Loan delivery data validation: every delivered field matched to the signed closing file before the loan is sold
Clean deliveries, fewer purchase suspense items and buyback risks.
See one case, screen by screen ↓Checking, before a closed mortgage is sold, that every field in the delivery file agrees with the signed documents — the note, the security instrument, the Closing Disclosure, the appraisal, the final underwriting findings and the MI certificate. Loan Delivery Data Validation does that for every field on every loan in a batch, and ranks what differs by what it would cost.
How to choose AI solutions for lending →
Updated
When a closed loan is sold to Fannie Mae, Freddie Mac or a private investor, the investor buys the data, not the folder. A few hundred delivered fields per loan have to agree with what the borrower signed: rate, amount, first payment and maturity on the note, the MIN on the security instrument, the appraised value, the score and DTI from the final findings, the MI certificate. The document custodian certifies the original note against that same data.
Today a post-closing team stares and compares the loan system’s export against the PDFs, loan by loan, in the last days before a commitment expires. The investor’s edits help with shape and eligibility — a MIN check digit, a missing project type — but not with truth. A first payment date a month off the note still agrees with a 360-month term. An occupancy of primary residence is valid even when the signed 1003 says investment property. Those reach the investor as suspense items, price adjustments or a buyback request years later.
Estimated split of an analyst’s hours on one delivery batch, by hand and with the solution.
Six agents draw a line from every delivered field to its signed document and run the investor's checks before sale.
Monday 12 October at Calder Point Mortgage. Batch DB-1012 — 48 closed 30-year fixed loans, $17,936,400 — has to fill a mandatory agency cash commitment that expires on Friday. Here is what happens, screen by screen, in the working solution.
Danielle opens the match: the 48 loans of DB-1012 as a row of borrower names, and the selected loan’s deciding fields as lines — the delivery file on the left, exactly as exported, the signed closing file on the right. Commitment CW-26-4471 is $18,000,000 and expires Fri 16 Oct, in 4 days. Nothing is matched yet: 0 of 13,728 fields.
“The investor runs its checks on Friday. Run them now.”
She presses Match all 48 loans. The delivery file reader maps 286 fields a loan to the page each must agree with; the closing document reader reads 48 notes, 48 security instruments, 62 riders and 48 Closing Disclosures; the appraisal and underwriting reader reads the appraisals, final findings, transmittals and credit reports. The field matcher recomputes LTV, CLTV, DTI, P&I and maturity on every loan and finds 11 mismatches on 8 loans.
“Correction drafter drafted 10 corrections and 1 QC referral.”
On the Brandt loan the delivery file says first payment 12/01/2026 and maturity 11/01/2056; the note says 11/01/2026 and 10/01/2056. Closing moved from 1 Oct to 25 Sep, the note was redrawn, the loan system kept December. Both pairs agree with a 360-month term, so the edits pass — the line snaps red on the custodian’s note certification. Three more lines break on the same loan: the property value is the $515,000 sales price, not the $509,000 appraised value; the 12% MI certificate was never linked; and the project type is blank where the appraisal says PUD.
“4 loans, $1,608,650, would stop at purchase and miss Friday; 3 would carry a repurchase trigger for years.”
The delivery file says Primary residence. The signed final application, page 4, marks an investment property with $2,150 a month in expected rent, and the appraiser found a tenant with a lease through 31 May 2027. The loan was underwritten, locked and priced as a primary residence. This is a repurchase trigger, and it is not a typo.
Documents that contradict each other are never corrected by the agents — the loan goes to a person in QC.
Danielle presses Send to QC, take out of batch. The loan is referred to Owen Whitaker in QC with both pages, and DB-1012 becomes 47 loans, $17,596,400 — 97.8% of the commitment, inside the ±5% tolerance, so no pair-off.
Each Brandt correction is drafted from its page: first payment 11/01/2026 from Note p. 1; value $509,000 with LTV and CLTV of 80.94% from Appraisal p. 3; 12% coverage, certificate HM-5513082 from Harborline Mortgage Insurance; PUD with $42 a month in dues. She accepts all four, and each broken line turns green in place.
