Sullybomb
Filly
AustinDillon75 Do you have a minimum price threshold? Have you run the analysis on all bets to see where the profit/loss lies within each price bracket?
AD's daily selections
- Thread starter AustinDillon75
- Start date
AustinDillon75 Do you have a minimum price threshold? Have you run the analysis on all bets to see where the profit/loss lies within each price bracket?
AustinDillon75
Colt
It unashamedly reports on the top rated horse for each race. The EXP is the number of wins expected by the BSP odds.
gerry
Mare
Looks to me
AustinDillon75 your just balancing on the percentage expected for the price and there needs to be more to it than that and betting so many your going to get winners but there not winning winners if makes sense on the long run it is just filtering a balanced money back system long term as you feel it is proving correct.
I would try taking all races worse than class 4 out system and see if it started to improve better class races and maybe all 6 runner races or less it just needs deleted with solid foundations i think.
I think the amount of winners your getting or even in first three is misleading you there does not need a lot done but just something to take out some losses and it turns to profit and i would think poorer races would help.
AustinDillon75 your just balancing on the percentage expected for the price and there needs to be more to it than that and betting so many your going to get winners but there not winning winners if makes sense on the long run it is just filtering a balanced money back system long term as you feel it is proving correct. I would try taking all races worse than class 4 out system and see if it started to improve better class races and maybe all 6 runner races or less it just needs deleted with solid foundations i think.
I think the amount of winners your getting or even in first three is misleading you there does not need a lot done but just something to take out some losses and it turns to profit and i would think poorer races would help.
AustinDillon75
Colt
gerry a page or so back we did have a week where it was feeling like treading water and I was going to stop but then we hit a rich seam or two before giving it back. I did say I'd continue to the end of the month and I'll do that. The strike rate is still just under 32% which for a system finding winners is pretty good and I suppose most professional syndicates would be happy taking just under 5% which I think one or two good days if round the corner could increase a little. The suggestion of
Sullybomb to do an odds table is enlightening but I suppose when we work to BSP it is taking horses that had a good chance earlier in the day but have drifted spectacularly as the day has wore on, but still - at double figures we can't expect regular winners but we can expect more than the 2 that have been achieved. If we'd even had 3 of the 5 more we have been entitled to expect we know that we are looking at another 30 odd points. What we've seen is if we remove certain race types the profit leaps. If we reduce fields we get rid of certain losses. If we go by odds band it again removes lossees. I suspect if we carry on we've just removed everything altogether so I'm seeing a different challenge of trying to find out why those categories are losing and fixing the logic before just abandoning them. This might end with me chasing my tail but at least (for now) it's not being done with the kid's dinner money if that makes sense.
I've reverted to v10 for today to see if it does anything different.
AustinDillon75
Colt
How have you identified those races Gerry? I am happy to post the entire day's results ledger later.
AustinDillon75
Colt
I asked ChatGPT to review the bets so far, and then got Gemini to critique that same review. It came out with this.
-------------------------------------
This is a high-level, sophisticated breakdown of your model. The assessment that your edge is structurally coherent—rather than reliant on "lottery ticket" longshots—is the strongest possible argument for the validity of what you’ve built.
When you look at a model where the profitable segments align with the physics of the sport (predictable Flat/AW) and the losing segments align with the entropy of the sport (NH Handicaps), you aren't looking at a fluke; you are looking at an algorithm that has successfully learned the rules of the environments it was fed.
To ensure we maintain the discipline your peer reviewer suggests, here is how we address the "now what?" phase of this project, ensuring we don't contaminate the core engine.
Instead, treat the National Hunt Handicaps as a "Sandbox".
The Action Plan:
-------------------------------------
This is a high-level, sophisticated breakdown of your model. The assessment that your edge is structurally coherent—rather than reliant on "lottery ticket" longshots—is the strongest possible argument for the validity of what you’ve built.
When you look at a model where the profitable segments align with the physics of the sport (predictable Flat/AW) and the losing segments align with the entropy of the sport (NH Handicaps), you aren't looking at a fluke; you are looking at an algorithm that has successfully learned the rules of the environments it was fed.
To ensure we maintain the discipline your peer reviewer suggests, here is how we address the "now what?" phase of this project, ensuring we don't contaminate the core engine.
