Summer updates to RateThat.Dog — new models & Irish racing
We've launched Irish greyhound racing, built a new enhanced model using Claude's Fable, and gone live with dynamic dog rankings and speed ratings. Here's what's new.
Irish racing is live — with caveats
Our models are trained on years of British data. Irish racing is a new frontier.
We've launched Irish greyhound racing on ratethat.dog. Our prediction models have spent months learning from British form, trap bias, and track characteristics — so we're the first to admit they'll need retraining with live Irish data to reach the standard you see on British racecards.
We've already started improving after a cold start. The early results show promise, and we're confident that Irish tracks will deliver plenty of winners over the long term. For now, treat the predictions as experimental. We're actively tweaking the model, and we'll keep you updated as the system learns.
The good news: the platform itself is ready. Racecards, dog profiles, track pages, and predictions are all live for Irish racing — just with lower confidence initially. Watch the predictions improve in real time as the data accumulates.
Fable model in production — 28% strike rate in backtest
We've built a new model using Claude's latest Fable and it's performing exceptionally.
We just built an enhanced prediction model using Claude's latest Fable model, and the backtest results are exceptional. Even including Irish races in the dataset (where we're still learning), we're getting a **blanket 28% strike rate**. That's significantly better than our current production model.
On specific track and distance combinations — where the model has the most data — we're seeing even stronger performance. Take Central Park: we're hitting around 33% win rates on Hot Dogs candidates. And we haven't even started cutting the model down to the distance/grade/track combos where greyhound models really excel.
The Fable model is currently building in the background on live data, accumulating real-race outcomes to reach statistical significance. Within the next few days and weeks, it will start getting selected by our pick engine when the evidence is strong enough. We'll run both models in parallel during the validation phase.
Multi-model selection — the best pick for each race
Our engine now selects the best model for each distance/grade/track combination. This means that as Fable accumulates live data, it will automatically start replacing the current model where it's proven itself. You won't need to flip a switch — the system picks the strongest model for that specific scenario.
In practice, this means you might see slightly different picks in high-Fable-confidence races compared to races where we still prefer the current production model. The model selection happens behind the scenes; on the racecard, you just see the best pick.
This approach lets us deploy new models safely. Fable starts with zero live data. By the time it reaches statistical significance and starts getting selected, we've run thousands of validation bets. No nasty surprises.
Dog rankings now live — moving daily
We've launched dynamic dog rankings based on the latest performance data. Every day the rankings move as dogs race and the model updates. Scroll through the rankings page to see the current top performers — you'll notice the list changes daily.
This is useful for spotting dogs that are on a sharp upswing (climbing quickly = recent form is strong) or those trending downward. The rankings are one thing, but the trajectory is often the signal. A dog climbing 10 spots in a week is worth watching.
You can already access dog rankings on individual race pages and the rankings hub. Expect to see these integrated more deeply into track analysis and trainer pages over the next few weeks.
Speed ratings now live — find the quick dogs
Speed ratings are now live across the platform. Use them to find dogs that are extraordinarily quick. On the dog selector, you can filter by speed rating — and on individual race pages, you'll see the speed breakdown for every runner in the race.
Speed is one dimension of the prediction model (it feeds into composite score), but seeing it in isolation is useful. A dog that looks average on composite but exceptional on speed might be valuable if the race setup suits front-runners. Conversely, a quick dog with a poor record at this distance deserves caution.
The speed ratings are distance-relative — a 120 speed at 480m is different from a 120 at 305m. The context is always shown, so you can calibrate expectations for the race distance.
What's next?
The Irish model will keep improving — we'll retrain it as the data sample builds. Fable will gradually take over selections as its performance validates. The rankings and speed ratings will feed into more of the platform — expect to see better track/trainer/condition-specific analysis soon.
We're also in the early stages of rolling out some new systems and strategies built on the richer feature set we now have. You'll hear about those soon.
Thanks for being here while we build this. Keep an eye on the blog for updates.
Frequently asked questions
How good are the Irish racing predictions?
They're improving daily, but starting from a cold base. The models were trained on British data, so they need Irish form to calibrate. Initial results are promising, but treat them as experimental for now. The predictions will improve as we accumulate Irish data.
What's the Fable model? When can I use it?
It's a new prediction model we built using Claude's Fable. Backtest shows 28% strike rate (significantly better than the current model). It's building on live data now and will start getting selected for races where it has statistical confidence — likely within days or weeks.
Will my saved systems still work?
Yes. The model swap happens behind the scenes. Your saved filters and systems continue to work; they'll just be based on slightly different underlying picks as Fable gets selected. No action needed from you.
What are dog rankings and how do I use them?
Dynamic rankings based on the latest performance data — updated daily as dogs race. Use them to spot dogs on form upswings (climbing rapidly) or downswings. Check the rankings page to see current standings.
What are speed ratings and why do they matter?
Speed ratings are 0-120 relative to the field. They feed into composite score but seeing them in isolation is useful — you can spot quick dogs that might be valuable at distance-specific races. Available on the dog selector and all race pages.
Is Irish racing free to use?
Yes — racecards, predictions, and all data pages for Irish racing are available to all users. Like British racing, some advanced analysis is locked behind subscription tiers.