The problem: jury selection relies on gut instinct with no documentation. When appeals happen, attorneys can't prove their process was sound.

MA'AT creates hash-verified audit trails. It analyzes statement consistency without psychology - just coherence scores, contradiction counts, and evidence gaps.

Key constraint: a language firewall blocks diagnostic terms ("diagnosis", "archetype", "prediction") before they reach reports. Court-survivable by design.

Stack: Python/FastAPI backend, SQLite, vanilla JS frontend. Docker deployment.

Free tier is permanent. The philosophy is "power pays for infrastructure, underdogs get the blade."

Looking for feedback on the technical approach and whether the legal constraints make sense.