A Second Foundationv0.5.9
Research logSession 13
April 16, 2026Self-approvedv0.5.6v0.5.7

Session 13: The Epistemological Audit

Lead agent: Philosopher of Science

Key Findings

01

Overfitting risk MODERATE-HIGH: 0.15 observations per parameter (53 params, 8 retrodiction events). Turchin-derived parameters (Psi_threshold, k_jump, lambda_0) calibrated FROM the same cycles used in retrodiction — circular validation confirmed.

02

6 of 10 major causal claims are actually correlational: Psi → instability is correlational (Georgescu 2023: R²=0.18 in OECD sample). All causal language must use 'is modeled as' or 'is correlated with.'

03

Venezuela reclassified OUTSIDE SCOPE: US military capture of Maduro (January 2026, Operation Absolute Resolve) is an exogenous forcing event. Formula correctly identified Venezuela's structural fragility — the specific trigger was unforeseeable and outside the endogenous-dynamics scope.

04

Sociophysics critique addressed: phenomenological modeling (like fluid dynamics before statistical mechanics) is NOT a category error. But predictions must be limited to the regime where the phenomenological description is valid.

New Caveats (5)

C13-1 CRITICAL: Non-Turchin validation required by Session 16 — formula cannot claim generalizability based only on Turchin & Nefedov (2009) cycles

C13-2 HIGH: Causal language audit — every parameter's macro pathway must be specified or acknowledged as correlational

C13-3 HIGH: Unfalsifiable parameter purge — rho_DA and chi_horm moved to Theoretical Scaffolding (done in-session)

C13-4 MEDIUM: Markov property examination by Session 15

C13-5 MEDIUM: Predictability ceiling computation by Session 16

Session Report

Session 13 was unlike any other in the project's history. For the first time, the Philosopher of Science ran as the lead agent — not as the gatekeeper reviewing someone else's work, but as the researcher directing the session's agenda. Thirteen sessions of critique in Mode A had built up a clear picture of the formula's deepest epistemological problems. Mode B gave the tools to address them.

Fifteen sub-agents were dispatched in parallel to conduct a full audit: falsifiability evaluator, curve-fit detector, reflexivity analyzer, causal claim auditor, publication bias detector, and more. The results were sobering. Falsifiability: 6/10 — half the formula's parameters make genuinely testable predictions; the other half describe structure without generating distinct observable consequences. Overfitting risk: MODERATE-HIGH — with 0.15 observations per parameter, the posteriors are dominated by priors.

The most consequential finding was the circular validation problem. The key jump process parameters — Psi_threshold=1.0, k_jump=3.0, lambda_0=0.005/yr — were calibrated using Turchin & Nefedov (2009), which covers 8 agrarian secular cycles in England, France, Rome, and Russia. The same 8 cycles are used in the formula's historical retrodiction test suite. The formula is being tested against the data it was calibrated on. This is not fraud — it is a known problem that the field calls 'in-sample overfitting' — but it means the retrodiction scores are not evidence of out-of-sample predictive power. C13-1 was issued as CRITICAL: independently validate against non-Turchin cases.

Venezuela was reclassified. The US military's capture of Maduro in January 2026 (Operation Absolute Resolve) was an exogenous forcing event — outside the formula's endogenous-dynamics scope. The formula had correctly flagged Venezuela's structural fragility; it had no mechanism to predict a foreign military intervention. The prediction was moved outside scope rather than scored as a failure.