Plannotator drops sem-jiff structural diff library

Pairing AST-based diffing with lightweight classification models is the right pragmatic middle ground for modern agentic code review. Rather than throwing massive LLMs at diff inspection or drowning developers in raw formatting churn, hybrid deterministic parsers with targeted AI tie-breakers deliver high recall without hallucinating false edits.

• –Combines Tree-sitter AST parsing across 12 languages with optional Jev decision models to identify extracted, inlined, moved, and reformatted code blocks.
• –Benchmark evaluations report 89% recall with TypeSafe AI's Jev and Clef tie-breakers, outperforming standalone AST tools like RefactoringMiner while maintaining a sub-3% false positive rate.
• –Reduces human review fatigue in agentic coding loops by filtering out trivial refactoring artifacts produced by automated coding assistants.
• –Available as @plannotator/sem-jiff on npm, running locally or powering the Sem tab inside Plannotator's review surface.