Every trader who’s ever gone on a good run has felt it: the quiet certainty that they’ve figured it out. That feeling is worthless as evidence. The market hands out winning streaks to skilled traders and coin-flippers alike, and it never tells you which one you are. The only thing that can is a log big enough to outvote luck.
Why you can’t feel whether an edge is real
Human intuition is catastrophically bad at separating skill from variance, and it fails in a specific, predictable direction: we credit skill for wins and blame luck for losses. A green week feels earned; a red week feels stolen. Both feelings are unreliable narrators.
The reason is that a real edge and a lucky streak produce identical short-term results. If your strategy has a genuine positive expectancy, it will still throw losing weeks. If it has no edge at all, it will still throw winning weeks. Over ten or twenty trades the two are indistinguishable by outcome — and your confidence is calibrated to outcomes, not to statistics. This is exactly how traders talk themselves into scaling a dead strategy right before it takes the account down.
You can’t reason your way past this bias in the moment. You have to outsource the judgment to the data.
The sample size that separates signal from variance
An edge is a claim about a distribution, and you can’t verify a distribution from a handful of draws. The core problem is that small samples are dominated by variance. A strategy with a modest true edge can easily show a strongly negative — or misleadingly spectacular — first twenty trades purely by chance.
What this means in practice:
- A dozen trades tell you essentially nothing about your true expectancy.
- A few dozen start to constrain the range, but loosely.
- It takes a genuinely large, consistent sample before the noise averages out enough to trust the center.
There’s no universal magic number — the noisier your strategy and the closer its edge is to zero, the more trades you need. But the direction is always the same: more logged trades, taken the same way, tighten the estimate. This is also why cherry-picking hurts you. If you only log the trades you’re proud of, or trade five different strategies and blend them, every trade is a draw from a different distribution and the sample never converges on anything. Prove one edge at a time, and log all of it — winners, losers, and the embarrassing ones especially.
Deriving expectancy and a confidence interval from your log
Once the trades exist, the proof is arithmetic. Your log gives you win rate, average win, and average loss, and from those you get expectancy — the average result per trade. Positive and stable across a large sample is what “real edge” actually means. Walk the calculation through an expectancy calculator, and if the mechanics are unfamiliar, the expectancy primer breaks it down.
Do the work in R-multiples so the number survives changes in position size — read the R-multiple explainer if that’s new. An expectancy of +0.4R means nothing changes when you move from a small challenge account to a larger funded one; the edge travels with you.
But the point estimate is only half the proof. The other half is the confidence interval — a range that says, given your sample size, where your true expectancy probably sits. A +0.4R expectancy from thirty trades and a +0.4R expectancy from three hundred trades are not the same claim. The first has a wide band that may still dip below zero; the second has a tight band you can bet size on. The interval is what converts “I’m up” into “I have an edge.”
Reading the range, not just the point estimate
The single most important habit in proving an edge is to read the range, not the center. A point estimate flatters you — it names one hopeful number. The interval keeps you honest by showing how much that number could be luck.
Here’s how to act on the band:
- Entire interval above zero — you have statistical grounds to call the edge real. The worse-case end of your range is still profitable.
- Interval straddling zero — you don’t yet have proof, no matter how good the center looks. Keep trading the same way and let the sample grow before you scale.
- Interval below zero — the data is telling you what the P&L already did. Stop and rebuild the thesis.
A Wilson confidence interval is the right tool here precisely because it behaves well on the small samples traders actually have — it doesn’t overclaim early, and it tightens honestly as trades accumulate. Shibiki computes exactly this as a live edge-health band per strategy, so you’re always reading the range instead of a single seductive number.
Measuring edge continuously as new trades post
Proving an edge is not a one-time certification — it’s a standing measurement. Markets change, and a proven edge can decay, so the interval has to keep updating with every trade rather than being calculated once and framed on the wall.
The obstacle to continuous measurement is friction. If proving your edge requires exporting spreadsheets and recomputing statistics by hand, you’ll do it twice and then stop — right when you most need the ongoing signal. The fix is to remove the manual step entirely. Shibiki’s auto-journaling captures each fill straight from the broker, so the sample grows on its own and the edge-health band recomputes as trades post. You watch the interval tighten (or widen) in real time instead of trusting a stale snapshot.
That continuity is what turns a journal from a diary into an instrument. Instead of feeling whether you still have an edge, you can see the current range at any moment — and on a prop account, where scaling into a firm’s limits on a phantom edge is how accounts die, that difference is the whole game.
Related: Expectancy calculator · Expectancy explained · R-multiple explained