Playbook-1, specified before it exists
a frozen contract, release gates that are tested code, three teacher pilots, one bootstrap demonstration that says how hard the bar is, and no trained model. the claim comes before the run
writing
Working notes on evaluating and training agents on real legal work โ what the environment measures, what the models miss, and what changes.
a frozen contract, release gates that are tested code, three teacher pilots, one bootstrap demonstration that says how hard the bar is, and no trained model. the claim comes before the run
twelve synthetic matters, a scripted counterparty, 121 deterministic gates, five measured models, and the adversarial audit that found 84 blocker-grade false positives in my own instrument