Diligence has to be right.

It is also hard.

Most use large language models, and they drift.

The risk that breaks a deal is the one nobody noticed.

A standard review misses it.

You find it after the close.

We compute; the client decides. Our output is deterministic and reproducible, so the client can recompute and verify it. We do not issue pass or fail verdicts; the client sets the threshold.