UBC 2023 Final Day 1: Match 3, Game 6 - Score-based offense vs defense

Exciting matches today!

One moment I found interesting was in (Day 1) Match 3, Game 6 where there were a few decisions in sequence where Sander makes (apparent) errors by playing offensive (i.e., priming and blitzing) moves on his side instead of working on his (four) back checkers. Marc (on commentary) found the decisions interesting and was musing about why XG preferred connecting plays over attacking. Marc eventually attributes match score as the main probable cause (gammons being worth less for Sander at 3-away).

The action happens on the YouTube stream starting at 4:00:00 through 4:05:30.

Here is the last position in the series that had the most dramatic effect (this particular move happens at 4:01:37).


Sander plays 11/2* here where XG (2-ply?) prefers, at this score, the play 24/18 21/18.


Here is a rollout of the top four ++ moves at the score [edit]4-away/3-away 3-away/4-away:

    1. Rollout¹    11/2*                        eq:+0.925
      Player:   77.98% (G:45.00% B:9.54%)
      Opponent: 22.02% (G:4.80% B:0.42%)
      Confidence: ±0.011 (+0.914..+0.936) - [100.0%]

    2. Rollout¹    24/18 21/18                  eq:+0.894 (-0.031)
      Player:   78.85% (G:38.40% B:7.55%)
      Opponent: 21.15% (G:3.25% B:0.25%)
      Confidence: ±0.012 (+0.882..+0.907) - [0.0%]

    3. Rollout¹    24/15                        eq:+0.879 (-0.046)
      Player:   78.37% (G:40.64% B:10.12%)
      Opponent: 21.63% (G:3.99% B:0.31%)
      Confidence: ±0.014 (+0.866..+0.893) - [0.0%]

    4. Rollout¹    21/15 18/15                  eq:+0.851 (-0.074)
      Player:   77.50% (G:42.88% B:12.71%)
      Opponent: 22.50% (G:3.44% B:0.33%)
      Confidence: ±0.014 (+0.837..+0.866) - [0.0%]
      Duration: 2 inutes 08 seconds

    ¹1296 Games rolled with Variance Reduction.
      Moves: 3-ply, cube decisions: XG Roller

And here are the top two from above at the (more) normal score of 7-away/7-away:

    1. Rollout¹    11/2*                        eq:+1.180
      Player:   78.82% (G:41.86% B:3.39%)
      Opponent: 21.18% (G:4.45% B:0.33%)
      Confidence: ±0.009 (+1.172..+1.189) - [100.0%]

    2. Rollout¹    24/18 21/18                  eq:+1.106 (-0.074)
      Player:   80.13% (G:31.23% B:2.19%)
      Opponent: 19.87% (G:3.47% B:0.17%)
      Confidence: ±0.006 (+1.100..+1.112) - [0.0%]

So, while the score does have a significant effect, it is not enough to make 11/2* wrong (and Sander apparently knows what he is doing :stuck_out_tongue: ).

(Note: I noticed that the stream analysis for 11/2* was only 2-ply after doing the rollouts, which made the result a bit less exciting… but I made it this far so figured I would post anyway ¯\_(ツ)_/¯ )

(Note 2: XG++ says 11/2* a tossup (-0.010) BUT, on my system/setting, XG doesn’t even consider this move to be worth the extra ++ effort and defaults only to 3-ply)

1 Like

I remember watching that and feeling skeptical that match score was the explanation. Interesting to see that it does have a significant effect. Your instincts were good to check on that. The final analysis for UBC uses ++ I think, so he should get credit for hitting even if the initial eval was wrong.

Commentators are sometimes saying a move is inferior because it’s -.003 on 2-ply… they should say they are tied or we don’t know.

I did think in general Sander was playing too much for the attack and not enough for connectivity. Once you get a mostly-won front position, it’s time to make sure you don’t get stuck in the rear.

(Note: I’m not aware of any widely accepted convention, but I would call it 3-away/4-away, from the bottom player’s perspective. But as long as you’re consistent or it’s obvious, not a big deal.)
(Note 2: If the score was switched to 4-away/3-away, hitting would be even more right!)

Ahh, typo; thanks for calling it out.

1 Like