EDGE — Domain Evaluation Lens Generator

Built on the Fresta Lens Framework — a unified structural theory of evaluation, entropy, and system coherence.

EDGE generates structured, domain-specific evaluation criteria using a three-order structural analysis engine. Given any domain (e.g. gaming laptop, noise-cancelling headphones, electric car), it produces a lens — a JSON file containing weighted metrics, interdependency maps, systemic stress factors, and buyer profiles.

The underlying formula:

E_total = E0 + E_upstream + E_inherited + P_structure

Component	Description
E0	1st order — Local, intrinsic, measurable metrics
E_upstream	2nd order — Bottlenecks & synergies between metrics
E_inherited	3rd order — Systemic stress (infrastructure, saturation, cycles)
P_structure	Structural penalty derived from the above

---

How It Works

Domain input
    │
    ▼
Knowledge gathering (Wikipedia + web scraping)
    │
    ▼
1st Order — local metrics (E0)
    │
    ▼
2nd Order — bottlenecks & synergies (E_upstream)
    │
    ▼
3rd Order — systemic stress (E_inherited + P_structure)
    │
    ▼
Profile generation (buyer types inferred from systemic factors)
    │
    ▼
Human summary
    │
    ▼
Lens JSON  →  ./lenses/<domain>_lens.json

---

Requirements

Python 3.9+

Required

pip install requests beautifulsoup4

Optional (for knowledge gathering & LLM backends)

pip install wikipedia-api googlesearch-python ollama

LLM Backend

EDGE works with any OpenAI-compatible local LLM server, such as:

• LM Studio
• llama.cpp server
• Ollama (automatic fallback)

---

Usage

python edge.py

You will be prompted to enter a domain. EDGE then runs the full pipeline and saves the lens to ./lenses/<domain>_lens.json.

Environment Variables

Variable	Default	Description
EDGE_LLM_URL	http://127.0.0.1:1234	Base URL of the LLM server
EDGE_LLM_MODEL	meta-llama-3-8b-instruct	Model name to use

Example:

EDGE_LLM_URL=http://127.0.0.1:11434 EDGE_LLM_MODEL=llama3 python edge.py

---

Output Format

{
  "domain": "gaming laptop",
  "generated_at": "2025-03-01T12:00:00Z",
  "formula": "E_total = E0 + E_upstream + E_inherited + P_structure",
  "block_weights": { "E0": 0.40, "E_upstream": 0.35, "E_inherited": 0.25 },
  "global_admission_threshold": 0.85,
  "metrics_1st_order": [ ... ],
  "relations_2nd_order": [ ... ],
  "stress_3rd_order": [ ... ],
  "profiles": [ ... ],
  "summary": "..."
}

Metrics (1st order)

{
  "name": "thermal_management",
  "order": 1,
  "direction": "maximize",
  "weight": 0.18,
  "threshold": 0.6,
  "justification": "Sustained performance depends on thermal headroom."
}

Relations (2nd order)

{
  "type": "bottleneck",
  "metrics": ["gpu_performance", "display_refresh_rate"],
  "intensity": 0.75,
  "penalty_or_bonus": 0.6,
  "explanation": "High refresh rate is wasted if the GPU cannot push frames."
}

Systemic Stress (3rd order)

{
  "type": "cycle",
  "name": "gpu_hype_cycles",
  "inherited_stress": 0.45,
  "penalty": 0.30,
  "explanation": "Marketing around GPU generations distorts perceived value."
}

Profiles

{
  "id": "competitive_esports",
  "label": "Competitive Esports Player",
  "description": "Prioritizes frame rate and input latency above all else.",
  "weight_adjustments": {
    "display_refresh_rate": 1.8,
    "input_latency": 1.7,
    "battery_life": 0.5
  }
}

---

Graceful Degradation

EDGE is designed to work even when dependencies are missing:

• No wikipedia → skips Wikipedia lookup
• No googlesearch / beautifulsoup4 → skips web scraping
• No LLM server reachable → uses built-in default fallback values for all three orders

---

Theoretical Background

EDGE is the first practical implementation of the Fresta Lens Framework.

📄 Read about the framework →
📚 Full 5-volume work on Zenodo →

---

Support

This project is developed independently, without institutional funding.
If it's useful to you, consider supporting its continued development.

☕ Support on Ko-fi →
⭐ Or simply star the repo — it helps more than you'd think.

→ See all ways to help

---

License

EDGE is released under the MIT License.
The underlying Fresta Framework uses a dual license (AGPLv3 + Enterprise). See FRESTA.md for details.