How to Read the Performance Page (and Protect Your Bankroll)

Our performance page publishes every result, including the bad ones. That honesty is only useful if you know how to read it. This guide walks through the numbers on that page, then covers the part that decides whether a good model helps you at all: how much you stake, and how you behave after a losing run.

1. Start with the honest number, not the best one

The first thing you see is the return of the current engine with a confidence interval next to it, for example +0.4% ± 3.5 points. Those two numbers belong together. The percentage is what happened; the interval is how much of it could be noise. When the interval crosses zero, the honest reading is "not distinguishable from zero yet", not "we are profitable".

You will also see a provisional label until a version has enough days and enough settled bets behind it. That label exists so that a lucky first month cannot look like proof. A period is a sample, not a promise.

2. Hit rate is not profit

A market can win 70% of the time and still lose money, and another can win 45% of the time and make money. What decides it is the price. This is why every table on the page ranks by return, not by hit rate, and why the strongest cells are sorted by the lower bound of return rather than the headline figure.

3. Why playing every selection drains a bankroll

On a busy day the board can carry dozens of selections. If you stake the same amount on every one of them, your risk for the day is set by the fixture calendar rather than by you. Two quiet days and one heavy day produce completely different exposure with the same "flat stake" habit.

The bankroll cards on the page make this visible: the rule that risks a fixed share per bet behaves very differently from the rule that splits one daily budget across that day's bets, even though both are called "flat". Same picks, same odds — different survival.

4. Stake discipline in three decisions

• Decide the daily share first. A single number, for example 5% or 10% of the bankroll for the whole day, split across that day's selections.
• Cap the single bet. Even on a quiet day, one selection should not carry the whole daily budget. A per-bet ceiling is what keeps one result from mattering too much.
• Recalculate from the bankroll, not from the loss. Stake sizes follow the balance at the start of the day, not the amount you are down.

The cards on the performance page are built exactly this way, and the assumption is stated under them: every published bet placed at the odds we published, staked by the rule, none skipped. Your own results will differ if you skip selections or get different prices.

5. Why loss-doubling systems end in ruin

The most common amateur system is doubling the stake after a loss to "recover" it. The arithmetic is not an opinion; it is fixed. At even money, starting from 1 unit, a run of losses requires 1, 2, 4, 8, 16, 32, 64, 128 units. Eight losses in a row cost 255 units to chase a 1 unit profit. In football markets, eight consecutive losses are not rare over a season — they are expected at some point.

ExPrysm never publishes a staking plan that increases after a loss, and the bankroll cards on the performance page are all fraction-of-bankroll rules: they get smaller after losses, not bigger. That is the difference between surviving a bad month and ending it.

6. How to use the strength table

The page shows where we have been strong and where we have been weak, by league and market cell. Use it as a process tool, not as a shopping list:

• Read the sample first. A cell with 30 settled bets tells you far less than a cell with 300, however good the number looks.
• Prefer stability over peaks. A cell that is quietly positive across months is worth more than one that spiked in a single month.
• Expect cells to change. Squads, referees and schedules change; a strong cell can cool down. Re-read the table over time instead of memorising it once.
• Do not treat a weak cell as a signal to bet the other side. Weak means we have not shown an edge there; it does not mean the opposite has one.

7. Timing, lineups and closing prices

The second tab covers the part most pages hide. Closing line value compares the price we published with the price the market closed at: consistently better prices mean we were early on real information, and it is a slower but more honest signal than a hit-rate streak. The lineup check re-runs a match about an hour before kickoff, when confirmed lineups arrive, and can drop a published selection. A dropped selection is shown struck through and is not counted as a win or a loss anywhere on the page.

8. Calibration: what "confidence" actually promises

The calibration curve compares predicted probability with what actually happened. If we say 60% and those selections land about 60% of the time, the model is calibrated. Calibration is necessary but not sufficient: it proves the numbers are honest, not that they beat the price. That is why the page also shows return by confidence threshold — and why filtering for "higher confidence" does not automatically improve return.

9. Amateur mistakes the page is designed to expose

• Judging by one day. A single day is noise. Look at the monthly table and the version timeline instead.
• Chasing after a losing day by raising stakes. See section 5.
• Counting pending or voided bets as wins. Only settled bets count; a push returns the stake.
• Mixing accumulators with measurement. Multiplying selections multiplies variance; the page measures single bets.
• Keeping no records. If you skip selections or take different prices, our numbers are not your numbers. Write down what you actually placed.

10. This is a long-haul measurement

Football markets are efficient enough that any real edge is small and slow. Sixty to ninety days is a reasonable window to judge a version; three to five days is not a window at all. The page is built to make that visible: negative months stay on the table, weak cells stay in the list, and every claim is derived from published results rather than written by hand.

Read it like a track record, size your stakes so that a bad month is survivable, and treat any single day — good or bad — as one sample.

— ExPrysm Team