The week badge is computed last, and it should be read last.

The ordering

Five layers, and the sequence is a constraint rather than a suggestion.

The scorecard is built as a descent from inputs to verdict, and reading it in that order is what makes the verdict mean anything. Raw probabilities come first because they are the closest thing on the tab to observed fact — hit rates carried across from the weekly summary, unprocessed. Branch expectancy comes second, because it is the first place probability meets payoff structure and the first place a branch can be seen earning or failing to earn its risk. Blended expectancy comes third, because it is branch expectancy weighted by the profile actually in force. Profile comparison comes fourth, because it is only meaningful once the active blend has been read on its own terms. Rolling and cumulative context comes fifth, because it is the layer that decides whether anything above it is signal or a single loud week. The badge is downstream of all five. Reading it first is not a shortcut to the same conclusion; it is arriving at a conclusion with the reasoning deleted.

Why the badge is not the answer

A status colour cannot distinguish between four different weeks.

Consider what a single green badge is compatible with. It is compatible with all four branches positive and stable, which is the week it appears to describe. It is equally compatible with one branch far above its usual contribution and another deeply negative, netting to a comfortable number. It is compatible with a blend that is green only because the profile happens to weight the branch that got lucky. And it is compatible with a green reading that is the third consecutive step down from a much stronger cumulative position — improvement and deterioration produce the same colour when they cross the same threshold from opposite directions. The badge is a threshold test on one number, and threshold tests on aggregates cannot preserve the information that made the aggregate. This is not a flaw in the badge; it is what a badge is for. The error is treating a designed compression as a source.

Layers one and two

Probabilities before expectancy, because expectancy inherits their errors.

Branch expectancy is a function of hit rate and payoff structure, which means every defect in the probability layer arrives intact in the expectancy layer wearing a more authoritative costume. A hit rate computed over eleven trades is a number; it is not yet evidence, and the expectancy built on it is not evidence either. Reading probabilities first is how the operator learns the sample they are about to reason from — how many trades, distributed across which branches, in what proportion to a normal week. The probability bridge exists precisely so these values are carried from the weekly summary rather than retyped, and the standing instruction is not to type over formula-driven probabilities unless a test is being run deliberately. When a branch expectancy reads strangely, the discipline is to look up at the probabilities that produced it before looking sideways at other branches, because the overwhelming majority of strange expectancy readings are thin-sample readings rather than genuine structural changes.

Layers three and four

The blend is the practical system; the counterfactual is not the system at all.

Blended expectancy is the number that describes what the operator actually has, because it is branch expectancy weighted by the profile in force. It is the practical system EV, and it is the number the week is fairly judged on. Profile comparison sits one layer below it for a specific reason: it computes what other weight structures would have produced over the same trades, and the answer is always interesting and rarely actionable. Read in order, the comparison is context — it tells the operator whether the active profile remains competitive across weeks. Read first, it becomes a temptation, because there is nearly always some alternative weighting that would have done better over any single window, and finding it feels like discovering an improvement rather than fitting a curve to seven days. The ordering protects against a specific failure: mistaking the best profile for last week for the best profile for next week.

Layer five

Frozen and cumulative blocks exist so history cannot be quietly rewritten.

The last layer is the one that decides whether any of the above is persistent, and it depends on a piece of machinery that is easy to overlook. Rolling formulas recompute as new weeks land, which is correct behaviour for a rolling window and catastrophic behaviour for a historical record — a formula that recalculates the past means last quarter's interpretation silently changes every time this week's data arrives, and there is then no fixed thing against which drift can be measured. The frozen and cumulative blocks store and stabilise expectancy outputs so that longer-run review has something that stays still. The reading that matters here is the relationship between the two: short-term expectancy spiking while cumulative expectancy deteriorates is a warning that the rolling layer alone cannot produce, because the rolling layer would simply show a good week.

• Cumulative and rolling aligned positively — edge is stable, and the week is evidence.
• Short-term spike, cumulative decay — the week is noise on a declining structure.
• Frozen blocks are what make the comparison possible; without them the past moves.

The key idea

The badge tells the operator where to look, never why.

The drill-down rule for the dashboard applies with equal force one level down: when a status deteriorates, it identifies a location rather than a cause, and the cause is always in the layers the status compressed. Was it probability, payoff, branch mix, or friction? Four questions, four different repairs, one identical colour. An operator who reads the badge first will find themselves reasoning backward from a conclusion, which is a posture that reliably generates a story rather than a diagnosis. An operator who reads in order arrives at the badge already knowing what it is going to say and, more usefully, already knowing which of the four it is. The sequence costs a few extra minutes each week, and what it buys is the difference between knowing that something is wrong and knowing what to change.