THE EXTRACTION SPEC Understanding AI HQ free worksheet, version 1.0 (2026-11-12) Fill it in online, print it or get the PDF: Understanding AI HQ website, Learn > Worksheets HOW TO USE IT 1. Pick one document type and three real examples you're allowed to use. 2. List every field, its format and a rule for missing values. 3. Run the spec on all three examples and record what your checks found. ================================================================ 1. THE DOCUMENT ================================================================ DOCUMENT TYPE > HOW I'LL GET THE TEXT OUT, AND WHAT I CHECKED IN THE CONVERSION > ================================================================ 2. FIELDS ================================================================ WHAT TO EXTRACT | Field | Format | If missing ------- | ----- | ------ | ---------- Field 1 | ____ | ____ | ____ Field 2 | ____ | ____ | ____ Field 3 | ____ | ____ | ____ Field 4 | ____ | ____ | ____ Field 5 | ____ | ____ | ____ RULES > > ================================================================ 3. CHECKS ================================================================ ARITHMETIC OR SCRIPT CHECKS > > WHAT I'LL SPOT-CHECK BY HAND > > TEST RESULTS | What the checks found --------- | --------------------- Example 1 | ____ Example 2 | ____ Example 3 | ____ ---------------------------------------------------------------- A schema fixes the shape, not the facts. Check numbers with arithmetic and text against the original. Don't upload documents to a tool your organisation hasn't approved. ################################################################ # EXAMPLE ANSWERS ################################################################ THE EXTRACTION SPEC Understanding AI HQ free worksheet, version 1.0 (2026-11-12) Fill it in online, print it or get the PDF: Understanding AI HQ website, Learn > Worksheets HOW TO USE IT 1. Pick one document type and three real examples you're allowed to use. 2. List every field, its format and a rule for missing values. 3. Run the spec on all three examples and record what your checks found. EXAMPLE ANSWERS. The example is the fictional invoice from the lesson. Example 1's result is from our real run on 12 November 2026; the others are for you to fill in. ================================================================ 1. THE DOCUMENT ================================================================ DOCUMENT TYPE > Supplier invoices (Word and PDF) HOW I'LL GET THE TEXT OUT, AND WHAT I CHECKED IN THE CONVERSION > MarkItDown. Checked: table rows intact; header row shifted down one, which is fine. ================================================================ 2. FIELDS ================================================================ WHAT TO EXTRACT | Field | Format | If missing ------- | ------------ | --------------------------------- | -------------------------------- Field 1 | supplier | text | null Field 2 | invoice_date | as written | null Field 3 | due_date | as written | null; never calculate from terms Field 4 | lines | item, qty, unit price, line total | empty list Field 5 | total | number, no £ sign | null RULES > Use only what the document says. If a field isn't stated, use null. Never calculate or guess a value. Reply with JSON only. ================================================================ 3. CHECKS ================================================================ ARITHMETIC OR SCRIPT CHECKS > qty x price = line total; lines add up to subtotal; subtotal + VAT = total. WHAT I'LL SPOT-CHECK BY HAND > Supplier name and invoice number against the original. TEST RESULTS | What the checks found --------- | --------------------- Example 1 | All checks passed. Example 2 | (your result) Example 3 | (your result) ---------------------------------------------------------------- A schema fixes the shape, not the facts. Check numbers with arithmetic and text against the original. Don't upload documents to a tool your organisation hasn't approved.