SignalProphet
The significance gate

Why the track record won't call a lucky streak an edge

This is the most important page in the docs. The track record only publishes a verdict of “edge” once the sample is large enough — and consistent enough — to actually mean something. Until then it says “still building.” Here is exactly what that means.

The problem it solves

A handful of winning trades proves nothing. Flip a coin eight times and you can easily get six heads — that isn’t a biased coin, it’s just variance. The same is true of trades: a short winning run can look impressive and mean nothing. The classic trick of the industry is to take that lucky run and annualize it into a huge headline number. SignalProphet refuses to do that.

So before the scorecard is allowed to say the analyst has an edge, the result has to clear a statistical gate. If it can’t, the honest answer is displayed instead: the edge isn’t proven yet.

What the gate actually checks

The gate looks at the realized per-trade returns — the actual outcome of each closed paper trade — and asks whether the average is distinguishable from zero, or whether it could just be noise. A verdict of “demonstrated edge” requires all four of these to be true at once:

  • The average trade is actually positive.An unprofitable record can never pass, no matter how it’s sliced.
  • A significance test rejects “pure luck.” A two-sided t-test on the per-trade returns has to say the positive average is unlikely to be a fluke, at a pre-committed confidence level (its alpha).
  • A confidence interval on the average excludes zero.A bootstrap (resampling the trades thousands of times) builds a range for the true average return, and that whole range has to sit above zero — a second, independent check that the edge isn’t a rounding error.
  • There are enough trades. There is a hard minimum trade count— a floor — below which “edge” is never declared, period. This is the piece that makes “eight winning trades is luck, not proof” true by construction: on a tiny sample, a low-variance run can produce a deceptively strong test statistic, so statistics alone are not allowed to certify an edge on a handful of trades.

Miss any one of those and the verdict is “insufficient evidence.”

What each honest verdict means
Demonstrated edge. All four conditions above held. Only now does the scorecard claim an edge — and it shows the numbers behind it (the sample size, the significance, and the confidence interval), not just a headline.
Still building / insufficient evidence.The average may even be positive, but the sample is still too small to rule out luck — so the page tells you roughly how many more closed trades it needs before the question can be answered. This is the default state, and there is nothing wrong with it: it’s the record being honest.
No positive edge.The average trade isn’t positive. The record says so plainly rather than hunting for a flattering slice.
It never annualizes a lucky streak.Below the significance bar, no forward-looking or annualized number is manufactured from a small sample. “Still building” means exactly that — and it’s shown in amber, the same way the daily read shows “no edge.”

Why two checks, not one

The t-test and the bootstrap confidence interval are two different ways of asking the same question — “could this average just be zero?” — and the gate insists both agree. The trade-count floor is a third, blunter guardrail that a small sample can’t argue its way around. Belt, suspenders, and a hard floor. The point isn’t mathematical elegance; it’s that the record should be hard to fool, including by us.

See the gate in action
The public track record shows the live verdict — including, honestly, whenever the record is still too thin to claim anything. That page is the single quantitative claim we make anywhere.