%% SCRAP: experiments/02-experiments/README
%% SOURCE: docs/working/experiments/02-experiments/README.md
%% STATUS: CURRENT
%% FITS: experiments/ch-framework
%% EDITORIAL: lifted — prose rewritten to press voice

\section{Experimental Framework Overview}

StarForth's experimental programme applies rigorous design-of-experiments methodology
to the physics-driven adaptive runtime, yielding a body of evidence that grounds
every performance claim in reproducible, statistically validated data. Four
experimental series structure the work.

\begin{itemize}
  \item \textbf{Factorial DoE} — Full $2^k$ factorial designs sweep all binary
        combinations of feedback-loop configurations, revealing main effects and
        interaction terms that single-factor studies cannot detect.
  \item \textbf{Heartbeat DoE} — Focused experiments on the heartbeat subsystem
        measure temporal stability, load-response coupling, and convergence
        dynamics of the adaptive tick mechanism.
  \item \textbf{James Law} — Window-scaling experiments validate the conservation
        invariant $\Lambda \times (\mathrm{DoF}+1) = W$ across multiple window
        sizes, degrees of freedom, and workload shapes.
  \item \textbf{Physics Optimization} — Targeted parameter sweeps quantify the
        marginal contribution of individual physics-engine parameters, establishing
        a tuning baseline for production builds.
\end{itemize}

\subsection{Experiment Lifecycle}

Every experiment follows a five-phase protocol:

\begin{enumerate}
  \item \textbf{Design} — state the hypothesis, identify factors, and specify
        response variables before collecting any data.
  \item \textbf{Protocol} — write a formal execution guide documenting methodology
        and success criteria.
  \item \textbf{Execute} — run the experiment and collect data without mid-course
        adjustments.
  \item \textbf{Analyse} — apply statistical methods (ANOVA, Levene's test,
        credible-interval estimation) to the collected measurements.
  \item \textbf{Document} — write a summary and promote publication-ready results
        to the research archive.
\end{enumerate}

\subsection{Key Result: Deterministic Behaviour}

Across 90 experimental runs in the physics-engine validation campaign, StarForth
achieves \textbf{0\% algorithmic variance}, confirming formally proven deterministic
behaviour of the physics-driven adaptive runtime. The $2^7$ factorial campaign
extends this over 38{,}400 runs with consistent convergence to the lowest-CV
steady state.

\subsection{Statistical Methodology}

Each campaign employs factorial designs for multi-factor analysis, heartbeat-driven
data collection at millisecond resolution, and statistical validation via ANOVA and
Levene's test. Experimental protocols are reproducible: run scripts, R analysis
code, and raw CSV data are archived alongside results.

%% TODO(bob): confirm canonical published path for the DoE metrics schema and
%%   publication-results documents referenced in the source