evals.levyo.co.uk

Levyo Evals Admin

Internal review surfaces for Levyo evals, labels, and benchmark evidence. Use this hub as the stable entry point for current and future eval packs.

Eval OS

Flywheel cockpit: runtime, evals, Promptfoo Cloud proof, release gates, cycles, governance, and tracked repo inventory.

flywheel v1 Repo Inventory Promptfoo Cloud

Open Eval OS

Benchmark v0

50-case benchmark surface: 30 text-only core cases, 20 image-attached cases, runtime outputs, and recovery proof.

ready for review 50 cases 50/50 outputs

Open benchmark

Benchmark Labels

Dominik labeling queue for all 50 Benchmark v0 cases with Headline, Reference Use, Wearability/Taste, Broken Image Text, Intent Fulfillment and Hard-Fail/Production.

label here 50 cases six heads

Open 50 labels

M8 20 Label Queue

Dedicated labeling view for the 20 real Levyo runtime shirts generated from the M8 image-intent calibration pack.

label here 20 cases autosave

Open 20 labels

Promptfoo

Real Promptfoo benchmark report for the 50 locked Benchmark v0 cases, plus the no-spend media packet and scorecard.

50 cases eval report

Open Promptfoo

Calibration

Review the six candidate Dominik label blocks before they are used for Judge calibration or root-cause work.

review needed 6 sources label origin

Open calibration

M5 Media View

Image-visible no-spend review packet for the 50-case benchmark. Supporting evidence, not a Promptfoo substitute.

50 cases media packet

Open media view

Image Addition Labels

Existing Dominik labeling surface for image-addition review artifacts generated through the internal eval path.

existing surface label review

Open labels

Design Reviews

Admin-only queue for Levyo Maker design-review feedback and publish responses.

Firebase admin review queue

Open queue

Updated 2026-06-29T01:00:00.000Z. Static eval/admin surface only; no customer-live flow, Shopify surface, API spend, or registry promotion is changed by these pages.