Defensive gates are supposed to shrink deployment. Check that they did.

The assumption worth testing

Everything downstream treats compression as a fact.

The simulated drawdown bands were generated with braking modelled in. The benchmark's legitimacy rests on live and simulated populations sharing transition mechanics. The comparison workflow's cross-checks assume the gate did what gates do. Each of these inherits a single premise — that when the account entered a defensive gate, deployment actually compressed — and none of them verifies it. If the premise fails, the failure is invisible in all of them and shows up only as a set of comparisons that mysteriously stop making sense.

The instrument

Tier usage conditioned on gate, not tier usage in aggregate.

The question cannot be answered by overall tier distribution, because an account can show a perfectly reasonable annual tier profile while deploying badly inside every defensive gate — the aggregate averages the two behaviours together. What answers it is the tier-by-gate breakdown: for each capital state the account occupied, which tiers were used while it was there. That table makes the ladder's execution visible as a shape, and a healthy shape steps down. A flat one means the account carried similar deployment through progressively tighter authority, which is the ladder existing on paper only.

The mechanism that flattens it

Override, used often enough to stop feeling like an exception.

Deployment does not usually exceed a gate cap through error. It does so through the manual override that permits full-tier risk beyond what smart capacity would allow — a facility that exists for genuine exceptions and degrades, through repetition, into a default. The override stress table is built specifically to catch this, counting how often forcing full tier risk would exceed the authorised pool. Its existence in the benchmark is itself informative: the designers expected this failure and instrumented for it, because an override that requires no justification and leaves no count is indistinguishable from a higher risk setting.

Why it hides

Flattening the ladder improves results right up until it does not.

The behaviour is self-concealing in the worst way: deploying more heavily through defensive gates produces better outcomes most of the time, because most drawdowns recover, and the heavier deployment recovers them faster. The operator receives repeated confirmation that the override was correct. What the behaviour has actually done is remove the protection that matters in the minority of drawdowns that keep deepening — and those are the ones the ladder was built for. A practice that is rewarded in the common case and catastrophic in the rare one will always feel validated by experience, which is precisely why it is audited by table rather than by recollection.

What a finding routes to

Deployment behaviour, not the ladder's design.

A confirmed flattening is a compliance finding and it goes to the deployment record: which overrides were logged, with what stated reasons, in which gates, over what period. It does not go to the ladder's thresholds. The instinct to respond by adjusting the gate boundaries — reasoning that the caps were evidently too tight, since deviating from them worked — inverts the entire logic, converting an audit failure into a justification for lowering the standard that detected it. Ladder changes have their own route through the sandbox and the promotion gauntlet, and 'the operator kept overriding it' is not evidence admissible on that route.

• Aggregate tier usage cannot answer this; the gate-conditioned breakdown can.
• Override counts are the corroborating evidence, and the benchmark instruments them deliberately.
• The finding is about compliance. Ladder redesign is a separate question with a separate process.

The key idea

An automatic protection nobody verifies is a protection on trust.

The gate ladder's value comes entirely from being mechanical — it decided how much to de-risk while everyone was calm, and it binds without negotiation. But it binds through the operator's deployment decisions, which means the mechanism has a human step in it, and human steps are where automatic systems quietly become advisory. The tier-by-gate record is a short table that answers whether the most important promise in the risk stack was kept. Not checking it is how a governed account discovers, in its worst drawdown, that it had been ungoverned for a year.