Experiments

stops	algorithm	expanded	generated	cost_min	gap_pct	time_ms
6	UCS	95.5 ± 16.5	106.5 ± 22.2	224.8 ± 54.9	0 ± 0	0.6274 ± 0.133
6	Greedy	64.7 ± 16.4	70.8 ± 16.2	239.4 ± 56.7	6.655 ± 5.72	2.346 ± 0.699
6	A* (max-incoming h)	51.3 ± 17.9	86.1 ± 19.2	224.8 ± 54.9	0 ± 0	1.936 ± 0.411
6	A* (sum/wait h)	34 ± 20.4	63.7 ± 21.6	224.8 ± 54.9	0 ± 0	1.823 ± 0.824
7	UCS	184.3 ± 23.1	202.6 ± 31.2	258.7 ± 63.1	0 ± 0	1.447 ± 0.287
7	Greedy	136.2 ± 51.6	143.3 ± 52.1	265.2 ± 66.4	2.433 ± 3.09	5.242 ± 2.12
7	A* (max-incoming h)	123.9 ± 17.1	187.9 ± 22.2	258.7 ± 63.1	0 ± 0	4.983 ± 0.895
7	A* (sum/wait h)	92.1 ± 27.3	159.4 ± 28.3	258.7 ± 63.1	0 ± 0	5.485 ± 2.52
8	UCS	356.7 ± 53.3	425.4 ± 75.2	301.3 ± 57.4	0 ± 0	2.798 ± 0.582
8	Greedy	326.4 ± 104	333.2 ± 105	311.6 ± 55.7	3.651 ± 3.85	10.53 ± 3.23
8	A* (max-incoming h)	213.8 ± 70.7	363.6 ± 58.9	301.3 ± 57.4	0 ± 0	8.534 ± 1.51
8	A* (sum/wait h)	109.3 ± 60.8	227.4 ± 68.9	301.3 ± 57.4	0 ± 0	6.675 ± 2.21
9	UCS	777.1 ± 108	889.2 ± 145	289.9 ± 56.3	0 ± 0	6.422 ± 1.16
9	Greedy	775.2 ± 221	784.1 ± 223	304.2 ± 53.9	5.424 ± 5.1	24.59 ± 7.35
9	A* (max-incoming h)	478 ± 77.5	776.4 ± 126	289.9 ± 56.3	0 ± 0	19.02 ± 3.47
9	A* (sum/wait h)	254.3 ± 56.9	526.7 ± 73.7	289.9 ± 56.3	0 ± 0	15.07 ± 2.28
10	UCS	1519 ± 161	1717 ± 301	334.4 ± 36.6	0 ± 0	13.55 ± 1.49
10	Greedy	1312 ± 440	1322 ± 441	349.8 ± 40.5	4.612 ± 4.21	42.08 ± 13.6
10	A* (max-incoming h)	865.4 ± 180	1439 ± 184	334.4 ± 36.6	0 ± 0	36.08 ± 4.12
10	A* (sum/wait h)	368.7 ± 249	847.7 ± 293	334.4 ± 36.6	0 ± 0	24.69 ± 9.73

A* vs brute force (6-8 stops): 30/30 identical optimal costs.

orders	solver	assignments	assignments median	backtracks	checks	time_ms	nn_edges	solved/unsat/limit
15	BT	30.1 ± 28.4	16.5	15.1 ± 28.4	53.3 ± 70.2	0.08177 ± 0.0695	2.1 ± 3.08	10/0/0
15	BT+FC	28.8 ± 27.4	16.0	13.8 ± 27.4	0 ± 0	0.07832 ± 0.0578	2.1 ± 3.08	10/0/0
15	BT+FC+AC3	28.6 ± 27.5	15.5	13.6 ± 27.5	0 ± 0	0.3828 ± 0.861	2.1 ± 3.08	10/0/0
15	BT+FC+AC3 +degree order	15 ± 0	15.0	0 ± 0	0 ± 0	0.06435 ± 0.00557	2.1 ± 3.08	10/0/0
20	BT	445.5 ± 1.23e+03	24.0	425.5 ± 1.23e+03	1038 ± 2.92e+03	1.158 ± 3.14	4.5 ± 4.13	10/0/0
20	BT+FC	192.4 ± 503	23.5	172.4 ± 503	0 ± 0	0.4694 ± 1.02	4.5 ± 4.13	10/0/0
20	BT+FC+AC3	140.1 ± 349	23.0	120.1 ± 349	0 ± 0	0.3777 ± 0.842	4.5 ± 4.13	10/0/0
20	BT+FC+AC3 +degree order	20.2 ± 0.4	20.0	0.2 ± 0.4	0 ± 0	0.1188 ± 0.0756	4.5 ± 4.13	10/0/0
30	BT	4.089e+04 ± 7.96e+04	165.5	4.087e+04 ± 7.96e+04	1.08e+05 ± 2.1e+05	99.93 ± 194	7.9 ± 6.47	8/0/2
30	BT+FC	4.051e+04 ± 7.98e+04	105.5	4.049e+04 ± 7.98e+04	0 ± 0	117.9 ± 238	7.9 ± 6.47	8/0/2
30	BT+FC+AC3	2.203e+04 ± 5.95e+04	104.5	2.2e+04 ± 5.95e+04	0 ± 0	66.14 ± 164	7.9 ± 6.47	8/1/1
30	BT+FC+AC3 +degree order	4475 ± 1.32e+04	36.5	4448 ± 1.32e+04	0 ± 0	11.56 ± 33.6	7.9 ± 6.47	9/1/0

scenario	model	replay_late	replay_late_pct	replay_late_minutes	planned_late	planned_route_min	learned_rules	plans ok
standard windows 30-120 min	NN (MLP)	0.31 ± 0.486	1.55 ± 2.43	7.794 ± 18.3	0 ± 0	1172 ± 75.2	0 ± 0	10/10
standard windows 30-120 min	hist-avg	0.3933 ± 0.499	1.967 ± 2.5	6.985 ± 11	0 ± 0	1172 ± 75.5	0 ± 0	10/10
standard windows 30-120 min	distance-linear	0.3067 ± 0.46	1.533 ± 2.3	5.061 ± 9.86	0 ± 0	1159 ± 74.1	0 ± 0	10/10
tight windows 15-30 min	NN (MLP)	1.043 ± 0.69	5.217 ± 3.45	9.115 ± 6.72	0 ± 0	1300 ± 71	0.6 ± 1.2	10/10
tight windows 15-30 min	hist-avg	1.01 ± 1.04	5.05 ± 5.19	10.76 ± 12.7	0 ± 0	1299 ± 67.7	0.4 ± 0.917	10/10
tight windows 15-30 min	distance-linear	1.617 ± 1.21	8.083 ± 6.07	17.52 ± 15.1	0.1 ± 0.3	1288 ± 70.6	0.8 ± 1.66	9/10

Replay pool back-offs: {'hour+-1': 3099, 'all-hours': 667}