Loan compliance testing on every closed loan, each result tied to the page that proves it
Every rule tested on every loan, each result tied to its page.
See one case, screen by screen ↓Software that checks closed loans against the consumer-lending rules — APR and finance charge accuracy, TRID timing and tolerance, RESPA, ECOA and flood notices, higher-priced and high-cost tests — and records each result with its evidence. Loan Compliance Testing has agents read the whole file, test every rule that applies and group the fails into findings for the testing manager to confirm.
How to choose AI solutions for lending →
Updated
Second-line testers prove, loan by loan, that each closed mortgage met the rules: the APR and finance charge disclosed accurately, the Loan Estimate and Closing Disclosure delivered on time, fee changes inside tolerance, the RESPA, ECOA and flood notices sent, and no higher-priced or high-cost line crossed. By hand, that is a checklist of 40-odd rules per file — so the team pulls a small sample at quarter end and works through it months after the loans closed.
A systemic error does not wait for the sample. One fee code mapped wrong in the loan system keeps producing understated disclosures on every loan it touches, and each month it runs adds borrowers who may need to be re-disclosed or reimbursed. And because the fails are never traced to a branch, channel or system change, the same finding comes back next quarter.
Estimated split: about 100 days for quarterly sample testing by hand; 19 days for the Whitcombe loan in the working solution — closed 14 Aug, F-01 put to the testing manager on 2 Sep.
Five specialist agents test every closed loan and recompute every APR; the testing manager confirms each finding.
Daniel and Rosa Whitcombe closed a $412,000 conventional 30-year fixed purchase through a wholesale broker on 14 August. Here is what the August run found on their loan — and on six others like it — screen by screen, in the working solution.
The month-end run takes all 1,284 August closings — retail, wholesale and correspondent, in Ohio, Indiana and Kentucky. By 05:52 the wall is full: 31,107 tests pass, 318 fail, 167 need a person, and the fails are already grouped into 8 findings. Last quarter's sample would have looked at 64 loans.
“7 wholesale loans fail APR and finance charge. Same cause on every one — discount points and lock fees coded seller-paid since the fee-table update of 28 Jul.”
Graham opens WH-2026-08-3372. The file reader took 14 documents, 212 pages and the loan system record and pulled 214 values with their page; the rule selector chose 26 of the 212 rules — conventional purchase, first lien, Ohio, closed 14 Aug. The result: 19 pass, 4 fail, 1 review, 2 not applicable, tested in 41 seconds.
“Disclosed 6.875%, recalculated 7.031% — understated by 0.156, beyond the 0.125% tolerance.” — Closing Disclosure, page 5
The calculator never takes the disclosed APR as its answer. It classifies every fee on page 2 of the Closing Disclosure as a finance charge or not, rebuilds the amount financed and solves the actuarial APR from 360 payments of $2,638.08. The disclosure counted $10,423.27 of prepaid finance charges; the calculator counts $16,723.27 — the $5,150.00 of discount points and the $1,150.00 rate-lock extension fee were left out.
Amount financed $395,276.73 · actuarial APR 7.031% · finance charge understated $6,300.00, against a $100 tolerance.
The data-integrity test puts the loan system's fee screen beside the Closing Disclosure. Fee codes 802 and 811 — the discount points and the lock extension — are marked “Paid by: Seller” in the loan system, while the Closing Disclosure shows them borrower-paid. That is why the disclosure left them out of the finance charge.
The loan system's own change log: fee table v14 for wholesale, effective 28 Jul 2026, changed by Pricing Operations.
One test stays amber: qualified mortgage points and fees. Counting the 1.25 discount points they are 3.54%, over the 3% limit; if the points are bona fide they drop out and it is 2.29%. The undiscounted rate of 6.990% is within one point of APOR at 6.420% — but only the rate sheet can show the points really bought the rate down.
The review goes to a tester. She attaches the wholesale rate sheet of 17 Jun 2026, page 4, which shows the 1.25 points bought the rate from 6.990% to 6.625%, picks “Pass — evidence confirms it” and signs. Had she chosen fail, it would have gone to the testing manager to confirm as a finding.
