kalanosdev/lerobot-grader

Kalanos LeRobot Dataset Grader

Enter any Hugging Face LeRobot dataset ref (e.g. lerobot/aloha_sim_insertion_human) and get its Kalanos RDQ (Robotics Data Quality) report card.

• —If the dataset has already been graded, the Space fetches the published grade from [grade.kalanos.ai](https://grade.kalanos.ai).
• —If not, it runs the metadata-only grader on the dataset's public HF metadata (meta/info.json plus repo facts) and shows a provisional report card.

Metadata-only (v0). Scores reflect declared metadata quality, not measured sensor data. sync and outliers are unmeasured (shown as "—"). See the METADATA-ONLY note on every report.

Links

• —Registry — all graded datasets: https://grade.kalanos.ai/datasets
• —A report page: https://grade.kalanos.ai/d/hf/lerobot/alohasiminsertion_human
• —Project: https://kalanos.ai · Code: https://github.com/KalanosAI

Files

• —app.py — Gradio UI, published-grade lookup, and the metadata-grade wrapper
• —grading.py — canonical metadata-only scoring, shared verbatim with the batch grader so a grade shown here matches the one published on grade.kalanos.ai
• —requirements.txt — dependencies

Config

• —KALANOS_API (optional var) — base URL for lookups, default https://grade.kalanos.ai
• —INGEST_API_KEY (optional secret) — enables the "Publish to grade.kalanos.ai" button
• —SIZE_WALK_DEADLINE_S (optional var) — cap in seconds for the repo-size lookup on huge repos; default 12 (size never affects the score)