Pin it once. Parse every PDF after.

Build a parser from one sample of a recurring PDF layout. Every later document with that layout comes back as JSON, with each value correct or flagged, never guessed.

{
  "status": "needs_review",
  "data": {
    "order_number": "PO-24117",
    "total": "1840.50",
    "delivery_date": null
  },
  "fields": {
    "delivery_date": { "status": "not_found" }
  }
}