Reading execution evidence: the scenarios that fool operators.

The premise

Execution evidence arrives tangled, and untangling is a skill.

The lab's outputs — MAE, MFE, capture, adherence, branch integrity, fees — rarely point the same direction at once. A week hands you profitable trades with weak adherence, controlled losses with terrible capture, a Normal trade that behaved like a trend. Each combination has a correct reading and a tempting wrong one, and the tempting one usually flatters. The scenario guide exists because these combinations recur — they're not edge cases but the ordinary texture of live evidence — and because the cost of misreading them compounds: a misread scenario becomes a mislabeled journal row becomes a corrupted branch statistic becomes a deployment decision standing on it.

The dangerous winner

Profitable with weak adherence is contaminated evidence, not success.

The guide's sharpest entry: plan adherence is weak but results are green. The manual's verdict is unequivocal — this is dangerous; the result was profitable, but not clean evidence; do not reward the behavior; log the violation and treat the trade as contaminated process evidence. The reasoning cuts to what the evidence system is for. Every journal row is a vote about whether the system works as designed — and a trade that violated the design while winning is a vote about nothing, or worse, a vote for the violation. Rewarding it (mentally, or by letting it pad the clean statistics) trains exactly the drift the whole governance stack exists to prevent. The CP3 doctrine says the same thing from the other side: profitable violations should not be counted as clean system evidence. The profit is real and welcome. The evidence is quarantined.

The honest loser

Red with high adherence and controlled MAE is usually weather.

The mirror scenario: a red week with high adherence and controlled MAE. The tempting read is failure — the week lost, something must change. The manual's read is patience: this may be normal variance rather than system failure, and the prescribed move is to check CP3, the Weekly Scorecard, the gate/brake state, and the benchmark context before changing rules. The logic is the deviation doctrine in miniature. Clean process plus controlled risk plus a losing outcome is exactly what a healthy positive-expectancy system produces some fraction of the time — the benchmark quantifies how often — and a week like that carries no repair instruction at all. The scenario's real danger isn't the loss; it's the repair urge, because rules edited in response to variance never accumulate the sample that would have vindicated them.

• Adherence and MAE control are the process reads; the outcome is the variance read. When process is clean, the outcome gets benchmark context before judgment.
• The scorecard's RED-stable cell prescribes the same order: reduce per doctrine and diagnose — never improvise fixes.
• A clean red week is, evidentially, worth more than a dirty green one. The record should reflect that inversion.

The mismatch and the data floor

Branch integrity and QA: the scenarios about the record itself.

Two more entries guard the evidence layer directly. The branch mismatch — a Normal trade that caught a trend, a No-Partial that failed survival — tempts retroactive relabeling to make the statistics tidy; the rule is the opposite: review branch classification before entry, and never retrofit labels after the outcome, because a label changed to fit a result converts evidence into decoration. The lab's branch-integrity flag exists to surface these cases rather than bury them. And beneath everything sits the QA floor: the Trade_Log's data-QA column must read OK before any output is trusted, exact branch labels are mandatory (a misspelled branch is invisible to every rollup), hit flags come from the actual price path rather than the final outcome, and fees include swap when incurred. Bad data produces false diagnosis — the manual's five words that justify the entire QA section.

The key idea

The scenarios are one discipline wearing five costumes.

Every entry in the guide reduces to the same separation: process evidence and outcome evidence are different things, and each gets judged by its own standard. Profit doesn't launder a violation; loss doesn't indict a clean process; a label describes the plan, not the result; and none of it means anything on a corrupted record. Operators who internalize the separation read the same tangled weeks as everyone else and file systematically truer rows — and since every downstream engine eats those rows, the reading discipline at this layer is quietly load-bearing for the whole stack.