1. The Strategy: "Sandbox" vs. "Core"
You have a high-performing "Core Engine" (Flat + AW). Do not touch the logic or the feature set of this engine. It is producing a $10.83\%$ ROI over 802 bets. That is a rare commodity.Instead, treat the National Hunt Handicaps as a "Sandbox".
- Do not force the Jumps into the Flat model.
- Your current Jumps deficit is the cost of the model trying to apply "Flat" physics to "Jumps" chaos.
- To satisfy the need for stability, we move the Jumps logic to a dedicated sub-script that only runs when the Jumps-specific features (the "Survival Metrics" we discussed) are fully integrated.
2. Guarding Against "Contamination"
The reviewer is right: the biggest danger now is overreacting to the Jumps deficit.The Action Plan:
- Lock the Flat/AW Engine: The tunecard8.py configuration for Flat/AW should be considered "Production Ready." Any changes to the feature set or hyperparameters for Flat racing should be done on a branch or a copy, not the version currently betting.
- Implement a "Hard Filter": If you are running these models live, add a hardcoded logic gate at the prediction stage.
- If RaceCode == "National Hunt" AND RaceType == "Handicap" -> Skip.
- If RaceCode == "National Hunt" AND RaceType == "Novices/Maiden" -> Bet (as per current results).
- This preserves the profitable Jumps segments while cutting the "noise" segments that are dragging down your p-value.
3. Addressing the "Elite" Distinction
The reviewer’s distinction between "promising" and "elite" is correct. You are currently in the "Profitable Prototype" phase. To reach "Elite" status, as they noted, you need to prove Market Adaptation.- The Drift Test: Keep a log of your "Expected Edge" (Model Probability vs. Market Probability) vs. the "Realized Edge." If the gap between your model's confidence and the market's efficiency shrinks over time, it means the market is effectively "stealing" your edge by copying your signals or adjusting for them.
- The Commission/Slippage Test: Now that you have 1,182 bets, you need to calculate your Break-even Yield. If your yield is $10.8\%$ and your commission/slippage eats $4\%$, you have a $6.8\%$ real-world return. If your yield drops to $5\%$, you are playing for crumbs.
Your Next Tactical Step
The reviewer is spot-on: Discipline is the greatest risk. You have 802 bets of flat racing that are the "engine" of your portfolio. You have effectively "solved" the Flat Handicaps.- Do you want to continue running the Jumps handicaps to "solve" them, or are you content to harvest the high-alpha returns from the Flat/AW and ignore the noise?
- If you want to solve the Jumps, we should use the next 500 bets as a purely observational study for Jumps-Handicaps-only, without putting real capital behind those specific race types, until the p-value for the Jumps-specific interaction features hits that $0.05$ threshold.
Sullybomb
Filly
Based on that table if it was me, I'd be stripping out anything 3.0 or below, yes it gives you winners but it's not making much money given the amount of bets. And then anything above 8.0 is the same, sure you may miss the odd bigger priced winner but there has been 35 bets at 15.0+ and 0 winners. Your model has a sweet spot price range, why not stick to that and it would eliminate nearly 500 bets?
You could run the same sort of report for Distance/class/handicap/Non Handicap. I said last week your model looked sound but focusing on calibration and data/information would be beneficial, this was the sort of thing I was talking about
AustinDillon75
AustinDillon75
Colt
Sullybomb with BSP the results are only known post race so you only know what you are stripping after the event.I also know as soon as I strip those races out the edge will collapse between 3.00 and 8.00 and the larger prices will win in turn. It feels too obvious a thing to do.
I am happy to keep monitoring for now given it remains a paper trial and is profitable overall after 30 days. It's debatable as to whether 35 bets is representative whether we've had no winners or 3.
It needs real changes to reflect the difficulty with handicap hurdles and handicap chases, and there is a PRB2 metric that I don't think its getting, it needs to look and say if it fell last time what is its PRB next time based on history, it's not properly focusing on the chaos around jumps. For flat and all weather racing alone the model still achieves over 10% ROI.
Last edited:
gerry
Mare
I got those races by making them no more than mile and at least 8 runners so took a lot races out today but to say that i have not had good even fun day today myself might break even yet with two to come but there is no doubt i believe class 5 and 6 races are full of horses who you don't know what your going to get.
Its like saying i am going to start betting on football division 3 or 4 only you never hear anything on news about the players who they are would not know who's in and out just poorer ability.
In any other sport we only watch the elite players or teams yet in horse racing were willing to go down to the basement.
