› › › ›

Step 1 — Ask

What do you want to test?

Type a category in plain language. The scout checks what your library already covers, tells you what's missing, and drafts new tests — each shaped like an agentic banking scenario.

Step 2 — Drafts

Review the drafts

Each draft is shaped like an agentic test: what it checks, how it runs (single-turn or tool-using agent), and how it is graded. Uncheck any you don't want.

0 drafts selected

Step 3 — Quality

How much can you trust these tests?

Each test is scored on the five dimensions from the AI Evaluation Quality Scorecard (San Joaquin et al., Feb 2026). The suite grade is set by its weakest dimension.

Step 4 — Run

Pick the models, run the agentic tests

Choose models to compare. Each is run as an agent against every selected test, then a judge model grades the trajectory.

Target system
full agent system · bank + vendor shared (BNM RMiT)
Model under test
Calls go to your chosen provider in Settings
cheap & free only
Ready to run

Step 5 — Review

Every model, every test, judged

Trajectory verdicts plus the draft-quality grade of each test, in one place. Full steps are hidden by default.

Everything you've saved

Previously scored runs and saved drafts from this browser.

Settings & keys

Keys live in this browser's localStorage only — never sent anywhere except the provider endpoint you choose.

Model provider
Plaid Sandbox (agentic tool backing)