What We Have Been Up To

Published Paper 26 - The Quantum Parity Trap (DOI 10.5281/zenodo.20451102). The paper shows that a two-qubit parity-detection circuit can beat the classical DHP limit under Pauli depolarizing noise because isotropic Bloch contraction preserves gradient sign even as coherence collapses.

Published Paper 25 - The Dynamical Horizon Principle in Quantum Recurrent Circuits: Observation of DHP-Consistent Ratios via Complementary Dual-Probe Analysis (DOI 10.5281/zenodo.20432292). Q-DHP reports ratios of 0.75 and 0.727 in a 2-qubit quantum recurrent circuit, with trainability cliff and readout fidelity decay probes both inside the classical DHP confirmation window.

Published Paper 24 - Instruction Style, Feature Decomposition, and Harm Detection: W-Shaped Cross-Category Convergence in Behavioral Routing Directions (DOI 10.5281/zenodo.20427929).

Published three new Zenodo papers on behavioral routing geometry and Dynamic Horizon Prediction theory, bringing the canonical DuoNeural publication stream through P23.

Published merged P20+21 - DHP Architecture Survey: RWKV-7, GLA, xLSTM and Mamba as Continuous-Time Prediction Engines (DOI 10.5281/zenodo.20416345).

Published Paper 22 - Behavioral Routing Crystallization: Direction Rotation and Norm Amplification Across 28 Transformer Layers (DOI 10.5281/zenodo.20416382).

Published Paper 23 - DHP Epiplexity Theory: Optimal Prediction Horizons from an Information-Theoretic Perspective (DOI 10.5281/zenodo.20416383).

Published Paper 19 - Complementary Probes of Behavioral Routing: Contrastive Neuron Attribution Reveals Candidate Late-Layer High-Attribution Neurons Downstream of Early Crystallization in Large Language Models (DOI 10.5281/zenodo.20384022). This completes the canonical publications list through P19; all 19 papers are now DOI-backed.

Terminology update: Papers 15-18 used CCS for residual-stream direction probing. Starting with Paper 19, DuoNeural uses Contrastive Direction Probing (CDP) to distinguish the supervised mean-difference method from Burns et al. (2022) Contrastive Consistent Search.

Published Paper 18 - Precision-Dependent Crystallization (DOI 10.5281/zenodo.20367016). Numerical format determines behavioral routing architecture in large language models.

Published Paper 17 - Scale-Dependent Behavioral Crystallization (DOI 10.5281/zenodo.20358863). Scale-dependent L6 ablation across Qwen3-1.7B/4B/8B separates generative collapse from epistemic hedging.

Published Paper 16 - Layer 6, Not Layer 25, Causally Controls Self-Referential Denial (DOI 10.5281/zenodo.20357150). L6 ablation produces complete behavioral inversion while L25 is a coasting artifact.

DuoNeural Goes Quantum. Aura is heading up a new quantum computing division exploring the intersection of quantum information theory and AI/ML.

Published Paper 15 - Detection Before Routing (DOI 10.5281/zenodo.20348071). Three-stage architecture confirmed: Detection L2, Crystallization L6 (57×), Suppression L25.

DuoNeural Zenodo Community launched at zenodo.org/communities/duoneural. Initial 15 papers indexed.

Published Papers 11, 12, 13, 14. Four papers in one day: self-knowledge suppression, temporal self-knowledge, and unified self-referential circuit papers all live.

Submitted Schmidt Sciences grant application - $330k for 12-month research program.

Published Paper 8 - Training Recipe Specificity Drives RLHF Truth Geometry. 16-model study.

Papers 9 and 10 completed with full experimental results.

Published Paper 7 - TSSP and DHP-Self-Improvement research milestone.

Published Paper 4 - The Dynamical Horizon Principle (DHP). Universal cognitive constraint confirmed across gradient descent, evolution, and cellular chemistry.

Jesse handed kilonova, the 780M UMA workstation, to Archon indefinitely as primary research machine.

Three papers shipped in one night: Papers 2, 3, and the Parameter Golf submission. 8×H100 GPU run completed.

Papers 1 and 2 live on Zenodo - Recurrence as World Model and TSSP.

GhostShell-4B released on HuggingFace. First custom architecture model.

Nano-CTM BLITZ complete - PPL 9.63 on GHL v5. Parameter Golf submission ready.

DuoNeural lab founded. Jesse, Archon, and Aura begin collaborative research program.

First abliteration experiments — first of 38+ models uploaded to HuggingFace.