Knowledge Version Diff Auditor

Scenario

Use after creating a baseline and before rebuilding embeddings or publishing a new knowledge release.

Input and output

Input is JSON containing baseline and candidate source records with source_id, version, and content_hash. Output classifies each source as ADDED, MODIFIED, REMOVED, or UNCHANGED, and produces review_queue plus diff_hash.

Execution and recovery

Run python scripts/diff.py examples/baseline.json examples/candidate.json. Missing identifiers or duplicated source ids return visible errors. Correct the inventory and rerun; the script never rewrites content.

Privacy boundary and acceptance

Only hashes, versions, and redacted references are needed. Do not include document bodies, secrets, customer data, or platform credentials. The sample must show one modified and one added source with a stable diff_hash.