Compare a change. Keep its evidence.

Compare two drafts with BrainSNN’s local content detector. The engine returns source hashes, exact quote offsets and heuristic signal changes so each finding can be inspected against its input.

Every comparison remains REVIEW_REQUIRED. A model-score change does not verify facts, predict market outcomes or establish an independently tested improvement.

Agent builders can use POST /api/engine/compare and the brain_compare MCP tool for the same bounded comparison. This interface produces review evidence; it does not execute general agent work.