pCwOrM/werr_open_decisions

WERR Open Decisions Benchmark Dataset

Official benchmark dataset accompanying the research paper: "Universal Fractal Natural Language Decision Map: Real-Time Edge Triage Across Heterogeneous Domains" arXiv:2609.25498

📌 Dataset Summary

werr_open_decisions contains 1,205 verified decision trajectories and over 3,200 evaluation questions evaluated using the WERR (Waves & Errors) machine-native zero-VRAM reflex engine and the production answerr.me platform.

Unlike conventional neural networks storing millions of static tensor parameters, each decision is procedurally synthesized from a 24-byte Mandelbrot coordinate seed $\Theta = (cx, cy, \text{zoom})$ evaluating non-linear escape dynamics along the chaotic boundary of the Mandelbrot set $\mathbb{M}$.

Decision Primitives:

1. 1.`noul` (Boolean / Binary): Instant binary triage (e.g. permit / reject, anomaly detection, panic trip).
2. 2.`choice` (Categorical): Multi-class routing across discrete destination microservices.
3. 3.`score` (Ordinal / Continuous): Multi-tier risk severity scoring (0–3 scale).

🚀 Usage

from datasets import load_dataset

# Load directly from Hugging Face Hub
dataset = load_dataset("pCwOrM/werr_open_decisions")

# Inspect the first decision
print(dataset["train"][0])

📊 Dataset Structure

Each JSONL record represents an operational decision event:

• —id: Monotonic event identifier.
• —seed: 24-byte coordinate seed (cx, cy, zoom) on the complex plane $\mathbb{C}$.
• —state_summary: Operational state features (user role, error rates, frequency, telemetry).
• —questions: Array of strongly-typed decision probes (noul, choice, score) with computed confidence and probability distributions.
• —latency_ms: Measured end-to-end evaluation latency on CPU hardware (sub-2ms reflex execution).

🏛️ Citation

@article{dagli2026universal,
  title={Universal Fractal Natural Language Decision Map: Real-Time Edge Triage Across Heterogeneous Domains},
  author={Da{\u{g}}l{\i}, Volkan and Da{\u{g}}l{\i}, Zerrin and Da{\u{g}}l{\i}, Da{\u{g}}han},
  journal={arXiv preprint arXiv:2609.25498},
  year={2026},
  url={https://arxiv.org/abs/2609.25498}
}

📜 License

Licensed under the Business Source License 1.1 (BSL 1.1). Converts to Apache 2.0 on 2030-01-01. Free for academic, educational, and research benchmarking. Corporate Licensor: ITouch Systems (ITouch Bilişim Sistemleri Ltd. Şti., Çukurova Teknokent). Contact: ask@answerr.me / info@itouch.com.tr.