Methodology · · 6 min read
Calibration Is the Product
A single accuracy number across every audience and every question is a marketing artifact, not a methodology. Real calibration is per-project, per-audience, per-question — and it ships with every deliverable.
Why one number can't be honest
Synthetic fidelity depends on the audience being modeled, the question being asked, and the data the model was grounded against. A vendor reporting eighty-eight percent accuracy is reporting one number across one methodology choice, usually correlation against a specific survey type. That number tells you almost nothing about how the model will perform on your audience and your question.
We report fidelity per project, against your data, on the questions you care about. The number is smaller in scope. It is also more useful.
What a calibration appendix actually contains
A persona-level grounding ledger. A per-question benchmark trace. A coverage-gap declaration for audiences where the data was thin. A divergence log documenting every place the model disagreed with traditional research, with our reading of why.
This appendix is not the marketing version. It is the working document insights and procurement teams ask for, and it ships with every engagement. The customers who read it first are the ones who get the most out of the platform.