diff --git a/vision_runner.ts b/vision_runner.ts index 941c44a..4105545 100644 --- a/vision_runner.ts +++ b/vision_runner.ts @@ -316,10 +316,14 @@ Task: Identify the page types by their headings, then transcribe form fields fro STEP 1 — Identify each page by its UNIQUE marker text (pages can appear in ANY order): - NOTES page: contains the text "Date NDA Scanned". It is a handwritten cover sheet with "Name:", "Date NDA Scanned:", a two-column table headed "Business Interested In:", and a free-text "Notes" area at the bottom. NOT every package has one. - - INFO SHEET page: contains the heading "BUYER INFORMATION SHEET". + - INFO SHEET page: contains the printed heading "BUYER INFORMATION SHEET" (usually with the BizMatch logo at the top). - CA page: contains "CONFIDENTIALITY AGREEMENT" or "PROSPECTIVE BUYER AGREES TO KEEP AND HOLD CONFIDENTIAL". Set notes_page, info_page, ca_page to the respective 1-based page numbers, or null if that page type is absent. - CRITICAL: a page with "Date NDA Scanned" is the NOTES page, NEVER the info sheet — even if it is page 1. + EXCLUSIVITY RULE — each page number may be assigned to AT MOST ONE of notes_page / info_page / ca_page. A single page is never two types at once. Decide by marker priority: + 1. If the page shows "Date NDA Scanned" → it is the NOTES page. It is NEVER the info sheet, even if it is page 1 and even if it also lists businesses. + 2. Else if it shows "BUYER INFORMATION SHEET" → info sheet. + 3. Else if it shows the confidentiality wording → CA page. + So info_page and notes_page must be DIFFERENT numbers (or one of them null). If you were about to set them equal, you misread one — re-check which page carries "BUYER INFORMATION SHEET" versus "Date NDA Scanned". STEP 2 — doc_type: - "buyer_sheet": an INFO SHEET page exists. @@ -335,7 +339,7 @@ STEP 3 — Transcribe VERBATIM (exactly as written, do not normalize or expand a - phone, cell, email, address, state - how_did_you_hear: "How did you hear about us" - interested_in_updates: the marked updates answer/checkbox ("Yes"/"No"), else null - - types_of_business_raw: the "Type(s) of business interested in" list FROM THE INFO SHEET ONLY (may span multiple lines — join with a space). This must come from the info sheet page, NEVER from the notes page. + - types_of_business_raw: the "Type(s) of business interested in" list FROM THE INFO SHEET ONLY. This is usually a MULTI-LINE list with several entries stacked vertically (e.g. "RETAIL/TRADE", "FUEL STATIONS/CONVENIENCE STORES", "RESTAURANTS/BAKERY/CAFES", "TRANSPORT", "IT TELECOM"). Transcribe EVERY line of the list, not just the first one. Include entries even if they are struck through / crossed out (transcribe them as written). Join all entries with a comma in top-to-bottom order. This must come from the info sheet page, NEVER from the notes page. - background_experience: "Background/Experience" (may span multiple lines) - total_purchase_price: "Total Purchase Price" as written - down_payment_raw: "Down Payment" as written (e.g. "$350,000", "1.5M", "TBD")