HMDA and 1071 data validation on every record, every field matched to its page in the loan file
Every record checked against the file, not a sample, before filing.
See one case, screen by screen ↓It is checking each reportable field on the loan application register — and, from 2028 for covered lenders, the Section 1071 small-business register — against the loan file before filing. HMDA and 1071 Data Validation does this on every record, not a sample: agents find the page each field comes from, apply the examiners' tolerances and show only the fields that break.
How to choose AI solutions for lending →
Updated
Every record on the loan application register carries dozens of reportable fields — action taken date, income relied on, rate spread, census tract, the demographic fields — and each one has to match the loan file. A data reviewer checks them one field at a time, opening the file in more than one system to find the page that holds the answer. There is time for a sample each quarter and a rush in February before the 1 March deadline.
The examiner samples too. For a register of 501 to 100,000 records they start with 35 and go to 79 when a field shows errors. Four errors in one field among the 79 mean that field is corrected and resubmitted across the whole register. The tolerances are narrow: three days on the dates, $1,000 on the loan amount, rounding on income, nothing on rate spread or demographics. And from 1 January 2028 the larger small-business lenders take on a second register under Section 1071.
Estimated split for one quarter's register review at a mid-size lender — by hand on a sample, and with the solution on every record.
Six specialist agents match every reportable field to its page in the loan file; the HMDA data manager corrects and signs.
Kestrel Valley Bank's Q3 2026 run finished at 06:00 on Monday 12 October. The first record with a break is the Ellery home purchase, denied for debt-to-income. Here is what happens next, screen by screen, in the working solution.
Renata opens Field match. The Q3 run read 1,846 records and made 158,756 field checks: 214 breaks, 129 already resolved, 37 within tolerance, 48 open. First in the queue is Marcus and Dana Ellery — $412,000, home purchase, denied, Eastgate branch — with every register field drawn by a line to the page it came from.
“Every reportable field on every record, drawn to the page in the loan file it came from. Two fields break on the Ellery application.”
Re-running the match shows what the agents did: the register reader found 86 fields that apply to a denied purchase; the loan file reader opened 31 documents and 212 pages and found a source page for 86 of 86; rate spread is NA for a denial and the address sits in tract 42041011802. 83 fields match, and the application date is one day apart — inside the 3-day tolerance, so it is not an error.
The last line is a person: “Waiting for Renata Okafor — 2 corrections to sign.”
The register holds 03/14/2026, the day the file was moved to Suspended in the origination system for missing conditions. The credit decision was signed on 03/20 and the adverse action notice is dated 03/21. Seven days apart is outside the 3-day tolerance, so an examiner counts it as an error. Open page shows the notice itself, the date highlighted.
“Kestrel Valley HMDA procedure §4.2: a denial is reported on the date of the adverse action notice.”
Renata presses Correct to 03/21. The correction states what changes, on which record and on the evidence of which document, and the reason goes to the audit trail: “Denial reported on the adverse action notice date (§4.2); the register held the Suspended status date.” She signs with her name and role, and the line rejoins.
The register reports $148,000 — base pay plus a $13,000 bonus. The underwriter excluded the bonus for under two years of history and relied on $135,000: $9,250 a month base salary and $2,000 a month for the co-borrower. The 47% debt-to-income on the register already uses $135,000.
“Regulation C: report the gross annual income relied on in making the credit decision. Tolerance: rounding to the nearest thousand only.”
Renata corrects income to 135 and signs. Both breaks now show green with the corrected value beside the old one, and the panel records who signed and when. The run strip moves to 131 resolved and 46 open, and the next record with a break is one click away.
“Marcus and Dana Ellery is clean · 46 breaks open.”
Error rates lays out the 2026 register to date — 5,318 records — field by branch, as first entered. A 79-record sample would have ordered a resubmission 37 times in 100 as first entered; income alone 23 in 100, action taken date 12 in 100. After the check it is under 1 in 100. Draw a sample to see what the examiner would find, before and after.
Eastgate carries 41 of the 127 action-taken-date errors this year — 32% of the errors on 16% of the records. In 38 of the 41 the register holds the date the file was moved to Suspended; the decision came 4 to 12 days later. Renata sends the finding to Priya with the 41 records attached.
“Could officers leave the status until the decision is signed?”
412 small-business applications from Q3, 23 fields under the 2026 rule, 9,476 field checks, 63 breaks. Halvorsen Provisions LLC stated revenue of $980,000; Desmond verified $1,214,300 from the 2025 return. Verified revenue is reported, and above $1 million it is not a small business under the 2026 rule, so the application leaves the register — one of 9 like it.
The filing edits run on the corrected register: syntactical and validity edits with 0 failed, macro-quality passed, and 3 quality edits to confirm — values unusual but true, such as an applicant age above 90. Once they are confirmed, the register goes to Glen Whitaker, who signs it off before it is filed.
The filing edit runner never files: the compliance testing manager submits.
The same quarter, seen by the four people who carry it — what their quarter looked like, and what it looks like now.
Six specialist agents match every reportable field to its page in the loan file; the HMDA data manager corrects and signs.
Reads the HMDA register and the 1071 practice register from the origination systems every night, one record per application, and notes which fields apply — a denial has no rate spread.
