The most expensive number in betting software is a win rate. It is the first figure anybody quotes, it is almost always high, and on its own it tells you close to nothing about what a strategy will do to your money.
These are the checks worth running on any backtest — yours, ours, or one you saw on YouTube — before it changes how you play.
A high win rate hides what the bad sessions cost
You can set a win rate to almost any number you like just by choosing how deep a recovery you are willing to run. Deeper ladder, higher win rate, bigger disaster when it fails. The number goes up and nothing has improved.
Take a strategy that wins 1 unit on 95% of its sessions and loses 30 on the other 5%. Play a hundred sessions:
| 95 winning sessions, +1 unit each | +95 units |
| 5 losing sessions, -30 units each | -150 units |
| Where you actually are | -55 units |
Ninety-five percent of your nights end well. You are down fifty-five units. Both of those are true at the same time, and only the first one makes it into the video.
The question is never "how often does it win?" It is "what do the losing ones cost, and how many good nights does it take to pay for one bad night?" If the answer is ninety-five, you are one unlucky session away from wiping out everything the strategy has ever earned you.
Do that sum on any strategy you are considering. Winning sessions times what they make, against losing sessions times what they cost. It is primary-school arithmetic and it disqualifies most of what gets published.
Could you actually survive it?
This is the check that decides whether a strategy is playable by you, and it is the one most often skipped entirely.
A strategy can have shallow dips and be played all evening. Another one, ending in exactly the same place, can need a 255-unit recovery and empty you inside an hour. Same destination, completely different journey — and you only complete the journey if you are still at the table.
Get these three numbers before you play anything:
- The worst drawdown in the whole backtest — how far it fell from a high point before recovering.
- The deepest losing run, and what the ladder needed at its very worst moment.
- The bankroll it assumed, against the bankroll you actually intend to bring.
If the deepest run in the test needed 255 units and you are turning up with 100, that backtest does not describe you. You will be stopped out long before the strategy gets the chance to do what it did on the screen. Sizing a bankroll properly works through the arithmetic.
Is the sample big enough to tell luck from signal?
Over a short run, ordinary randomness swings far harder than any edge either of you has. That is the whole story of a single session, and it is just as true of a small backtest.
Which means a few hundred spins cannot tell a good strategy from a lucky one. Not "probably cannot" — cannot, in principle.
| Spins tested | Ordinary swing | What the wheel takes |
|---|---|---|
| 100 | ±10 units | 2.7 units |
| 1,000 | ±32 units | 27 units |
| 10,000 | ±100 units | 270 units |
| 100,000 | ±316 units | 2,700 units |
Watch the two columns change places. At 100 spins the noise is nearly four times the wheel's cut, and any result is meaningless. By 100,000 the cut is nine times the noise and there is nowhere left to hide.
Anything claiming a real result on a few hundred spins is reading its own luck. This is exactly why the free million-spin dataset ships with every purchase — it is not a bonus, it is the minimum sample at which most of these questions have answers at all.
Does the selection survive without the progression?
The most useful check on this list, and the most skipped.
A staking plan can make almost any selection rule produce a rising curve for a long time, because it converts frequent small wins into that shape and buries the cost in rare deep drops. Test them together and you cannot tell which one produced the result.
So pull them apart:
- Run the selection flat — one unit, every trigger, no progression at all.
- See how it does. If it bleeds at roughly the rate the wheel charges, the selection is neutral. That is normal and it is fine.
- If it bleeds much faster than that, the selection is actively working against you and no staking plan will rescue it.
- Only then add the progression — understanding that you are choosing a risk profile, not adding an edge.
If a strategy looks strong with a progression and ordinary without one, the progression is the entire result. That is not a reason to bin it — but you are choosing a shape, not a winner, and you must size your bankroll for the rare heavy loss rather than the typical night.
What data was it tuned on?
A rule with enough adjustable knobs will fit any dataset you point it at, including one made of pure noise. That is not a flaw in the rule; it is what searching does.
- Tune on one slice, prove it on another. Fit the parameters on the first half of the data, then test them untouched on the second half.
- Distrust sharp peaks. If a setting is superb at 7 and poor at 6 and 8, that is an artefact of the search, not a discovery. Real effects sit on broad plateaus.
- Count what you tried. If you swept 400 combinations and kept the winner, it looks good partly because it was the best of 400.
Does the trigger fire often enough to matter?
A rule that waits for a rare pattern can look wonderful and still be useless, for two reasons.
The sample of actual bets is far smaller than the sample of spins, so the result is much noisier than the spin count suggests. Ten thousand spins with a trigger that fires once every 300 is a 33-bet sample — statistically worthless, however impressive the header number looks.
And sitting through 300 spins to place one bet is not something anyone actually does. Long before the strategy triggers, boredom will have you betting something else.
Always look at bets placed, never spins examined.
The short version
Before you act on any backtest, including one of ours:
- What do the losing sessions cost, and how many wins pay for one?
- What is the worst drawdown, and does my bankroll cover it?
- How many bets — not spins — does the sample contain?
- Does the selection hold up flat, without the progression?
- Was it tuned and tested on different data?
- How often does the trigger actually fire?
Six questions, a couple of minutes in the Strategy Builder. Most published results fail at least two of them, and finding that out on a laptop costs nothing at all.
What VIPER does with these
Every check above needs a number that most tools simply do not report, which is why we built the testing side around them rather than around a headline win rate.
- The backtester reports the whole distribution, not the highlight: worst drawdown, deepest losing run, what the ladder needed at its peak, and bets placed as distinct from spins examined. The figures that answer questions two, three and six are on the screen, not left for you to work out.
- The free million-spin dataset ships with every purchase, because question three has no answer at all below tens of thousands of spins.
- Flat and progression are separable, so you can run the selection on its own before you let a staking plan flatter it.
- The dials are real dials. Anything the AI Wizard exposes as a parameter is verified against the generated code, so a sweep tunes what you think it is tuning.
And when a strategy does survive all six, bankroll protection is what makes the backtest mean something at the table — because a tested strategy played on a bankroll that cannot absorb its worst run is still a losing evening waiting to happen.
— Team Viper