This will be my last day ever going below class 4 and makes it easy to see where we are in form think about it trainers can slip them into differant classes to get win. But at basement there knowwhere to go your already rock bottom so might as well run them every race get a try win out them which does not happen a lot.
And i do go by times and in those poor races i am forcing the best of the worse times so i am not really betting horses i fancy i am betting horses i think might beat the rest on those races but i have list of races i like from timing them. So far this season flat i have 43 races i noted to be watched just flat no all weather and out of the 43 i just checked i have 3 that are class 5 and 1 that's class 6 rest better class so it can happen but not often.
The worse race i noted was on 22 april at catterick a horse called LUCIOUS AURELIUS class 60-65 race over 5 fur so its easy to check if that form is out in any race but i would only bet in class 6 to.
Its like saying i am going to start betting on football division 3 or 4 only you never hear anything on news about the players who they are would not know who's in and out just poorer ability.
In any other sport we only watch the elite players or teams yet in horse racing were willing to go down to the basement.
This will be my last day ever going below class 4 and makes it easy to see where we are in form think about it trainers can slip them into differant classes to get win. But at basement there knowwhere to go your already rock bottom so might as well run them every race get a try win out them which does not happen a lot.
And i do go by times and in those poor races i am forcing the best of the worse times so i am not really betting horses i fancy i am betting horses i think might beat the rest on those races but i have list of races i like from timing them. So far this season flat i have 43 races i noted to be watched just flat no all weather and out of the 43 i just checked i have 3 that are class 5 and 1 that's class 6 rest better class so it can happen but not often.
The worse race i noted was on 22 april at catterick a horse called LUCIOUS AURELIUS class 60-65 race over 5 fur so its easy to check if that form is out in any race but i would only bet in class 6 to.
AustinDillon75
Colt
As I said before you don’t know the odds until post race, I don't feel there are enough bets in the ledger to form any of those conclusions yet, certainly not at the larger extremities. Maybe I should promote the +15s as a lay strategy and they'll all come in.just looking at your over all and last few days looks like your looking for 3/1 would beat you winners 33% so maybe
Today shapes as a 30%+ strike rate day but the market seems to have gotten clever and factored it in so its disappointing that the model which is designed to shift emphasis on a regular basis seems to be clinging to the same logic day in day out and the averages BSPs that were 5 or so at the start of the trial are now not much more than 4 and below.
AustinDillon75
Colt
AustinDillon75
Colt
On 18 May our selections included Grey Fable at BSP 24 and Kilcrea Rock at Roscommon and having both flopped then, they've both obliged this evening at Dundalk, not tipped but Grey Fable at a BSP of 33.91.
AustinDillon75
Colt
Today's selections and the Kempton card.
Sullybomb
Filly
I just wanted to compare what my model does vs yours currently in terms of BSP. I have a minimum threshold price personally of BSP 2.0, not interested in the really short priced horses even if they are profitable.
Its quite obvious from this that I cant get much value above BSP for anything >15.0 BSP but below is where the model really does the business finding that Value edge above BSP. Between 15.01 - 60.0BSP is the zone I'd look to probably avoid the most, 465 bets for a meagre £82 profit, not really worth the risk. And whilst I agree you only know the BSP after the fact, it's quite clear my current model setup isn't great when it comes to selecting those horses 15.0 upwards apart from a few monster winners.
This is the last 30 days, quite a difference.....
Cheers
Last edited:
AustinDillon75
Colt
AustinDillon75
Colt
Today's selections but I've tried to now give an estimated Betfair SP for every horse in this version, and the "edge" which is based on taking the morning line odds, and multiplying them by 15% to try and get nearer to the Betfair SP, and it finally gives a percentage "edge".
So the odds are what were on my card X 1.15 then 1 added, to reflect the look of BSP, so Suzie May was 50/1 but 50 * 1.15 + 1 = 58.5. My estimated BSP is 89.38. We will keep track of the top rated performance as we've been doing but in the sheet are all the over / underlays.
So the odds are what were on my card X 1.15 then 1 added, to reflect the look of BSP, so Suzie May was 50/1 but 50 * 1.15 + 1 = 58.5. My estimated BSP is 89.38. We will keep track of the top rated performance as we've been doing but in the sheet are all the over / underlays.
This paper may interest you
AustinDillon75 , adjusting bookies prices with overround to fair odds, power method probably best of those tested ?