How to keep a betting log: what to record and why

A log exists to answer one question: is what has happened so far a method or chance? The balance alone doesn't say. You need the number of bets and the odds, because those are what set how much noise surrounds the result.

The scale of that noise is brutal: after a hundred bets at 2.00 the margin of error on return is ±19.6 points. An observed +15% is consistent with a true −4%. Without knowing how many bets you've placed, you can't even interpret that +15%.

The eight fields you need

• The date and time of the bet — not of the match: it tells you how far ahead of kick-off you played.
• The event and the market, written so they still make sense in six months.
• The odds at the moment you bet. It is the most important field after the outcome, and the one most often skipped.
• The stake. You need it for yield, which is profit divided by turnover.
• The operator. Without it you'll never know whether part of your result depends on where you play.
• The outcome: won, lost or refunded. A refund isn't a draw: it removes the bet from the count rather than counting it as zero.
• The reason, in one line. It isn't for today's you, it's for the you rereading in three months.
• ⚠ The closing price, if you can capture it. It is what makes the check in the next section possible.

Why a refund shouldn't count as a draw

A refunded bet — a Draw No Bet that ended level, a void event — isn't a neutral result: it is a bet that didn't happen. Counting it as «zero» inflates the denominator of your yield and makes the return look lower than it is.

It is how our own log treats it: the bet leaves the count, and an accumulator's total odds are recalculated by dividing out the refunded selection's odds.

The two things a log shows and memory doesn't

The first is chasing. After a run of losses, stakes rise: it is the most common behaviour and the hardest to notice from inside, because each individual increase feels reasonable. In a log it shows up immediately, by checking whether average stake grows after the bad days.

The second is memory's selection. Memory keeps the episodes that confirm — the time the streak held — and lets the others go. A log keeps those too, and that is the whole difference.

How to read it, once it exists

Three cuts, in this order. By odds band: almost everyone has a different return on short prices and long ones, and often it is the only true thing that emerges. By market: the 1X2 and Under/Over don't cost the same, and which of the two is cheaper changes from operator to operator. By operator: the same market carries different margins from one book to another, and on some leagues the difference runs to several points. ⚠ None of the three is a view in our log: you get them by exporting to CSV, which is one button, and grouping in a spreadsheet.

⚠ And in each cut, check how many bets sit inside it before reading the percentage. Splitting a hundred bets across five odds bands leaves twenty per band, and twenty bets measure nothing.

When a log starts to say something

Not soon. With a true 2% edge at 2.00 it takes around 9,600 bets for return to separate from zero; with a 5% edge, 1,533. Those numbers put «I'm doing well» back into scale.

Which doesn't make the log useless in the meantime — quite the opposite: the two things above, chasing and memory's selection, show up far earlier, and they are the ones you can actually act on.

⚠ The OddSonar log currently keeps six of them: date, sport, event, operator, stake, odds, plus the wallet and the outcome. It has no field for the time, the market, the reason or the closing price — those go at the end of the event field or in a separate sheet. What it does by itself is settle bets against the results, so at least the sums don't depend on what you remember.