“Corrected from Note p. 1 by Danielle Okafor · the investor’s check now passes.”
Every loan has the full comparison: delivered value, signed value and the page, grouped by loan terms, property and value, borrower and credit, insurance, parties and registry. On Lindqvist the application was taken by phone as “Lindquist”; the note, open beside the fields, is signed “Katherine A. Lindqvist”. The custodian cannot certify a note whose borrower name differs.
Herrera’s note rate was delivered from the original lock — 6.750% — while the note signed after the relock carries 6.875%. Gutierrez’s MIN fails its check digit. Osei’s DTI of 38.2% came from underwriting submission 5; the final submission 7 reads 44.6%. Abernathy’s score of 724 is from the first report; the re-pull of 18 Sep reads 716, one bucket down at −0.375 points. Where a fix moves price, its dollar effect on the grid of 9 Oct sits beside it. She accepts the six drafted corrections together.
The corrected file is a new version; the exported file is never overwritten. The pre-delivery edits run again on it: 0 fatal edits, 0 warnings, and the note certification preview shows all 47 notes agree with the delivered data. The screen also says what the edits alone would have seen: 2 of the 11 mismatches.
“The Brandt dates agree with a 360-month term and the value gives a tidy 80.00% LTV — every edit passes. Only the note and the appraisal show they are wrong.”
Danielle signs with the meaning “Reviewed and approved for delivery”; Marcus Feld releases 47 loans, $17,596,400, for the agency cash window. Out go DB-1012_ULDD_v2.xml for import, the certification list for Northpine, the correction log and the QC referral. Net price effect of the corrections: −$804 — against a $772,250 pair-off without the check.
The same batch, seen by the four people who carry it — what their month-end looked like, and what it looks like now.
Six agents draw a line from every delivered field to its signed document and run the investor's checks before sale.
Reads the ULDD delivery file (MISMO 3.0 XML) exported by the loan system and maps every delivered field to the document and page it must agree with, using the field map for the investor.
Reads the note, security instrument and riders, the Closing Disclosure, the title commitment and the MI certificate — amount, rate, term, dates, payment, names as signed, address, MIN, occupancy covenant, price and dues — citing the page for every value.
Reads the appraisal (value, effective date, project type and dues, occupant), the automated underwriting findings, the transmittal summary and the credit report.
Compares every delivered field with its document value under the investor’s rounding rules, and recomputes LTV and CLTV from the lower of price and appraised value, P&I from amount, rate and term, and maturity from the first payment date.
Applies each corrected value to the investor’s price grid and eligibility rules, ranks the mismatch — blocks purchase, buyback risk, changes price, data warning — and states the price effect in dollars.
Drafts each correction from the cited document, refers contradictions to QC, writes a new version of the delivery file once corrections are accepted, re-runs the edits and the custodian preview, and drafts the release summary.
Accepts each correction; the manager releases. The agents propose; a named person decides.
The analyst and the manager can ask in plain words — or tell it which loan to send to QC. Answers point to the page they rest on.
Why is the Brandt loan-to-value wrong?
The delivery file carries $515,000 as the property value — the contract price on the Closing Disclosure. The appraisal of 2 Sep values the house at $509,000. Value is the lower of the two, so the LTV is 80.94%, not 80.00%. That needs mortgage insurance — Harborline’s 12% certificate HM-5513082 is on file but was never linked — and moves the price by −0.250 points, $1,030.
Which loans would the custodian refuse to certify?
Three notes would not certify against the file as exported: Brandt (note first payment 11/01/2026, file 12/01/2026), Herrera (note 6.875%, file 6.750%) and Lindqvist (note “Lindqvist”, file “Lindquist”). The investor’s edits pass all three. Each has a correction drafted from the note.
What do the corrections do to price?
Three corrections move price on the grid of 9 Oct: Brandt LTV 80.00% → 80.94%, −$1,030; Abernathy score 724 → 716, −$1,110; Herrera rate 6.750% → 6.875%, +$1,336. Net −$804. Delivering the file as it is would instead leave four loans, $1,608,650, in suspense past Friday — a $772,250 pair-off on the commitment — and three loans sold with buyback risk.
