Results

Every number here is computed from recorded experiment runs when the site is built, never typed. A rung that has not run says so; a rung with too few seeds says so; a rung measured on different data is not set beside the others as though it were comparable.

Metric: average_precision. Interval: 95% t-interval across seeds, mean ± t(0.975, n−1)·s/√n. The protocol requires at least 5 seeds per rung.

0 Base rate

not yet run

1 Degree-only null

not yet run

2 Non-graph model

not yet run

2b Graph-topology GBM

not yet run

3 Shallow KGE

not yet run

4 Relation-agnostic GNN

not yet run

5 Relation-aware GNN

not yet run