The analysis

I've now had a chance to analyse the effect of the bug.

Firstly, in deciding whether or not a horse is a selection, the program looks at all of the previous races relating to that horse. The bug is due to the default method that dates are handled. I ask it for all of the races before 07/11/2009, and instead of giving me the races before 7th November, it gives me the races before 11th July. It expects the American date format. However, it handles most dates correctly. It knows that 23/09/2009 has to be 23rd September, because there isn't a 23rd month.

The bug only effects 11 days every month; the days before the 13th, but not the day that is equal to the month. From the 13th onwards, the system is a loss maker. For those 11 days each month, the system (with the bug) is hugely profitable. Then I noticed something even more exciting; there is a large number of selections that are made by the system WITH the bug as well as the system WITHOUT the bug. By removing the bugless selections, the system is even more profitable.

That's not all though, I discovered that if I removed the selections that are made on days where the month is higher than the day, the system becomes even more profitable, because the others make a loss. So it makes big profits on 12th January, and big losses on 1st December.

I had discovered, by pure chance, the perfect system. I'd only have to place bets on about 70 days a year, and have fewer selections to lay on those days. I'd also be fantastically rich, and that would be useful as I would spend the second half of every month in a different sunny country, where I'd have my own apartment.

Then I spotted a flaw in my plan. Maybe you have too?