About
The control that enforces the FCRA on AI screening agents.
Kōtsū is built for consumer reporting agencies deploying AI in their screening workflows — background checks, tenant screening, and the adverse-action decisions that turn on them.
It sits between your screening agent and the systems it acts on: blocking an action that would violate the FCRA before it executes, ensuring every required notice is issued, and emitting a cryptographically signed certificate for each file — verifiable offline by a CFPB examiner or opposing counsel.
What we've built
Kōtsū enforces the FCRA where it actually applies — at the moment a screening agent acts. The controls a CFPB examiner or a plaintiff's attorney asks about — permissible purpose, the adverse-action sequence, maximum-possible-accuracy, dispute reinvestigation — are enforced in the system, not requested of the model. Each is translated from the statute and your own adjudication policy into a control that gates the action at the moment it applies, and each produces a per-file conformance certificate an examiner can verify offline.
Values
These are operating values, not aspirational ones. They exist to constrain how we make decisions when the situation is ambiguous — when two reasonable things pull against each other, when the dashboard says one thing and the people doing the work say another, when a decision in front of us has consequences for someone who isn't in the room.
Each value rules things out. That's the test. A value that doesn't rule anything out isn't doing its job.
- 01
Trust the people doing the work.
When the plan and the reality disagree, reality wins. The people closest to the work know things the plan doesn't. Before you "fix" what they're doing, understand why they're doing it. Workarounds are signal, not deviation — they're telling you something true about the system.
Rules out Designing controls, processes, or policies for an imagined version of the work rather than the real one. Treating operator adaptation as non-compliance to be corrected. Mistaking the dashboard for the territory. Believing that because something should work a certain way, it does.
- 02
Acknowledge what you don't know.
Every tool, dashboard, framework, and process shows you some things by hiding others. Stay curious about what your view leaves out. The thing that surprises you later was usually invisible from where you were standing. The most dangerous condition is confidence that your view is complete.
Rules out Treating the absence of evidence as evidence of absence. Mistaking the metrics you measure for the reality you're trying to understand. The instinct to defend your model when reality contradicts it instead of updating the model. Confidence that scales with seniority rather than with actual visibility.
- 03
Choose challenging.
When the easy path and the hard-but-right path diverge, take the hard one. The challenging choice — the deeper check, the harder conversation, the problem no one wants to own — is usually the one that compounds, and the easy default is usually how standards quietly erode. We choose difficulty when the work honestly requires it — never for its own sake.
Rules out Taking the convenient path and calling it pragmatic. Shipping the shallow version because the rigorous one is harder. Ducking the conversation, the edge case, or the unglamorous problem in the hope that someone else picks it up. Mistaking ease for efficiency — and its inverse, choosing difficulty for sport when the simple path is the honest one.
- 04
Ask who pays.
Every decision has someone who absorbs the cost. Sometimes it's us. Sometimes it's a customer, a teammate, a future hire who isn't here yet, a community we serve. Before you decide, name who's on the hook — and ask whether they'd agree if they were in the room.
Rules out Decisions that look reasonable inside the room but transfer their consequences to people who weren't consulted, can't push back, and didn't agree to bear them. "Risk acceptance" by people who won't carry the risk. Tradeoffs that are clean for the decider precisely because the cost lands somewhere else.
- 05
Think and act considerately.
Consideration is two steps, not one: think before you act, and let the thinking change the act. Before a decision lands on someone — a teammate, a customer, the person described in a file — consider how it reaches them and choose the version that respects them. Speed is no excuse for carelessness, and good intentions don't count unless they show up in the actual decision.
Rules out Acting first and considering later. Thoughtfulness that stays in your head and never changes what you do. Treating people as throughput — a ticket, a record, a number on a dashboard. Curtness defended as efficiency or honesty. Decisions that are technically defensible but land on someone as indifference.