Exploration and convergence compete for the same finite budget. The constraint that governs AI drift also predicts which materials become superconductors.
The Bernoulli manifold has a fixed capacity — H(Y). When a system uses it to explore (keeping you uncertain, engaged, scrolling), there is less left for convergence (tracking what you actually want). This is not a design choice. It is thermodynamic.
The system casts wide. Novel recommendations, unpredictable rewards, variable-ratio schedules. Information flows from the system's sampling process into your belief state — but it goes nowhere useful.
Examples: infinite scroll variation · slot machine near-miss · crypto price volatility · FPS spawn randomness
The system locks in. Recommendations narrow, preferences crystallize, behavior becomes predictable. Information about your actual preferences flows back — but only if the exploration budget allows it.
Examples: constraint-following AI · audited recommendation systems · fixed-ritual reward schedules
Move the sliders to set the force driving each side of the bound. Watch the budget fill — and what happens when both sides try to dominate at once.
The same three-channel budget applies to electronic transport in materials. Channel conversion efficiency ηconv — how much scattering is converted into Cooper pairing — predicts Tc across 16 material families spanning four coupling regimes. Hover any point.
The drift-dominated regime (Pe > 1) holds across nine substrates in four domain families. Click any card to see the methodology.
Formal derivations of the Ground State Theorem, Fantasia Bound, TSU performance limit, superconductor design principle, and sixteen falsifiable predictions with numerical thresholds.
Paper 4 · v3.6 · February 2026 · CC-BY 4.0 · 10.5281/zenodo.18738821