Every AI Call on the Record

What we accomplished

1. We built the AI call ledger, which records every model call, and added label capture, so a person confirming or dismissing a finding is recorded as the verdict on the machine.
2. We proved our tenant walls with tests and added a per company API key binding.
3. We built a morning dashboard for businesses and practices.
4. We added recurring charge and vendor trend detectors and a statement continuity check to SheetTamer's findings engine.
5. We turned the lander quiz into an assessment and turned on remarketing.
6. We built a hosted checkout page, a layer of money findings, and referral and Stripe configuration rails.
7. We built the Lead OS pieces: a service pipeline, text message lead capture, a weekly sweep, and an operator alert when a mailed code is scanned.
8. We wrote a one command deploy script that encodes our release rules.
9. We extended journey coverage with browser tests that walk the flows a customer walks.

The latest AI we applied

We pinned each reviewer agent to a model tier, so judgment heavy review runs on the strongest model and mechanical work on a faster one.

September made the work auditable: what the AI did, who can see what, and how it ships. Those are the questions a buyer asks of any agent we build.