---
license: mit
language:
- es
- en
- pt
- ro
- fr
tags:
- structural-engineering
- construction
- eurocode
- AISC
- BIM
- fine-tuning
- ORPHEON
task_categories:
- text-generation
size_categories:
- 100K-1M
---

# 🏗️ ORPHEON Structural Dataset v4

## 264,494 ejemplos | 17 tools | 48 categorías

### Correcciones en v4
- ✅ **132 respuestas adversariales únicas** (antes 2)
- ✅ **7 MCP tools añadidas** (search_web, get_normative, run_agent, validate_project, get_project_summary, get_project_quantities, calculate_duration_days)
- ✅ **17 herramientas totales** (antes 11)
- ✅ **100% language coherence** ES/EN/PT/RO
- ✅ **Timber realistico**: 57% passed, 15% failed
- ✅ **0 categorías artificiales** de 2000 exactos
- ✅ **query_bim_element reemplazado** por lookup_section

### Tools disponibles
search_web, get_normative, run_agent, validate_project, get_project_summary, get_project_quantities, calculate_duration_days, lookup_section, verify_bending, design_rc_section, verify_timber_bending, verify_bolted_connection, get_seismic_params, verify_shear, classify_section, check_deflection, calculate_footing

### Validación
- 100% tool_calls verificados contra código real
- Quality Score ≥ 0.7
- Cross-validation 3+ fuentes
- Anti-overfitting: 1000+ templates únicos

### Files
| File | Records |
|------|---------|
| data/final_dataset_combined.jsonl | 264,494 |
| data/final_dataset_train.jsonl | 211,595 |
| data/final_dataset_val.jsonl | 26,449 |
| data/final_dataset_test.jsonl | 26,450 |

### Uso
from datasets import load_dataset
dataset = load_dataset("rapman/orpheon-structural-dataset", split="train")