A healthy blend can contain a branch that is quietly failing.

The mechanism

Weighting means a strong branch can outvote a failing one.

Blended EV combines branch expectancies by the live weight profile, which is the correct way to describe what the account is actually doing. It is also, structurally, a mechanism by which one component can compensate for another. A branch running well above its designed contribution raises the blend enough to keep the week tagged green while a second branch sits below zero. Nothing has malfunctioned — the blend is reporting the account faithfully. What it cannot report is that the account's health is resting on one component behaving unusually well. That is not a flaw in the blend — a weighted average that also disclosed its own dispersion would no longer be a summary — but it does mean the figure answers a narrower question than it appears to.

Why it goes unnoticed

A green tag is not a prompt to investigate anything.

The masking would be harmless if it triggered a look. It does the opposite. Review attention is naturally allocated by the headline reading, so a green week is closed quickly and a red one is examined — which means the specific configuration where a failing branch hides behind a strong one receives the least scrutiny available. The pattern can persist for a quarter or more without anyone behaving unreasonably, because at every point the visible evidence said the operation was fine and the invisible evidence was never requested. By the time a red week finally forces the decomposition open, the failing branch has usually been failing for long enough that its recent history is the only history available.

The two directions it fails in

The concealed branch is a problem; so is the concealing one.

The obvious reading is that a deteriorating branch is being missed, and it is. The second reading matters as much: a branch performing far above its assigned role is itself a finding, not a bonus. Each branch exists to do a specific job — Normal as the core compounding branch, Trend No-Partial as the true fat-tail accelerator, Overflow as the stability branch — and a branch substantially exceeding its role is behaving differently from its specification. Whatever is producing that is unlikely to persist, and the blend's health is currently a function of it continuing.

What actually catches it

Per-branch figures read every week, whatever the tag says.

The defence is procedural rather than analytical: the branch column is read on green weeks, not only on red ones. The engine already computes branch-level weekly EV alongside the blend, so the finding costs no additional machinery — it costs the discipline to look at a decomposition when the headline gave no reason to. The rolling 4W, 6W and 12W windows are where the pattern becomes unambiguous, because a single week of dispersion is ordinary and a quarter of the same branch carrying the blend is not.

• Green weeks receive the least scrutiny and are where masking survives.
• A branch far above its role is a finding, not a bonus — it is off-specification.
• One week of dispersion is normal; a quarter of the same carrier is not.

The consequence for deployment

Weight profiles derived from a masked week are wrong in a specific direction.

The practical cost surfaces at the next weighting decision. If the blend is read as healthy and weights are adjusted toward what appears to be working, the account concentrates further into the branch already carrying it — increasing exposure to the single component whose unusual performance was never examined. The masking therefore does not merely delay a finding; it actively biases the next allocation toward the fragility. That is why this is a weekly reading obligation rather than a quarterly review item. A quarterly cadence would catch the finding after the allocation it should have prevented.

The key idea

The blend answers what happened. Only the decomposition answers whether it repeats.

Blended expectancy is the right figure for setting posture and the wrong figure for forming beliefs about the operation. It is a summary, and the entire value of a summary is that it discards composition. Reading it as though it also carried composition is the error, and it is invited by the fact that the blend appears first, appears larger, and carries the tag. The decomposition is one column away and is skipped precisely on the weeks where it would have been most informative.