The scorecard does not read trades. It reads probabilities, and something upstream produced them.

The chain

Five stages, and the scorecard is the fourth thing to touch the evidence.

Execution produces trade-level truth at the broker and in the capture sheets: pair, session, setup, branch, result, fees, tags. The journal and its weekly summary compress that into branch counts and weekly probabilities — the step where a week of individual trades becomes a handful of decimals. The scorecard converts those decimals into branch expectancy, blends by the active profile weights, and grades the week. Dashboards and governance panels then read the grade, and the structural engine takes the longer-horizon view to decide whether a trend is real. Each stage discards detail the next one does not need, which is what makes the chain workable and also what makes it lossy in a specific and knowable direction.

The conversion

Branch probabilities meet stored constants, and the constants are assumptions rather than measurements.

Turning a probability into an expectancy requires knowing what the outcome is worth when it occurs, and those values live in the workbook as named constants rather than being recomputed each week. The expected value of a runner after a partial and protection trigger is one such constant; the expected full-size value under no-partial and trail behaviour is another. They are estimates of branch behaviour, held stable so that week-to-week comparisons measure the trading rather than the model. That stability is a feature and it carries an obligation: the constants describe how the branches are supposed to behave, and if actual behaviour drifts away from them the scorecard will keep converting cleanly and quietly measuring the wrong thing.

The blend

Branch expectancies are weighted by the active profile, and the weights must sum to one.

The overall figure is the weighted combination of the four branch expectancies, and the weights are the strategy's intended mix expressed as numbers. The sum check exists because a profile that does not total one produces a blended figure that is not an expectancy at all — it is a scaled quantity that will still print, still grade, and still look plausible. This is one of the few places in the workbook where a wrong number is silent rather than obvious, which is why the check is explicit and why it should be read rather than assumed after any edit to the profile.

• Branch expectancy answers whether a branch is working; blended expectancy answers whether the system is.
• The weights are the intended mix, not the observed one — divergence between them is its own finding.
• If the weights do not sum correctly, every downstream status is computed from a meaningless number.

The upstream dependency

Every classification decision is made before the workbook sees anything, and it cannot be audited from inside.

Because the input is already compressed, the scorecard has no way to check whether a trade was labelled correctly. If Normal trades were classified as trend, both branches' probabilities are wrong, both expectancies are wrong, and the blended figure is wrong in a direction nothing in the workbook can detect. The grade will still compute, still fall in a band, and still carry the same authority. This is the single largest source of quiet error in the whole loop, and the only defence is upstream: consistent labelling in the journal, checked at the point of entry, with the branch integrity flags in the execution lab as a secondary net.

Why the grading step is non-negotiable

Three systems consume the output, and an ungraded week starves all of them.

The grade is not a report to be admired. Gate review consumes it as state confirmation. The regime engine consumes its structural components on a monthly cadence. Quota governance consumes the branch counts to check that participation minimums are being met. A week that goes ungraded does not just leave a gap in a spreadsheet — it removes an input from three governance processes, and the gap will not announce itself because each of those processes simply proceeds with one fewer observation. That is why the weekly discipline is fixed rather than best-effort, and why catching up two weeks at a time is worse than it looks.

The key idea

Knowing the workbook's position in the chain is what stops it being asked the wrong questions.

A tool that consumes probabilities can tell you that a branch's expectancy is thin. It cannot tell you which trades made it thin, because it never saw trades. It cannot tell you whether the thinness is mechanical, behavioural or a classification error, because those distinctions were compressed away one stage earlier. Understanding that boundary is what routes each question to the layer that can answer it, and it is the difference between a review that ends with a diagnosis and one that ends with an operator staring at a red cell.