One dropdown controls every diagnostic on the tab, and that is the architecture.

The routing rail

The selected entity is a global variable, not a filter on a view.

The Live Intake block is described as the control and routing rail, and the distinction from a filter is worth drawing precisely. A filter narrows what is displayed while the underlying figures stay what they were. A routing rail changes what the figures are: choosing an entity repopulates the observed efficiency ratio, the profit factor, average risk, expectancy, and hit rate, and every calculated diagnostic below is computed from that repopulated set. There is no cell on the tab that means something entity-independent. This is why the selector is placed at the top of the intake block rather than tucked beside a chart — it is not a convenience control, it is the statement of what the entire surface is currently about. An operator who reads a final readout without registering which entity is selected has read a well-formed sentence about an unspecified subject.

Why not all seven at once

A surface that shows everything invites a conclusion about nothing.

The alternative design — seven entities rendered side by side — is tempting because it appears to save time, and it fails in a specific way. The diagnostics on this tab are not simple magnitudes that can be scanned across a row; they are classes, matrix states, functional roles and narrative stances, each of which requires reading a modifier and a context line to interpret correctly. Presented seven abreast they would be read as a ranking, because that is what humans do with parallel columns, and a ranking of narrative diagnostic states is not a meaningful object. Restricting the surface to one entity forces the interpretation to complete before the next one begins. The cost is that a full pass takes seven readings instead of one glance, and the design accepts that cost deliberately: the seven readings are each valid, and the one glance would not have been.

The confirmation step

Confirming the intake is populated and sensible is not a formality.

The interpretation flow names a step between selecting the entity and reading anything: confirm the intake values are populated and numerically sensible. It sits there because the cockpit is a translation surface rather than a calculation surface, and a translation surface will translate whatever it is given without complaint. A blank routed value, a stale figure left from a previous selection, or an average risk reading that belongs to a different period will all produce a complete, confident, fully formatted final readout. Nothing downstream can detect it, because every downstream category is doing exactly its job on the inputs present. The check is cheap — five numbers, glanced at for plausibility — and it is the only place in the chain where a routing failure can be caught. Skipping it does not produce an error message; it produces a briefing about a fiction.

Six categories, one wiring rule each

Every row on the tab declares how it was computed.

The calculated diagnostics are organised into six categories by wiring method rather than by subject: direct spectrum and lookup reads, directional comparisons between two values, ten-band normalised scores, semantic meaning lookups, rule-composition narrative metrics, and weighted composite roll-ups. The stated purpose of the organisation is to prevent formula chaos and keep the cockpit auditable, and it does something else that matters more day to day. Because a row's category is visible, the operator knows what kind of claim they are reading before they read it. A Category 1 output is a classification of one number against a fixed spectrum and is as reliable as that number. A Category 6 output is a weighted composite of normalised sub-scores and inherits every weakness in every input. Both are rendered as a short readable phrase, and without the category they would look equally authoritative.

Order inside the pass

Classes first, gaps second, composites only when the helpers are stable.

The pass has an internal sequence and it mirrors the wiring hierarchy. Category 1 spectrum and class outputs are read first because they depend on a single input each and are the most directly checkable. Category 2 directional gaps come next, identifying whether live deployment is helping or diluting the benchmark reading. Categories 3 and 6 — the normalised scores and composite roll-ups — are read only after the helper scores are stable and validated, because a composite built on an unsettled helper is a confident number with no support underneath it. Category 5 narrative metrics supply the operational sentence, and the final readout comes last. The instruction attached to the readout is the same one that governs every summary in the system: it summarises the condition and is not a standalone override of governance rules.

• Select, then confirm the five intake values before reading anything below.
• Category 1 and 2 are checkable directly; 3 and 6 inherit every upstream weakness.
• The final readout is a briefing, never an authorisation.

The key idea

Scoping a surface to one subject is what makes its statements falsifiable.

A diagnostic that describes the whole system at once cannot be wrong in any specific way — there is always some part of the system that fits the description, and the reading survives every outcome. A diagnostic scoped to Trend No-Partial in a named week makes a claim that a later week can contradict, which is the property that lets the engine be checked against reality and corrected. The single selector is the mechanism that enforces the scope. It looks like an interface decision and it is an epistemic one: the cockpit is built so that everything it says is about something in particular, and can therefore turn out to have been mistaken.