# Results are here

## Posted on May 18, 2012 at 10:24:15 PM by Craig G

According to my best information - which includes computer simulation AND real-world data - in spec 7 games, the winning point is scored by the server very, very close to 88% of the time.

So Bennett nailed it, and Haitian Gary, IWasSam and Tiger were essentially correct correct in their reasoning. Just a matter of getting a more accurate figure for the percentage.

The way I encountered this was that I was recently designing some new 64-bit assembly code with the goal of simulating 1 million games in 1/10th of a second. (but 1/7th is more likely) I started off just doing the 'win phase' of the game without any playoffs, and all teams rated equal and no serve/receive bias. To handle 1 million cycles I creating a 'random buffer' of 15 million values, which is more than enough. The idea is that after each point is played, we advance the 'buffer pointer' by one to get ready for the next point. However, in one particular test, I neglected to include the code to advance the pointer at the conclusion of each game. This resulted in the first point of the next game using the same random result that had been used for the last point of the previous game.

That gave me a surprising result of:

1 - 204442
2 - 121374
3 - 138753
4 - 124290
5 - 101867
6 - 102296
7 - 89121
8 - 117857

Instead of the usual near symmetry like:

1 - 163338
2 - 162834
etc

What's up with that? Well, after ruling out bugs and logic errors in my code, I suspected that something was biasing the outcome of the first point. Putting in a pair of point counters for S and R on the first point confirmed that yes, the server was winning the first point 88% of the time. Which meant that the previous point - the one that determined the game-winner of the previous game - was also won by the server at 88%. As a double check, I put in the instruction to correctly advance the pointer for the random buffer. Back to 'normalcy'. As a triple check, I wrote a quick and dirty version of the non-incrementing simulation in C++. Same 88% result. As a quadruple check, I took note of some recent spec 7 games at Dania which yielded 63 - 8 or 88.7%. Case closed.

Incidentally, for those who thought that the figure might be lower in a 9-point game, you were right. The sim says 83.5%. i don't have the patience to tally actual games for those, but since the sim was so close on spec 7, I choose to believe that figure too.

Anyway, after all that, it is actually easy to believe those figures. Since the only way for the receiver to win the game point is to come on the court with 5 or 6, it simply will not happen that often.

However, it is kind of bizarre to think that if you and a friend tuned in to Fort Pierce just as the server began what turned out to be the last point of the game, then even with no scores, graphics, or announcing, and no way to be certain who actually won the game or even the point, the fact that the game was obviously now over would mean that you could bet your friend that the server had won, and you would be correct 88% of the time.

Finally, re handicapping considerations, other than underscoring the importance of winning points on serve - which we already know - this seems to have little significance. But on further review, I do think it has something to offer. That's slightly complex and would require a separate topic if anyone is interested.