Four external modules, four questions that don't overlap.

The division

Each module owns one question and declines the others.

The Weekly Trading Scorecard asks whether the system is still producing acceptable expectancy under the current branch mix — a weekly cockpit converting branch probabilities into EV and a GREEN/YELLOW/RED status. The MAE/MFE Lab asks whether opportunity was translated into realized R, through capture, giveback, excursion, and friction. The Variant Attribution Module asks which management structures produced the period's result. The Advanced EV Analytics Lab asks what would happen under different assumptions — and is explicitly R&D rather than live authority. Health, execution, attribution, and hypothesis. Four questions, no overlap.

The overlap that isn't

Three of them will produce a number for almost any question you ask.

The practical hazard is that these modules share inputs, which means each can be made to say something about territory it does not own. The Scorecard will show branch EV, which looks like attribution. The MAE/MFE Lab reports by branch, which looks like health. Attribution reports average R, which looks like expectancy. In each case the number is real and the inference is unsupported, because the module computed it for a different purpose and applies none of the qualifications the owning instrument would. A number's existence in a workbook is not a claim that the workbook is authoritative about it.

The sample-size divide

The instruments need different amounts of evidence to say anything.

A related and less obvious distinction: these four modules reach reliability at different rates. The Scorecard's weekly EV read is designed to be produced every week and interpreted with explicit stability weighting, because a single week is a small sample. Execution-quality metrics stabilize relatively quickly, since capture and giveback are measured per trade and do not depend on outcome rarity. Attribution by variant needs considerably more, because it partitions an already-partitioned sample. And the EV Lab's scenario work is hypothetical throughout, so its sample question is about the inputs rather than the output.

• Do not overinterpret branch or timeframe outputs with tiny sample size — the lab's own instruction.
• Partitioning multiplies the sample requirement: branch × variant × regime fragments fast.
• The Scorecard's stability weighting exists precisely because weekly EV is noisy by construction.

The R&D boundary again

One of the four is not allowed to conclude anything about today.

The Advanced EV Analytics Lab sits in a different category from its three neighbours and the architecture is explicit about it: the EV Lab and sensitivity tools are research and validation layers, not live authority, and they sit last in the nine-position module-disagreement order. Its outputs are claims about hypotheticals, and a claim about a hypothetical cannot outrank a measurement of the actual. When the Lab and the Scorecard disagree about a profile, the Lab has proposed something and the Scorecard has reported something, and only one of those is evidence about the account as it currently stands.

The order they get read in

Health, then execution, then attribution — and hypothesis only after all three.

The questions nest, which gives the review a natural sequence. Establish whether the system is healthy; if it isn't, establish whether the problem is execution quality or edge; if it is execution, establish which management structures are responsible; and only with all of that settled does it make sense to model alternatives. Running the sequence backwards — starting in the EV Lab because it is the most interesting instrument — produces a proposed change with no established problem for it to solve, which is the exact failure the sandbox boundary was drawn to prevent.

Why four instead of one

A single omnibus analytics workbook would be worse, not simpler.

It is fair to ask why these are four files rather than one comprehensive analytics module. The answer is the same one that governs the rest of the ecosystem: separation is what makes each output trustworthy within its actual competence. A combined workbook would present health, execution, attribution, and hypothesis in one surface with equal visual authority, and the R&D outputs would sit beside live evidence looking exactly as credible. The boundaries between these questions are load-bearing, and file boundaries are the cheapest available way to enforce them.