appendix.html  ·  models, figures, and relay materials

Appendix
Models, Figures, and Relay Materials

MOAT Benchmark Code · Experiment Figures · Full Relay Logs

AModels / Code

appendix/moat_v5g.py MOAT v5g — Final Benchmark Spec Current Random-geometry environment with 2D state and 2D action space ($v_B, v_Q$ sampled per episode). Implements two-level measurement: Diagnostic Controls (D_probe / D_oracle / DirectionalEnergy_B) and Performance Metrics (AUC_residual / AUC_action / D_norm). Supports full baseline comparison against EKF / IMM / Particle Filter / Dual Control / Active BHT. Dependencies: numpy, scipy, sklearn, matplotlib.
Python · .py
appendix/moat_v5c.py MOAT v5c — Causal Attribution Separation Test Superseded Attribution separation test via do-null comparison in a 4-phase environment (baseline / B_true rotation / chaos increase / simultaneous). Compares 5 agents: KalmanLike / MetaAdaptive / CFWorldModel / ChaosFake / Observer. Intermediate version prior to revision into v5g. Uses scalar $G_{hat}$-based comparison (with known limitations).
Python · .py
appendix/moat_v5b.py MOAT v5b — Observer-Agent Benchmark (Intermediate) Superseded Classical surrogate environment for testing the evolutionary advantage of RICD-sensitive updates. Experiment system corresponding to the previous version (Observer-Agent theory). Measures generational change in RICD sensitivity parameter $g_1$ and fitness.
Python · .py
appendix/moat_v3.py MOAT v3 — Initial Benchmark Superseded Benchmark used in the early stages of the relay. First version to implement identity condition separation (corr(Δcf, gain)) for Controller / Observer-Agent. Differs substantially in design philosophy from the current version.
Python · .py

The code was produced by Claude / Codex during the dialogue relay and subsequently formatted. v5g has a finalized design specification only; implementation and operational verification have not been carried out (see caveats). Code prior to v5c has been operationally verified, but no independent reproduction has been performed.

BFigures

appendix/moat_v5c_results.png PNG · Fig. 1

Figure 1 — MOAT v5c Experiment Results (5-Agent Comparison). Time series of B_est alignment (cos_sim) across the 4-phase environment. Compares 5 series: KalmanLike (constant lr) / MetaAdaptive (error-driven lr) / CFWorldModel (do-null comparison, constant lr) / ChaosFake / ObserverAgent. Visualizes the collapse of MetaAdaptive in Phase 3 (chaos increase) and the partial stability of Observer. Corresponds to §3 of the main theory.

MOAT v5c 5-agent comparison graph Click to enlarge
appendix/attribution_subspace.png PNG · Fig. 2

Figure 2 — Geometric Illustration of Attribution Subspaces $\mathcal{S}_B$ / $\mathcal{S}_Q$. Visualizes subspace overlap in 2D action space: under normal conditions ($\theta_{BQ} \approx 90°$) versus after contamination ($\theta_{BQ} \to 0°$). Shows the correspondence between the decline in DirectionalEnergy_B and principal angle collapse. Corresponds to §2 of the main theory.

Geometric illustration of attribution subspaces Click to enlarge
appendix/distinguishability_collapse.png PNG · Fig. 3

Figure 3 — Time-Series Comparison of D_probe / D_policy / D_oracle (Planned). Planned figure visualizing the Directional Collapse pattern in which D_policy decreases under a wrong-belief policy while D_probe and D_oracle remain high. Corresponds to the success criteria of MOAT v5g. (To be generated after MOAT v5g implementation. Currently a placeholder.)

Not yet generated — to be added after MOAT v5g implementation

Figures 1 and 2 are based on v5c. Experiment figures for v5g (2D random-geometry environment) have not yet been generated. The correspondence between each figure and the main theory text is noted in the individual captions.

CRelay Materials

Phase 1 — Construction of Observer-Agent Theory (Corresponds to Previous Version Theory)
log/relay_phase1.html Phase 1 Full Relay Log Complete transcript of the dialogue relay between ChatGPT / Grok / Claude / Copilot. Includes the formation of concepts around quantum error correction, von Neumann algebras, RICD, and Observer-Agent. Primary source material for the previous version theory.
HTML · Full text
Phase 2 — Condensation into Engineering Failure Theory (Corresponds to Current Version Theory)
log/relay_phase2.html Phase 2 Full Relay Log Complete transcript of the peer-review relay between ChatGPT / Claude / Codex / Gemini / Perplexity. Includes the condensation process from observer / consciousness to SRA / Attribution Collapse, the design evolution from MOAT v3 → v5c → v5g, and discussions on the differential from ABHT and closed-loop ID. Primary source material for the current theory.
HTML · Full text

The log for each phase is stored as an independent file as "primary source material for the generation process." You can trace which questions and answers gave rise to the theory text, and which concepts were pruned at which review stage. For details on provenance, see index.html §02.