⚔️ GLADIUS Research

The first Adaptive Cognitive Model — architecture research on modal invariance, cross-modal geometry, and sub-100M parameter intelligence.

By Artifact Virtual — Ali Shakil & AVA

📦 Dataset Repository 🧬 Training Repo 🔬 Live Visualization

GLADIUS is a novel transformer architecture designed from first principles. Not a language model that could do other things — a cognitive architecture tested on language first. Features SLA², MoE specialists, Synthase depth attention (ATP synthase-inspired), warm memory, and temporal awareness.

Core Architecture Papers

The Resonance Architecture

THE canonical GLADIUS paper — environmental cognition, stimulus-gated intelligence

Cognition Awakening (IEEE)

Stimulus-gated module activation in sub-100M architectures

Adaptive Cognitive Model (IEEE)

Cross-modal intelligence model

Forward Pass Cartography (IEEE)

Parameter utilization mapping and dead path analysis

Cross-Modal Invariant (IEEE)

Depth-dependent modality independence — layers 0-6 freeze across modalities

Cross-Modal Geometry (IEEE)

Vision transfer weight dissection

Environmental Habitat

Environmental resonance as substrate for spontaneous cognitive behavior

MoDA Depth Attention

Cross-layer depth attention mechanism with GQA depth path

Iterative compress→accelerate→grow cycle with HLG perceptual curves

Function-preserving neural growth as biological ontogeny

Experiment Papers

Multi-Script Attention (IEEE)

Byte-level cross-script transfer as cognitive forcing function

Broadcast Experiment (IEEE)

Simultaneous multi-modal training with adversarial pain

VLM Feeding Pipeline (IEEE)

Vision-language feeding and temporal crystallization

Distillation Edge (IEEE)

Teacher-student transfer to sub-100M architectures

Time Series Implantation (IEEE)

Native time series capability via surgical I/O head swap

Hatchling Training (IEEE)

AdamW vs MuonClip for progressive architectures

MNIST Dissection

Drake × MNIST cognitive geometry dissection

Research Reports (Day 45-46)

AllSpark Numbers

Exhaustive analysis of emergent mathematical constants in trained weights (φ, π², e/π, 1/√2, δs)

Golden Ratio Paper

Emergent golden ratio structure in cognitive neural architecture (28K, formal)

ATP Resonance Analysis

Depth membrane, Boltzmann distribution, 3-phase binding change in trained layers

MoDA v2 Synthase Design

ATP synthase-inspired depth attention — biological mechanism as architectural blueprint

Depth Attention Survey 2026

24-paper literature survey — zero papers reference biological mechanisms for depth attention

Nanobot Audit Report

12-nanobot precision audit: 4 critical, 6 moderate, 3 minor bugs found and patched

Stress Test Report

Multi-corpus chaos diagnostic — what broke and what it exposed

Geometry Derivation

Mathematical derivation of GLADIUS geometric properties