Cedar Parcel Lab: fully generated operational learning data. One parcel per row. No personal data.
parcel_id: text unique key. dispatch_date: ISO date for chronological ordering, not a predictor.
distance_km: numeric kilometres known at dispatch; blanks are missing. load_ratio: numeric planned load known at dispatch; blanks missing.
zone: nominal operational category known at dispatch. actual_hours: numeric delivery duration outcome. late: 1 iff actual_hours > 24, else 0.
delivered_at: later illustrative date (day precision, not exact elapsed-time arithmetic). post_delivery_rating: synthetic later rating; excluded.
Pilot: 60 rows, first 40 train, last 20 test. Main: 180 rows, 108 train, 36 validation, 36 final test. All fields are fictional.
Reference models are learning implementations, not a service-level, causal, fairness or production-readiness claim.