Defect Leakage: Why the Findings That Reach an Investor Are the Ones QC Was Least Equipped to Catch
Every quality control programme misses defects. This is not a controversial claim and it is not, by itself, an indictment. Post-closing review operates on a sample, under a deadline, against guidelines that change. Some proportion of what is wrong will not be found.
The useful question is not whether defects escape. It is whether the ones that escape are random.
They are not. And the pattern in what leaks is more informative than the leakage rate.
What the aggregate numbers conceal
The industry's post-closing critical defect rate sat at 1.50% for calendar year 2025, effectively unchanged from 1.52% the year before, with the fourth quarter falling to 1.38% from 1.79% in the third. [1] Read alone, that reads like a system in equilibrium.
The composition tells a different story. Legal, Regulatory and Compliance returned to the top defect category in Q4 2025 at 24.66%, rising roughly 30% from 18.97% and marking its third consecutive quarterly increase. Income and Employment sat at 21.52%. Across the full year, Borrower and Mortgage Eligibility rose 291.58% and Credit rose 166.13% year over year. [1]
A flat headline rate covering that much internal movement is not equilibrium. It is a reallocation of where failure occurs, and reallocation is the thing a sampling programme is slowest to detect.
Leakage is structural, not careless
It is tempting to attribute escaped defects to reviewer error. Occasionally that is the cause. More often the defect was never within reach of the method.
Three structural reasons account for most of it.
Sampling. Post-closing review examines a fraction of production. A defect concentrated in a narrow slice of originations — one channel, one product, one underwriter, one state — can be statistically invisible at portfolio level while being systematic within that slice. Random sampling is designed to estimate a rate. It is poorly suited to finding a pocket.
Single-document review. Where review is organised document by document, defects that exist only in the relationship between documents have no natural point of detection. Nobody is looking at the pair. The application is unremarkable. The pay stub is unremarkable. Only the comparison fails, and no step in the process performs it.
Undocumented rules. Where the test lives in a reviewer's judgement rather than in a written, versioned rule, it is applied inconsistently by definition. Two reviewers reach two conclusions, and neither can be re-examined later, because neither was written down at the time.
These are properties of how the work is organised. No amount of diligence within the method corrects them.
The evidence that leakage is not random
If escaped defects were randomly distributed, the defects surfacing later — at investor review, at audit, in a repurchase demand — would resemble the ones QC catches. They do not.
Analysis of GSE repurchase activity found income-related and appraisal-related issues together accounted for 57% of repurchase demands across the eighteen months from April 2023 to October 2024, at an average cost to the seller of approximately $32,288 per loan against a request rate of 0.49%. [2]
Income and appraisal. Both require holding documents against one another. Both are exactly the class of finding that single-document review cannot produce.
The concentration is the finding. Two categories accounting for well over half of realised losses is not the signature of random escape. It is the signature of a method with a consistent blind spot, and the blind spot is cross-document validation.
Why the leak is closing more slowly than the headline suggests
Repurchase volumes have fallen materially from their 2022 peak. Freddie Mac's total repurchase dollar volume dropped 54% between the second and fourth quarters of 2023, from $594 million to $276 million; Fannie Mae's declined 21%, from $444 million to $349 million. [3]
This is real improvement and it deserves to be stated plainly rather than argued around. Part of it reflects better origination-era credit, part improved seller remediation, part a shift in agency posture.
What it does not represent is a change in which defects escape. Falling volume tells us fewer loans are being demanded back. The composition data tells us the ones that are demanded back still cluster in the same categories. A smaller leak through the same hole is progress, not a repair.
What closing it actually requires
Three changes, in order of difficulty.
Move from document review to file review. The unit of analysis has to be the loan, not the page. A finding is produced when values drawn from different documents fail to agree under a stated rule. That is the change aimed at the categories the 57% is concentrated in — which is not the same as claiming it prevents 57% of demands. How much of that share is reachable by cross-document checking is an open question, and the measurement plan below is how it would be answered rather than asserted.
Write the rules down and version them. Experienced reviewers apply consistent standards and document their reasoning; the problem is not that professional judgement is unauditable. It is that a standard held only in individual judgement is hard to apply identically across reviewers, hard to improve deliberately, and slow to evidence at scale when an examiner asks. Versioning matters as much as writing it down: a file reviewed in March must be demonstrably assessed under March's rules, not today's.
Add targeted selection to the random sample — not instead of it. These do different jobs. Random selection produces a defensible estimate of the defect rate across production, and Fannie Mae requires it; discretionary selection weighted toward channels, products or originators showing early signal is what finds concentrations. [3] Discretionary reviews supplement the random sample and do not replace it. A random sample can detect a pocket — whether it does depends on the pocket's size and the sample's design, so this is a question of probability rather than a blind spot.
The borrower's position in this
Defect leakage is normally framed as a lender's exposure, and the repurchase figures make that framing easy. It is worth stating the other half.
A compliance defect that escapes review is, at the borrower's end, a disclosure that arrived outside the required window. An income defect that escapes is a loan approved on an assessment that the file does not support. An eligibility defect is a borrower placed in a product they did not qualify for.
The borrower has no visibility into any of this and no mechanism to check it. Quality control is the only stage at which anyone examines the file on their behalf as well as the investor's. That it currently leaks hardest in the categories that bear most directly on the borrower is not a detail. It is the strongest argument for fixing it.
Sources
- ACES Quality Management, Mortgage QC Industry Trends Report, Q4 and CY 2025. Critical defect rate 1.50% for CY 2025 against 1.52% for CY 2024. Category figures are shares of critical defects, not defect incidence, and are drawn from the report's sample of reviewed loans.
- STRATMOR Group, Unpacking the drivers and costs of GSE repurchase demands, reported by National Mortgage News. The $32,288 figure is a study estimate of cost per repurchase demand, not per completed repurchase; income and appraisal together account for 57% of demands in that study. A category share does not establish that those demands were preventable.
- Fannie Mae Selling Guide, D1-3-01, Lender Post-Closing Quality Control Review Process. Requires both random and discretionary selection; discretionary reviews supplement the random sample rather than replace it.