EV above the model is a candidate for alpha, not a finding.

The claim and the caveat

Alpha begins there. It is not established there.

Live expectancy above the benchmark is a necessary condition for alpha and nowhere near a sufficient one, and the page's two sentences carry that entire distinction. Read alone, the first sentence licenses a trader to conclude they have found something on a single favourable reading. The second sentence exists to prevent exactly that, and the mechanism it specifies is confirmation by three independent metrics before the reading is treated as a finding. Two sentences carrying that much weight is a risk in itself, which is why the confirmation requirement is implemented as a check in the workflow rather than left as a caution in the copy.

What each confirmation rules out

Three specific ways to exceed the model without producing alpha.

Each confirming metric closes off a distinct false positive, and all three are ordinary rather than exotic:

• RAPF — rules out excess earned by luck rather than repeatable profit quality.
• RAER — rules out excess bought by deploying more risk than the model assumed.
• ACCEL — rules out excess produced by a machine that is already decaying underneath it.

The most common false positive

Excess earned by quietly running hotter than the benchmark assumed.

Of the three, the risk-inflation case is by far the most frequent and the least likely to be noticed. The benchmark simulated a system at authorised tier; the live account ran somewhat above it, through override, drift, or a tier that was never formally demoted. Live EV duly exceeds the model — and the excess is compensation for additional risk rather than additional skill. RAER catches it because efficiency asks what was converted per unit of risk deployed, which is precisely the ratio that stays flat when the outperformance is purchased rather than earned.

Why three and not one

The false positives are independent, so the checks must be too.

A single composite confirmation would fail here for the same reason a composite metric fails generally: the three false positives are unrelated to each other, and an account can exhibit exactly one of them while looking healthy on the others. A blended check would average a severe failure on one axis against comfortable readings on two, producing a passable score for an account with a specific, identifiable, disqualifying problem. Independence in the failure modes requires independence in the checks.

The direction of the burden

The reading is treated as guilty until confirmed, and deliberately so.

Note where the burden of proof sits. An EV reading above the model does not get provisional credit pending review — it is a claim, and it earns nothing until the confirmations pass. That asymmetry is intentional, because the cost of the two errors is unequal: wrongly dismissing genuine alpha costs a period of unnecessary caution, while wrongly crediting a false positive leads to sizing up on an edge that is not there. The second error is the one that damages accounts, so the process is tilted against it.

What confirmed alpha licenses

Less than traders expect, and it is not a size increase.

Even fully confirmed, the finding does not authorise deploying more. Sizing authority comes from the gate ladder and the tier model, which read drawdown and evidence depth — not from outperformance. What confirmed alpha licenses is narrower: continuing to run the system with justified confidence, and treating the branch producing it as a candidate for tier review through the ordinary evidence process. The distinction matters because the intuitive response to discovering alpha is to press it, and pressing it is how it gets converted back into variance. The gap between what a finding licenses and what it feels like it licenses is one of the more reliable places for a well-run account to be damaged, precisely because the finding was genuine.