Capital Manifold: Planetary Balance-Sheet Network Dynamics, Siphon Forensics & Macro-Financial Graph Engine

1. The Planetary Epistemic Crisis

Modern macroeconomics models money as an aggregate scalar quantity rather than a directed topological manifold. Beneath official national accounts, capital moves through a labyrinth of multi-tier institutional balance sheets, offshore currency swaps, shadow banking conduits, and dark maritime logistics arbitrage:

• $575 Trillion in OTC derivatives commitments remains off-balance-sheet.
• $67 Trillion in shadow banking operates outside direct central bank oversight.
• $180 Billion/year leaks into the IMF's global Net Errors & Omissions (NEO) statistical discrepancy.

Capital Manifold models planetary capital flows as a multi-tier, directed, double-entry financial network built on the Procedural Graph Engine (pgm-engine) and DuckDB. It treats balance sheets as state variables ($S$), cash and credit flows as time derivatives ($dS/dt$), and institutional conduits as nodes constrained by Kirchhoff flow conservation.

2. Quantitative Signal Generation & Numerai Applications

When combined with Thalamocortical Systems' continuous-time dynamical models, Capital Manifold provides a clean, invariant-bounded signal generation engine for quantitative tournaments (such as Numerai):

1. Continuous Phase-Space Drift (CfC): Real market returns do not pause between discrete token steps. Continuous-time neural ODEs trace price action as smooth geometric drift across non-stationary regimes.
2. Siamese Volatility Gating (DR-S-TWE): Sudden liquidity drawdowns, central bank policy shocks, and earnings gaps trigger surprise alarms ($g_t \to 0$), immediately routing predictions through discrete shock-absorbing representations without crashing continuous gradients.
3. Orthogonal Dictionary Neutralization: By applying PhaseSAE parameter partitioning, feature representations are decomposed into orthogonal phase slices. In the Numerai tournament, this directly neutralizes common risk-factor exposures, maximizing True Contribution (TC) and Feature Neutral Correlation (FNC).
4. Double-Entry Kirchhoff Invariance: Unlike unconstrained regression models that invent capital out of thin air, every modeled transaction enforces $\sum \dot{M}_{\text{in}} = \sum \dot{M}_{\text{out}}$.

3. Four-Tier Monetary Architecture