← Back

Fast Track: where playing to win lost (seed 69)

The 80-game run of "own less the leader" against the champion had 9 games that red won only as the champion. This note takes one of them, seed 69, apart. A short table of the other eight is at the end.

How it was run (fasttrack/exp_pivot.roc):

Board values are each player's own, as the champion values its pieces: Steve's square ranking (100 points a place) plus the base bonus of 1000 a step.

Who won

red plays as winner red home on
the champion red, on turn 10 turn 10
the chaser blue, on turn 17 turn 18

The pivot: red's turn 6

Red holds 7 2 9 6 6. Board values before the turn:

red blue green purple
15600 5500 5700 18400 (the leader)

Both strategies play the same cards: a 6 brings a piece out of red's pen to L0, and then the 7 is split. The only difference is how the 7 splits:

For the two players in the capture, red and purple:

red purple
before 15600 18400
champion's pick (B3, L2) 21100 (+5500) 18400 (0)
chaser's pick (DS, FT, capture) 20300 (+4700) 13000 (−5400)

So the capture costs red 800 of its own value and takes 5400 off purple.

How each strategy scores the two lines:

Every line red had

All 30 lines of red's turn 6, sorted by the champion's score. Each pair of rows with the same squares is one position the search failed to merge; that is a separate issue, covered at the end.

line pieces red purple (leader) others champion's score chaser's score
champion's pick red@red.R0 red@red.HP2 -> red@red.B3 red@red.L2 +5500 0 21100 2700
red@red.R0 red@red.HP2 -> red@red.L2 red@red.B3 +5500 0 21100 2700
red@red.R0 red@red.HP2 -> red@red.B1 red@red.L4 +5200 0 20800 2400
red@red.R0 red@red.HP2 -> red@red.L4 red@red.B1 +5200 0 20800 2400
red@red.R0 red@red.HP2 -> red@red.B2 red@red.L3 +4700 0 20300 1900
red@red.R0 red@red.HP2 -> red@red.L3 red@red.B2 +4700 0 20300 1900
chaser's pick purple@red.FT red@red.R0 red@red.HP2 -> red@red.FT purple@purple.HP2 red@red.DS +4700 −5400 20300 7300
red@red.R0 red@red.HP2 -> red@red.B3 red@blue.R2 +4700 0 20300 1900
red@red.R0 red@red.HP2 -> red@blue.R2 red@red.B3 +4700 0 20300 1900
red@red.R0 red@red.HP2 -> red@red.B2 red@blue.R1 +4600 0 20200 1800
red@red.R0 red@red.HP2 -> red@blue.R1 red@red.B2 +4600 0 20200 1800
red@red.R0 red@red.HP2 -> red@bullseye red@red.BR +4300 0 19900 1500
red@red.R0 red@red.HP2 -> red@red.DS red@bullseye +4200 0 19800 1400
blue@blue.R0 red@red.R0 red@red.HP2 -> red@red.B1 red@blue.R0 blue@blue.HP4 +3100 0 blue −5500 18700 300
blue@blue.R0 red@red.R0 red@red.HP2 -> red@blue.R0 blue@blue.HP4 red@red.B1 +3100 0 blue −5500 18700 300
red@red.R0 red@red.HP2 -> red@red.DS red@red.L0 +2900 0 18500 100
red@red.HP2 -> red@blue.R1 +2200 0 17800 −600
red@red.HP2 -> red@red.L2 +2000 0 17600 −800
red@red.R0 red@red.HP2 -> red@red.DS red@blue.BR +1700 0 17300 −1100
red@red.R0 red@red.HP2 -> red@blue.BR red@red.DS +1700 0 17300 −1100
red@red.R0 red@red.HP2 -> red@red.BR red@blue.DS +1500 0 17100 −1300
red@red.R0 red@red.HP2 -> red@blue.DS red@red.BR +1500 0 17100 −1300
red@red.HP2 -> red@blue.R2 +1200 0 16800 −1600
red@red.HP2 -> red@blue.HH +1000 0 16600 −1800
red@red.HP2 -> red@blue.L1 +700 0 16300 −2100
red@red.HP2 -> red@blue.R3 +300 0 15900 −2500
red@red.R0 red@red.HP2 -> red@red.BR red@blue.R4 +300 0 15900 −2500
red@red.R0 red@red.HP2 -> red@blue.R4 red@red.BR +300 0 15900 −2500
red@red.R0 red@red.HP2 -> red@red.DS red@blue.R4 +200 0 15800 −2600
red@red.R0 -> red@red.DS +100 0 15700 −2700

One line captures blue instead, costing blue 5500. The chaser scores it only 300: blue is not the leader, so taking blue's piece is worth nothing to it.