Send the Pham loan to QC and take it out of this batch
Done. The delivery file says primary residence, but the signed 1003 marks an investment property with $2,150 a month in expected rent, and the appraiser found a tenant with a lease through 31 May 2027. I referred loan 2026-08761 to Owen Whitaker in QC with both pages. The batch is now 47 loans, $17,596,400 — 97.8% of the commitment, inside the ±5% tolerance.
13 screens from the working solution, on its sample data. Pick one to see it large.
What a delivery is checked against is a setting, versioned and on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Investor rules | Agency cash window · ULDD Phase 5 | Agency cash window (ULDD Phase 5) · Agency MBS swap (ULDD Phase 5) · Meridian Correspondent Capital (jumbo bulk) |
| Price grid | Grid of 9 Oct 2026 | Grid of 9 Oct 2026 · Grid of 2 Oct 2026 |
| Commitment tolerance | 5% | 1% to 10%, set with − and +; a shortfall beyond it pairs off |
| Check a batch before release | On | Release only with 0 fatal edits and 0 certification exceptions |
| Delivered fields | All 286 fields | All 286 fields · Must-match list (64 fields) |
| Underwriting findings | Final submission only | Final submission only · Final submission, warn on earlier |
| Rounding | Investor rounding | LTV and CLTV to two decimals · DTI within half a point · dates exact |
| Occupancy or income contradictions | Always referred to QC | Never corrected by the agents |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample delivery batch · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Milliman Mortgage Repurchase Index, 2026 Q1: lifetime repurchase probability 0.168% Fannie Mae, 0.252% Freddie Mac). Price effects come from the sample lender’s configured price grid and are illustrative, not a published pricing matrix. People, lenders, borrowers and investors named on this page are fictional — characters and sample data in the working solution.
Checking, before a closed mortgage is sold, that every field in the delivery file agrees with the signed documents — the note, the security instrument, the Closing Disclosure, the appraisal, the final underwriting findings and the MI certificate. Loan Delivery Data Validation does that for every field on every loan in a batch, and ranks what differs by what it would cost.
The edits check that data is well-formed and eligible, such as a MIN check digit or a missing project type. This checks that it is what was signed. In the working solution, 9 of 11 mismatches in one batch passed the edits — a first payment date a month off the note, a sales price delivered as the appraised value, a DTI from an earlier underwriting run. It also re-runs the edits on the corrected file.
Not on its own. The agents draft each correction with its page; the loan delivery analyst accepts it or keeps the delivered value with a reason and a second person. Accepted corrections make a new version of the file — the exported file is never overwritten.
The agents never correct it. An occupancy or income contradiction — like a loan priced as a primary residence whose signed 1003 says investment property — is referred to the QC manager with both pages, and the loan is taken out of the batch.
Yes. The price and eligibility checker applies each corrected value to your investor’s price grid — LTV, score, occupancy and other buckets — and shows the dollar effect, alongside the commitment fill and the pair-off a blocked loan would cause. It never changes a price or a commitment.
The ULDD delivery file exported by your loan system, and each loan’s signed closing package: note, security instrument and riders, Closing Disclosure, title commitment, MI certificate, appraisal, final findings, transmittal summary and credit report. Investor rules and price grids are settings.
Yes. The analyst accepts every correction and signs the data; the secondary marketing manager releases the batch, only with 0 fatal edits and 0 certification exceptions. Every agent step and every decision is on the audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your investors and their rules, your price grids, your approvers and a sample of delivered loans with their closing files — and it goes live once every quality gate has passed. We will walk you through it on your own loans first.
An income worksheet in minutes, every figure traced to the return.
1-min demoMortgage Condition ManagementConditions cleared the day the document arrives, not days later.
1-min demoMortgage QC AutomationEvery loan reviewed before it is sold, not a sample after.
1-min demoWe’ll run Loan Delivery Data Validation on a sample of your own delivered loans and their closing files.