The findings analyst groups the fails by rule and cause and ranks them by severity and loans affected. F-01, APR and finance charge understated, is Critical: 7 loans, 28 failed tests, $38,940 understated, all on the wholesale desk — every affected loan listed with its disclosed and recalculated APR, and a proposed owner, Victor Ames.
Remediation proposed: correct the fee-table mapping; re-disclose or reimburse the 7 borrowers under the restitution policy; re-test wholesale loans closing in September.
The heat map shows F-01 only in the wholesale column. The change timeline explains it: the rate sheet added a 10-day lock extension on 21 Jul, fee table v14 went live on 28 Jul, the first wholesale lock under it came on 29 Jul, and the 7 affected loans closed between 3 and 22 Aug. Three findings are back from the Q2 sample — F-04, F-06 and F-08 — at a higher count now that every loan is tested.
Graham checks the evidence — every failed test cites its page and value, the APR and finance charge were recomputed on all 7 loans, the root cause and owner are proposed — and signs. Nineteen days after the Whitcombe loan closed, F-01 goes to the Head of Wholesale Lending, not to the next quarterly sample.
“Your signature means: I reviewed the evidence and confirm this finding. It goes to the owner and onto workpaper CT-2026-08.”
Workpaper CT-2026-08 sets out the objective, the scope — every loan, no sample — the method, results by rule family and the eight findings, prepared by Nadia Okonkwo. Signing it locks it; later changes go in an addendum, and it is kept seven years for audit and the examiners.
The same August run, seen by the four people who carry it — what their quarter looked like, and what it looks like now.
Five specialist agents test every closed loan and recompute every APR; the testing manager confirms each finding.
Reads the Loan Estimates, Closing Disclosures, note, appraisal receipt, notices, flood determination and the loan system fee screen, and pulls every value the rules need — fees with payer and section, dates of issue, delivery and receipt, rates — with its page and position.
Chooses which of the 212 rules in the library apply to each loan, by product, lien position, purpose, occupancy, state, channel and application and closing dates.
Recomputes the APR, finance charge, amount financed, total interest percentage, points and fees and rate spread against APOR for the lock week, and the Loan Estimate to Closing Disclosure tolerance buckets — then compares them with the disclosure and shows the working.
Counts business days for every delivery and receipt — the specific definition for Closing Disclosure receipt and rescission, the general definition for the Loan Estimate and notices, Sundays and federal holidays excluded.
Groups failed tests into findings, ranks them by severity and loans affected, looks for a common branch, channel, product, settlement agent or system change, proposes an owner and drafts the workpaper.
Confirms findings and signs workpapers; testers resolve reviews. The agents propose; a named person decides.
The testing manager and the testers can ask in plain words — or tell it what to change. Answers point back to the page and the test they came from.
Why did the APR test fail on the Whitcombe loan?
The disclosure counted $10,423.27 of prepaid finance charges and showed an APR of 6.875%. The Closing Disclosure also shows $5,150.00 discount points and a $1,150.00 lock extension fee paid by the borrowers — the loan system has both as seller-paid. Counting them, the amount financed is $395,276.73 and the APR 7.031%: 0.156 over, beyond the 0.125 tolerance. Finance charge understated by $6,300.00.
Which other loans have the same fee-coding problem?
7 loans, all wholesale, all locked after the fee-table update of 28 Jul — Oduya, Tran, Brandt, Lachlan, Whitcombe, Marlow and Verhoeven, closed 3 to 22 Aug. $38,940 in all. No retail or correspondent loan is affected — their fee table did not change.
Test every wholesale loan the day after it closes
Done. From tonight every wholesale loan is tested the night after it closes, on top of the monthly run. Had this been on in August, F-01 would have shown up on 4 Aug, the night after the first affected loan closed — before the other 6. It is in Settings as a watch rule and on the audit trail.
Is the Whitcombe loan still a qualified mortgage?
