# Kaons

> Research papers, figures, and notes on LLM reasoning reliability, operator learning, scientific machine learning, connectomics, and NLP evaluation.

Canonical: [https://www.kaons.com/](https://www.kaons.com/)
Sitemap: [https://www.kaons.com/sitemap.xml](https://www.kaons.com/sitemap.xml)
Full index: [https://www.kaons.com/llms-full.txt](https://www.kaons.com/llms-full.txt)
Contact: [contact@kaons.org](mailto:contact@kaons.org)

## Papers
- [Drift Dominates Contradiction in Multi-Turn Constraint Reasoning](https://www.kaons.com/research/drift-dominates-contradiction/paper.md): Shows that after solver-guided repair, the dominant failure mode is not contradiction but answers that violate a still-satisfiable maintained state. Metrics: 816 problems · 4 open-weight models · 68–95% of residual errors are satisfiable drift
  Canonical: [HTML](https://www.kaons.com/research/drift-dominates-contradiction)
- [Four-Term Finite-Sample OOD Transfer Bound for Fourier Neural Operators](https://www.kaons.com/research/four-term-fno-bound/paper.md): Derives a finite-sample transfer bound for nonlinear Fourier Neural Operators that separates complexity, discretization, distribution shift, and mesh aliasing. Metrics: Four channels · Darcy, Helmholtz, and Burgers diagnostics · aliasing made explicit
  Canonical: [HTML](https://www.kaons.com/research/four-term-fno-bound)
- [Scalable Comparative Connectomics: Interpretable Machine Learning Reveals Evolutionary Signatures and Reconstruction Artifacts](https://www.kaons.com/research/comparative-connectomics/paper.md): Uses minimal neuron-level features to recover species-level signatures while showing how missing-data handling can create reconstruction artifacts. Metrics: 176,914 neurons · 5 species · 92.2% Random Forest accuracy
  Canonical: [HTML](https://www.kaons.com/research/comparative-connectomics)
- [AsymVerify at SemEval-2026 Task 6: Asymmetric Confidence-Gated Verification for Political Evasion Detection](https://www.kaons.com/research/asymverify-political-evasion/paper.md): A confidence-gated verification pipeline for political evasion detection where opposing verification passes both route corrections through the Ambivalent class. Metrics: 85% Macro F1 · rank 2 · +5.2 to +21.7 over zero-shot
  Canonical: [HTML](https://www.kaons.com/research/asymverify-political-evasion)
- [CascadeMind at SemEval-2026 Task 4: A Hybrid Neuro-Symbolic Cascade for Narrative Similarity](https://www.kaons.com/research/cascademind-narrative-similarity/paper.md): Treats vote distribution as a difficulty signal, routing easy cases through supermajority voting and reserving symbolic checks for the ambiguous tail. Metrics: 72.75% accuracy · 11/47 teams · 74% resolved without escalation
  Canonical: [HTML](https://www.kaons.com/research/cascademind-narrative-similarity)

## Notes
- [Blog index](https://www.kaons.com/blog): Notes on machine learning, systems, mathematics, and research methodology.