Clarification on leaderboard aggregation formulas (AS Score, Correctness, fast_1, G-Mean Speedup)

[Lolo1222]

Hello, thanks for releasing PassNet.

I have a reproducibility question about the leaderboard / Table 1 aggregation formulas.

From the paper and code, I understand the within-sample computation of ES(t) and AS. However, I could not find the exact cross-sample aggregation used to produce the final leaderboard metrics.

Could you please clarify:

1. How is model-level AS Score aggregated from the 200 evaluation samples?
2. Are Correctness and fast_1 computed globally over all 2,060 subgraphs, or averaged from per-sample values?

Thanks a lot.