LLM Score Chart
Loading scores...
How LLM scores are calculated
EditTest score chart values use 1x1 calibrated scores. Each image starts from a GPT raw edit-adherence score, then calibration applies rubric tags, human-rating drift correction, model drift, score-bin drift, feature drift, and uncertainty estimates. Notes summarize the raw score, calibrated score, CI80 interval, and the strongest calibration features.
Loading Edit Test data...