Data Contamination Python Packages

traceguard

Point-in-time correct LLM instrumentation — tracing, version pinning and look-ahead-bias protection for research pipelines. pip install traceguard

corpus-assay

Fast, auditable n-gram detection of benchmark contamination in LLM training corpora (Rust-accelerated).

siftan

Reproducible n-gram screening of language-model training data for overlap with evaluation benchmarks.

verascan

Detect train-eval data contamination with exact, fuzzy, and semantic matching. Built for LLM fine-tuning, RAG evaluation, and reliable benchmarking.

contamination-audit

Scan evaluation data for n-gram overlap, canaries, answer patterns, hashes, and optional embedding similarity.

overlapy

Python package developed to evaluate textual overlap (N-Grams) between two volumes of text.