Three interpretation modes are selected before any result is read.

Three switches

Each one changes the interpretation without changing a single trade.

The expectancy calculation mode decides how closed evidence becomes a daily reading — weighted by baseline branch weights, averaged flat across closed trades, weighted by represented branch identity, or forced manually for unusual situations. The throttle output mode decides whether an incomplete cycle may produce promotable evidence at all. The throughput override mode decides how fresh deployment is sized around open exposure. Between them they can move a cycle's reported expectancy, change whether that expectancy is allowed to travel, and change how much gets deployed on it — all while the ledger underneath remains character-for-character identical. That is a large amount of interpretive latitude sitting behind three dropdowns.

The failure

A mode selected after the result is a conclusion wearing a method's clothes.

The mechanism is ordinary and does not require bad faith. An operator reads a disappointing weighted expectancy, recalls that flat averaging exists and is a legitimate reality check, switches to it, and finds a friendlier number. Every step is defensible in isolation and the mode genuinely is legitimate — the manual describes it as a quick reality check of realised cycle output. What makes the sequence corrupt is that the switch was prompted by the first result rather than by a question about which interpretation fits the cycle. Repeated across a quarter, this produces a reported expectancy that is systematically the maximum across available interpretations, which is a statistic about the operator's mode selection rather than about the account.

Why ordering fixes it

Choosing in ignorance of the outcome is what makes the choice about the question.

The protection is not that the operator becomes more principled, but that at step four the results do not yet exist to be selected among. The only information available for choosing is structural: how many branches have closed evidence this cycle, whether trades remain open, whether the reading is intended for promotion or for end-of-day monitoring. Those are exactly the considerations each mode's guidance names — branch weighting when branch mix discipline matters, branch-weighted closed as the best partial-cycle analytical mode, closed-only context for monitoring rather than casual promotion. A choice made on those grounds is a choice about the situation. The same choice made two steps later is a choice about the number, and nothing in the workbook can tell the two apart afterwards.

The default that should hold

The strict combination is the live default and departures are the thing to notice.

The named defaults are the strict ones: branch-weighted evidence, complete cycles required before expectancy may travel, and smart capacity for sizing. Together they describe the system operating as designed, and the honest consequence is that they produce fewer promotable readings and more conservative deployment than the alternatives — which is precisely why they are the defaults rather than an option. The useful discipline is therefore not to avoid the other modes but to treat any departure as an event with a reason attached. A cycle run in flat averaging with closed-only context permitted is a cycle interpreted under a relaxed configuration, and the reading it produced should be remembered as such rather than filed alongside readings taken under the default.

The one that leaves a trace

The throughput mode is logged; the other two are not.

There is an asymmetry here worth flagging plainly. The throughput override mode travels in the packet and is recorded in the decision log, so its use is visible in review and countable across cycles. The expectancy calculation mode and the throttle output mode are console-side controls that shape what the packet contains without themselves being part of it. Their effect is therefore embedded in the transferred status with no record of which interpretation produced it. This is not an argument against them, and it does mean the discipline around those two rests entirely on the checklist's ordering rather than on any downstream audit. A note in the cycle's own record of which modes were in force costs one line and closes the gap.

• The three modes change interpretation, transfer and sizing — never the evidence.
• Chosen at step four the decision is about the cycle; chosen at step seven it is about the number.
• Only the throughput mode is logged; the other two leave no trace at all.

The key idea

Where a choice sits in a sequence determines what the choice can be about.

This is the same protection the simulation layer gets from declaring a configuration before a run and the same one the profile comparison gets from requiring persistence across windows. In each case the mechanism is identical: fix the interpretive decision before the outcome is visible, so that the decision cannot be a response to the outcome. Nothing about the operator's judgement is distrusted in the abstract. What is distrusted is judgement exercised after a number has appeared, which is a different and demonstrably worse instrument than the same judgement exercised a few minutes earlier.