A map from the question you have to the instrument that answers it.

Why an index is the right shape

Nobody arrives with a capability. They arrive with a question.

An architecture diagram answers what exists; it does not answer where to go. Operating questions arrive in a specific and fairly stable form — is this allowed, how much is left, why did that trade give back so much, is this stretch normal, which branch is carrying me, should I change something. Each has one instrument that owns it and at least one that will answer plausibly without owning it. The failure mode is not confusion but false resolution: the wrong instrument returns a real number, the operator acts on it, and nothing indicates the question was answered somewhere it does not live.

The permission family

Anything phrased as may I resolves at the throttle.

Questions about whether a trade is allowed, what size is authorised, or whether there is capacity for one more position all terminate in the same place, and they are the only family whose answer is binding. The pre-trade checklist handles the categorical part — approved asset, valid session, stop and target inside plan — and the throttle handles the quantitative part through the gate row, tier ceiling and authorised pool. The characteristic misdirection is to answer this family from conviction about the setup, which is a real signal about a different question entirely and has no authority over deployment.

The explanation family

Anything phrased as why sits downstream of the decision and cannot amend it.

Why expectancy moved, why a trade gave back its excursion, why the branch mix is drifting, whether the operation is strengthening or decaying — these belong to the MAE/MFE Lab, the structural engine and the regime classifier, and they share an important property: none of them can change what is currently permitted. Their findings enter the next authorisation through the evidence they contribute, after review. An operator who takes a compelling diagnostic finding as grounds to exceed the current gate has moved an answer from the explanation family into the permission family, where it has no standing.

The comparison family

Anything phrased as is this normal needs an external reference.

Whether a drawdown is unusually deep, whether a run of losses is remarkable, whether target completion is fast because the method is strong or because risk is excessive — none of these can be answered from the live record alone, because a single path carries no distribution to be judged against. That is the benchmark's territory, and the structural engine supplies the complementary internal read. The habitual misdirection is to answer from recent win rate, which describes the same path the question was asked about and therefore cannot say whether it is normal. A path cannot be its own reference frame, and the reason this error survives experience is that the win rate is genuinely informative about something else and arrives faster than the benchmark does.

• Permission questions bind; explanation questions do not, until the next cycle.
• Comparison questions need an external distribution — a live path cannot judge itself.
• The wrong instrument returns a real figure, which is why misdirection is hard to spot.

The change family

Anything phrased as should I change resolves in the sandbox, never in production.

Proposals to adjust a trail coefficient, a branch weight, an exit rule or a gate threshold form their own family with a single owner and a formal procedure: defined, tested, stress-tested, compared with baseline, reviewed for variance and drawdown effects, approved, and deliberately promoted. The instrument that must never answer this family is last month's result, and the reason is measurement rather than caution — a rule changed in response to a recent outcome resets the observation count on the thing being evaluated, so the question that prompted the change becomes permanently unanswerable.

The key idea

Knowing where a question does not live is the more useful half of the map.

Every instrument here will return a number for almost any question put to it, and the number will be correct about something. The value of the index is therefore less in finding the right destination — which experience supplies fairly quickly — than in recognising the four or five habitual misdirections that survive experience, because they feel like answers. Setup conviction answering a permission question and account total answering a branch question are not beginner errors; they are the default reflexes the architecture was assembled to interrupt. Each of them was a reasonable heuristic before the instruments existed, which is exactly why they persist after.