Neither panic below median nor worship a lucky P90.

One error, two directions

Both halves treat a percentile position as new information.

The instruction pairs two behaviours that look unrelated: despairing below median and celebrating a top-decile month. They are the same mistake. In each case the operator has read a sample's position within a distribution they already knew existed, and treated it as though something had been learned. Nothing has — a distribution with a P90 band is a distribution that produces P90 outcomes at a known rate, and observing one is confirmation rather than news. Pairing them in a single instruction is deliberate, because a trader who has learned to stay calm below median and has not learned to stay calm above it has fixed the less dangerous half.

The below-median half

Half of all outcomes are below median. That is what median means.

Stated plainly the arithmetic is obvious and it is routinely ignored in practice. A system performing exactly as designed will spend roughly half its periods below its own median, and a trader who treats each of those as evidence of a problem will conclude their system is broken approximately half the time. The reading that matters is not whether live sits below median but whether it sits below the adverse band — that is the threshold where the sample has left what the evidence can explain, and it is a genuinely different event.

The P90 half

An exceptional month is the more dangerous of the two.

Below-median stretches produce discomfort and occasionally an abandoned system. Top-decile months produce something worse, because they arrive feeling like evidence of mastery and they carry a specific action attached: size up. A P90 result is a favourable ordering of the same trades the system was always going to produce, and treating it as a new capability leads directly to deploying larger size into an edge that has not changed. The asymmetry of consequence is why the instruction bothers to name the upside case at all.

What does license action

Position outside the bands, or a structural reading — never a feeling.

Three readings warrant a response, and none of them is a percentile in the middle of the distribution:

• Live below the adverse band — the sample cannot explain it; investigate before deploying further.
• Live above the favourable band — also unexplained; check for a process change before crediting skill.
• A structural drift reading — the machine is changing, independent of where this month landed.

Why this is hard

The instruction asks you to ignore the most vivid information available.

It is worth acknowledging the difficulty rather than repeating the rule more firmly. Recent results are experienced, watched, and felt; the distribution is an abstraction on a chart. Asking an operator to weight the abstraction over the experience is asking them to override the evidence their nervous system considers most reliable. This is not achievable through conviction, which is exactly why the benchmark is drawn in advance and consulted as a document rather than recalled as a belief. The difficulty also explains why the instruction is stated as a rule rather than as advice: advice is evaluated in the moment by the person it is aimed at, which is exactly the evaluation the rule exists to pre-empt.

The habit that makes it work

Read the position before reading the P&L.

One practical sequencing change carries most of the benefit. Reading percentile position first and currency second means the interpretive frame arrives before the emotional content, and the number is understood as a location within a known distribution rather than as a verdict. Reversed — P&L first, distribution second — the percentile becomes something the operator checks in order to feel better or worse about a conclusion already reached, which is the same instrument doing none of the work. The sequencing change costs nothing and is unusually durable, because unlike most discipline habits it does not require resisting anything — it only requires reading two numbers in the other order.