York 5.30

[tony spencer]

Race overview​

This is a big-field York sprint where:

• There should be plenty of early pace.
• A strong gallop is likely.
• Horses that can sit just behind the speed and finish are often favoured over those involved in a speed duel.

That fits the profile your report was highlighting.

---

Rock Opera​

This is still my first choice.

Why?​

Your report gave him:

• FinishStrength leader.
• Best model + setup blend.
• Strong late finishing profile.
• A race shape that should play to his strengths.

This isn't simply a horse with good ratings—it's one where your FinishStrength model and setup model agree, which is something we've found valuable.
The current betting market also has him among the leading contenders, which is consistent with your analysis.

---

American Bay​

This is the horse I'd fear most.

Positives​

• Strong draw/pace combination.
• Expected to race in a good tactical position.
• Doesn't need a perfect race to get involved.
• Strong overall suitability in your report.

If the pace isn't quite as strong as expected, I could easily see American Bay getting first run on the closers.

---

Mister Sox​

One I'd include in wider bets.
He's respected in the current market and has the profile of a horse capable of running well in a race of this nature.

---

Fiscal Policy​

Interesting outsider.
If the race develops into a real stamina test over the final furlong, he's capable of passing beaten horses late.

---

Pace scenario​

This is how I see it:

• Several runners go hard from the gates.
• The leaders don't get an easy time.
• American Bay gets a lovely stalking position.
• Rock Opera is produced late.
• The race is decided in the final 100 yards.

---

My betting rankings​

Rank	Horse	Confidence
	Rock Opera
	American Bay
	Mister Sox
4	Fiscal Policy

Betting view​

This isn't one of my strongest races of the day because of the field size and the number of runners with chances. However, if I had to play it, I'd stick with Rock Opera because he's the runner where your FinishStrength model and model/setup blend both point in the same direction. American Bay is the main danger, particularly if the race doesn't collapse quite as much as expected.

 

[tony spencer]

One of the things I do is break down the reports for each race. what went well what didnt and why ? This is the analysis of race i got badly wrong.

 

[tony spencer]

Race 3 York

 

[tony spencer]

This was one of the better races

 

[TheBluesBrother]

T tony spencer

I haven't analysed a race after the fact, so, using Copilot and my speed-figure database, here are the results.

Mike.

5:30 York
Model vs Actual Race — Key Insights
Your model’s top pick: Almarada Prince (170)
Finished 4th, but was hampered late — your model absolutely found the right horse.

Magic Boy (165)
Your model had him 2nd best — he won.

Havana Rum (163)
Your model had him 4th — he finished 3rd, and from the disadvantaged far side.

Betties Bay (164)
Your model had him 3rd — he finished 7th, but ran on the weak near side.

Rock Opera (160)
Your model had him 5th — he finished 9th, but was denied a run.

Conclusion
Your model nailed the race:

Top 2 in your ratings finished 1st and 4th (with trouble).

Four of your top 5 beat the majority of the field.

Track bias + trouble in running explain the deviations.

This is exactly what a strong ratings model looks like — it consistently identifies the right horses even when the race circumstances distort the finishing order.

 

[tony spencer]

T tony spencer

I haven't analysed a race after the fact, so, using Copilot and my speed-figure database, here are the results.

Mike.

5:30 York
Model vs Actual Race — Key Insights
Your model’s top pick: Almarada Prince (170)
Finished 4th, but was hampered late — your model absolutely found the right horse.

Magic Boy (165)
Your model had him 2nd best — he won.

Havana Rum (163)
Your model had him 4th — he finished 3rd, and from the disadvantaged far side.

Betties Bay (164)
Your model had him 3rd — he finished 7th, but ran on the weak near side.

Rock Opera (160)
Your model had him 5th — he finished 9th, but was denied a run.

Conclusion
Your model nailed the race:

Top 2 in your ratings finished 1st and 4th (with trouble).

Four of your top 5 beat the majority of the field.

Track bias + trouble in running explain the deviations.

This is exactly what a strong ratings model looks like — it consistently identifies the right horses even when the race circumstances distort the finishing order.

Well done good set of results

 

[tony spencer]

Well done good set of results

T tony spencer

I haven't analysed a race after the fact, so, using Copilot and my speed-figure database, here are the results.

Mike.

5:30 York
Model vs Actual Race — Key Insights
Your model’s top pick: Almarada Prince (170)
Finished 4th, but was hampered late — your model absolutely found the right horse.

Magic Boy (165)
Your model had him 2nd best — he won.

Havana Rum (163)
Your model had him 4th — he finished 3rd, and from the disadvantaged far side.

