An acceptable average MAE can conceal a cluster in the worst quadrant.

The map

Six regions, each with a different meaning and a different repair.

Every closed trade lands somewhere on a grid of adverse excursion against favourable excursion, and the grid has named regions because the regions correspond to distinct causes. Low pain with high opportunity is the clean lifecycle worth studying and replicating. High pain with high opportunity means the idea was right and the entry was late — a timing repair, not a selection one. Low pain with low opportunity means the entry was fine and the market was not paying, which points at regime and branch choice rather than execution. High pain with low opportunity is the region to reduce exposure to until something changes. A separate flag marks the trades that captured most of what they were offered, because those are the repeatable ones.

How the average lies

A mean of 0.45R can be twenty comfortable trades or ten comfortable and ten severe.

This is the mechanical point and it is worth stating in numbers. A branch whose trades all cluster around 0.45R adverse excursion is behaving predictably, and its stop placement is roughly correct. A branch whose trades split into a group near 0.15R and a group near 0.80R has the same mean and a completely different problem: something distinguishes those two groups — session, pair, time of entry, whether the setup was a pullback or a breakout — and the average is the one statistic guaranteed to hide it. The lab's distribution view exists because that split is common, and because the repair for a bimodal branch is to find the discriminator rather than to move the stop.

Why each branch occupies a different region

The four branches are supposed to produce four different maps, and comparing them directly is the error.

Normal is a static branch aiming at a fixed target, so a healthy Normal population concentrates in the low-pain, moderate-opportunity band and should look boring. Trend Partial banks early and lets a remainder run, which produces a two-lobed map by construction. Trend No-Partial holds full size until the move proves itself, so its adverse excursions are structurally deeper and its favourable tail structurally longer — the same map on Normal would be an emergency. Overflow should look tight and unremarkable, and when it starts producing wide excursions it has usually stopped being a diversification branch and started being recovery trading. The maps are the branch contracts made visible.

• A healthy Normal map is narrow and dull; width in Normal is the finding.
• A healthy TNP map is wide with a long favourable tail; narrowness in TNP means the branch is not being allowed to work.
• Trend Partial is two-lobed by design — do not read the lobes as inconsistency.

The concentration test

Ask what share of the branch sits in the worst region, not what the branch averaged.

The practical review question is a count rather than a mean: how many trades landed in the high-pain, low-opportunity region this month, and were they concentrated in one pair, one session or one week? Three such trades out of forty is ordinary. Three out of nine, all on the same instrument, is a finding that a mean of the whole month would have absorbed without trace. This is also the read that survives small samples best, because counting how many trades fell into a bad region is a more robust operation than averaging a quantity across a thin population.

Where it hands off

The quadrant names the cause; the repair engine prices the fix.

The map is deliberately diagnostic rather than prescriptive. Knowing that a branch has drifted into the late-but-correct region tells you the entry is the problem and not the exit, which narrows the candidate repairs from six to about two. It does not tell you whether tightening entry timing is worth more expectancy than raising capture, because that is an arithmetic question the sensitivity engine answers with numbers. The order is fixed and it matters: locate the cause on the map, then price the candidate repairs, then commit to one. Reversing it produces a fix chosen because it was easy to compute.

The key idea

Averages are for reporting. Distributions are for diagnosis.

The temptation in a weekly review is to read six aggregate numbers and form a view, because six numbers fit in a glance and a distribution does not. The cost of that shortcut is that every genuinely actionable execution finding this lab can produce lives in the shape rather than the centre. A branch does not usually fail by drifting a little worse on average. It fails by developing a cluster, and a cluster is precisely the thing a mean is designed to smooth away.