Turning a Normal trade into a trend trade mid-flight corrupts evidence that was already collected.

How it happens

Nobody decides to drift. The trade simply keeps going and the plan quietly stops applying.

The mechanism is almost never a decision. A Normal trade is entered with a static target and no trail, it reaches that target area, and the move looks like it has more in it. Closing at the declared level now feels like leaving money on the table, so the target is extended, and once it is extended the trade needs management it was never given rules for. At no point did anyone choose to reclassify. Each step was small, locally reasonable, and made in response to information that arrived after the branch was declared — which is the exact information the declaration was designed to be immune to.

What actually breaks

The damage is to the branch's statistics, and it is retroactive in effect.

The trade itself might end well. What does not end well is Normal's record, which now contains an outcome produced by trend management. That single row moves the branch's average, widens its excursion distribution, lifts its far-rung hit rates, and inflates its expectancy — all in ways that suggest Normal is capable of something it is not. Decisions taken later against that record are the real cost. The blend weights are set from branch expectancy. The seven-lens assessment reads branch expectancy. The scorecard's weekly questions ask whether Normal produced acceptable expectancy. Every one of those consumes a number that a drifted trade has quietly moved.

Why it cannot be fixed afterwards

Reclassifying the trade at the weekend fixes the row and destroys something more important.

The instinct on discovering a drifted trade is to relabel it as the branch it was actually managed as, which puts the outcome in the right statistical home. That is the wrong repair and the reason is worth stating carefully. Branch statistics are supposed to describe what happens when a branch is selected and executed, which means the population has to be defined by the decision rather than by the behaviour. Relabelling by behaviour turns every branch record into a sorted-by-outcome collection, and the branches converge on describing results instead of describing strategies. The correct handling is to leave the label, record the departure, and treat the row as contaminated evidence for both branches rather than clean evidence for either.

• Branch populations are defined by the declaration, never by what the trade ended up doing.
• A drifted trade is flagged, not reassigned — reassignment corrupts the definition itself.
• Contaminated rows should be excludable from analysis, which requires the flag to exist.

The disguise version

Choosing a variant in order to make one branch behave like another is the same failure, arranged in advance.

Drift has a premeditated cousin. Selecting a delayed-protection posture on a static branch, specifically so the trade can run further, produces a Normal trade managed toward trend behaviour with paperwork that says everything is in order. The matrix names this directly: a variant must not be used to disguise one branch as another. The tell is the reasoning — if the argument for the posture is that this particular setup deserves to be treated like a trend trade, then the branch selection was wrong at entry and the honest correction is to reclassify before the trade exists or to skip it. A variant tunes management within an identity. It cannot smuggle in a different one.

The detection

Drift shows in the branch's shape before it shows in its results, which is the only useful warning.

Because drifted trades often end profitably, expectancy is a slow detector. Shape is faster. A static branch that begins producing occasional far-rung hits it has no mechanism to produce is announcing something, and so is a widening favourable excursion distribution on a branch whose exit is supposed to be fixed. The branch integrity flag in the execution lab exists for this, and the variant usage audit catches the premeditated version by showing postures being selected at rates the branch does not warrant. Both are cheap to read weekly and both fire well before the contamination has moved anything that matters.

The key idea

A branch is a promise about how a trade will be managed, and a promise revised mid-trade was never one.

The four branches are not four kinds of market opportunity. They are four exit architectures, chosen before the position exists precisely because that is the last moment at which the choice can be made without reference to how the trade is going. Every protection the architecture offers depends on that timing. Allowing the identity to be revised once price is moving does not give the operator flexibility; it gives them a system in which the branch label records what they wished they had chosen, and no statistic built on it can be trusted afterwards.