How raw inputs become scores — and why an experiment of one is built the way it is.

Why N=1

Population studies find the average effect across a group — real signal, but a statistical composite that may not resemble any individual. The N=1 approach turns that limitation into a feature: one subject means every measurement is directly relevant, with no between-person noise. The trade-off is external validity — you cannot generalise these results to you, and that is fine. The value is the framework, not the conclusions.

The correlation engine

The core analytical layer is a rolling correlation engine that continuously computes Pearson r across metric pairs (sleep, recovery, nutrition, training, glucose). Running ~23 pairs at once creates a multiple-comparisons problem — the more tests you run, the more likely one is a false positive. So a Benjamini-Hochberg FDR correction controls the false-discovery rate across all simultaneous comparisons, and a minimum sample size is required before any correlation is surfaced.

A finding is only reported with its p-value alongside r, and an FDR-adjusted q-value used for filtering. Example: a confirmed pattern at p=0.003, n=47 paired days, BH-FDR q=0.014 — survives correction.

Confidence vocabulary

Everything is correlative, never causal. Fewer than 12 observations is a preliminary pattern; fewer than 30 is low confidence. The character model rolls these into a Level (1–100) across five tiers — Foundation → Momentum → Discipline → Mastery → Elite — over seven pillars: Sleep, Movement, Nutrition, Metabolic, Mind, Relationships, and Consistency.

The model never computes in prose — it interprets pre-computed numbers only. N=1

The Evidence