A fixed protocol for reading percentile charts.

Step one

Verify the ruler before measuring anything.

Every percentile read begins with a question about the chart, not the account: is this benchmark still measuring the system I'm actually running? The assumption register is the checklist — branch weights, trade rhythm, risk model, cost assumptions, starting conditions. If live operation has drifted from the register in any documented way, that drift is noted before the read, because it contaminates the comparison in a knowable direction. The Operating Standard is blunt about this: do not compare live data to the benchmark unless the live system still operates under its assumptions, or the differences are explicitly documented. Thirty seconds of verification protects everything that follows.

Step two

Locate — one vertical slice, one placement, no narrative.

Take the current week or month, find that time-slice of the fan, and place the live value in it: above P75, P50–P75, P25–P50, P10–P25, P5–P10, below P5. Write the placement down before interpreting it — the protocol deliberately separates measurement from meaning, because the two contaminate each other when done together. What's forbidden at this step: reading the chart horizontally to extrapolate a trajectory, comparing against the band you were in last quarter to construct momentum, or averaging placements across metrics into a vibe. One slice, one metric, one placement. The equity read, the return read, and the drawdown read each get their own.

Step three

Contextualize with the survival numbers the fan doesn't show.

The equity fan has a silent flaw: locked paths sit frozen in its lower reaches, and completed paths sit capped at its top, and the fan itself won't tell you how many of each. So the protocol forces the context read: alongside the placement, note the benchmark's completion share and lock share at this horizon, and the drawdown-band placement from the drawdown distribution. This is where a flattering equity read gets its risk bill — above-median equity with drawdown outside its own bands is the fast-but-risky scenario, and the playbook's response is audit, not celebration. Equity placement without survival context is exactly half a read.

Steps four and five

Persistence check, then the doctrine's response — never your own.

Step four asks the only question that separates information from action: is this placement new, or is it a pattern? One month anywhere inside P10–P90 is weather; the same adverse band three reviews running is climate, and the deviation-classification doctrine takes over. Step five closes the loop by taking the response from the playbook rather than composing one: healthy placements get documentation of what's working, weak ones get the diagnostic review at the prescribed depth, sub-P5 gets the formal audit, drawdown-adverse gets gate respect and no recovery overrides. The protocol's authorship rule is absolute — the operator executes responses; the doctrine writes them.

• Single placements inform. Only persistent placements act.
• Responses escalate with adverse persistence, never with a single month's discomfort.
• If the honest step-one answer was 'the ruler is stale,' the response is a benchmark refresh — not a performance conclusion.

The key idea

The protocol exists so every read is the same read.

Nothing in the five steps is individually clever — verify, locate, contextualize, check persistence, respond by doctrine. The power is entirely in the fixedness: a read performed identically in a euphoric month and a bleeding one produces comparable records, and comparable records are what persistence detection, deviation classification, and the edge-change threshold are all built on. Percentile literacy isn't the ability to read the chart. It's the discipline of reading it the same way every time.