Driving capture toward one hundred per cent would destroy the branches it was meant to protect.

The tradeoff

Capture and continuation room are the same quantity, measured with opposite signs.

Every unit of room given to a trade so it can continue is a unit of favourable excursion that may not survive to the close. That is not a management failure, it is the mechanism by which continuation is possible at all — a position cannot run without being allowed to retrace. Tightening protection converts potential continuation into realised outcome, which lifts capture. Loosening it does the reverse. The two cannot be improved simultaneously because they are the same decision viewed from either end, and any repair framed as raising capture without cost is describing an operation the market does not offer.

Why the peak is not in the same place for every branch

Four branches make four different promises about continuation, so four different capture targets follow.

Normal aims at a fixed target with a static exit and no trail. Its continuation requirement is small, its expectancy peak sits well to the tight side, and a high capture figure there is genuine evidence of clean management. Trend No-Partial holds full size specifically so that a small number of trades can run a long way; its peak sits far to the loose side, and a capture figure that would be excellent on Normal is, on TNP, evidence the branch has been throttled into something it was not designed to be. Trend Partial sits between them by construction, banking some and running the rest. A single capture target applied across all four would push at least two branches off their peak in the name of an improvement.

• A capture target is only meaningful next to the branch it applies to.
• TNP monetising a modest share of offered excursion can be the branch working correctly.
• The same figure on Normal would usually mean the exit is being managed badly.

The visible-metric problem

Capture is easy to measure and the cost of raising it is not, so the pressure runs one way.

The reason this needs saying at all is asymmetry of visibility. Giveback shows up in the log as a number that went down, and that improvement is legible in every weekly review. The continuation that no longer happens leaves no trace anywhere — the trade that would have run to 3.2R and now closes at 1.4R does not record the 3.2R, because the protection was triggered before the excursion existed. So one side of the tradeoff is measured every week and the other side is structurally invisible, and an operator optimising honestly against the visible evidence will drift tighter, month after month, with every individual step supported by data.

What detects the drift

The branch's own excursion signature is the check, because tightening changes its shape before it changes its mean.

Because the cost is invisible in capture, it has to be caught somewhere else. The branch's favourable excursion distribution is where it shows: a TNP population that used to produce a thin right tail and now produces none has been managed out of its identity, and this is visible as a change in shape well before the branch's expectancy falls enough to trigger a status change. The checkpoint hit rates carry the same signal in a cleaner form — the share of trades reaching the far rungs is the direct measurement of whether the tail is still being reached. When capture is rising and far-rung hit rates are falling, that is not two findings. It is one.

What the target actually is

The correct capture figure is the one the branch's contract implies, and it is stated in advance.

Since capture has no maximum, it needs a stated target, and the target belongs to the branch rather than to the operator's ambition. Setting it in advance is what makes a subsequent reading interpretable — above target is a signal to check whether continuation is being cut, below target is a signal to check the exit rules, and neither reading is available if the goal was simply 'higher'. This is also what keeps the repair engine honest, because a driver diagnostic can only flag capture as drag if there is a target to be short of. Against a maximum, every reading is short of it forever.

The key idea

A metric with no ceiling will eventually be optimised past the thing it was measuring.

This is the general form of the problem and it is worth carrying beyond this one ratio. Capture was introduced to detect a real leak: opportunity generated and then surrendered. It does that well within a band. Pushed beyond the band it becomes an instruction to stop generating the opportunity in the first place, at which point the metric is perfect and the system no longer has the behaviour the metric existed to protect. Every efficiency ratio in the lab carries some version of this, which is why none of them is read alone and why all of them are read against a branch.