A profile that would have won last week has not earned next week.

What it computes

The same trades, re-weighted — and only the trades that were taken.

The comparison block reweights realised branch expectancy under alternative profiles and reports the resulting blended figure against the baseline and the active structure. Its legitimate use is stated narrowly: it shows whether a different branch mix would have improved expectancy, and it is read to confirm that the active profile remains competitive across weeks. That word — competitive — is doing precise work. The question the comparison is built to answer is not which profile is best, it is whether the one in force has stopped being defensible. Those are different tests with different evidence bars, and the second one is answerable from a single tab while the first is not. A profile that is consistently near the top of the comparison across many weeks is competitive. A profile that is beaten in one week by a structure that has never been near the top before has learned nothing about itself.

The first omission

The counterfactual reports the return and not the path.

Reweighting toward a branch that performed well raises the blended expectancy figure, and it also raises variance, because the branches are not interchangeable in their distributions. Trend No-Partial is the true fat-tail accelerator and buys geometric growth with high variance; Overflow is the stability branch. A profile that shifts weight toward the accelerator will, over a good week, look straightforwardly superior on a single scalar, and the drawdown path it would have produced is not on the tab. The explicit instruction is to ask whether drawdown would have worsened before treating the improvement as real. This is not a caution about being careful. It is a statement that the comparison is computing one dimension of a two-dimensional outcome, and the missing dimension is the one that governs survival.

The second omission

The alternative profile never had to ask the gate for permission.

This is the omission most likely to be missed, because it is structural rather than statistical. The comparison arithmetic reweights outcomes; it does not re-run deployment authority. A more aggressive profile implies larger risk deployment, and larger risk deployment has to be available under the gate that was actually in force that week. If the account was in a compressed gate, the exposure the winning counterfactual assumes was not permitted, and the profile did not merely perform better — it played a different game with a different rulebook. Comparing a governed result against an ungoverned one and concluding the ungoverned one is superior is a category error, and it is the specific error the ask-the-gate test is designed to catch. The question is not whether the alternative profile would have made more. It is whether the alternative profile could have been run at all.

The sample problem

A week is the decision unit for results and the wrong unit for structure.

A profile is a structural commitment that governs how risk is distributed across every subsequent week, and structural commitments require structural evidence. One week contains too few trades per branch for branch-level expectancy to have stabilised, which means the comparison's inputs are themselves noisy and the ranking it produces is substantially a ranking of which branch got lucky. The warning sign named in the interpretation logic is an alternative profile dominating for many weeks or months — the plural is the whole content of the rule. Switching on a single week has a further property that makes it worse than merely uninformative: it is systematically biased toward whichever branch is currently in a favourable stretch, which is on average the branch closest to reverting. The operator who reweights every time the comparison changes its mind has built a strategy that buys after strength and sells after weakness, at the profile level, forever.

What promotion looks like

Persistence, then path, then permission — and the default is to do nothing.

A genuine profile change is slow and boring by construction, and it has a shape. The alternative must dominate across enough windows that the result cannot be one branch's good stretch. The drawdown implication must be examined rather than assumed, because a profile that improves expectancy while deepening the equity trough may be worse in every sense the operator cares about. The exposure must be one the gate ladder would have permitted through the period being tested, not merely one that is permitted today. And when all three hold, the change is still made deliberately and logged, because a profile change resets the interpretation baseline for everything downstream. The default answer to the comparison is to leave the profile alone, and that default should be defeated by evidence rather than by the tab having produced a number.

• Dominance across many windows, not a single favourable week.
• Drawdown path examined — expectancy is one dimension of a two-dimensional result.
• Exposure checkable against the gate that was actually in force.

The key idea

The comparison is a monitor on the active profile, not a search for a better one.

Read as a search, the block will always return a result, because over any finite window some weighting is optimal and the machinery will find it. Read as a monitor, it answers a much more useful question: has the profile in force stopped being defensible? That question has a stable answer, it degrades gracefully with noisy inputs, and it points somewhere specific when it turns negative. The distinction is the difference between a tool that keeps an operator honest about a structural decision they have already made and a tool that invites them to remake that decision every seven days on the strength of the most recent noise.