This section provides a live view of the modelling pipeline’s performance and status, sourced directly from the latest validation artefacts and provenance records.
| Component | Current status | Latest evidence | Caveat |
|---|---|---|---|
| Stage 1a: AADT model | Implemented | Local R²: 0.831, Spatial R²: 0.792, N=17412 | Estimates traffic where direct counts are unavailable |
| Stage 2: GLM collision-risk model | Implemented | Pseudo-R²: 0.347 | Interpretable exposure-adjusted baseline |
| Stage 2: XGBoost collision-risk model | Implemented across 5 seeds | Mean Pseudo-R²: 0.323, Seeds: [42, 43, 44, 45, 46] | Flexible comparison/diagnostic model; weaker interpretability than GLM |
| Final scored output | Implemented | 2,167,557 scored links | Current scoring grain is one row per road link, not link-year |
- Latest model run: 2026-05-03T13:25:09.900667+00:00
- Git SHA:
301a766b - Sources read: rank_stability_provenance.json, aadt_validation_summary.parquet
“These figures describe the latest model artefacts available at documentation render time. They indicate the current analytical state of the exposure-adjusted risk pipeline; they should not be read as operational approval for safety decisions.”