shadow-eval-dashboard

Backend for a regulator-facing dashboard that surfaces continuous shadow-eval disagreement rates between the production model and a challenger model.

Routes

Mounted under /api/plugins/shadow-eval-dashboard/.

| Route | Description | | ------------------------ | ---------------------------------------- | | /disagreement-rate | Aggregate rate over the last N days. | | /trends | Per-day disagreement rate trend. | | /disagreements | Paged list of flagged samples. |

Configuration

| Env var | Required | Description | | --------------------- | -------- | ------------------------------------ | | SHADOW_EVAL_DB_URL | yes | DSN for shadow eval sample store. |

UI

This scaffold ships backend-only. When a UI bundle is added, build it with the workbench design system (React + shadcn/ui), set has_ui: true in dryade.json, and add a matching ui_bundle_hash.

Tier

enterprise