Evaluation Workbench
ExamplesEvaluate selected rows, switch local rule providers and criteria, and recover from simulated rate limits, timeouts, save failures and cancellation without AI calls.
Row evaluation with bounded concurrency, cancellation, stale results after editing, and review filtering with mock and authenticated live Jev modes, saved results, and detailed distributions.
Change the width and theme, then interact with the components.
Evaluation studio
Evaluate answers against a reference, then focus on the rows that need attention.
0/3 complete · Needs review 3
Score and confidence are different. The 0.8 threshold is a demo policy; calibrate production thresholds on labeled data.
bun add @reopt-ai/opt-datagridimport { DataGrid, type DataGridColumn } from "@reopt-ai/opt-datagrid";
import { useDataGridRowEvaluation } from "@reopt-ai/opt-datagrid/evaluation";Evaluate selected rows, switch local rule providers and criteria, and recover from simulated rate limits, timeouts, save failures and cancellation without AI calls.
A table a real model writes while you watch: YAML becomes patches, changes wait for review, edits are never overwritten, and formulas, sorting and per-row AI columns take one line.
Starter spreadsheet grid with editable columns and commit feedback.