The layers are not stacked. They are axes, and they multiply.

The three choices

Every EV reading requires a lens, a window and a family — and the lab makes all three explicit.

The control panel is where the three choices are made, and it is deliberately arranged so none of them can be made by accident. The branch lens selects whose expectancy is being measured: one of the four primitives, one of the composites, or the all-blended whole. The date scope selects the window it is measured over: all-time, last N, a named week or month, a custom range. The metric family selects what kind of question is being asked of that sample — whether the edge is statistically credible, whether it is leaking through mechanics, whether behaviour is helping or hurting it, whether it scales, where it is strongest by context, whether it reconciles in native units. Change any one and the printed number changes. That is not instability. It is the instrument working.

Why the shape matters

A ladder has a top. A cube has no privileged cell.

If the layers were a ladder, there would be a most-refined number and the discipline would be to climb toward it. Because they are axes, there is no such number — only the cell that answers the question actually being asked. All-blended all-time realised EV is not a rough draft of Normal-branch last-twenty-trades heat-adjusted EV; they are answers to unrelated questions and neither supersedes the other. This is why the lab refuses to print a single headline figure and call it the truth, and why an operator who wants one number will keep finding the workbook uncooperative. The uncooperativeness is the design.

The reading order

Statistical truth is asked first, because it decides whether the other families mean anything.

Axes being independent does not make the order arbitrary. The family axis has a fixed entry point: statistical truth comes first, because it establishes whether the selected sample supports any conclusion at all. A mechanical-leakage read on a sample below the minimum gate is a decorated coin flip, and so is a behavioural read, and so is a context read. Only after credibility is established does routing to a diagnostic family become worthwhile. The same logic applies within the branch axis in a softer form — the all-blended baseline is normally read before an isolated lens, because a branch figure means little without the whole it is being compared against.

The combinatorial trap

Every filter you add spends sample, and the axes spend it multiplicatively.

The obvious use of three axes is to combine them aggressively: this branch, this session, this timeframe, this ATR state, over the last three weeks. Each restriction is individually reasonable and the intersection is often four trades. The resulting figure prints with the same confident formatting as a two-hundred-trade read, which is exactly the danger — precision is a display property here, not a statistical one. The minimum sample gate exists as the structural answer, greying out reads it does not trust rather than letting them look decision-grade. The operator's own version of the same rule is to move one or two filters at a time and to notice when the trade count is the number that changed most.

• Filters restrict multiplicatively; sample size falls faster than the number of choices suggests.
• The sample gate is a trust threshold, not a suggestion — below it, no family read is decision-grade.
• When a narrow cell looks extraordinary, widen one axis before believing it.

When cells disagree

Disagreement between two cells localises the problem, which is the whole point of asking twice.

The lab earns its keep in the disagreements. Branch EV healthy while blended EV sags is a weighting problem, not an edge problem — the branches are performing and the mix is wrong. Trade-level EV healthy while session-level EV decays is a timing or fatigue problem living in sequence rather than in setup quality. Gross EV strong while net EV is thin is friction, and it will not be repaired by better selection. A single headline number would have averaged all three of these into one mildly disappointing figure and sent the repair effort at whichever part of the machine the operator happened to suspect. Two cells that disagree name the part.

The key idea

'What is my EV' is an underspecified question, and the lab's job is to make that obvious.

The practical output of the axis model is a habit: never quote an EV figure without its lens and its window, and never accept one from yourself without the same. The number is meaningless as a bare quantity and precise as a coordinate. That habit costs a few extra words in a review note and removes an entire class of error in which an operator remembers a flattering cell, repeats it as the system's expectancy, and sizes a decision against a figure that described three weeks of one branch.