What is calibrated Decision Mode?
Calibrated Decision Mode evaluates a specified rubric against supplied row context and returns probabilities for possible outcomes. A team can set thresholds to keep, reject, or review uncertain rows in a workflow. Calibration helps interpret scores across cases, but a probability is not a guarantee and depends on the evidence and rubric.
How is it different from generated copy?
Sorzio's Decision Mode uses typesafe-ai/jev-1.14.0, a non-generative evaluation model. It is suited to a defined question and answer set, while generative AI produces text or structured content. Give the evaluator evidence it can assess and specify how to handle missing or ambiguous context.
| Use | Output | Next step |
|---|---|---|
| Decision Mode | Probabilities for a defined rubric | Apply reviewed thresholds |
| Generative AI | New text or structured content | Review facts and formatting |
The base fee is $1.00 per 1,000 rows plus 1.25 times input token cost, approximately $1.01 per 1,000 in the reference example. Full-run checkpoints can filter downstream work, while test runs mark verdicts without stopping later test steps. Read the ICP scoring guide for a worked example.