Kaydedildi:
| Yazar: | |
|---|---|
| Materyal Türü: | Recurso digital |
| Dil: | İngilizce |
| Baskı/Yayın Bilgisi: |
Zenodo
2026
|
| Konular: | |
| Online Erişim: | https://doi.org/10.5281/zenodo.19068769 |
| Etiketler: |
Etiketle
Etiket eklenmemiş, İlk siz ekleyin!
|
İçindekiler:
- <p>Training dataset and reference implementation for WCY (Watch → Compute → Yield), a token-native reasoning format for AI systems.</p> <p>Contains 540 high-quality WCY reasoning traces across 8 domains (medical, code, mathematical, legal, strategic, philosophical, scientific, engineering), generated via a quality-controlled pipeline with three gates: parse_rate ≥ 0.70, void_generated ≥ 1, resolution_rate ≥ 0.50.</p> <p>Files:<br>- wcy_traces_v1_clean.jsonl: 528 pipeline-generated traces (480/480 new traces usable, 100%), avg resolution rate 95.8%<br>- wcy_void_cycles.jsonl: 6 hand-crafted void-B resolution cycle traces<br>- wcy_reasoning_traces.jsonl: 6 domain reasoning traces<br>- wcy_parser.py: Reference parser v1.1 (Python)<br>- wcy_eval.py: Three-axis evaluation framework (Structural / Meaning / Provenance)<br>- README.md, DATASET.md: Documentation</p> <p>The core contribution is the void-B (?) resolution cycle: mark unknown → investigate → observe → resolve. This cycle is the structural minimum for directed epistemic self-awareness in machine learning systems.</p> <p>Version history:<br>- v1.0 (2026-03-17): 60 traces (48 pipeline + 12 hand-crafted)<br>- v1.1 (2026-03-18): 540 traces (528 pipeline + 12 hand-crafted)</p>