The cost of eyeballing is not wrong distances. It is that nothing can ever be graded.

The concession

On any individual trade, the eye can win — and that is not the comparison that matters.

It is worth granting this at full strength rather than arguing against a weak version. An operator who has watched an instrument for two years carries genuine pattern knowledge that no lookup table encodes, and there are trades where they will place a stop better than the coefficient would. If the question were which method places a better stop on a given trade, the answer would be contested. But that is not the operating question. The operating question is which method produces a book that can be improved next quarter, and on that question there is no contest at all.

Why the sample collapses

Grading requires grouping, and grouping requires that the parameter took repeated values.

The mechanics are simple and decisive. To answer whether a stop policy is too tight, a review groups trades by the policy they were managed under and compares stop-out rates, capture and giveback across the groups. That operation needs the parameter to have taken a small number of repeated values across a meaningful sample. An eyeballed book has a parameter that took a hundred approximately unique values, none of which is recorded as a value at all. There is nothing to group by. The trades are still evidence about the operator's overall performance and they carry no information about distance policy whatsoever.

The drift the eye cannot see in itself

Eyeballed distances move with recent outcomes, in the direction that feels safest.

Beyond being unanalysable, the eyeballed parameter is not random — it drifts systematically. After a run of stop-outs the eye places stops wider, because the recent memory is of being shaken out. After a run of clean wins it places them tighter, because the recent memory is of moves that ran cleanly. Both adjustments feel like learning and both are recency weighting. The result is a parameter that is loosest after losses, when the account can least afford the wider risk, and tightest after wins, when conditions are most likely to still be favourable. A fixed coefficient is not smarter than the eye; it simply does not move for reasons that have nothing to do with the instrument.

• The drift is directional and follows the last few outcomes, not the market's state.
• It is invisible from inside, because each individual adjustment feels justified.
• A coefficient can be wrong; it cannot be wrong differently on Tuesday than on Monday.

What the structure supplies

Four inputs the eye does not have access to, and one it has too much of.

The calibration reads volatility in relation to the trigger timeframe, the authority timeframe, the asset class and the engine mode. The eye reads the chart in front of it, which is the trigger timeframe alone. It has no view of what the higher timeframe is doing to the noise envelope, no reference for how this asset class typically behaves relative to others, and no awareness of which engine mode is active. What it does have in abundance is the last few hours of price action, weighted far more heavily than a quarter of history would justify. The structure is not adding intuition — it is adding three references and removing one bias.

The honest boundary

A coefficient can still be the wrong coefficient, and the review is what catches it.

None of this claims the calibrated distance is correct. It claims the calibrated distance is legible, which is a different and more useful property. A coefficient assigned to the wrong band will produce a measurably poor stop-out profile, that profile will show up grouped by band in the review, and the assignment can then be corrected with evidence. The eyeballed alternative fails silently in the same situation — the stop-outs happen, they feel like bad luck, and there is no parameter to point at. Being wrong in a way that can be discovered is the entire advantage on offer.

The key idea

Consistency is not a virtue in itself. It is the precondition for learning anything.

The argument for calibration is often made as though consistency were intrinsically good, which invites the reasonable objection that consistently doing the wrong thing is not an improvement. The real case is narrower. Consistency is what converts a sequence of trades into a sample, and a sample is the only thing an operator can interrogate. Without it, every conclusion about stop policy is an anecdote about the trades that happened to be memorable, and the operator's beliefs about their own distances become unfalsifiable — which is the point at which they stop improving.