Research

Document intelligence is not Mortgage Intelligence.

Reading a document and understanding a mortgage are different problems with different failure modes. Most mortgage AI has solved the first well, and reports it as though it solved the second. Establishing the difference, and measuring it on real mortgage work, is the research program.

The thesis

One question is about a document. The other is about the file.

An extraction system answers: what does this document say? A mortgage system answers: is this file sound, under the rules that apply to it, and can you show why? The first question is about one document. The second is about the relationships between all of them, checked against a body of policy that changes over time.

A system can be excellent at the first and structurally incapable of the second, and from a demonstration the difference is invisible. Both produce a screen of extracted fields with high confidence numbers beside them. The difference shows up later, in what escapes.

Two different questions, with different failure modes.

OCR and IDP extraction

Solves turning documents into fields, at volume and low cost. Does not solve whether the fields agree with one another, or with policy.

General-purpose AI

Solves reading unusual documents and answering questions. Does not solve reproducibility: ask twice and the reasoning may differ, and it cannot be pinned to a version.

Sample-based manual review

Solves judgment, context, and the cases no rule anticipates. Bounded by hours available: a sample estimates a rate; it does not find a pocket.

What the problem actually requires

Full-file coverage, cross-document checks as a first-class operation, deterministic disposition, versioned rules, and evidence at value level.

Published

Writing from the research program

Only topics with published work are listed. Benchmark results will be published with the method, the document set and a comparison anyone can reproduce, or not at all. The evaluation harness itself stays internal.

Mortgage Intelligence

Fifty Years of Mortgage Technology, and What Each Wave Left for the Reviewer

Five waves of mortgage technology since the 1970s, what each automated, and which review tasks each one left behind.

6 min read

Document intelligence versus mortgage reasoning

Mortgage Quality Control Is a Validation Problem, Not an Extraction Problem

Reading a document and checking a loan file are different problems. Extraction accuracy is a prerequisite for validation, not a substitute for it.

6 min read Document intelligence versus mortgage reasoning

Why a Mortgage File Has Fifty Documents to Establish About Twelve Facts

Dozens of documents establish a much smaller set of facts. Organising a file by fact rather than by document is what makes contradictions visible.

6 min read

Evaluation

How to Evaluate a Mortgage AI Claim

Six questions that separate a checkable AI claim from an unverifiable one, with the failure modes each question is designed to surface.

7 min read

AI governance in mortgage

Explainability Is Not a Feature of AI in Mortgage. It Is the Condition of Using It.

What a reviewer must be able to reconstruct about a finding years later, and why an adverse-action notice and a QC finding are different obligations.

6 min read

The wider library, across origination, documents, compliance, the secondary market and quality control operations, is under Resources.

The honest test is files whose outcomes you know.

Comparisons written by vendors are worth very little, including this one. Bring a set of files with known outcomes and see what a full-file, cross-document, versioned check finds that a sample did not.

Request a working session Explore QC