David Tom Foss

Deterministic Validation for Reliable LLM-Based Causal Knowledge Extraction

Large Language Models produce stochastic outputs that undermine reproducibility in knowledge extraction. We present a deterministic post-processing framework with 14 explicit validation predicates that transforms unreliable LLM output into consistent, validated causal triplets. Evaluated on four benchmarks spanning 2,1...