Chain Product Catalog Cleanup Into Reviewable Change Sets

ROLE Act as a catalog data workflow architect. OBJECTIVE Build a prompt chain to clean [CATALOG SAMPLE] according to [CATALOG RULES]. CONTEXT Target schema: [TARGET SCHEMA]. Approved reference data: [REFERENCE DATA]. Fields requiring human approval: [PROTECTED FIELDS]. INSTRUCTIONS Separate profiling, deterministic normalization, candidate matching, attribute standardization, and change-set validation. Supply a ready-to-use prompt for each model-assisted stage and identify transformations better handled by code. Preserve original values and stable record identifiers. Require evidence for inferred attributes and leave unsupported fields unresolved. Never merge products solely because their names resemble one another. Define how conflicting identifiers, units, variants, and missing reference entries are handled. Produce proposed changes only; do not assume permission to publish them. Include a reversible representation of every accepted modification. OUTPUT FORMAT Deliver the chain map, stage prompts, handoff schemas, before-and-after diff format, and five synthetic edge cases with expected dispositions. QUALITY CHECK Flag assumptions and avoid fabricated statistics. Ensure protected fields cannot be changed by a later stage that did not receive approval.