GLIDE (Generated Label Inference & Debiasing Engine) is a Python library for rigorous evaluation of GenAI systems.

• LLM judges are biased. Proxy evaluators systematically over- or under-estimate true performance.
• Human annotation is expensive. Full labeling at scale is rarely feasible.
• GLIDE bridges the gap. Thanks to prediction-powered inference, it combines both to produce valid, unbiased performance estimates.

• Unbiased performance and risk estimates
• Smart annotation budget allocation to minimise human labeling
• Guaranteed confidence intervals regardless of proxy quality
• Works for any metric, any AI system, any unstructured output requiring annotation