HangJang/c4-cognitive-adapters

C4 Cognitive Adapters

Production C4 classifiers for cognitive state detection from text.

This repository contains adapters for the C4 (Complete Cognitive Coordinate System) framework - a formal model of cognition as navigation through a 27-state space (Z₃³).

Model Performance

Production BERT model: 93.5% average accuracy

• Time (T) axis: 94.3% accuracy
• Scale/Dimension (D) axis: 93.9% accuracy
• Agency (A) axis: 92.5% accuracy

Trained on 11K consensus-labeled samples with 4-model agreement (67.4% consensus rate).

Architecture

• Base model: microsoft/mdeberta-v3-base (768 hidden size)
• Adapters: 12 LoRA adapters
• Classification heads: 65 heads across cognitive dimensions
• Core adapter: T/D/A axes classification (9 classes)

C4 Framework

C4 models cognition through three orthogonal axes:

• Time: Past / Present / Future
• Scale: Concrete / Abstract / Meta
• Agency: Self / Other / System

Resources

• 📚 GitHub Repository: https://github.com/cognitive-functors/adaptive-topology
• 🌐 Interactive Demo: https://c4cognitive.com
• 📖 Documentation: See papers/UNIFIED-GEOMETRIC-COGNITION-en.md in the repo
• 🔬 Research: ~5 years of development, 24+ papers, 11 theorems (10 verified in Agda)

Usage

from transformers import AutoModelForSequenceClassification, AutoTokenizer

# Load model and tokenizer
model = AutoModelForSequenceClassification.from_pretrained(
    "HangJang/c4-cognitive-adapters",
    subfolder="pytorch"
)
tokenizer = AutoTokenizer.from_pretrained("microsoft/mdeberta-v3-base")

# Classify text
text = "I remember planning for the future"
inputs = tokenizer(text, return_tensors="pt", max_length=256, truncation=True)
outputs = model(**inputs)
# Returns: Time=Past, Scale=Meta, Agency=Self (coordinates 0,2,0)

Installation

pip install transformers torch

Model Files

• pytorch/ - PyTorch adapter weights (.pt files)
• onnx/ - ONNX format for production deployment
• metadata.json - Model metadata and adapter specifications

Limitations

• Input text limited to 256 tokens
• Optimized for English and Russian texts
• Requires proper context for accurate classification

License

Triple license: Apache-2.0-NC / AGPL-3.0 / Commercial See: https://github.com/cognitive-functors/adaptive-topology/blob/main/LICENSE

Citation

@software{c4_cognitive_adapters,
  title = {C4 Complete Cognitive Coordinate System},
  author = {Selyutin, Ilya and Kovalev, Nikolai},
  year = {2026},
  url = {https://github.com/cognitive-functors/adaptive-topology}
}

Contact

For questions and collaboration:

• GitHub Issues: https://github.com/cognitive-functors/adaptive-topology/issues
• Email: See CITATION.cff in the repository