Betties Bay (164)
Your model had him 3rd — he finished 7th, but ran on the weak near side.

Rock Opera (160)
Your model had him 5th — he finished 9th, but was denied a run.

Conclusion
Your model nailed the race:

Top 2 in your ratings finished 1st and 4th (with trouble).

Four of your top 5 beat the majority of the field.

Track bias + trouble in running explain the deviations.

This is exactly what a strong ratings model looks like — it consistently identifies the right horses even when the race circumstances distort the finishing order.

 

[tony spencer]

Mike, the amount of work that must go into producing ratings every single day is considerable, and the fact that you make them freely available to everyone is incredibly generous. There aren't many people prepared to put that amount of time and expertise into something purely to help other racing enthusiasts.


I particularly like the consistency of the figures and the depth of information you've included. It's obvious they've been developed with a lot of thought and experience.


I do have a few questions, if you don't mind.


You provide the last six speed ratings for each horse. In your experience:

• Which of those six ratings tends to be the most predictive?
• Is the latest figure usually the one to trust most, or do you find an average works better?
• Have you ever tested using the average of the best two ratings from the last six?
• Have you compared different approaches such as:
  • last run only,
  • average of the last three,
  • average of the best two,
  • weighted average favouring recent runs,
  • highest recent rating?
• Have you done much modelling or back-testing to see which approach produces the best strike rate or profitability?
• Do you find the ratings are more reliable under similar race conditions, such as the same trip, going or class?
• Are there situations where you deliberately ignore a horse's highest rating because you feel it's an outlier?

I'm always interested in how experienced rating compilers validate their figures, so I'd be fascinated to hear your thoughts on what you've found works best over the years.


Thanks again for all the work you put into producing these every day—it's very much appreciated.


Kind regards,


Tony

 

[tacker]

The 5.30 y race yesterday ended up with a worthy winner in MAGIC BOY, a horse that sometimes appears to be a little too keen in his races but the jockey got him to settle here and seemingly improved for that tactic, only won by a nk so won't unduly punished for that performance.

 

[tony spencer]

The 5.30 y race yesterday ended up with a worthy winner in MAGIC BOY, a horse that sometimes appears to be a little too keen in his races but the jockey got him to settle here and seemingly improved for that tactic, only won by a nk so won't unduly punished for that performance.

Yes changed the way he was ridden.

 

[TheBluesBrother]

I do have a few questions, if you don't mind.

Which of those six ratings tends to be the most predictive?

When I run my model, it's the top speed figure of the last 6, it would be nice to only use the last 2 if they were true run races.

Is the latest figure usually the one to trust most, or do you find an average works better?

I never use averages.

Have you ever tested using the average of the best two ratings from the last six?

No, I haven't.

Have you compared different approaches such as:
last run only,
average of the last three,
average of the best two,
weighted average favouring recent runs,
highest recent rating?

The hours I spend each day compiling speed figures (4 hours per day on average) does not give me any time to do analysis, I already go down too many rabbit holes to start doing that.

I'm always interested in how experienced rating compilers validate their figures, so I'd be fascinated to hear your thoughts on what you've found works best over the years.

My speed figure method is unique to me, unlike the professional compilers, I do not use static pounds per length,
every racecourse and distance has its own pounds per length figures, and every rating produced is rail-adjusted.

Mike.

 

[pete]

Have you ever tested using the average of the best two ratings from the last six?

I have landed on "the average of best three from last 6"
I find this works well in the lower grade handicaps that I tend to target.

I am also still working on higher grade racing, where I suspect but have not yet confirmed that a "ceiling" type rating may be the best option (best rating of the last 12 months)

I think a lot of the higher grade handicap horses tend to win or be aimed at a couple of races each year ...

Of course, if you are looking at younger horses, the most recent may be better ... as they are likely still on the upswing.

I don't think there is a "right or wrong" answer, but I think each race has to be looked at within the confines of that race and the runners in it.

If you are looking at a 4yo+ ... an average may work well as there will be solid form to look at.
If it is a 3yo handicap, look for an upswing in numbers combined with a good performance as the horse may be still improving ..

