Mall,

As to large numbers trumping anecdotal evidence I agree with you 137%.

But the bounce theory -- playing a horse to run 'X' number of points or more off the last race -- has many possible scenarios under which we may say that it should play out, such as:

1) Horse just ran new top
2) Horse just ran new top at least 'X' points greater than prior top
3) Horse of age 'X' just ran either scenario 1 or 2
4) Horse just ran an absolute number greater than 'X'
5) Horse with a particularly good(bad) previous pattern (of age 'X') just ran a top under scenario 1 or 2 or ran an absolute number such as in scenario 4.

Then there are, of course, the trainer angles -- trainers whose horses supposedly don't bounce (Baffert, Lake, Dollase, etc.).

There are probably a dozen distinct scenarios with enough data points for us to draw conclusions about expected performance.

The question we want answered for each scenario is what the expected performance of the horse is in the next race (or, similarly, what % of the time does a horse 'bounce' 'X' number of points) and how does the horse's performance shake out relative to the horse's odds.

If I had been as rigorous over time in studying the actual performance of horses versus my prediction as I am in my day-to-day work I would have a better fact base upon which to pontificate, but I don't. And, since you would be a fool to reveal the eleven herbs and spices you use in your secret recipe I guess the string ends here. Chop.