Discipline — Bankroll and Bet Tracking

Make the uncertainty visible before it becomes a financial problem.

Theory: define the money and the limits

A betting bankroll is money specifically set aside for betting risk, separate from living expenses and essential savings. If losing it would damage your ability to meet obligations, it is not an appropriate experimental bankroll.

For learning, a paper record is enough. You can practise estimating probabilities, recognising price changes and reviewing decisions without placing a bet.

If you choose to use money, set the amount, exposure limits and stopping rules before the session. Planned deposits or withdrawals belong in a separate ledger. Adding money impulsively to recover losses is a different decision from a considered change to a discretionary budget.

No staking method turns a negative expectation into a positive one.

Kelly shows how sensitive staking is to probability

For a single back bet at fixed decimal odds o, with a known win probability p, no extra costs and a log-growth objective, the familiar Kelly fraction is:

f = (p × o − 1) / (o − 1).

Use no back stake when the expression is zero or negative. The simple formula assumes a binary win/loss settlement and valid inputs; a set of related bets needs joint treatment.

With a hypothetical 25% chance at 5.00, full Kelly is 6.25% of the bankroll. Quarter-Kelly would use one quarter of that fraction, about 1.56%.

The arithmetic is not a recommendation to stake that amount. You do not know p exactly. If the true chance were 20%, the same price would offer no edge before costs. A small error in the estimate can change the appropriate decision from a sizeable stake to no bet.

Fractional Kelly reduces exposure relative to the full calculation. It does not guarantee a tolerable drawdown, a particular growth rate or survival. The probability estimates and correlations still matter.

Put exposure limits above the calculator

A calculated stake does not override your budget. Decide what loss you can tolerate on one bet, one race and one session. Consider the combined tickets supporting the same horse or track assumption.

A win bet and a multi-race bet may both fail when one runner underperforms. The same applies to several horses selected because you think the inside rail is unusually good. Their risks can be connected even across races.

There is no universal safe percentage. Smaller exposure may be appropriate when the estimate is weak, the price uncertain or your recent process unreliable. Passing remains available even when a calculator returns a positive number.

Drawdowns depend on the whole setup

A drawdown is the fall from a previous bankroll high. Its possible size depends on odds, frequency, stake fractions, correlations, true edge and errors in your estimates.

A claim such as “expect a 25% drawdown once a year” is incomplete without those assumptions. Even a positive-expectation method can have a long bad run. A negative-expectation method can look convincing for a while.

Precommit to review and stop conditions that fit your actual situation. If something seems wrong, reduce or stop exposure while investigating. There is no obligation to keep betting to demonstrate faith in a model.

A pause can be triggered by a broken data feed or repeated impulsive bets as well as losses. You should not wait for a large drawdown to notice a process problem.

Theory: use the tracker as a feedback loop

A bet log gives you the financial record. A decision log tells you why that record happened. You need both if the aim is to learn.

Keep a betting and bankroll record in a tracker such as The Long Edge Tracker or a spreadsheet. Keep the pre-race decision note alongside it if your version does not have separate fields for each part of the assessment. The method is more important than whether every item occupies its own screen.

Record the event, selection, market, time, stake, accepted price, settlement and net profit or loss. Keep deposits and withdrawals separate from betting P/L.

Then add the learning fields: your pre-race probability, any model baseline, minimum acceptable price and the reason for the decision. Tag the main idea as ability, readiness or conditions, with a more specific note where useful.

Calibration: do your percentages mean what you think?

Over a sufficiently informative sample, horses you assign similar chances should win at roughly those frequencies. If the group you call 30% repeatedly wins much less often, your estimates may be too high or your sample may differ from what you assumed.

Show the number of observations and uncertainty beside the comparison. A few bins from a small sample cannot certify calibration. A well-calibrated forecast can also be uninformative, so compare a proper probability score and a useful baseline on the same races.

If you want to evaluate the whole forecasting method, record whole-race lines or a predefined set of practice races. A bet-only log tells you about your selected bets and can miss errors elsewhere in the field.

Find strengths without inventing them

Group results by a few categories chosen for a reason: jurisdiction, surface, distance group, market or the type of adjustment you made.

Ask whether the pattern repeats in later data. “I am strong on turf milers” is a hypothesis if it emerges from the log, not a licence to scale up immediately. Small slices and large-price winners can create a flattering map.

Compare your adjusted forecasts with the unadjusted baseline. Perhaps your pace work helps while your imagined intent signals make you overconfident. That is actionable learning even before you can demonstrate a profitable strategy.

CLV and execution

For locked-price bets, record a comparable closing benchmark where available, with costs and market differences understood. For tote bets, distinguish the displayed estimate at decision time from the final dividend.

A good price relative to a benchmark is useful evidence about execution, not a guarantee of profit. A correctly chosen horse at a badly executed price can still be a poor bet.

Accountability and useful friction

Make yourself write the probability and minimum price before submitting. If the price changes, compare it with the recorded threshold. Do not silently rewrite the threshold to permit the bet.

If you change an estimate, preserve the earlier version and the reason. If a field is missing, leave it missing rather than fill it from memory after the result.

This small amount of friction slows the decisions most likely to need slowing. It also makes the next review shorter because you are reading a record instead of reconstructing a mood.

Practical: test the risk and start the record

Quiz: risk and records

Up next: put the whole method to work on a race you have not already seen.