Your Quantitative Betting Plan

Why a greyhound betting plan needs a routine

Many readers feel energised by a book like this, then drift back to old habits within days: a tipster's message, a flashy card, a quick bet on the favourite. The chapter argues this happens not because the ideas were wrong but because nothing was planned for the boring middle stretch after setup.

The numbers are also stacked against you. A standard six-runner UK greyhound race carries roughly a 125 per cent overround, around a 25 per cent structural margin for the bookmaker built into prices before a dog leaves the trap. No clever weekend overcomes that. Only weeks of careful, consistent work give a system a chance, which is why the rest of the chapter reads as an operating manual rather than a list of borrowed best practices.

The first thirty days: setup before live betting

Resist betting on day one. The first month is about getting the plumbing right, split into four rough weeks. Week one covers data: choose a primary results source and a second source for verification, confirm coverage of the tracks and grades you want, settle on a history window such as three years, and pick a simple, durable store like a local SQLite database or a well-named folder of CSV files.

Week two covers your stack: one analysis environment kept for at least a quarter, a schedule that brings results in within an hour of the final race, and three written sanity checks such as matching row counts, no missing trap numbers and no duplicate races. Week three builds the simplest possible rating and checks it against the favourite strike rate, which sits around 33 per cent in graded UK races and ranges from 31.6 to 42 per cent by venue.

Paper betting, stake rules and your bet log

Week four is about going live carefully. Place at least fifty paper bets through the model without real money, recording the price you would have taken, the stake and your reasoning in one line. Decide your unit size, often 1 to 2 per cent of bank, write it down and do not change it during your first three months of real betting.

Set up a bet log covering date, race, selection, price, stake, model probability, market probability, result, and profit or loss. By day thirty you should have five things in place: a dataset, a working stack, a logged model, a paper-betting trail and a written stake plan. If any one is missing, the advice is simple: do not go live yet.

A weekly review of about ninety minutes

The biggest threat to a system is not a bad model but a missed weekend, because a few unlogged weeks and you lose the thread. So the weekly review is kept to roughly ninety minutes spread across seven days. Monday brings the weekend's results in and runs the validation rules. Tuesday generates the week's cards and lists value races with model price against market price. Wednesday places midweek bets, logging each before placing it.

Thursday settles those bets and updates running profit and loss plus a four-week moving strike rate. Friday refreshes the weekend cards, looking for the largest edge rather than the biggest races. Saturday places weekend bets at identical plan stakes. Sunday settles the week with one page of notes on what the model got right, what it missed and one thing to investigate, the step the chapter calls most important.

A monthly audit calendar for your model

Weekly reviews keep you current; monthly audits keep you honest. Run one on the first weekend of each month, ideally before that month's cards are priced, with a different focus each time. Examples include reviewing yearly profit and recalibrating stakes, checking which inputs earn their place, auditing trap bias (Trap 1 winning roughly 18 to 19 per cent against a theoretical 16.6 per cent), and comparing model probabilities with the market's.

Later months re-run a small four-week backtest, review data quality, stress-test stakes against the worst losing run, check for overfitting by hiding the latest month, audit venue-level performance, question recent improvements, and compare realised strike rate with model probability, since a 20 per cent shot should win about 20 per cent of the time. December pauses live betting for two days to plan the year ahead.

When to pause, refine or scale back

The chapter separates three responses. Pause is a short stop: three losing sessions while probabilities still line up means a 24 to 48 hour break, failing validation rules means pausing until three clean weeks of data, and hunting extra bets outside the model means two weekends off. Refine is a planned change: a strike rate well below model probability for two weeks calls for a week of paper betting and a retest.

Scale back is a deliberate cut in risk. If the bank drops more than 20 per cent from its peak, halve the unit for a full month; if bets cluster on short-priced favourites, reduce stakes until longer-priced value is also taken. Each action needs a defined length, because an open-ended pause becomes drift. Keep the system boring, log everything and accept losing weeks, since discipline on a bad week is what lets an edge be taken.