Maturity gating: an output that cannot yet be true stays dark.

The principle

Evidence sufficiency is a precondition for display, not a caveat on it.

The QA doctrine defines maturity gating as outputs activating only when the timeframe has enough evidence, and states its purpose bluntly: to prevent a one-week sample from pretending it can prove monthly, quarterly, annual, or twelve-week structural behaviour. The design choice inside that sentence is *activation* rather than annotation. The alternative — computing the value and marking it provisional — is what most systems do, and it fails predictably, because a number on screen with a caveat beside it is a number that will be read and a caveat that will not. Withholding the output removes the temptation rather than labelling it.

Blanks, not zeros

The distinction that keeps the whole discipline honest.

An immature output must be blank rather than zero, and the difference is not cosmetic. A zero is a value: it enters averages, it plots on charts, it satisfies formulas that check for numeric content, and it silently drags every aggregate that touches it. A blank is an absence: it propagates as an absence, refuses to be averaged, and forces any downstream calculation to acknowledge that the evidence is not there. Systems that populate immature cells with zeros produce dashboards that look complete and are quietly wrong in a direction that always flatters early data — the missing months read as neutral rather than as unknown. The blank is what makes the gating structural rather than decorative.

• A zero says the reading is neutral. A blank says the reading does not exist. Only one of those is true.
• Zeros in immature cells bias every rolling window that spans them, and the bias is invisible.
• This is the same rule the SDE pipeline applies to its own early tables — the discipline is system-wide.

Why the impatience is the point

The gate binds hardest exactly when the operator most wants the number.

Maturity gating is at its most useful and least welcome in the first months of a new system, a new branch, or a rebuilt workbook — precisely the period in which an operator most wants structural confirmation that the thing works. The gate's answer is that it cannot know yet, delivered by simply not producing the output. This is uncomfortable and it is the entire value: the early period is when a favourable reading would be most persuasive and least warranted, and when acting on one would set a direction that months of subsequent data would then have to argue against. A system that will tell you what you want to hear at week three has not protected you from anything.

What it does not do

Maturity is about quantity of evidence, not quality of it.

A matured output is not a verified one. The gate confirms that enough periods have accumulated for a structural claim to be arithmetically supportable; it says nothing about whether the underlying data was recorded correctly, routed from the right source layer, or computed by a formula that does what it claims. Those are separate principles with separate mechanisms — source-of-truth routing, black-box testing, and the data QA status stamps. An operator who reads a matured output as a validated one has collapsed two independent checks into one and will trust a well-aged number built on a broken input. Maturity is necessary and nowhere near sufficient.

The key idea

The honest response to insufficient evidence is silence, not a smaller number.

Every alternative to withholding — a provisional flag, a confidence interval, a greyed-out value — leaves a number on the screen, and a number on the screen gets used. Gating accepts a less complete dashboard in exchange for the guarantee that everything visible on it is entitled to be there.