The gate and the benchmark read the same drawdown and ask different things of it.

Two instruments

A location and a distribution are different kinds of answer.

The gate's answer is categorical and absolute: minus fifteen percent is the buffer state, and it would be the buffer state for any account with that reading, regardless of strategy, history or how the drawdown was reached. The benchmark's answer is relative and probabilistic: minus fifteen percent sits at some percentile among the modelled outcomes for a system with this configuration, and the same figure could be unremarkable for one setup and an outlier for another. Neither is more correct. They answer questions that happen to take the same input, and the temptation to reconcile them into a single verdict is the error this brief is about.

The divergence that tempts

A deep drawdown that is entirely normal for the model is still a deep drawdown.

The first informative case is a reading that compresses the gate while sitting comfortably inside the modelled bands. The benchmark is saying something true and useful: a system with this branch mix and this variance profile produces drawdowns of this depth routinely, so nothing structural is indicated and the correct posture is patience rather than repair. What it is not saying is that the account has more capital than the gate believes. The damage is real regardless of how expected it was, the recovery arithmetic is unchanged, and the ceiling stands. The benchmark's reassurance is about diagnosis — do not go looking for a broken system — and it has no purchase on deployment at all.

The divergence that warns

A shallow drawdown outside the bands is a finding the gate cannot produce.

The reverse case is more valuable and easier to miss. An account sitting at a modest drawdown that nonetheless falls outside what the model expects is telling the operator that something about the live system no longer matches the one that was simulated — the branch mix has drifted, execution has degraded, or the configuration the benchmark was built from is stale. The gate reports a mild state and permits a near-full ladder, entirely correctly, because the account genuinely is only slightly below its peak. The warning exists only in the comparison, and an operator reading the gate alone would see a healthy capital state and no reason to look further. This is the case that justifies running both instruments.

Why they must not be merged

A deployment figure that consulted the benchmark would be arguable.

The obvious efficiency is to fold the benchmark reading into the gate — permit a higher ceiling when the drawdown is normal for the model, compress harder when it is not. It should be resisted, and the reason is the property the gate exists to have. The gate's authority rests on being a lookup with no parameters at the moment of decision: one number, one boundary, the same answer for everyone. A percentile is the output of a simulation built from a configuration that was chosen, seeded and regenerated on some cadence, which makes it a considered estimate rather than a fact about the account. Routing it into deployment would put a modelled quantity upstream of capital authority, and every conversation about the ceiling would become a conversation about the model.

Reading them in order

Gate first for what is permitted, benchmark second for what it means.

The sequence matters for the same reason it matters on the EV scorecard. Read the gate first and the deployment question is settled before any interpretation begins, which prevents the interpretation from being recruited into an argument about size. Then read the benchmark comparison, whose entire output is a posture: is this drawdown behaving as a system like this one should, or has something changed? That answer directs review effort — toward patience, toward the structural pipelines, or toward the configuration the benchmark itself was built from. Two instruments, two questions, and a fixed order that keeps the informative one from becoming a lever on the binding one.

• Normal-for-the-model does not add capital; the gate's answer stands unchanged.
• Shallow-but-abnormal is a warning available nowhere else in the system.
• Read the gate first so the benchmark cannot be recruited into a sizing argument.

The key idea

Two instruments on one input is redundancy only if they ask the same question.

It looks wasteful to feed a single number into two separate readings, and it would be if both were producing verdicts of the same kind. They are not. One converts the number into permission and one converts it into meaning, and a system that had only the first would deploy correctly while understanding nothing about why. Keeping the questions distinct — and keeping the answers in separate places with separate authority — is what allows the account to be governed strictly and diagnosed generously at the same time.