 

[tony spencer]

Thanks Mike, that's really interesting.
I should say first that I don't think my speed ratings were as sophisticated as yours. From what you've described over the years, your ratings involve a huge amount of work that mine simply didn't. Mine didn't have individual pounds-per-length values for every course and distance, they weren't rail-adjusted, and I certainly wasn't putting in the level of work that goes into calculating going allowances and all the other variables that you account for. Producing the ratings themselves is a specialist skill, and I have a lot of respect for the effort that goes into yours.
My work was more about trying to discover the best way of using a set of ratings once they had been produced. Rather than assuming one number was always best, I built machine-learning models using different representations of the previous six ratings to see which contained the most predictive information.
Some of the inputs I tested included:

• Latest rating only.
• Best rating from the last six (ceiling ability).
• Average of the last three.
• Average of all six.
• Average of the best two.
• Average of the best three.
• Weighted and exponential moving averages that gave more emphasis to recent runs.
• Trend over the last few runs.
• Linear prediction of the next rating.
• Consistency measures, such as standard deviation, to distinguish reliable horses from inconsistent ones.
• Time since the peak figure and time since the last run.

I also looked at contextual factors such as distance, class, going, surface and race conditions, because I suspected the same mathematical approach wouldn't necessarily suit every type of horse.
One thing that surprised me was that there wasn't a single method that dominated everything. A horse's best recent figure, a weighted average and the recent trend often carried different pieces of information, and models generally performed better when they combined those features rather than relying on one rating alone.
I also came to the conclusion that the answer probably depends on the horse. An improving three-year-old may be better represented by its latest figures, whereas an exposed older handicapper may be better represented by an average or by its established ceiling ability.
Your comment about only trusting the last couple of figures if they came from genuinely run races also makes a lot of sense. That's something I never tested directly, but I think it could be very important because the quality of the race shape probably affects the value of the speed figure itself.
In truth, I was trying to find the best way to use the ratings, whereas you've invested years refining the ratings themselves. They're really two different problems.

Which of those six ratings tends to be the most predictive?

When I run my model, it's the top speed figure of the last 6, it would be nice to only use the last 2 if they were true run races.

Is the latest figure usually the one to trust most, or do you find an average works better?

I never use averages.

Have you ever tested using the average of the best two ratings from the last six?

No, I haven't.

Have you compared different approaches such as:
last run only,
average of the last three,
average of the best two,
weighted average favouring recent runs,
highest recent rating?

The hours I spend each day compiling speed figures (4 hours per day on average) does not give me any time to do analysis, I already go down too many rabbit holes to start doing that.

I'm always interested in how experienced rating compilers validate their figures, so I'd be fascinated to hear your thoughts on what you've found works best over the years.

My speed figure method is unique to me, unlike the professional compilers, I do not use static pounds per length,
every racecourse and distance has its own pounds per length figures, and every rating produced is rail-adjusted.

Mike.

Thanks Mike, that's really interesting.
I should say first that I don't think my speed ratings were as sophisticated as yours. From what you've described over the years, your ratings involve a huge amount of work that mine simply didn't. Mine didn't have individual pounds-per-length values for every course and distance, they weren't rail-adjusted, and I certainly wasn't putting in the level of work that goes into calculating going allowances and all the other variables that you account for. Producing the ratings themselves is a specialist skill, and I have a lot of respect for the effort that goes into yours.
My work was more about trying to discover the best way of using a set of ratings once they had been produced. Rather than assuming one number was always best, I built machine-learning models using different representations of the previous six ratings to see which contained the most predictive information.
Some of the inputs I tested included:

• Latest rating only.
• Best rating from the last six (ceiling ability).
• Average of the last three.
• Average of all six.
• Average of the best two.
• Average of the best three.
• Weighted and exponential moving averages that gave more emphasis to recent runs.
• Trend over the last few runs.
• Linear prediction of the next rating.
• Consistency measures, such as standard deviation, to distinguish reliable horses from inconsistent ones.
• Time since the peak figure and time since the last run.

I also looked at contextual factors such as distance, class, going, surface and race conditions, because I suspected the same mathematical approach wouldn't necessarily suit every type of horse.
One thing that surprised me was that there wasn't a single method that dominated everything. A horse's best recent figure, a weighted average and the recent trend often carried different pieces of information, and models generally performed better when they combined those features rather than relying on one rating alone.
I also came to the conclusion that the answer probably depends on the horse. An improving three-year-old may be better represented by its latest figures, whereas an exposed older handicapper may be better represented by an average or by its established ceiling ability.
Your comment about only trusting the last couple of figures if they came from genuinely run races also makes a lot of sense. That's something I never tested directly, but I think it could be very important because the quality of the race shape probably affects the value of the speed figure itself.
In truth, I was trying to find the best way to use the ratings, whereas you've invested years refining the ratings themselves. They're really two different problems.

 

[tony spencer]

Mike if I get time and you want to send me youre database Zip it up as much data as you have got? I will produce some model inputs and we can model it and see how it performs? I need to match the data with mine so we can get some Sp data and finishing data. Put we could see if we can find some profit angles ?