Leaderboard¶

The leaderboard is the entry index for complete planning procedures on PASSAGE v1.0, scored with passage-eval. Its checked-in index is src/passage-eval/results/leaderboard.csv in the source checkout. Public rows omit host and user names.

Compare the six primary metrics on matching tasks: median and p90 cost regret, average symmetric surface distance (ASSD), percentile Hausdorff distance, success rate and obstacle-violation rate. Compare latency separately by hardware, runtime and timing boundary. See the evaluation protocol.

Paper results¶

The paper results cover the complete 40,000-scene 256×256 public test split under elevation. They are exploratory after prior test exposure. Reporting a measurement in the paper does not make it a submitted or verified leaderboard entry.

Entry status¶

• Submitted means the submitter declared a score and its receipt.
• Verified means the leaderboard operator reviewed the measurement evidence and admitted the identified result under the applicable procedure.

A local score, an all-green test admission gate, or an evaluator setting named verified-candidate-run does not grant that status. Verification is also separate from reproducible-entry qualification or confirmatory statistical evidence. The Hugging Face Hub's badge for evaluations run in HF Jobs uses a separate definition.

How to add an entry¶

Evaluate your planner to produce local scores and receipts. The repository's CLI does not publish official entries, and the site has no submission service. Maintainers can regenerate this view from the checked-in index with make docs-prepare.