Explanation
How to interpret AI-detection evidence
Read classifications, fractions, confidence, and segments without invented thresholds.
Pangram returns overall and segment-level evidence. Fractions describe the reported composition; they are not a certainty score. Confidence belongs to the returned evidence, not to a universal claim about authorship. Keep the result with other evidence and avoid inventing thresholds the API does not publish. Pangram CLI preserves Pangram's labels and ordering.