Skip to content
Multigrid

Consumer, travel and closing documents

Vehicle titles, folios, itineraries and mortgage closing statements — where a named form has numbered lines that must reconcile.

The documents in this cluster arrive one at a time, from a person, in whatever condition a person keeps paper in. A photographed title with a thumb over the corner, a hotel folio slid under a door at 6am, a confirmation email forwarded three times. None of them is an API, and none of them was designed to be read by a machine. What they do have — and what a generic extraction pipeline throws away — is that each one is a named form with an internal arithmetic or an external standard behind it. A VIN carries a check digit. A folio carries a running balance. An auction receipt carries a premium that is a stated function of the hammer price. A manifest states its own total.

So these pages are about the invariant rather than the schema. The schema is the easy half and it is covered elsewhere; the useful half is knowing which field can be checked against which other field, which disagreement is a real error, and which disagreement is the document being correct in a way your validator did not anticipate. Where a page does the arithmetic, the inputs are labelled as assumptions in the sentence that uses them.

Extracting Structured Fields From a Vehicle Registration Document

How to validate an extracted VIN arithmetically using the published check-digit formula, and what the check digit still cannot catch.

10 min read

Extracting Structured Fields From a Vehicle Title Document

Why title brands, odometer disclosure status and lien release each need a tri-state field rather than a string, and why absence is not a negative.

10 min read

Extracting Structured Data From an Auction Sale Receipt

Why the check that buyer's premium equals a stated percentage of hammer price fails on almost every real receipt, and what to check instead.

10 min read

Extracting Structured Data From a Pawn Shop or Consignment Receipt

Why the first extracted field must be whether the document is a loan or an agency agreement, and why the redemption deadline should be read rather than computed.

10 min read

Extracting Structured Fields From a Moving Company Bill of Lading

Why the condition-code legend printed on the mover's own form must be extracted before the inventory rows, and why origin and destination notes are one paired record.

11 min read

Extracting Structured Fields From a Restaurant Reservation Confirmation

Why a booking time must be stored as wall time plus a zone inferred from the venue address, and never normalised to UTC at extraction.

10 min read

Extracting Structured Fields From a Flight Itinerary Beyond Calendar Events

Fare class, baggage allowance and ticket number are three fields an itinerary states ambiguously, and each has a different resolution rule.

11 min read

Extracting Room Charges and Taxes From a Hotel Folio

Why nightly rate times nights almost never equals the room total, and what the running balance column lets you check instead.

10 min read

Extracting Structured Fields From a Car Rental Agreement

Why fuel policy needs an enumerated field, why a mileage cap needs a period, and why an unticked waiver box is not a decline.

10 min read

Extracting a Structured Route From a Delivery Manifest

Why the sum of per-stop package counts legitimately disagrees with the manifest total, and how to tell which disagreements are extraction errors.

10 min read

Extracting Structured Fields From a Proof of Delivery Document

Why a proof of delivery needs a delivery-status enum rather than a nullable signature field, and how exception notations and photo references change the schema.

9 min read

Extracting Structured Fields From a Building Permit Inspection Card

How to model the multi-row inspection log a permit card accumulates over a project, and the sequencing check that flags an inspection recorded before its prerequisite.

9 min read

Extracting Structured Data From a Zoning Compliance Letter

Parcel identifiers have no check digit, permitted use is a three-way distinction rather than a boolean, and the code citation is what makes the extraction worth anything.

9 min read

Extracting Structured Fields From a Purchase and Sale Agreement

Financing, inspection and appraisal contingencies each carry their own deadline computed from a defined basis date, and merging them into one text field destroys every one of those deadlines.

10 min read

Extracting Structured Fields From a Property Management Inspection Checklist

A room-by-room checklist is a table of tables, and the schema has to keep each room as its own record so a move-in form can be diffed against a move-out form.

9 min read

Extracting Structured Fields From a Home Warranty Claim Form

The coverage-relevant content of a home warranty claim is cause language inside a technician's free-text diagnosis, and the covered item must map to the specific contract's list rather than a generic taxonomy.

9 min read

Extracting a Structured Ingredient and Dosage Table From a Supplement Facts Panel

A blank percent Daily Value cell on a Supplement Facts panel is correct data, not a missing field, and proprietary blends withhold ingredient amounts by regulation rather than by accident.

9 min read

Extracting Structured Fields From an Employment Verification Letter

Why a letter stating a current annual salary is genuinely ambiguous between base and total compensation, and how to record that ambiguity instead of resolving it silently.

9 min read

Extracting Structured Fields From a Background Check Report

A background report is several independent searches bound into one PDF, each with its own status and scope, and a single overall-status field misrepresents every partial result.

10 min read

Extracting Structured Data From a Reference Check Form

A numeric rating and the comment qualifying it are one fact, and a rating stored without its scale definition and direction is not interpretable at all.

9 min read

Extracting Structured Fields From a Non-Profit Donation Receipt

Why the goods-or-services disclosure on a charitable receipt is a three-state field, and how its absence changes what the extracted record can be used for.

9 min read

Extracting the Payment Schedule From a Multi-Year Pledge Agreement

How to pull an installment table out of a multi-year pledge and check it against the stated total, including the two cases where an exact match is the wrong test.

9 min read

Extracting Structured Fields From a Membership Application Form

How to tell the tier somebody selected from the rate table printed next to it, and why the obvious cross-check between monthly and annual pricing is wrong.

9 min read

Extracting Structured Fields From a Conference Badge Scan Export

What to do when the QR payload and the printed name on a scanned badge disagree, and which of the two a schema should treat as authoritative.

9 min read

Extracting Structured Fields From a Warranty Claim Denial Letter

How to capture the exclusion a denial letter cites and the appeal window it grants, and why the deadline should be computed downstream rather than extracted.

9 min read

Extracting Structured Fields From a Loan Payoff Statement

How to read principal, accrued interest and per-diem off a payoff statement, and project the figure forward to a closing date that is not the good-through date.

10 min read

Extracting Structured Fields From a Mortgage Amortization Schedule

How to recompute an amortization schedule from principal, rate and term so that every extracted row can be checked, and which checks catch a row dropped at a page break.

10 min read

Extracting Structured Fields From a Closing Disclosure

Why a Closing Disclosure should be addressed by its prescribed section letters rather than by layout inference, and how the five money columns on page two destroy a naive extraction.

10 min read

Extracting Structured Fields From a HUD-1 Settlement Statement

How the HUD-1's numbered line series work as an addressing scheme, and the three footing identities that let an extracted statement check itself.

10 min read

Designing a Schema for a Document Type You Only See a Few Times a Year

An argument that bespoke schemas for rare document types fail on testability rather than on build cost, and that a general capture schema plus a human is the right default.

10 min read

Writing Regression Tests for an Extraction Prompt Before You Touch It

A worked golden-set harness that scores an extraction prompt per field rather than per document, so a fix for one field cannot silently break another.

11 min read

Other topics