Can I ask how you came up with your Class Table # — Charles Lucas

Good class numbers come from good par times.
Good par times come from good class numbers.
They work together.

I have my own very sophisticated software to do this. While it is somewhat automated, every track-surface-distance is hand-crafted. (Yes, it is a joy to do.)

Stabilizing Pars and Classes
1. For class levels, I grab a year of races at one track across all distances. (But only dirt races.)

2. I have the software create all the pars based upon the most recent levels.
It does this in about 20 seconds.

3. Then I lock in the pars.
(This is not the final job; just temporary.

4. Then, using these pars, I have it assign levels to all the classes.
(Again, this is not the final job; just temporary.

5. There is an indicator of the change in $10k claimers. (Or some other class that is closest to $10k.)

I note the amount of the change.
What I want is for that error to be shrinking.
If it gets larger, there is a problem either in some classes or some pars that must be corrected before making good levels or good pars.



I repeat steps 2 thru 5 until the number stops changing.

Making the final Class Levels
1. I look at the classes and see where they are out of whack.
Each type of race is done by itself.

2. Always begin with the most common class type.
Place them and lock them in.
Make the others fit in.
Imagine you have a solid $18k level of 103.8 and a $12k level of 101.9.
You've got a $15k with only one race that indicates it ran like $22k.

Don't agonize.
Just put it between 101.9 and 103.8.
Averagae = 102.85.

3. Do this for all types and all levels.

4. Don't bother with OCs or Starters. They are just too volatile from year to year and race to race.

5. After all the class levels are set, look at the $10k claimer.
Adjust it by as much as necessary (up or down) to make it a "100.00."

6. Adjust all the other classes by the same amount.


Next, you build the actual par times.
1. Again, start with the most commonly run sprint distances.
Here's the real challenge for you.

It may take you 3 or 4 seasons to get your classes right, so just start with the class types you have confidence in. Logically, claiming races for older males.

(On that topic, females are an adjustment. Don't try to make specific pars for them. You'll find they're all over the place.)

Make your pars by taking all the races that fit (i.e. clm, older males, fast tracks).

2. Weight each race based upon where the winner first challenged for the lead.
Currently, I use this weighting system:

• 9 for 1st call
• 5 for 2nd call
• 2 for stretch call
• 1 for finish

Races that run late are just naturally going to be slower. Way slower sometimes.

Build a par for all the sprints that you have enough races to assess.

3. Determine the Track's "Pattern."
Quirin had a nice table in his book where he outline how tracks we typified by A-B-C, etc. with each letter representing a pattern.

It's up to you to decide what pattern should be used for sprints and routes at each track.

You do this once you have at least 2 sprint distances.

4. After you have those, fill in the others in the order of how many races or where they fit.

Make sure they match the pattern.

5. Follow the same process for routes.

THAT'S IT.

If this seems like too much, just buy my pars. :)
(They come with several different CSV formats.)

2023 HorseStreet Par Times

Just to be clear, about what I said elsewhere, I am saying that every attempt I ever did to make a computational daily variant failed badly when compared with using simple ratings derived from par for the track.

As most are aware, the Average Daily Variant is flawed, simply because the horses that run on Wednesday are (as a general rule) not as strong as those who run on Saturday.

The logic has typically been to apply an adjustment to each day of the week.

However, even that fails because not all Saturdays are the same caliber of races.


The solution would be to build a Strength of Field Rating for every single race, and then address the issue on a daily basis, asking the question, "How strong was this day?"

However, the challenge is not over there because the very definition of the fast/slow-ness of each race is in question.

This is best evidenced is found in the improvement of my own par times.
A little over a decade ago I committed to LONG-TERM IMPROVEMENT in my par making.

I designed something I called The Speed Reliability Index. (SRI)

Year over year I began making changes.

The first thing I noticed was that the *1ST CHALLENGE CALL of the eventual winner of a race made a huge difference in the final time.

IOW, winners that challenged for the lead at the 1st call resulted in faster final times than winners that did not get their heads within a length of the leader at the stretch call.

Of course, this is all very logical.

When a horse is on the lead wire-to-wire the time will be faster than when he's (say) 8 lengths back at the first call.

I began weighting the races in the sample using a 4-3-2-1 approach.
That is, races where the winner of the race mounted their 1st challenge at the 1st call were weighted at "4," while winners who did not get within 1 length until the last fraction were weighted as a "1."


The improvement was both immediate and drastic.
So much so that each year I changed the weightings to even heavier weighting of the EARLIER CHALLENGERS.

It went from 4-3-2-1 to the current 9-5-2-1!

On a higher level, the impact of this change can be seen in our internal Flagship Rating, simply called "RTG" in our software.

Note that RTG is a "Composite Factor;" a combination of 32 other factors, many of which are Composites themselves.

Speed Ratings make up the biggest part of RTG.
___
Think of RTG as being like the BRIS PRIME POWER rating, as well as HDW's PSR and CPW (an older version of PSR).

Back when I started, PSR, CPW and BRIS PP, were all hitting about 29-30% winners on the top ranked horse. (Now about 31%.)

RTG was at 24.6%.
Today, RTG stands at slightly over 31%, with the highest return of the 4 listed numbers. (That's because it does not correlate with the tote quite as much as the other numbers.)


So, what do you do?
The answer seems to lay in weighting each race by the running style of the winner.

But even that has problems.
It is simply difficult to look at a day with 8 races on the card:
4 sprints
2 routes
1 4f race
1 turf race

And make a viable daily variant.

