Branch quality is a permission input. It is never an authority.

The ordering

Two questions, asked in a fixed sequence, and only one of them is negotiable.

Deployment resolves two separate questions and the system keeps them separate on purpose. How much risk is available is answered by the gate, which is derived from drawdown state and caps the maximum tier the throttle may reach. Which branch that available risk should go to is answered here, by branch quality. The sequence is not interchangeable, because the second question operates entirely inside the answer to the first. A branch reading well does not enlarge the envelope; it competes for a share of an envelope whose size was already fixed elsewhere. Collapsing the two into a single judgement — this branch looks strong, therefore commit more — is the specific failure the ordering exists to prevent, and it is attractive precisely because the branch evidence is detailed and quantitative and feels like it should be able to settle the matter.

Why drawdown outranks it

Branch quality is tactical evidence, and the gate governs survivability.

The two readings are not merely different in subject; they differ in what they are protecting. Branch quality is a statement about where edge currently sits, assembled from recent expectancy and dispersion, and it is genuinely tactical — it will look different in six weeks and should. The gate is a statement about how much loss the account has already absorbed and therefore how much room remains before recovery arithmetic becomes punishing. That second quantity is not an opinion about the market and does not improve because a branch is performing. The doctrine that follows runs throughout the system: a favourable expectancy read, a constructive structural posture, or a strong branch score cannot outrank the gate, because tactical evidence never redefines drawdown authority. Permitting it to would mean that the periods when the account can least afford to be wrong are exactly the periods when the strongest-looking evidence is allowed to expand exposure.

Composite risk

Branch scores are read together, because they draw on one budget.

A subtler version of the same error appears when several branches read well at once. Each branch's evidence is assembled independently and each may justify its own case, but they are not independently funded — every branch draws from a single deployment envelope, and a set of individually reasonable increases can compose into total exposure the gate never authorised. The engine's separate branch blocks exist so behaviours can be compared without mixing, which is exactly right for reading and exactly wrong for summing. The composite view is what governs: what matters is total risk currently deployed across all branches against the ceiling, and a branch's claim is a claim on a share of that total rather than an addition to it. Strong evidence across several branches is a reason to reallocate between them, and it remains not a reason to deploy more in aggregate.

Where it does decide

Under a compressed gate the allocation question becomes more consequential, not less.

It would be easy to read the ordering as making branch quality unimportant during restriction, and the opposite is true. When the envelope is small, the cost of putting the available risk into the wrong branch is proportionally higher, because there is less capacity to absorb the error and fewer opportunities before the state is re-read. This is the regime in which the engine earns its place most clearly: the gate has decided that little will be deployed, and the branch evidence decides that the little which is deployed goes to the branch with the most persistent case rather than the branch the operator most enjoys trading. Habit is the silent allocator during good conditions, when everything is affordable. During restriction it is expensive, and the branch blocks are what replace it with something that can be inspected.

• The gate answers how much. The engine answers which. Never the reverse.
• Individually justified branch increases can still breach the composite ceiling.
• Restriction raises the value of good allocation — it does not suspend it.

The argument that always loses

"This branch is clearly working" is true, relevant, and insufficient.

The case for overriding a compressed gate on branch evidence is genuinely reasonable-sounding, which is why it needs a structural answer rather than a persuasive one. It usually runs: the branch shows positive expectancy across every window, contained dispersion, and clean attribution, so restricting it is leaving demonstrated edge unharvested. Every clause is likely true. It is still insufficient, because the gate is not making a claim about the branch — it is making a claim about the account's remaining capacity to be wrong, and those two claims do not compete. The situation the doctrine anticipates is the account in a compressed gate with a genuinely strong branch, because that is precisely when the override is most tempting and most damaging. A rule that only binds when the evidence is weak is not a rule; it is a preference that has not yet been tested.

The key idea

Two good answers to two different questions do not combine into a bigger position.

The engine and the gate are both working correctly when they disagree — the branch really is strong and the account really is restricted, and neither reading is a mistake to be reconciled. What the hierarchy provides is a fixed rule for what happens next, so the disagreement resolves the same way every time regardless of how the operator feels about the branch that week. Deployment authority descends from drawdown state; branch quality distributes what descends. Keeping that boundary intact is what stops a good analytical tool from becoming the mechanism by which risk expands during exactly the periods it should be contracting.