"peps...@gmail.com" <pepstein5@gmail.com> writes:

> Suppose a racing formula is applied in a relevant situation.
> And suppose the players follow the formula.
> Does anyone know the maximum possible equity loss that can
> arise from the formulas?

Results for the Isight count, using Tom Keith's database (about 50000
positions used for calibration) and giving the GNU Backgammon ID:

Worst no double: CwAA8AAAAAAAAA with -0.722
Worst no redouble: EwAA8AAAAAAAAA with -0.769
Worst double: NQAAwBEAAAAAAA with -0.359
Worst redouble: IAAACAEAAAAAAA with -0.597
Worst pass: NQAAYAwAAAAAAA with -1.32
Worst take: CAAAJAAAAAAAAA with -0.778

Quite a freak show (Keeler/Gillogly does not fare better)! Similar or
even identical values for my own database (50000 positions used for
validation). (-:

No idea whether more extreme positions exist.

I also took the opportunity to calculate the probability for a whopper
(> 0.1 equity loss) for the Isight count and the Keeler/Gillogly count
(in parentheses):

To double or not: 1.36 % (2.88 %)
To redouble or not: 1.96 % (3.78 %)
To take or not: 5.72 % (6.81 %)

Best regards

Axel