The highest authority in the stack has a documented exception path.

Two gates

One is computed in the console; one is selected in the throttle.

The gate exists at two points in the chain and they are not the same object. The console computes it from current equity against the equity peak, producing a drawdown percentage and the band it falls in. The throttle receives it as a selected input in its decision block — a dropdown, not a formula — and everything downstream of that selection is computed from what the dropdown says rather than from what the console calculated. In the ordinary case they agree and the distinction is invisible. The clause exists because they are separable, and because separating them is occasionally correct.

When the calculation is wrong

The arithmetic is sound and its inputs are not always representative.

The drawdown computation is a subtraction, a division and a lookup, and none of those steps can misbehave. What can misbehave is the peak reference. A withdrawal lowers equity without any trading loss, so the computed drawdown deepens on evidence of nothing. A deposit does the reverse. A data correction, a broker adjustment, or a peak captured from a transient intraday figure rather than a settled balance all produce the same effect: a mathematically correct drawdown measured against a reference point that does not represent the account's trading history. In those cases the calculated gate is faithfully reporting a comparison nobody wants made, and the honest response is a documented divergence rather than deploying against a state the account is not in.

The direction that matters

Overriding toward restriction is cheap; overriding toward permission needs scrutiny.

The clause does not distinguish direction and the risk profile emphatically does. Selecting a deeper gate than the calculation produced applies the survivability rule by hand — the account deploys less than the arithmetic authorised, the cost is bounded forgone upside, and nothing the governance layer protects is at risk. Selecting a shallower gate raises the tier ceiling, which is the single thing the gate exists to prevent any other input from doing. And the person most motivated to find the calculation unrepresentative is an operator who has just computed an uncomfortable drawdown, which is precisely when the ceiling is doing its most valuable work. The documentation requirement carries the weight here: a written reason for deploying more after a drawdown reads very differently in review than it did while being composed.

What documentation is for

The record makes an exception countable, which is what keeps it exceptional.

A justification written into the decision log does not make an override correct, and that is not its function. Its function is to convert a discretionary act into a countable one. One documented gate divergence explaining a mid-quarter withdrawal is a piece of bookkeeping. Four of them across a difficult quarter, each individually reasonable, is a pattern with a name — and the pattern is only visible because the individual instances were recorded in a form that survives past the week in which they felt justified. This is the same mechanism the throughput override runs on: legality plus visibility, with the visibility doing the actual restraining. An override nobody can count is an input to the deployment chain that nobody ranked.

The better fix

A divergence that recurs is a signal to repair the reference, not to keep overriding.

The clause handles the case where the calculation is unrepresentative on a particular day. It is not a maintenance strategy. If the computed gate is regularly wrong in the same direction, the peak reference has drifted and the correct response is to establish what it should be and reset it, rather than to override the consequence each cycle. Capital movements are the usual cause and the usual remedy is to reconsider the peak at the moment of the movement, treating a deposit or withdrawal as an event that requires a deliberate decision about the reference. An override applied repeatedly to compensate for a stale input means the gate is now being set by hand every cycle, which is a materially different system from the one on paper.

• The gate is computed in one workbook and selected in another — they can differ.
• Toward restriction costs upside; toward permission removes the ceiling's whole purpose.
• Recurring divergence means repair the peak reference, not repeat the override.

The key idea

A rule with no exception path gets broken invisibly instead of bent visibly.

The system could have wired the gate so the throttle computed it directly and no divergence was possible. It would be a cleaner diagram and a worse instrument, because the cases where the calculation is unrepresentative are real and would then have to be handled by not following the directive at all — outside the workbook, outside the log, with nothing recorded. Permitting the exception and requiring it to be written down keeps the departure inside the system where it can be counted. That is the same reasoning that put a legal override on the throughput chain, applied at the one layer where it is most uncomfortable to admit it is needed.