Why charge per verdict instead of per token or per minute?
Because the value isn't model time — it's a decision you can defend in committee. A clean reject saves a week of underwriting; a confident pursue closes deals. We price on the artifact, not the inference.
What if the agent says reject and I disagree?
Every verdict cites the signal that drove it. You see the cap-rate delta, the zoning code, the tax burden number. Override is a one-click action that logs your rationale; we never silently downgrade.
How do you keep things accurate?
Hard breakers run in deterministic Python before the LLM ever speaks. The model can enrich narrative; it cannot upgrade past a structural problem. Source quality and missing data are surfaced explicitly.
Is this Texas-only forever?
For now, yes — depth over breadth. Texas CRE has its own zoning, tax, and submarket dynamics, and we tune for them. Florida and Arizona are next on the roadmap.