Expand description
Per-EdgeCategory weight multipliers — the calibratable w_τ from the
paper.
Each fine-grained edge type (130+ in weights.rs) carries a default
domain-prior weight. On top, every edge is scaled by a per-category
multiplier w_τ ∈ R_{≥0}, one per EdgeCategory variant (10 total).
These ten scalars are the parameters intended for offline calibration
(Bayesian opt / grid search) against a labeled corpus, as described in
paper Section 4.3 (Edge-Type Weight Calibration).
Default for every variant is 1.0 (no scaling — the fine-grained
prior weights from weights.rs apply unchanged).