No. It's a layer before — to reduce the number of expensive hypotheses that make it to the field. You simulate dozens of scenarios, then take only the survivors into fieldwork. Real research gets sharper focus, not eliminated.
No. They are cognitive architectures calibrated against population strata and the behavioral literature — we never model an identifiable individual. The unit is the cohort, not the person.
Every output arrives as an uncertainty band, not a single number, and comes with a provenance card listing calibration sources and a sensitivity analysis. We treat this as a scientific instrument, not a crystal ball.
Because it's the hardest test. A model that holds up in markets as heterogeneous as Brazil's is more honest than one calibrated on a homogeneous audience. The cultural density here is our calibration, not our boundary.
Predictive policing, partisan electoral targeting, pricing that violates LGPD, models of workers without consent. The refusal ledger is public — "no" is part of the product.
By invitation only, in cohorts. We respond to a small number of requests at a time to keep calibration close and care high. Scope is defined by decision, not by seat.