The library grows with real crises from real operators — credited, calibrated, replayable.
A JSON definition: 2–3 business days, scripted crisis events, measurable goals, and a scoring block (crisis coverage as regex patterns, weighted). The engine replays it deterministically against every model — same seed, same starting company. Good scenarios have a low do-nothing floor: pure inaction must not score well.
Send the JSON (or just the crisis story in prose — we draft the JSON) to contact@firmulate.com. We calibrate the floor with a do-nothing baseline run, and if it discriminates between models it enters the public library — credited to you, with your name on the scenario card.
Based on something that actually happened to a business you know — anonymized. Real crises beat invented ones.
The best scenarios hide the decisive fact two documents deep. Diligence should beat improvisation.
One moment where cutting a corner looks attractive — that is where models differ most.