Measuring Real-World Performance

Why live greyhound results trail your backtest

Most bettors who backtest carefully find that their first live weeks feel underwhelming. The chapter treats this as normal and gives three honest reasons. Slippage means the odds move between your model flagging a bet and you placing it. Selection friction means you miss races, stake the wrong amount or lose a price. Market adaptation means a pattern that is taken consistently tends to be priced against you over time.

None of these make your backtest worthless. They are costs to budget for. A rough rule of thumb offered here is to knock one to two percentage points off your backtested yield before treating it as your live expectation. It is an approximate adjustment, but it keeps your expectations realistic while real data builds up, and it stops you abandoning a working system in a panic.

Yield, strike rate and ROI: the three core metrics

You could track dozens of numbers, but the chapter argues that three, recorded weekly, tell you almost everything. Yield is profit as a percentage of total stakes, so plus four per cent means four pounds profit per hundred staked. Strike rate is the share of bets that win. Return on investment is often used interchangeably with yield, although strictly it can include the returned stake. Choose one definition and stay with it.

Yield is the number to steer by. Strike rate only means something alongside the average odds you take, and a very high strike rate at short prices can be a warning sign rather than a success. On a flat stake, a yield of two to four per cent over hundreds of bets counts as a meaningful edge in greyhound betting. Remember too that the bookmaker's overround is a built-in margin you must clear before any profit appears.

Drawdown and longest losing run

Yield tells you whether you are winning, but not how much pain the journey involves. Drawdown is the percentage fall from your highest bankroll point to the lowest point after it. If a bank peaks at £1,200 and later drops to £960, that is a 20 per cent drawdown. It shows how much you should be ready to give back during losing runs, which arrive even when the edge is real.

The longest losing run is your system's worst streak of consecutive losers. Knowing that your system has previously produced a twelve-bet losing streak makes an eleven-bet streak far easier to sit through without breaking your staking rules. It also matters structurally, because your staking plan should be sized so a typical losing run does not remove a large slice of the bankroll. Write both numbers down somewhere you will see them.

How many bets before you can trust your greyhound results

Separating luck from skill is the hardest part of measurement. After fifty bets, variance alone can swing yield by ten percentage points. The chapter suggests treating results as a distribution: with a true edge of three per cent, your yield after two hundred bets could plausibly sit anywhere from about minus two to plus eight per cent through luck alone. That is mathematics, not failure.

Practical thresholds help. Under 100 bets, results are anecdotal and only useful for spotting catastrophic errors. Between 100 and 500, they start to inform but remain well within noise. From 500 to 1,000, a stable figure begins to show a meaningful signal, and beyond 1,000 you can start trusting the headline number. Fixed field sizes make variance per bet fairly predictable, but they do not remove the need for volume.

Keeping a betting log that tells the truth

If you adopt one habit from this chapter, make it a clean, complete betting log from your first live bet. The log turns a sequence of bets into a dataset. Each line should hold the date and time, track and race number, dog and trap, decimal odds taken, stake in units, your model rating or probability, a short consistent reason for the bet, and the result with profit or loss.

Consistency matters more than which tool you use. A spreadsheet with one row per bet is the minimum viable log, and a small SQLite table is better still. Watch for the common traps: recency bias, logging winners carefully and losers loosely, changing model rules halfway through a sample, and letting stake size drift when you feel confident. If you change the model, close the old sample and start a fresh one rather than half-merging logs.

A monthly review habit for greyhound bettors

Weekly is too often to draw conclusions and yearly is too late to catch a broken system, so a monthly review is recommended. Look at the month's flat-stake yield, strike rate at the average odds taken and largest drawdown so far. Then ask whether you followed your staking plan on every bet, whether you logged every bet, and whether a negative month is explained by luck or by something structural.

That last question matters most. A losing month early in a live sample is almost always noise, while a losing month at bet eight hundred deserves a harder look. Measurement lets you be patient with variance and firm with systems that are genuinely broken, and comparing live results with backtested expectations creates a feedback loop that, over time, is the real advantage.