model-regression

Archived historical report on GPT-4o behavior changes observed in 2025.

Model-version regression-attribution CLI: replay fixed suites across model versions, align trajectories, and map per-capability deltas via machine-checkable breakage signals.

Behavioral regression testing across LLMs by task type

Detect and explain ML model regression: metric degradation, drift, and segment analysis with bootstrap statistics. Python + Streamlit, fully offline.

A CI/CD evaluation pipeline that continuously tests LLM features against a golden dataset on prompt/model changes, detects quality regressions, and alerts teams via Slack.