Evidence
How Thornveil proves it works
Every claim Thornveil makes about RigRun, Pyros, and Mycelium is backed by working code, measured benchmarks, and published research. Here's the evidence.
Research papers
Peer-reviewable papers documenting the algorithms behind Thornveil's products.
Selective-Buffer Streaming Safety for AI Coding Agents
cs.CR 8 pages J. Morgan, 2026
PDF
Submitted April 2026
Zeroth-Order Preference Optimization on 100B+ Quantized MoE Models via Live Inference API
cs.LG 7 pages J. Morgan, 2026
PDF
Submitted April 2026
TurboQuant-SGLang: Compressed KV Cache Attention as a Plugin Backend
cs.DC 6 pages J. Morgan, 2026
PDF
Submitted April 2026
Speculative Decoding Fails on Sparse MoE: A Negative Result and Practical Multi-Model Cascade Alternative
cs.SE 6 pages J. Morgan, 2026
PDF
Submitted April 2026
RigRun: Complete Local AI Infrastructure on a Single GPU
cs.AI 5 pages J. Morgan, 2026
PDF
Submitted April 2026
Protected IP you license, not rebuild
When you buy Thornveil, you get the benefit of patent-pending methods — not a wrapper around someone else's API. Full U.S. provisional applications, assigned to Thornveil LLC, available for license. All 27 provisionals must be converted to non-provisional within 12 months of filing.
Counter-UAS AI Safety AI Infrastructure AI Training Computer Vision Arabic NLP CBRN Mesh/P2P Agents Theory
001 Acoustic Direction-of-Arrival Estimation Using RPM-Indexed Self-Noise Cancellation on an Airborne Interceptor UAV
002 Classification-Gated Routing System for Language Model Inference with Downgrade Prevention
003 Learned Differentiable CLAHE Module for End-to-End Neural Network Image Enhancement
004 Anti-Bypass Spillage Detection for Secure AI Inference Pipelines
005 Hybrid Native-Synthetic Reasoning with Classification-Gated Thinking Depth Control
006 Multi-Agent Deep RL for Autonomous Counter-UAS Intercept Guidance
007 IFF Re-verification at Instant of Discharge with Pursue-Without-Engagement Fallback
008 Multi-Condition Continuous Engagement Authorization Revocation for Autonomous Interceptors
009 Heterogeneous ISA Platform for Safety-Critical Engagement Authorization
010 Multi-Signal Adaptive Confidence Scoring with Confabulation Detection
011 Deep RL for Fixed-Wing UAV Forward-Firing Gun Engagement of sUAS Targets
012 Hierarchical Triple-Key Error-Report Disambiguation in Arabic Diacritization
013 Contrastive Dual-Head Neural Corrector for Arabic Diacritization
014 Communication-Optimized Blade Geometry for Acoustic Data Transmission
015 Spatial Grid-Accelerated Contamination Zone Fusion for CBRN Sensor Networks
016 ISA 2.0 Canonical Data Model with CAIRN-Extended CoT for CBRN Networks
017 Tactical Mobile C2 for Armed Counter-UAS Interceptors via MAVLink
018 Selective-Buffer Streaming Safety Proxy for LLM Tool Calls
019 Closed-Loop Self-Improvement for Local Language Models
020 Multi-Stage Cascade Routing with Entropy-Based Escalation
021 Classification-Isolated Memory with Hardware-Backed Encryption
022 Distributed Adaptive AI Mesh with Predictive Expert Prefetching and PoI
023 Zero-Downtime Self-Improvement via Zeroth-Order Optimization
024 HawkStack Cross-Domain Tiny Neural Detection Framework
025 Agent Factory: Autonomous Meta-Agent Manufacturing with Closed-Loop DPO
026 Loss Surface Topology Model for Neural Architecture Design
027 PID Homeostasis Controller for Language Model Inference Quality
27
Patent-pending methods
10
Patent families
5
Research papers