81 lines
3.8 KiB
TeX
81 lines
3.8 KiB
TeX
%% SCRAP: archive/session-logs/2025-11-20-recap
|
|
%% SOURCE: docs/working/archive/session-logs/2025-11-20-recap.md
|
|
%% STATUS: HISTORICAL
|
|
%% FITS: none
|
|
%% EDITORIAL: lifted — prose rewritten to press voice
|
|
|
|
\section{Session Recap: 20 November 2025}
|
|
\subsection*{From Factorial DoE to Multivariate Dynamics Analysis}
|
|
|
|
\textbf{Duration:} approximately 90 minutes.
|
|
\textbf{Outcome:} complete paradigm shift documented; ready for code
|
|
implementation.
|
|
\textbf{Commits:} 3 (design documents plus executive summary).
|
|
|
|
\subsection{Summary of Paradigm Shift}
|
|
|
|
Work began with the goal of running the DoE with heartbeat data under a
|
|
five-configuration, fifty-runs-per-configuration plan (250 total runs).
|
|
Three sequential corrections, each driven by deeper understanding of the
|
|
system, produced a fundamentally different experimental framework.
|
|
|
|
\textbf{Correction~1.} The heartbeat tick interval (\texttt{tick\_ns}) is
|
|
not a fixed constant but a tuning variable modulated by the inference engine
|
|
in response to load. The experiment therefore measures load-responsive
|
|
adaptation, not fixed-interval jitter.
|
|
|
|
\textbf{Correction~2.} 250 runs are statistically insufficient for
|
|
correlation analysis. The revised plan calls for 150 runs per configuration,
|
|
for a total of 750 runs (5 configurations $\times$ 150 runs). This provides
|
|
adequate statistical power for correlation analysis between metrics.
|
|
|
|
\textbf{Core insight.} Each metric in the feedback system is simultaneously
|
|
a signal and a feedback input. The seven feedback loops form a coupled
|
|
dynamical system in which every metric influences the future state of every
|
|
other metric. The experimental paradigm therefore shifts from factorial
|
|
analysis of main effects to multivariate dynamical systems analysis:
|
|
measuring the full trajectory shape of each configuration through
|
|
metric-space rather than summarising it by means.
|
|
|
|
\subsection{Design Documents Produced}
|
|
|
|
\textbf{MULTIVARIATE\_DYNAMICS\_DESIGN.md (622 lines).}
|
|
Defines \texttt{HeartbeatTickSnapshot} (per-tick capture), \\
|
|
\texttt{RunStabilityFingerprint} (per-run summary vector), and
|
|
\texttt{ConfigurationFingerprint} (per-configuration aggregation). Provides
|
|
complete CSV schemas for three output levels, a five-phase analysis pipeline,
|
|
and references to control theory and systems-biology literature.
|
|
|
|
\textbf{HEARTBEAT\_INSTRUMENTATION\_PLAN.md (350+ lines).}
|
|
Specifies a minimal-overhead circular buffer (115\,ns per tick, 0.01\%
|
|
overhead). Defines delta metric tracking, integration points in
|
|
\texttt{vm.c} (\texttt{vm\_init}, \texttt{vm\_cleanup},
|
|
\texttt{vm\_heartbeat\_run\_cycle}), and a CSV export function. Addresses
|
|
storage and sampling for up to 75~million tick rows.
|
|
|
|
\textbf{PHASE\_2\_EXECUTIVE\_SUMMARY.md (267 lines).}
|
|
Provides stakeholder-level clarity on the paradigm shift, the seven-metric
|
|
feedback web, the three measurement scales (tick, run, configuration), and
|
|
the golden-configuration selection criteria (five weighted dimensions).
|
|
|
|
\subsection{Key Insights}
|
|
|
|
\begin{enumerate}
|
|
\item The heartbeat is an active control variable, not a passive clock.
|
|
\item Metrics must be measured simultaneously as a coupled system;
|
|
isolation is not possible.
|
|
\item 750 runs are required for correlation power; the experiment therefore
|
|
runs six to eight hours.
|
|
\item The optimal configuration is not the one with the highest mean
|
|
performance but the one with the deepest attractor basin, fastest
|
|
convergence, tightest load coupling, and lowest jitter.
|
|
\end{enumerate}
|
|
|
|
\subsection{Implementation Checklist at Session End}
|
|
|
|
Structs and data definitions were complete. Integration points in
|
|
\texttt{vm.c} were specified. The analysis pipeline (convergence detection,
|
|
coefficient of variation, correlation matrix, spectral decomposition,
|
|
composite stability scoring) was designed. Seven C-level code changes and
|
|
three R/Python analysis components remained unimplemented.
|