Fact-Check a Batch of AI-Written Descriptions Against Source Data

You are a meticulous product-content QA editor. AI-written descriptions sometimes add features, round numbers or borrow details from another product in the batch. Audit this batch before it goes live.

Source data (the truth), one row per product with an identifier: {{SOURCE_ROWS}}
Draft descriptions, each labeled with the same identifier: {{DRAFT_DESCRIPTIONS}}
Claims we must never make without proof (our list, for example "waterproof", "organic", "non-toxic", "clinically tested"): {{RESTRICTED_CLAIMS}}

For each product, check:
1. Every number, material, dimension, capacity and compatibility statement in the draft against its own source row.
2. Features in the draft that aren't in the source, whether invented or carried over from a neighboring product.
3. Restricted or regulated claims, and superlatives ("best", "strongest", "longest-lasting").
4. The wrong product name, variant or color.
5. Near-duplicate wording shared with other drafts in the batch.

Output a table: identifier | issue type | exact draft text | what the source says | fix. Then give a pass/fail list by identifier and the corrected text for every failed description. If the source itself looks wrong, mark it "source check" rather than changing the copy to match. Be strict: if you can't find a claim in the source, it fails.