Days are noise. Months are late. The week is where expectancy becomes readable.

The trade-off

Two properties move in opposite directions as the window grows.

Lengthening the aggregation window does two things at once. It raises the proportion of the reading attributable to the process rather than to variance, which is desirable and rises steeply at first. And it delays the moment the reading is available, which is undesirable and rises without limit. Any choice of window is a point on both curves simultaneously, so the question is not which window is most accurate — that is always the longest one — but where the product of readability and timeliness peaks. On a four-trade-per-cycle, four-cycle-per-week cadence, that point sits at a week.

Why not daily

A day contains too few trades for the average to describe anything.

At roughly four concurrent trades per cycle, a single day's outcome is a handful of draws from a wide distribution. The resulting figure is real and describes that day accurately; what it does not do is describe the process, because the variance term swamps the expectancy term at that count. Acting on daily expectancy therefore means acting on noise with the confidence appropriate to signal — which is the mechanism behind risk expansion after a good day and contraction after a bad one, and the specific behaviour the aggregation layer exists to make unnecessary.

Why not monthly

By the time a month reads clearly, four weeks of posture were set blind.

The opposite failure is subtler because monthly figures are genuinely more reliable. The problem is what happens in the interval. Deployment authority is decided every cycle whether or not evidence has been refreshed, so a monthly cadence does not mean waiting to decide — it means deciding four times on stale evidence and then learning. A structural change that begins in week one is acted upon for a month before it is visible, and the month that finally reveals it contains three weeks of decisions made without it.

What the week is not asked to do

The window sets posture; it does not settle whether the edge is real.

It is worth being precise about the claim, because the weekly window is often over-read. A week produces a readable posture signal — is the current expectancy state green, yellow or red, and is the branch mix behaving. It does not produce a verdict on whether the underlying edge exists, which needs far more observations and belongs to the rolling windows and the benchmark. Confusing the two turns a posture tag into a judgement about the method, which is the route to rewriting rules on a red week. The tag governs the next cycle's aggression, and that is its entire jurisdiction. Anything asked of it beyond that — whether the strategy works, whether a branch should be reweighted, whether a rule needs changing — belongs to instruments with longer windows and formal promotion procedures.

• Daily expectancy is real and describes the day, not the process.
• Monthly cadence means four cycles decided on stale evidence, not deferred decisions.
• A week tag governs next cycle's posture — it is not a verdict on the edge.

The rolling windows

Four, six and twelve weeks exist because one week is a point.

A single weekly figure has no direction, and direction is most of what matters. The 4W, 6W and 12W rolling windows sit alongside the current week for exactly that reason — they convert a sequence of posture readings into a trajectory without changing the decision cadence. This is how the layer serves both requirements at once: decisions stay weekly because that is where timeliness demands them, while interpretation borrows length from the rolling series because that is where readability lives. Neither requirement is compromised to satisfy the other. It also means a red week inside a rising 12W series and a red week inside a falling one are different findings, which the tag alone cannot express and the pair together can.

The key idea

The window was derived from the trading cadence, not chosen for convenience.

Weekly review is common enough in trading to look like a convention inherited from the calendar. Here it is a consequence: given four concurrent trades per cycle and four cycles per week, the week is the smallest unit containing enough observations for the expectancy term to exceed the variance term by a usable margin. Roughly sixteen closed positions is not a large sample by any statistical standard; it is simply the smallest one that beats a day by enough to justify deciding on it. An operator trading a materially different cadence would land on a different window, and the correct response would be to move the unit rather than to keep the week and read it more hopefully.