2024 · dataset · Lead & data pipeline

Corridor baseline atlas

Per-chokepoint transit baselines for the region's main straits — the reference layer the detectors score against.

Built with
  • DuckDB
  • Python
  • Parquet

Problem

Every anomaly method needs a baseline, and most are built ad hoc and thrown away. Without a shared, versioned reference for “ordinary” traffic, no two anomaly scores are comparable.

Approach

The atlas builds and versions a transit baseline per chokepoint — who passes, how fast, how often they report — so detectors score against a common reference rather than a private one.

  • Aggregate cleaned tracks into per-corridor rhythms.
  • Version each baseline so results are reproducible over time.
  • Expose the baselines as the reference layer for the detector.

Outcome

A reusable reference layer that makes anomaly scores comparable across analysts and across seasons.