Skip to main content

Overview

Run one explicit Decision provider against a small, versioned support-triage corpus. The JSON report keeps native and projected semantics visible and reports quality, calibration, latency, failures, retries, and an illustrative human review policy separately. The corpus demonstrates the mechanics. Replace it with representative, independently labeled application data before making a production decision. Execute the evaluator with a bundled preset name:
Other bundled presets work the same way, for example fastino with FASTINO_API_KEY set. A cold GLiDE model can make the first case slow or retried; see the Fastino guide for warming-aware timeouts and retries.

Example