I encountered an extreme DTC bottleneck in one of the 5-men fairy end-games I calculated: King + Bishop + Commoner against King + Nightrider. The stats for this end-game are:
Code: Select all
KBM_KH.dtc
WON.wtm 879,668,832
K capture 278,156,480
imm.conv. 469,199,456
other 410,469,376
0 731,929,356
0 65,626,988
1 896,664
2 268,028
3 155,080
4 122,900
5 124,704
6 120,892
7 123,168
8 121,744
9 118,088
10 117,920
11 112,816
12 109,440
13 100,492
14 99,780
15 95,136
16 96,992
17 97,340
18 99,396
19 107,572
20 119,656
21 131,564
22 137,048
23 142,068
24 143,928
25 144,848
26 133,756
27 123,152
28 114,728
29 112,080
30 110,368
31 108,136
32 98,352
33 90,064
34 81,288
35 66,320
36 53,976
37 51,912
38 53,576
39 56,728
40 57,280
41 60,248
42 64,000
43 63,312
44 62,208
45 52,440
46 46,144
47 44,448
48 41,176
49 36,188
50 30,296
51 28,160
52 23,104
53 16,952
54 13,072
55 8,512
56 6,404
57 5,392
58 4,320
59 4,136
60 3,584
61 3,576
62 2,904
63 2,752
64 2,232
65 2,824
66 4,320
67 4,608
68 5,256
69 5,632
70 6,512
71 8,080
72 10,032
100 35,072
120 185,528
140 72,248
160 43,168
180 44,560
200 84,956
220 499,428
240 1,210,036
260 2,516,044
280 8,922,756
300 288,592
301 117,912
302 41,816
303 16,256
304 5,296
305 1,272
306 216
307 40
WON.btm 347,224,148
stalemate 560
W check 143,965,980
LEGAL 770,975,460
TOTAL 914,941,440
As can be seen, 96% of all pseudo-legal positions is won with white to move, so this is a generally won end-game. As usual, the number of positions won for white with btm is much smaller than this (only 38%). In the other positions white is in check (illegal) or loses its Commoner through a trivial tactic (e.g. a Nightrider fork or skewer, leading to loss or trade), or has a hanging Bishop. But the wtm positions have a similar number of trivial wins, where a black piece was hanging, or the position was outright illegal (black was in check).
Of the positions that white can win with btm, about 7% are checkmates or immediate conversions (e.g. where black is in check, and its Nightrider is hanging). Almost none of those can be forced: the number of conversion-in-1 positions is 80 times lower.
The remarkable thing about this end-game is that there are only 2,232 DTC=64 positions, while 98% of the non-trivial wins with btm (i.e. DTC >= 1) has DTC > 64. The longest win has DTC=307, while the average DTC (of the non-trivially won btm positions) is 264.5 (220 if you count the trivial wins).
The DTC=64 positions thus form a very spectacular bottleneck; the number of positions that they would cut off from a win is 140,000 times their own number for btm positions, and about as much wtm positions on top of it.