Departing from the declared variant and winning is the most costly outcome available.

The loop

Improvise, succeed, generalise — and the matrix quietly becomes advisory.

The sequence is short and it repeats. A trade is running, something about it prompts a departure from its declared management, the departure works, and the outcome is filed as evidence that reading the situation live was better than following the rail. Nothing about that episode was measured — the counterfactual is unavailable, the sample is one, and the market conditions that made it work are not recorded. But the reinforcement is immediate and emotional and the counterargument is statistical and abstract, so the reinforcement wins. Repeated a handful of times, the operator has acquired a belief that the matrix is a default for uncertain situations rather than a rail, which is the precise state the module exists to prevent.

Why the outcome is not evidence

The comparison the operator makes in their head is against a version of the trade that does not exist.

The felt argument for a successful improvisation is a comparison: this worked, and following the rail would not have worked as well. The second half of that sentence is a fabrication. Nobody observed the rail version of the trade, because the trade only happened once. What is actually known is one outcome under one management, and what is being claimed is a difference between two. The variant rollups exist to supply that comparison properly, over many trades, with the declared posture recorded per trade — the variant-versus-Standard ratio is exactly the number a single improvised win pretends to be. One trade cannot produce it and no amount of vividness changes that.

What the log actually does

Recording the breach separates the outcome from the process, so both can be counted.

The mechanism is unglamorous. A breach is recorded as a departure with its outcome attached, which means the month's review can report both how the book performed and how often it was run as specified. Those are different numbers and they need to stay different. A month with strong results and eleven departures is not a good month; it is an unmeasured month that happened to end well, and the results cannot be attributed to the system because the system was not what ran. Keeping the count visible is what stops a run of favourable outcomes from being read as validation of a process that was partly abandoned.

• Outcome and adherence are reported separately, never netted against each other.
• A profitable month with many breaches is an evidence failure, not a success.
• The count is what makes the drift visible before it becomes the operator's default.

The contamination cost

Breaches damage the statistics of the variant they were logged under, not just the discipline.

There is a second cost that is easier to overlook. A trade declared under one posture and managed under another is recorded in the first posture's statistics with an outcome the first posture did not produce. Enough of those and the attribution rollups are describing a mixture, the variant-versus-Standard ratios drift toward each other, and the module loses its ability to tell which posture is actually performing. The damage is retrospective and cumulative: it degrades evidence that was already collected, and it cannot be repaired later because the record does not say which trades were mixed unless the breach was logged at the time. Logging is therefore doing double duty: it protects the discipline going forward and it protects the interpretability of the sample already gathered.

What a breach is not

A rule that no longer fits is a proposal, and proposals go through the scenario process.

None of this says the declared postures are beyond revision. Operators do notice real weaknesses in a rule while trading it, and that noticing is valuable. The distinction is where the revision happens. A weakness observed live becomes a candidate change, the candidate is defined as a scenario with expected behaviour, it is tested against evidence, and it is promoted or rejected on a persistence window. What it does not do is take effect during the trade in which it was noticed. The whole apparatus exists so that a good idea and an impulse — which feel identical in the moment — are separated by a process that only one of them survives.

The key idea

The dangerous departures are the ones that never generate a consequence.

A system's rules erode at their most forgiving points, not at their harshest. Nobody repeats an improvisation that cost them a large loss. Everybody repeats one that made them money, and each repetition makes the next departure feel more reasonable than the last. Recording the breach is a small, dull administrative act that does one specific thing: it ensures a favourable outcome cannot silently rewrite the rule that the outcome had nothing to do with.