Yeah, I understand that, and when I first read the phrase, "normalized times of past races" in the paper he wrote, I thought 'OK, he's talking about speed figures'.

But then in the presentation he gave, he talked repeatedly about factors like "normalized finishing position" and "normalized jockey performance", etc., so I thought, well maybe he's just referring to putting it on a 0.0 to 1.0 scale like all the other factors.

And the fact that he never said a single word in either the paper or his presentation about speed figures or performance figures or how he's deriving them (assuming he's either computing them on his own or downloading them from some sort of database which I knew was not TG, and probably wasn't Beyer), or a single word about parallel time charts, track variants or daily variants left me unclear as to what he was actually doing with running times. You would think that if he's going to give a talk on how go about creating a computer model, these things would be the first and foremost points of the discussion.

But instead, as the excerpt from the talk I posted shows, he spends a great deal of time and emphasizes pretty strongly that the first factor one should start out with if setting out to create a computer model is normalized [i]finishing position[/i](!). He says virtually nothing in his entire presentation about actually running times, be they raw times, speed figures, or performance figures (ground loss? wind?), unless he just figured it was a given, or maybe he made a deliberate decision not to disclose how he computed what would seem obvious to be the most important indicator of a horse's performance.

In any event, he must be doing it somehow, either directly by using speed figures, or indirectly through other factors (he mentions weight as a separate factor) in the model as Templeton stated, otherwise there's no way the model would be able to generate anywhere near the degree of accuracy it's getting in terms of its probability estimates.