Definition
Ratio of delay-minutes-per-train-mile absorbed by each train class compared to that class's share of network traffic, revealing whether the AI dispatcher systematically protects some service categories at the expense of others.
The Measurement Gap
AI dispatching systems make thousands of priority decisions daily, which train gets the meet, which train holds, which class absorbs the delay when two movements conflict. These decisions are invisible to On-Time Performance (OTP) because OTP measures aggregate schedule adherence, not the distribution of delay across service classes. A railroad can report strong OTP while one train class systematically absorbs disproportionate delay to protect another. Terminal dwell measures throughput at terminals but cannot see whether the dispatcher's upstream priority decisions loaded one class with delay before it ever reached the terminal. Train velocity measures network fluidity but treats all trains equally: it cannot reveal that intermodal trains move at planned speed while manifest freight absorbs all the meets and passes that make that intermodal performance possible. The failure mode is a pattern of systematic, invisible cross-subsidy between service classes that never appears on any dashboard because no legacy metric operates in the distribution dimension.
Decision Supported:
Is the AI dispatcher distributing delay equitably across train classes, or is it creating invisible winners and losers that aggregate OTP masks?
Formula
Plain language: For each train class (intermodal, manifest, unit train, passenger, local), compute delay minutes per train-mile. Divide each class's delay rate by the network-wide average delay rate. A ratio of 1.0 means that class absorbs its proportional share. Above 1.0 means disproportionate delay. Below 1.0 means the class is being protected.
Notation:
TCDEI_c = (Delay_c / TrainMiles_c) / (Delay_total / TrainMiles_total)
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Delay_c | Total delay minutes attributed to train class c in the measurement period | Minutes |
| TrainMiles_c | Total train-miles operated by class c | Miles |
| Delay_total | Total delay minutes across all classes | Minutes |
| TrainMiles_total | Total train-miles across all classes | Miles |
Report as a vector across all classes: the useful output is the distribution, not any single number. Values above 1.0 indicate that class absorbs excess delay; below 1.0 indicates protection.
Proposed formula, pending validation pilots.