Add the complexity of maiden races, first-time-starter races, off-track for the 1st few races... and a myriad of other complications... and you have a system that is ripe for over-inflating or under-inflating the races.
_____________
[SIZE="4"]That is why every test I have ever done leads me to NO VARIANT as the best approach.[/SIZE]



* 1ST CHALLENGE CALL must be credited to Jim Cramer, the great mind behind HDW's ratings.

Can't say they aren't worthwhile, but I've only been exposed to the average daily variant approach and every study I have ever done has shown that using a par without a daily variant works better.

Actually FAR better.

Now, just to be clear, I am not a guy who would put time into building a real, race-by-race variant that is dependent upon the true skill of the variant maker. I give honor to such people.

If you are that person... making an estimate of what each horse could do and comparing each to the other, considering post position, jockey impact (both good and bad), trips, and all the other complications, I am humbled by your skill.

If you try that and they don't work well, then I honor your effort.

But, alas, too much for me.

Really great questions.
________________

I did a little reading on "Agents" in AI". "Each of the 8 Agents has their own approach to the game". Is each approach somewhat predetermined by the AI programming? — Jim Pommier

The goal of the AI engine is to develop a system that allows each agent to become their own person, so to speak.

Imagine you're going to teach...
... a brand new horse player how to handicap. What do you teach him?

• You teach what you know and believe.
• This becomes his own personal starting point.
• He will begin with the factors you gave and proceed from there.

While he can - and most certainly will - eventually change how he handicaps, his roots will always be in his beginnings.

Will one Agent look at say Pace Shape while another looks at FTS's? Then each one builds from their experience in their particular approach? One agent may like horse #1, while another may like horses #2 and #3?

This is a very simplistic way to explain exactly how it works.

____________________

For example-- "Is there a vulnerable favorite in this race? Which horses' figure to improve or digress? Will only some Agents review and respond to the question, while other Agents do not based on their AI approach to handicapping? Thanks.

Again, precisely right.

There is one significant difference between HUMAN handicappers and AI AGENTS:
While humans are dedicated to improvement, there is a natural tendency to want to continue doing what they are doing now.

To some degree, we just naturally defend our current approach and resist changing it beyond minor improvements.

Agents are not encumbered by such complications.

HOWEVER, the will always be trapped (to some degree) relatively near their starting points; their POINT OF ORIGIN.

____________________
I've designed the 8 agents with HANDICAPPING PERSONALITIES that will have VERY GOOD HIGH contender hit rates.

There is one such agent whose approach will likely have the very highest contender hit rate. However, he will have a difficult time hitting price horses.

Each agent's tendency will be to stay somewhat close to his roots in contender selection, but his (or her) actual handicapping is completely free form in the sense that they have ALMOST ALL of the 3,000+ factors available to them.

Knowing that there's a vulnerable favorite, I can then pass the race or look at other entries — Jim Pommier

Yes, this is valuable information.

Also, does the AI take into consideration the trainer/jockey win and ITM percentage, class rating, earnings per start and the actual horse's win and ITM percentage? — Jim Pommier

It considers just about EVERYTHING.

Of course, it doesn't actually USE everything in every race.

BTW, there isn't actually ONE AI.

They are called AGENTS and there are 8 of them.

Each has their own approach to the game. You can select as many as 5 to be active in any given race.

Yes, @RanchWest - take care.

That's a big step towards understanding what people want and could use.
Thank you.

I'm trying to design something that allows for almost free-form notes.

I think that the key is to have them readily available - so you don't have to go looking.

Do you keep the notes?

WOW!

Won 4/1, $10.00
ytd, B 8 W 26.20, profit 18.20 — RanchWest

Nice play!
You've been performing consistently well.

Any chance of taking you on a percentage basis? :)

Good fortune to you with this.

Thanks Dave, but I think the term "excellent" is overly generous. — William Zayonce

Winning is a fragile state.
Anyone who can win consistently - and your picks over time have shown that you do that - is to be respected.

Even being close to breaking even is an indicator of (at least potential) greatness.

I salute you.

You really are excellent.

Got an ETA on that, Dave? — RanchWest

I've got a TARGET DATE but it keeps moving back.
The website is finished, as is the store. But no content yet.

Also need to hook the store to the new software to handle subscriptions and provisioning (of software on start up).

The stumbling block is finishing the software's output. Under the hood is 90% finished, but the output is barely started. The good news is that the output is the easiest part.

Meanwhile, in the background, the AI engine is doing its thing - developing artificial AGENTS who handicap and give opinions.

Really some cool stuff coming out of this WHEN I get it together.

Until then, it's all just vaporware and speculation.

There is a 200 & 300.
We lost a streaming site a while back.

They will be up when the new website & store go up.

I suppose there are another hundred hours or so.

What's your specific interest?

Update on this entire project.
The AI appears to be training so well that the new factors may well have to wait until a later version.

Shocked at how good this looks.

Pushing to make my self-imposed deadline.
Getting closer.

I'm familiar with the thread.

Thank you.

I'm not ready to make screenshots of the new software yet.

Still working on creating the factors.

We'll dump some of the old ones - maybe 2,000 or so - and add about 1,000 back.

Some of them are mind blowers.
Like one I got from AUS - "Ran like a favorite" (or 2nd choice, 3rd choice, etc).

Also, classing the races by mix of horses rather than a rigid class system.

Thus, a $16k claiming race that is loaded with horses who should be able to run much faster, will show as such. Hint: Using high-level factors to make those class distinction. ("High level" means PSR, BRIS Prime Power, etc.)

Oh, and there are now three factors that are similar to BRIS Prime Power. TWELVE if you use the HSH download!

The list goes on and on.

How about the HSH Getting Started video manual?