The documents of the bench. The field is live here too - place a lens.

The documents of the bench

FoQLens is a precision regulator for language models: the weights sit behind a glass and lenses open where the query needs to see. This is the documentation of the bench that tests it - the mechanism, the goals, the plan, every preregistration and every run.

Pick a document on the left. Each is a markdown file in the repository, and every link below goes to its source.

The idea

What the bench measures on

Around it

Preregistration

Runs - verdicts withdrawn