Every year, mortgage lenders send regulators a list of every home-loan application they handled: who applied, for how much, where the home is, and what the lender decided and when. HMDA data validation means checking that list against the loan files before you send it, so that every entry is true.
This guide is for anyone who owns, checks or pays for that list: HMDA and CRA officers, compliance and fair-lending teams, mortgage operations, internal audit, and the leaders deciding whether to automate the work. No background needed.
Imagine you run the mortgage desk at a community bank. In February, Marcus and Dana Ellery apply for a $412,000 loan to buy a house. In March you turn them down, and you mail them a letter dated March 21.
But your loan system records March 14, the day a clerk put the file on hold. That date goes on your list for the government.
Your list · one row
Applicants
M. and D. Ellery
Loan
$412,000
Decision
Denied
Date
03/14/2026
The letter · in the loan fileKestrel Valley Bank Date: March 21, 2026"We cannot approve your application."
7 days apartexaminers allow 3so an examiner counts it as an error.
No automatic check notices, because March 14 is a perfectly valid date. If an examiner finds four errors like this in the same column of a 79-row sample, you fix that column on every row and file again.
HMDA data validation catches that date before you file.
What HMDA data validation is, and why examiners care
The rules, the register, the exam and Section 1071, explained without selling anything. If you are new to HMDA, start here.
01 · In plain words
HMDA data validation in five minutes
HMDA is the Home Mortgage Disclosure Act, a law from 1975. It asks lenders to report every home-loan application they handle, so that regulators and the public can see whether lenders serve their communities fairly.28
The report is one file a year, called the loan application register, or LAR. Think of it as a spreadsheet: each row is one application, and each column is one fact about it, such as the loan amount, the applicant's income, the census tract of the home, the applicant's race and ethnicity, and what the lender decided and on which day.
Every one of those facts comes from a document in the loan file: the application, the income worksheet, the appraisal, the letter sent to the applicant. HMDA data validation means checking each fact on the register against the document it came from, before the register is filed. Some teams call this a HMDA scrub or LAR scrub.
The life of one application, in five steps
A family appliesThe Ellerys apply for a home loan in February.
The lender decidesThe bank denies it and mails a letter dated March 21.
A row goes on the registerUp to 110 fields, filled from the loan system. Here, the date reads March 14.
Someone checks the rowValidation: each field against its document. This is where the wrong date is caught, or missed.
The register is filed, then examinedDue March 1. Later an examiner compares a random sample of rows with the loan files.
Step 4 is the subject of this guide. Skip it, and step 5 finds the errors for you.
Four things that surprise people
The automatic checks do not check the truth
When you file, the government's system runs checks on every row. They confirm that each value is allowed and plausible. March 14 is an allowed date, so the wrong date passes every check.7
Examiners check against the paper
They pull a random sample of rows and compare every field with the loan file. Only small differences are forgiven: three days on a date, $1,000 on the loan amount, rounding on income.8
One field can force a full refile
For most lenders, four errors in one field across a 79-row sample means correcting that field on every row of the register and filing again.8
The data becomes public
A privacy-protected version of every lender's register is published each year. Advocates, researchers, journalists and examiners use it to study fair lending, so a wrong entry shapes how others see your bank.5
Four words you will see on this page
LAR
The loan application register: the yearly list of applications, one row each.
Edits
The automatic checks run when you file. They test format and plausibility, not truth.
Scrub
A review that compares the register with the loan files, by sample or in full.
Resubmission
Correcting one field on every row and filing again, when an examiner finds too many errors.
Who reports HMDA data, and what goes on the register
In shortBanks, credit unions and mortgage companies that make more than a handful of home loans in a metropolitan area usually report. The exact tests are below. Each application becomes one row with up to 110 fields, and 37 of them are the ones examiners check most closely.