Probably — a tester needs to confirm one fact. Counting the 1.25 discount points, points and fees are 3.54%, over the 3% limit; if the points are bona fide they drop out and it is 2.29%. The undiscounted rate is 6.990%, within one point of APOR 6.420% — the wholesale rate sheet for 17 Jun will show whether the points really bought the rate down.
13 screens from the working solution, on its sample data. Pick one to see it large.
What gets tested, how strictly and who decides — set by the testing manager, not built as a project.
| Setting | Default | Choose from |
|---|---|---|
| Which closed loans are tested | Every closed loan | Every closed loan · 5% random sample · 10% random sample · Risk-weighted sample |
| When a closed loan is tested | Nightly, the day after closing | Nightly · Weekly · Monthly |
| Rule library | v2026.3 · 212 rules | v2026.3 · 212 rules, or v2026.2 · 207 rules — a new version is tested on last month first |
| APR tolerance, regular transaction | 0.125 percentage point | Tighter allowed, never looser than Regulation Z; 0.250 for an irregular transaction |
| Finance charge tolerance, mortgage | Understated by no more than $100 | Set by Regulation Z; a refinance with right to cancel has its own $35 setting |
| Who confirms findings | Graham Feld · Testing Manager | Any testing manager |
| Who resolves reviews | Compliance testers | Compliance testers · Testing manager only |
| Keep signed workpapers | 7 years | 5 · 7 · 10 years |
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 lending data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (ACES Quality Management, Q4 2025 Mortgage QC Industry Trends — 24.66% of critical defects legal, regulatory and compliance). Rules cited: 12 CFR 1026.18, .19, .22, .32, .35, .43 (Regulation Z) · 12 CFR 1024.17, .20 (Regulation X) · 12 CFR 1002.14 (Regulation B). People, lenders, brokers and borrowers named on this page are fictional — characters and sample data in the working solution.
Software that checks closed loans against the consumer-lending rules — APR and finance charge accuracy, TRID timing and tolerance, RESPA, ECOA and flood notices, higher-priced and high-cost tests — and records each result with its evidence. Loan Compliance Testing has agents read the whole file, test every rule that applies and group the fails into findings for the testing manager to confirm.
Every closed loan, by default — in the working solution, all 1,284 August closings and 31,592 rule tests, against 64 loans in last quarter's sample. A random or risk-weighted sample is still a setting if you want one.
It recomputes them. The independent calculator classifies every fee as a finance charge or not, rebuilds the amount financed and solves the actuarial APR from the scheduled payment stream, then compares the result with the disclosure and shows the working. It never takes the disclosed APR as its answer.
The timing checker uses the specific business-day definition — all days but Sundays and federal holidays — for Closing Disclosure receipt and rescission, and the general definition for the Loan Estimate and notices. It applies the mailbox rule only when there is no receipt evidence, and says so.
The rules in your library. The working solution runs 212 rules — TILA loan calculations, TRID timing and tolerance, RESPA notices, Equal Credit Opportunity, higher-priced and high-cost, flood, data integrity and Ohio, Indiana and Kentucky overlays — and picks the ones that apply to each loan by product, lien, purpose, state and dates.
People do. Testers resolve every result the rules could not settle, with the evidence attached, and only the testing manager confirms a finding, with an electronic signature. The agents never change a loan record or contact a borrower.
Yes. Tolerances can be set tighter than Regulation Z, never looser, and new rules and rule changes go to the testing manager for approval. A new library version is tested on last month's loans first.
The Agentic Solution Engine builds and deploys it from your requirements — your rule checklists, workpaper format and a sample of closed loan files — and it goes live once every quality gate has passed. We will walk you through it on a month of your own closed loans first.
One question, all fifty states at once, every answer cited to the rule it comes from.
1-min demoRegulatory Change ManagementEvery new rule scoped to products and states, and owned the day it lands.
1-min demoHMDA and 1071 Data ValidationEvery record checked against the file, not a sample, before filing.
1-min demoWe’ll run Loan Compliance Testing on a month of your own closed loans.