FICTIONAL PRACTICE PACK: document review
Version: 2026-10-07. All details are made up.
Guide: https://animasai.co/documents-to-a-review-sheet

Read request-a.txt and attachment-b.txt. Use field-list.csv and review-blank.csv to keep the requested fields visible. Both text files label their page number. The unreadable marker is deliberate; this is not an OCR benchmark.

Copy this instruction into your existing AI tool with the two fictional documents and field-list.csv:

Prepare a field review sheet. For every requested field, copy the literal value and cite its document, page and supporting passage. Keep missing, unreadable, ambiguous and conflicting values visible. Neither document has priority over the other. Do not infer a date convention, choose an address, guess a reference number or invent an email. Leave approved_value and reviewer blank. Do not enter anything into a live system.

Compare with expected-review.csv after making the proposal. Expect all five fields: business name copied, date ambiguous, address conflicting, reference unreadable and email missing. Leave approved values empty until a person checks and resolves them. The answer key is an editorial example, not a measured model result.

For your own work, use only documents you are allowed to share. Provider source citations are pointers to check, not proof of correct extraction. Keep the source passage with every proposed value and count missing fields as well as filled ones.