same framing), you know which model wrote which so your bias picks the winner, nobody attacks the framing, and you never find out whether a plain one-line prompt would have done as well.

What it is

Run a question past several AI models at once, with assigned lenses, an adversarial red team, a blind judge and a no-stake governor, so you get decorrelated judgement instead of one model's confident average.

Delivered as a download: method documents, templates and standard-library Python scripts with tests that fail when the method's own logic is broken. Synthetic example data only.