Middle authority: the rung that explains but cannot command.

The position

Two boundaries, and the SDE sits between them.

MARS is built as an authority stack in which every layer has a declared job. Underneath the SDE sits the evidence layer — the Journal, the Weekly Summary, the EV Scorecard — which produces raw and computed truth about what actually happened. Above it sits the Gate and Brake system, which holds hard capital authority and decides what the account is permitted to do. The SDE receives from below and reports to neither. It takes weekly metric values that other components have already established and converts them into structural intelligence: whether expectancy is improving or degrading, whether drawdown is expanding or recovering, whether deployed risk is converting, whether profit quality is real, whether compounding velocity is gaining or losing. It publishes that reading and stops. The two boundaries are not symmetrical in the way they are enforced, and understanding the asymmetry is most of understanding the rung.

The lower boundary

The SDE does not alter raw data — and that is a formula rule, not a courtesy.

The engine never writes into the evidence it reads. It consumes downstream weekly metrics built from journal evidence, and if a number looks wrong the correct response is to fix it at source, not to adjust it in the diagnostic layer. This sounds obvious and is violated constantly in spreadsheet systems, usually with good intentions: an operator notices a metric that seems distorted by an outlier week, applies a small correction inside the diagnostic tab, and unknowingly creates a system in which two components disagree about what happened. From that point the engine is no longer diagnosing the machine — it is diagnosing a private, adjusted picture of the machine that exists nowhere else. The one-way boundary is what guarantees that a structural reading and the journal it derives from can always be reconciled.

• The engine reads five metric families — EV, DD, RAER, RAPF, ACCEL — and originates none of them.
• A disputed input is a source-layer problem. Correcting it downstream produces two truths and no way to tell which one is being read.
• Because the SDE recomputes nothing, any structural reading can be traced back to a specific week in the journal.

The upper boundary

If diagnostics and the gate disagree, the gate wins without discussion.

The manual states the rule flatly: if the SDE says conditions are constructive but the Gate Map reads Floor, Deep-Floor, Ground-Floor, or System Lock, the gate wins. Diagnostics cannot override capital protection. This is the clause that makes the rung safe to build on, because it settles in advance the single most tempting conflict the system can produce — a structural reading that is genuinely encouraging arriving during a period of genuine capital stress. Both readings are usually true at once. Expectancy really can be improving while drawdown is deep, and the improvement really is meaningful. What the rule denies is the inference: that an encouraging diagnosis constitutes permission to loosen the capital envelope. The gate governs a different question, on a different input, with a different consequence for being wrong, and the SDE has nothing to say about it.

What the rung is for

Influence over posture, not over permission.

Between those two boundaries there is real work. A structural diagnosis can and should influence operator confidence, review priority, the throttle conversation, and general risk posture — which sizing tier feels defensible inside the range the gate already allows, which branch deserves scrutiny this month, which metric family to open first on Saturday. None of that requires authority over capital. It requires only that the reading be trustworthy and that the operator know where it belongs. The distinction between posture and permission is the working definition of middle authority: the engine changes what the operator attends to and how confidently they act inside their envelope, and it never changes the envelope.

The key idea

A layer that cannot command is a layer that can be believed.

It is tempting to read the constraints as a demotion — an engine this sophisticated ought to be able to do something with what it finds. The opposite is true. Because the SDE cannot expand risk, an encouraging diagnosis carries no incentive to be encouraging. Because it cannot suppress the gate, a warning costs nothing to publish. A component with no stake in the decision is the only kind whose reading can be taken at face value, and that credibility is precisely what the narrow rung buys.