Do you filter out races with clearly broken BSPs?

[GaryCook]

Hey all — quick question for those doing large-scale greyhound backtesting or model training.

I’ve noticed a small number of races each year have one or more runners with extremely high BSPs — often 1000+, which are clearly unmatched or dead-runner scenarios. These can really skew metrics (e.g. P&L, Kelly staking, even price prediction).

It seems obvious we should remove runners at BSP = 1000+, but what about other cases?

Do you filter out entire races based on extreme prices?

Any heuristics for deciding when a race is too noisy or broken to trust?

Would love to hear how others handle this — especially if you’ve found a middle ground between throwing out too much vs letting bad data through.

Thanks in advance.

[ShaunWhite]

Hi Gary,

• Nothing should have a bsp bigger than 1000 so bin those races.

• Excl races with <6 finishers, the price data might contain a big change mid market due to an adjustment factor/withdrawal. If it's api data look for a market version change during the period you're looking at.

• Sensible overround on the BSP say 95-105.

• Sum of BSPs <500.

• Fav shorter than 6/1

With a big enough sample the few remaining oddities shouldn't skew the overall picture too much.

--------
Kelly, imo not applicable, small errors in est EV calc cause issues staking at big prices. Also markets are stake limited so bankroll management isn't usually the issue. Kelly was great for people having 5 bets a week and not wanting to go broke, or at the casino with accurate EV estimates, but exchange betting/trading is 100s or 1000s of bets a week and EV on any given selection is at best a finger in the wind.