BGonline.org Forums

Keith count errors, a first look

Posted By: Timothy Chow
Date: Monday, 4 January 2010, at 2:04 a.m.

I've written some code to read in the abridged database of about 50,000 non-contact races that Tom Keith has put online. It will take me some time to figure out what's going on with the cases that the Keith Count gets wrong, but here are some preliminary observations. I decided to look first at incorrect takes. In a lot of these positions, the taker has a lot of checkers stacked on the 1-point and/or the 2-point. The Keith Count is not properly counting the wastage of these checkers. Extremely bad takes can result if the taker has a big stack on the 1-point. Even if the 1-point is empty, very bad takes can occur if there is a big stack on the 2-point, as the first diagram below illustrates.

For an example of a bad take in a long race, see the second diagram below. The database says that the take is a 0.149 error (I haven't checked this with a rollout yet). I don't fully understand what's going wrong here, but note that if we use Walter Trice's "Rule of 62" (from Backgammon Boot Camp) with the raw pip count, ignoring wastage, then we get a very clear drop. So again the trouble seems to be inaccurate wastage accounting.

Money Game
Blue on roll. Cube action?

White	36
Blue	36
Position Id: /h0gAOC2HQAAAA Match Id: UQkAAAAAAAAA

Money Game
Blue on roll. Cube action?

White	102
Blue	86
Position Id: 6L6OAQA7eg0GAA Match Id: UQkAAAAAAAAA

Messages In This Thread

• Keith count errors, a first look
  Timothy Chow -- Monday, 4 January 2010, at 2:04 a.m.
  • Keith count errors, a first look
    Stick -- Monday, 4 January 2010, at 2:38 a.m.
  • Keith count errors, a first look
    Tom Keith -- Monday, 4 January 2010, at 4:36 p.m.