Larch (Larix spp.) fracture-parameter database and LEFM synthetic surrogate training data

This dataset supports the article "An Open, Larch-Focused Database of Conifer Fracture Parameters and LEFM-Based Synthetic Surrogate Models for Residual Bearing Capacity Prediction". It comprises: (1) a curated fracture-parameter database of conifers (Larix and Pinus), 47 records from 14 species and 18 sources, compiling Mode I/II fracture toughness (K_Ic, K_IIc) and fracture energies (G_f, G_IIc) with provenance, units, and verification grades; (2) 2,400 synthetic training samples from a full-factorial LEFM generator (normalised crack depth a/H, cross-section B×H, crack number n, spacing s) with the closed-form residual-capacity factor β; (3) trained surrogate models — XGBoost, Gaussian process regression, and a physics-informed neural network — with Python training/analysis code, requirements, and README. Data are provided as CSV/JSON; code runs under Python 3 with numpy, scikit-learn, and xgboost. The dataset enables sub-millisecond estimation of damage severity in cracked timber columns for structural health monitoring and reliability assessment.