What happened next

Board values at the start of each of red's turns, and every capture.

Red as the champion:

red's turn red blue green purple
7 21100 15600 12900 19500
8 26500 19100 16800 20100 red captures purple
9 28700 19100 19500 19100
10 32500 19100 23000 19100 red home: red wins

Red as the chaser:

red's turn red blue green purple
7 20300 15600 12900 12400
8 24900 19100 16800 12000 purple captures red
9 21900 19100 19500 14200
10 27400 19100 23000 18000
11 28700 21900 23200 19100
12 26500 24500 23200 19100
13 29400 27000 25500 19100 green captures red
14 27000 29300 29200 21900
15 27000 31600 31800 23700
16 27000 32700 31600 27000
17 29800 32200 32600 27000 blue home: blue wins
18 30900 33800 32400 31600 red home

Reading it

  1. The capture came instead of a piece tucked into base. The champion put a piece on B3, where no one can touch it. The chaser left red's pieces on DS and the fast-track square. Purple's pieces pass through red's zone on their way home.
  2. The capture was paid back almost at once. Two turns later purple captured red. Red went from 24900 to 21900, and red's lead over the field was gone by turn 14. Green captured red again on turn 13, and red sat at 27000 for three turns while blue and green went past.
  3. The champion took purple anyway, and for free. On turn 8, as the champion, red captured purple in passing, as part of its own best play. The chaser paid for a capture the champion got two turns later at no cost.
  4. The capture did not stop the leader from winning, because the leader changed. Purple was the leader at the pivot; it never threatened again. Blue, 5500 at the pivot and ignored by the chaser, won.

A caveat on cause. After the pivot the two games differ, and every later move is the champion's best play in a different position. The cards come in the same order, but they meet different boards. So the rest of each game is not caused by the pivot alone. What the pivot shows cleanly is the trade: red gave up a safe B3 for a capture that left two pieces exposed in an opponent's path.

The other eight games

The first point where red's two strategies part, and who won each game.

seed red's turn champion's pick chaser's pick (red / the leader) winner, red as champion winner, red as chaser
74 3 a piece to B1, keeping the A (+1400) captures green's only piece out, at green's DS (+2800 / −5600), spending the hoarded A red, turn 10 green, turn 14
21 8 a piece to B2 (+3700) captures blue on blue's R0 (+1800 / −5500) red, turn 12 purple, turn 15
22 2 B1 and a piece out (+8400) captures blue at BR, and a J trade sends green back (+5100 / −5700) red, turn 10 purple, turn 18
43 3 a piece to R3 (−300) green's piece to its pen with no red piece moving (0 / −2800) red, turn 16 purple, turn 18
55 2 out, to FT (+4600) out, onto purple at the bullseye (+4100 / −4100) red, turn 14 green, turn 17
60 4 out, to R0 (+5500) captures blue on blue's FT (+3100 / −4600) red, turn 10 blue, turn 13
76 6 B2, capturing purple on the way (+3800) captures blue on blue's R0 (−1000 / −5500) red, turn 14 blue, turn 15
79 2 a piece to R3 (−800) J trade with green's piece at the bullseye (−1600 / −2600) red, turn 12 green, turn 15

The same pattern shows up in most of them: a base square or a safe advance given up for a capture of the leader. The winner is usually a player the chaser was not chasing at the pivot: in 5 of these 9 games someone other than the pivot's leader won.

Seed 43 looks wrong. The chaser's line sends green's piece home while every red piece ends where it started. With K J J 3 3 there might be a legal sequence, a J trade and back, but I have not worked it out. It is worth checking before trusting anything that involves the J.

The search does not merge identical positions. The seed 69 table lists the same squares twice in reverse order: red to B3 and L2, then red to L2 and B3. The search merges lines by comparing the whole game record, and these two probably differ only in the order the pieces are listed. If so, it changes no decision but multiplies the work. Seed 74 had 1,257 lines. This is the first thing to measure for speed.