Every candidate repair has a price in R, and the engine's job is to print it.

The two-variable map

Adverse excursion on one axis, capture on the other, expectancy impact in the cells.

The engine holds the current sample fixed and asks what net expectancy would have been under single-variable improvements. Reducing average adverse excursion is the entry-quality axis: tighter location, less chasing, better stop placement relative to structure. Raising capture is the exit-quality axis: trail discipline, protection timing, fewer premature closes. Every cell in the grid is a combination of the two, and the value in it is the estimated change in net expectancy per trade. The output is not a prediction — it is a sensitivity, meaning it describes how much the machine responds to a lever rather than promising the lever will move.

Why the ranking surprises people

The driver that feels worst is often not the driver with the most expectancy attached.

Adverse excursion is the driver operators feel, because it is the part of the trade that hurts while it is happening. Capture is the driver they rationalise, because giving back open profit can always be explained as letting the trade breathe. The engine is indifferent to both experiences. It frequently reports that a sample already generating adequate favourable excursion has more expectancy available in the exit than in the entry, because the opportunity is being manufactured and then surrendered. That result is unpopular and usually correct, and it is precisely the finding a review conducted from memory would never reach.

How to use it honestly

Read the driver diagnostics first, then take the grid to the drivers that were flagged.

The engine will happily price a repair to a driver that is already performing at target, and the resulting number will look like an opportunity. It is not one. The correct sequence starts at the driver diagnostics, which rank the seven execution drivers against their branch-aware targets and flag the ones costing expectancy as drag or watch. Only those flagged drivers should be taken to the sensitivity grid. Doing it the other way around — scanning the grid for the largest number and working backwards — produces a repair aimed at whichever lever the model happens to be most sensitive to, which is a property of the model rather than of the trading.

• Diagnostics identify which drivers are failing; the grid prices only those.
• A large sensitivity on a healthy driver is a modelling artefact, not an opportunity.
• Sensitivities are estimates from the current sample and shift as the sample does.

The realism filter

The right cell is the largest improvement you can actually deliver next week, not the largest number on the grid.

The grid extends to improvements that would be transformative and are not available — halving adverse excursion, lifting capture twenty points. Reading the corner of the map and setting it as a target produces a repair that fails and a month of evidence that proves nothing, because the fix was never implemented at the size it was priced at. The disciplined read is to identify the most ambitious cell that corresponds to a change you can describe as a concrete rule adjustment, and to price that one. A tenth of an R of adverse excursion is a rule about where entries are permitted. Twenty points of capture usually is not a rule at all.

What it cannot price

The engine models the levers it was given, and adherence is not one of the mechanical ones.

Plan adherence sits in the driver list alongside the mechanical drivers, and it is the one the sensitivity grid handles least well. Mechanical improvements are continuous — excursion and capture can move by a tenth. Adherence is closer to a switch, and the expectancy consequences of breaking rules are not distributed like the consequences of entering slightly late. A sample containing three rule breaks will produce a sensitivity estimate for adherence that looks tidy and is describing a population of three events. When adherence is the flagged driver, the repair is a process one and the grid should be closed rather than consulted.

The key idea

Pricing the repair is what stops the review from choosing the most emotionally available fix.

Without a price, the repair selected after a difficult week is the one that addresses the memory that hurts most, which is almost always a loss and almost always an entry. With a price, the selection is arithmetic and the emotional weighting is removed from a decision it was never qualified to make. That is the entire contribution of this part of the lab: it does not make the operator better at execution, it makes them better at choosing which part of execution to work on, and over a year those are close to the same thing.