## Expressing Different Confidence Levels
Not all bets are equal. Some selections are based on stronger evidence, more reliable models, and larger estimated edges than others. Confidence-based staking attempts to capture this variation by betting more on higher-confidence selections.
## A Simple Variable Staking Scale
| Confidence Level | Unit Size | When to Use |
|---|---|---|
| Low (marginal value) | 0.5 units | EV 1–2%, uncertain probability estimate |
| Standard | 1.0 units | EV 2–5%, well-supported analysis |
| High | 1.5 units | EV 5%+, strong model signal + contextual support |
| Maximum | 2.0 units | EV 8%+, multiple confirming signals |
The maximum should rarely exceed 3× the minimum. Anything larger destroys the disciplined consistency that makes this approach work.
## The Danger of Subjective Confidence
The risk: "confidence" becomes a rationalisation for betting more on selections you emotionally prefer. True confidence-based staking requires your confidence level to be based on objective metrics (estimated EV%, model accuracy, CLV expectation) — not gut feeling.
## Linking Variable Staking to Edge
The most principled variable staking system links stake directly to estimated EV:
Stake = (Estimated EV%) / (Reference EV%) × Standard Unit
At standard unit = 1 and reference EV = 3%:
- 1.5% EV bet: stake = 0.5 units
- 3.0% EV bet: stake = 1.0 units
- 6.0% EV bet: stake = 2.0 units
This is a simplified proportional Kelly approach — mathematically grounded and defensible.
## Tracking Variable Staking Performance
When using variable stakes, always track both flat-stake ROI (to measure selection quality independently) and actual ROI (to measure the contribution of your staking decisions). If actual ROI consistently underperforms flat-stake ROI, your variable staking is adding noise rather than edge.
Create a free account to track your progress and save bookmarks.
