Dimension · Quality

Quality

Does the model do the application job correctly and reliably?

What this page will show

planned

Finboard's own gates, linked from its hosted Eval Ledger with the model and prompt provenance of each gate run. The Lab re-renders none of the ledger's numbers.

Tool selection Whether the assistant picks the right capability and binds its arguments.
Multi-turn context Whether context stays reliable across the turns of a conversation.
Narration Whether the narrated answer stays faithful, and how a failure is analysed.

The ledger labels each gate planned measured validated

Open the Eval Ledger ↗

Why it matters

Quality begins at application behaviour, not at a generic benchmark score. A quality gate must run the application's own prompt assembly to test that application, so Finboard's harness stays in Finboard, and the Lab links its record instead of re-running it. Apps share run identity and keep their own domain payloads; there is no single cross-app quality score.

First release

The first release of this page links Finboard's hosted Eval Ledger: its selection, multi-turn and narration gates, with the model and prompt provenance of each gate run.

run evidence

Quality has no payload schema yet, so the Lab shows no Quality value of its own: the app's record is linked below. Each committed run is listed here with its own reason. To read a bundle of your own, open it on the Performance page.

20261022T064227Z-finboard-runtime-direct absent — a lab Performance capture carries no Quality payload text retained false experiments/public/finboard/20261022T064227Z-finboard-runtime-direct/bundle.json

linked evidence

Finboard's Eval Ledger is the Quality record for Finboard. It labels each gate with the status vocabulary planned · measured · validated. The Lab links it and re-renders none of its numbers.