An open methodology for benchmarking data vendors against ground truth curated from your own inbox and CRM — weak supervision, entity resolution, per-field metrics, and cross-vendor overlap analysis.

An empirical variant annotation and data evaluation pipeline.

Synthetic Data Generation and Evaluation

Evaluación crítica de CTGAN en UCI Adult: fidelidad (Wasserstein, JSD, Frobenius), utilidad (TSTR) y privacidad (DCR, MIA). El modelo base genera 62 % de valores imposibles que Wasserstein y JSD no detectan; un recorte de una línea los elimina y conserva la utilidad.

Anthropic Engineering blog, converted to Markdown for use as agent context and RAG knowledge base