Finds the document and the page each field comes from: application, credit decision, adverse action notice, Closing Disclosure, appraisal, income worksheet, demographic form, tax return.
Compares every register value with its source, applies the examiners' tolerances and the bank's procedure rules, and writes each break with both values and the page.
Recomputes rate spread from the APR and the APOR table for the lock date; geocodes each address to state, county and census tract.
Runs the syntactical, validity, quality and macro-quality edits on the corrected register and drafts the explanations.
Counts errors by field, branch and officer, computes the chance an exam sample orders a resubmission, and writes root-cause findings.
Corrects and signs; testing manager signs off the filing. The agents propose; a named person decides.
The data manager can ask about any record, field, branch or the exam odds in plain words — or say what to correct. Every answer points to its source.
Why does the Ellery action taken date break?
The register says 03/14/2026 — the day the file was moved to Suspended for missing conditions. The credit decision was signed 03/20 and the adverse action notice is dated 03/21/2026. Your procedure §4.2 reports a denial on the notice date, and the FFIEC tolerance is 3 days, so 7 days is an exam error.
What are our odds of a resubmission if we filed today?
For a register of 5,318 records the examiner's sample goes up to 79, and 4 errors in one field order it resubmitted. As the fields were first entered, one sample would have ordered a resubmission 37 times in 100 — income alone 23 in 100, action taken date 12 in 100. With the open breaks the only errors left, it is under 1 in 100.
Which branch causes most action taken date errors?
Eastgate: 41 of the 127 this year — 32% of the errors on 16% of the records. In 38 of the 41 the register holds the date the file was moved to Suspended, 4 to 12 days before the decision. Fix the status step and the field stops breaking.
Is Halvorsen Provisions covered by Section 1071?
No. The application states revenue of $980,000, but the analyst verified $1,214,300 from the 2025 return. Under the 2026 rule a small business has gross annual revenue of $1 million or less, so this application leaves the practice register. Eight more Q3 applications are in the same position.
13 screens from the working solution, on its sample data. Pick one to see it large.
Tolerances, where each field comes from and who signs are settings. Changes apply to the next run and are kept as a new rule version.
| Setting | Default | Choose from |
|---|---|---|
| Application date tolerance | 3 days | Fewer days only — may only be tightened |
| Action taken date tolerance | 3 days, not across a year end | Fewer days only — may only be tightened |
| Loan amount and income | $1,000 · rounding to the nearest thousand | FFIEC transaction testing |
| Rate spread, demographics, denial reasons | Exact | No tolerance |
| Action taken date for denials | Date on the adverse action notice (§4.2) | On or off · your HMDA procedure |
| Income | Gross income relied on, from the underwriting income worksheet | On or off · your HMDA procedure |
| 1071 gross annual revenue | Verified figure when the analyst verified it | On or off · your HMDA procedure |
| Who signs | Data manager per correction · second reviewer on demographics · compliance testing manager before filing | Any named person |
| Run schedule | New records nightly · full run at quarter end | Nightly + quarter end · weekly + quarter end · quarter end only |
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 register · “target” = the design goal, measured in the live solution · “estimated” = our estimate (exam odds computed from the sample register’s 2026 error counts) · “typical” = published figures (FFIEC HMDA Examiner Transaction Testing Guidelines — 79-record sample, 4-error threshold, tolerances; CFPB enforcement, 2024 — $3.95M; CFPB Section 1071 final rule, May 2026 — compliance from 1 January 2028). People, lenders and businesses named on this page are fictional — characters and sample data in the working solution.
It is checking each reportable field on the loan application register — and, from 2028 for covered lenders, the Section 1071 small-business register — against the loan file before filing. HMDA and 1071 Data Validation does this on every record, not a sample: agents find the page each field comes from, apply the examiners' tolerances and show only the fields that break.
A scrub by hand checks a sample of files. Here every field on every record is matched to its source page: in the sample quarter, 1,846 records and 158,756 field checks. The examiner still samples, but from a register already checked field by field, with every correction signed.
Yes. Application date and action taken date within 3 calendar days (not across a year end), loan amount within $1,000 and income rounding to the nearest thousand are not errors; rate spread, demographics and denial reasons must be exact. Differences inside tolerance are closed with a note, and the tolerances can only be tightened.
No. The agents are read-only on the origination systems and never edit the register. Only a correction the HMDA data manager signs, with a reason, changes a value; she can also keep the register value and say why.
Yes. It runs the same field match on a practice small-business register under the 2026 rule — the statutory fields checked against the credit application, the tax return and the decision — and flags applications whose verified revenue is above $1 million, which are not covered.
The error-rate monitor counts errors by field, branch and loan officer, then computes the chance that an examiner's sample would find four errors in one field. You can draw a 79-record sample yourself and see the result as first entered and after the check.
Yes. Agents find breaks; the HMDA data manager corrects and signs each one, demographic corrections need a second reviewer, and the compliance testing manager signs off the register before a person files it. Agent findings and human decisions are recorded separately.
The Agentic Solution Engine builds and deploys it from your requirements — your HMDA procedure, your register extract and a sample of loan files — and it goes live once every quality gate has passed. We will walk you through it on your own register 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 demoFair Lending Exception MonitoringEvery exception reasoned at the time, not explained a quarter later.
1-min demoWe’ll run HMDA and 1071 Data Validation on a sample of your own loan files and register.