Attention fleet policy
What it computes
PyTorch attention policy, ONNX
Baseline and comparison design
Baseline: SPTF, LP and M4
Metric: throughput (t/h), higher is better
Pairing unit: seed, n = 30
Pre-registered acceptance criterion
same decision rule against SPTF, LP and M4 (>= 30 paired seeds)
SC-007-02 of 007-haulage-dispatch
Lane
Lanes: precompute, live
estimate
Runs live in the Simulate tab of
Documentation page: docs/methods/m05-attention-fleet-policy.md
Status: Not yet run