Definition
Ratio of downstream conflict events created by a reroute decision to the conflict events the reroute resolved, measuring whether AI dispatching decisions propagate or absorb network disruption.
The Measurement Gap
When the AI dispatcher reroutes a train to resolve a conflict, it eliminates meets, passes, or terminal approach delays at the original conflict point. But the rerouted train arrives on an alternate route where it may create new conflicts with trains that had clear paths before. OTP cannot see this because it measures whether individual trains arrive on time, not whether the dispatcher's decisions created a net reduction or amplification of conflict across the network. Train velocity captures network-wide throughput but cannot attribute throughput degradation to specific reroute decisions. Terminal dwell measures time spent at terminals but cannot trace extended dwell back to a cascade that originated with a reroute three territories upstream. The failure mode is a dispatcher that appears to resolve conflicts locally while systematically amplifying them globally: each reroute looks like a good decision in isolation, but the aggregate effect destabilizes the network.
Decision Supported:
When the AI dispatcher reroutes a train to resolve a conflict, does the reroute create more problems than it solves? Should the system's reroute threshold be adjusted?
Formula
Plain language: For each reroute decision, count the conflict events (meets, passes, terminal approach delays) on the alternate route within a downstream window that were not present before the reroute. Divide by the number of conflict events the reroute resolved on the original route.
Notation:
RCAF = Conflicts_created_downstream / Conflicts_resolved_upstream
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Conflicts_created_downstream | New meet/pass/terminal conflicts attributable to the rerouted train within the downstream assessment window (recommended: 6 hours post-reroute) | Count |
| Conflicts_resolved_upstream | Conflict events at the original conflict point that the reroute eliminated | Count |
Values below 1.0 indicate the reroute reduced net conflicts (good). Equal to 1.0 is break-even. Above 1.0 means the reroute amplified conflicts (bad). Average across all reroutes in the period.
Proposed formula, pending validation pilots.