Case study · Data
Market analytics and dispatch simulation
Implemented · 1,795 tests collected · local
This system watches a live freight market and turns thousands of postings into plain answers: which markets are strong, which lanes pay, which brokers are reliable and which loads deserve a call. It also runs a simulated fleet to test whether smarter routing actually beats picking the nearest load.
Overview
A dependency-free Python 3.12 system with a clean, test-enforced layered structure. It ingests a live load-board stream produced by a separate planning engine. Each upstream source is retained by content hash before anything is derived from it, and unseen inputs are replayed through the engine's own scorer, so economics have exactly one authority.
An append-only archive with idempotent re-ingest feeds about 45 command-line analyses, which produce roughly 60 versioned JSON document types: market and lane intelligence, broker profiles, load ratings, daily briefs, region maps and multi-leg network routes. A Next.js console renders those documents.
A deterministic discrete-event simulator dispatches a seeded fleet over live or replayed boards with no lookahead, explains every decision, and compares itself against a manual "nearest legal load" baseline. Type checking is strict, and the suite collects 1,795 tests.
Visual evidence
Dashboard, daily brief and region map. Recreated from the source with synthetic data; not a screenshot.