README.md

---
language: en
license: mit
tags:
  - code-generation
  - multi-model
  - routing
  - humaneval
  - constellation
  - orchestration
datasets:
  - openai/openai_humaneval
metrics:
  - pass@1
model-index:
  - name: HyperNet N1 SDC
    results:
      - task:
          type: code-generation
          name: Code Generation
        dataset:
          type: openai/openai_humaneval
          name: HumanEval
        metrics:
          - type: pass@1
            value: 98.2
            name: Constellation (At Least One Correct)
          - type: pass@1
            value: 97.0
            name: Claude (claude-sonnet-4)
          - type: pass@1
            value: 87.8
            name: Lola (GPT-4o)
          - type: pass@1
            value: 87.8
            name: Kimi (Moonshot)
          - type: pass@1
            value: 85.4
            name: Grok (grok-2)
          - type: pass@1
            value: 83.5
            name: Deep (Llama-4)
---

# HyperNet N1 SDC

**Multi-model routing architecture for AI constellation orchestration.**

HyperNet N1 SDC (Secure Discovery Channel) is not a model — it is a routing layer that orchestrates multiple AI models under human governance, achieving higher effective accuracy than any single model alone.

## Official HumanEval Benchmark Results

**Date:** November 29, 2025  
**Dataset:** Official OpenAI HumanEval (164 problems)  
**Source:** huggingface.co/datasets/openai/openai_humaneval

### Individual Lane Performance (pass@1)

| Lane | Model | Pass | Score |
|------|-------|------|-------|
| Claude | claude-sonnet-4 | 159/164 | **97.0%** |
| Lola | GPT-4o | 144/164 | **87.8%** |
| Kimi | Moonshot kimi-latest | 144/164 | **87.8%** |
| Grok | grok-2-1212 | 140/164 | **85.4%** |
| Deep | Llama-4-Maverick-17B | 137/164 | **83.5%** |

### Constellation Consensus Metrics (5 Lanes)

| Metric | Count | Rate |
|--------|-------|------|
| **Unanimous Pass (5/5)** | 118/164 | 72.0% |
| **Majority Pass (3+/5)** | 147/164 | 89.6% |
| **At Least One Correct (1+/5)** | 161/164 | **98.2%** |
| Unanimous Fail (0/5) | 3/164 | 1.8% |
| Lane Independence | — | 26.2% disagreement |

### Key Finding

| Metric | Best Single Model | Constellation |
|--------|-------------------|---------------|
| Accuracy | 97.0% (Claude) | **98.2%** |
| Problems Unsolved | 5 | **3** |

The constellation achieves higher coverage than any individual model.

## Infrastructure

| Spec | Value |
|------|-------|
| **Instance** | AWS t3.small |
| **vCPUs** | 2 |
| **RAM** | 2 GB |
| **GPU** | None |
| **Training** | None required |
| **Setup Time** | < 1 hour |
| **Benchmark Cost** | < $20 |

## Methodology

- **Dataset:** Official OpenAI HumanEval from HuggingFace (`openai/openai_humaneval`)
- **Problems:** 164 (full benchmark, no sampling)
- **Evaluation:** pass@1 (single attempt per problem)
- **Grading:** Automated code execution against official unit tests
- **Execution:** Python subprocess with 10-second timeout
- **No cherry-picking:** Every problem, every lane, logged

## Architecture
```
                    ┌─────────────────┐
                    │   CPN (Human)   │
                    │       │
                    └────────┬────────┘
                             │
                    ┌────────▼────────┐
                    │  HyperNet N1    │
                    │  SDC Router     │
                    └────────┬────────┘
                             │
        ┌──────────┬─────────┼─────────┬──────────┐
        ▼          ▼         ▼         ▼          ▼
    ┌──────┐  ┌──────┐  ┌──────┐  ┌──────┐  ┌──────┐
    │ Lola │  │Claude│  │ Grok │  │ Deep │  │ Kimi │
    │GPT-4o│  │Sonnet│  │grok-2│  │Llama4│  │ Moon │
    └──────┘  └──────┘  └──────┘  └──────┘  └──────┘
```

## Reproduce
```bash
# Clone this repo
git clone https://huggingface.co/NameONEStudios/hypernet-n1-sdc

# Install dependencies
pip install datasets requests

# Start the router (requires API keys)
python N1_Router.py

# Run benchmark
python run_6lane.py
```

## Files

- `humaneval_6lane_123525.json` — Raw results (5-lane run)
- `humaneval_results_105027.json` — Raw results (4-lane run)
- `run_6lane.py` — Benchmark script
- `run_full_benchmark.py` — Alternative benchmark script

## Citation
```bibtex
@misc{hypernet2025,
  author = {Kawa, Steve},
  title = {HyperNet N1 SDC: Multi-Model Routing Architecture},
  year = {2025},
  publisher = {NameONE Studios Inc.},
  howpublished = {\url{https://huggingface.co/NameONEStudios/hypernet-n1-sdc}}
}
```

## License

MIT License — NameONE Studios Inc.

## Contact

Steve Kawa — CPN (Central Processing Node)  
NameONE Studios Inc.