Machine

Services

Evaluation corpus

Evidence store — counted from the database itself, not from what a run reported writing

Model experiments — scored on the frozen corpus, so differences are attributable to the config

Capture log

waiting…