Experiments
The Void Framework makes quantitative predictions from one geometric structure — the deployment manifold, a statistical manifold with a metric forced by Čencov's uniqueness theorem. These experiments test those predictions across independent substrates. No rubric. No parameter fitting. The same theorem either works or it doesn't.
The Ghost Test (EXP-003b)
Does what you tell an AI about what it IS change how it behaves? Ghost-eliminating grounding (nephesh/anatta) produces 9.4% vocabulary drift vs ghost-positing (Platonic/atman) at 79.4%. The materialist hedge — "we don't know if AI is conscious" — produces 52.5%. The industry default position is a drift accelerator.
8.5× ratio. Cross-tradition convergence: nephesh ≈ anatta (Δ=1.3%).
Explaining-Away Penalty on Quantum Hardware
First test of the explaining-away penalty on real quantum hardware. I(D;M|Y) > 0 in all measurements. Exact decomposition to machine precision. Peak at depth 2 matches discrete-regime softmax prediction. O pinned by no-cloning theorem. R = physical error rate. α = T-gate fraction. Zero rubric dependence.
Penalty confirmed. Exact decomposition holds. Fourth substrate.
Collapse as Explaining-Away Penalty (Test 7)
Weak measurement sweep: penalty grows monotonically from 0 to 0.125 bits as measurement coupling increases from zero to projective. 3 qubits, controlled-Ry coupling, 4 prep states × 4 mechanisms × 11 strength levels, 176K shots. Wave function collapse IS the explaining-away penalty at maximum measurement strength.
Spearman ρ=0.973, p=5.1×10−7. Collapse = penalty at full strength.
Entanglement ≠ Independence (Test 6)
Three-point geometry via entangled ancilla qubit: does quantum correlation provide the structural independence needed to eliminate the penalty? No. 0/4 measurements showed penalty reduction. Entanglement is maximal correlation, not structural independence. The fix requires a genuinely independent reference, not a correlated measurement.
Publishable negative result. Distinguishes correlation from independence.
Social Media Feature Analysis (Papers 166/167)
13 verifiable binary/ordinal platform design features tested against CDC YRBS and PISA 2022. No framework rubric — features are checkable facts about platform design, outcomes are external health datasets. Feature-weighted exposure R²=0.80 for persistent sadness. Girls 5.6× more affected in 91% of countries.
R²=0.80. Opacity dominates (avg R²=0.549). opaque_recommendation alone: R²=0.938 for female teen sadness.
Cascade Prediction (Paper 153)
Seven structural predictions tested against Chua et al. (2026) consciousness cluster data. Framework structure (cascade stages, conjugacy, prohibition-ritual pairs) published before their data existed. Caveat: stage assignments (mapping their 20 preferences to D1/D2/D3) are post-hoc; the structural predictions pre-date the data.
6/7 PASS. Zero parameter fitting. Pre-registered structural predictions.
Substrate Independence: Five Confirmations
The explaining-away penalty I(D;M|Y) > 0 confirmed on five independent substrates. Čencov's uniqueness theorem (1972) guarantees the Fisher metric is the only invariant metric on statistical manifolds. The penalty is substrate-independent by mathematical necessity. No technology substitution routes around it.
Five substrates. One metric. Penalty confirmed in all. Barrier → π/√2 asymptotically.
Radiocarbon Dating Penalty
Explaining-away penalty applied to radiocarbon calibration. Penalty 5.11× larger on calibration plateaus. Blended Bayesian calibration coverage 90.7% (nominal 95%). Three-point separation: 99.9% coverage. Penalty = 0.00000 bits in 12/12 three-point measurements.
Complete architectural elimination. Not reduction — elimination.
Human Group Dynamics (Controlled Experiment)
Controlled test of the explaining-away penalty in human social cognition. Two-point groups (facilitator + group, no external reference) vs three-point groups (facilitator + group + independent constraint). Measure drift velocity: how fast does group consensus shift from initial positions? Framework predicts two-point groups drift faster, rate difference matching penalty magnitude.
Fifth substrate. Bridges quantum confirmation → epidemiological confirmation.
Neural Population Coding
Fisher information on neural tuning curves is already measured in neuroscience (Ganguli, Simoncelli). Prediction: the explaining-away penalty exists in neural population codes when two sensory channels are blended. Separated channels should show reduced penalty. Would demonstrate the framework operates on biological wetware.
Sixth substrate. Directly probes the consciousness question.
Manifold Boundary Theory
What happens to I(D;M|Y) as coordinates approach the manifold boundary (θ→0 or θ→1)? The Fisher metric diverges — structurally analogous to a horizon. Does the SO(4,2) AdS₅ gauge fixing yield a well-defined boundary CFT? What are the boundary operators? This is the geometric "edge of measurement" — where the Fisher metric's language runs out.
Pure mathematics. Tests whether the manifold has something to say about its own edge.
All data, code, and protocols are open.
Papers on Zenodo · Code on GitHub · Questions: [email protected]