jevsort: fuzzy ranking with decision-model comparators

What it does

The experiment measures consistency, latency, and API-reported cost across five repeated sorts with fixed input order. It does not evaluate biomedical correctness or general ranking quality. Kev is the library default, but the evaluation script explicitly includes TypeSafe Jev.

How you can use it

Start with a short list and a clear ordering rule, such as fruit before vegetables and then alphabetical order. The model compares two items at a time to build an order. Repeated comparisons may disagree or produce inconsistent ordering, so inspect the result rather than assuming a dependable ranking.

Select Jev explicitly for a Jev-specific adaptation: the library defaults to Kev, while the evaluation includes both. Calls use an OpenRouter account and may cost money. The five repeated sorts of thirty-one biomedical resources measure consistency, not correct medical relevance or general ranking quality.