Living Systems methods review — 24 October 2026

Run: python power.py
Test: python -m unittest -v test_power.py
Dependencies: Python 3 and SciPy (tested versions recorded in verification.json).

The primary analysis is a two-sided paired t test on independent, normally
distributed donor contrasts. Technical inserts are averaged within donor and
arm. tau is the SD of TRUE donor treatment contrasts, not donor intercepts.
With independent equal-variance insert errors, SD(D)^2=tau^2+2*sigma_w^2/m.
The inputs are illustrative. Do not transfer their units or variance to a
different endpoint, correlated replicates or a mixed-model analysis.

Power=P(T<-c)+P(T>c), T~noncentral-t(n-1,sqrt(n)*delta/SD(D)),
c=t_(n-1,1-alpha/2). This is exact under the model, not exact biological truth.
The tests use a separately computed conditional-normal/chi-square integral.
Fourteen evaluable donors meet 80%; eighteen meet 90%. Seventeen give 89.95%.
Attrition, endpoint selection and pilot uncertainty require separate planning.

This package does not contain new clock scores, IDAT reprocessing or a new
experimental result. See REVIEW.txt and the website methods page for scope.
