86.7 → 77.4

The prior method counted repeated predictions as independent arrivals and treated large early departures as on time. The corrected method keeps the last prediction within ten minutes of each arrival and uses a two-sided five-minute window.

Detection source: Reconciliation of the durable GTFS-Realtime delay archive against distinct trip, stop, and service-date arrivals.

Methods: archived-prediction-rows-v0 → settled-v1.

Independent approval pending: Open Data Steward independent verifier.

Revision: correction:reliability:2026-08-04:settled-v1:revision-1.

• screen: complete — /en/corrections and /fr/corrections
• api: complete — /api/public/v1/corrections
• download: complete — /api/public/v1/corrections?format=csv
• method: complete — /en/city, /fr/city, /en/methodology and /fr/methodology
• marketing: complete — controlled fact network-reliability-method channel derivatives
• grant: complete — apps/otranspo/docs/PUBLIC-CORRECTION-DRILL-2026-08-12.json
• archive: complete — apps/otranspo/docs/PUBLIC-CORRECTION-DRILL-2026-08-12.json

• Versioned JSON record
• Download CSV