The Home Mortgage Disclosure Act (1975) and its Regulation C (12 CFR Part 1003) require covered lenders to report one record a year for each mortgage application they take and each covered loan they purchase. That file is the loan/application register, or LAR. Its stated purposes are to show whether lenders serve their communities' housing needs, to guide public investment, and to help identify possible discriminatory lending patterns.28
Who must report HMDA data in 2026
Banks, savings associations, credit unions
All five tests
Assets above $59 million on December 31, 2025 (for 2026 data; it was $58 million for 2025)2
A home or branch office in a metropolitan statistical area (MSA)
At least one first-lien home purchase loan or refinancing on a 1–4 unit dwelling in the prior year
A federal connection: federally insured or regulated, or loans insured, guaranteed or intended for sale to Fannie Mae or Freddie Mac
25 closed-end loans in each of the two prior years, or 200 open-end lines1
Nondepository mortgage lenders
Two tests
A home or branch office in an MSA
The same volume test: 25 closed-end loans or 200 open-end lines in each of the two prior years1
Partial exemption. Insured banks and credit unions with fewer than 500 closed-end loans (or fewer than 500 open-end lines) in each of the two prior years can skip 26 of the 48 data points. This does not apply if their recent CRA ratings were poor.6
Inside the LAR: 48 data points, 110 fields, 37 key fields
Swipe sideways to see the full diagram.
Data points from Regulation C; 110 fields from the FFIEC Filing Instructions Guide for data collected since 2018; the 37 HMDA key data fields were designated by the Federal Reserve, FDIC and OCC as the most critical to the integrity of HMDA data analysis.6,7
What the data points cover
Identifiers
Universal loan identifier (ULI)
Legal entity identifier (LEI)
Application and outcome
Application date, loan type, purpose
Preapproval, occupancy, amount
Action taken and its date
Property and geography
Address, state, county
Census tract
Construction method, units
Applicant
Ethnicity, race and sex, and how collected
Age, income
Type of purchaser
Pricing
Rate spread, HOEPA status
Total loan costs, origination charges
Discount points, lender credits, interest rate
Underwriting
Credit score and model
Debt-to-income, combined loan-to-value
Denial reasons, property value
Loan features
Term, introductory rate period
Prepayment penalty, non-amortizing features
Reverse, open-end, business purpose
Channel
Application channel
Loan originator's NMLSR ID
AUS name and result
Section 1071 in one line. Section 1071 is the small-business counterpart to HMDA: covered lenders collect and report data on small-business credit applications, including whether the business is women-owned or minority-owned. Under the 2026 final rule it applies only to larger small-business lenders, from January 1, 2028. See section 6.
03 · From application to filing
Ten steps, many hands, and one certification at the end
In shortA row on the register is built over months, by many people and systems. Most errors enter at the start, in dates, income, demographics and geography, and nothing later in the process catches them automatically.
A LAR record is assembled over months by many people and systems. Any step can introduce an error that no filing edit will catch. The blue row is the register owner.
Who ↓ · Step →
1Application
2Demographics
3Decision
4Notice or closing
5LAR extract
6Geocode and spread
7Scrub
8Filing edits
9Certify and file
10Exam and public data
Applicant
Applies online, by phone or in person
Answers or declines the questions
Gets a notice or closes
Loan officer · processor
Takes the application; the date is set
Records answers; observes if in person and declined
Moves the file through statuses
Sends the adverse action notice, or closes
Underwriter
Decides; relies on income, DTI, CLTV, value
Final figures on the worksheet
Origination system
Status history starts
Fields keyed
Status dates stamped
Closing data
Maps fields to the LAR
HMDA software · tools
Imports the extract
Geocodes the tract; computes APR − APOR
Scrub reports
Pre-checks edits
Builds the file
HMDA data manager
Pulls the extract
Fixes tract and spread
Checks a sample against files
Clears or explains edits
Prepares the filing
Answers exam requests
Compliance testing · audit
Independent test
Signs off; certifies
Supports the exam
FFIEC · examiner · public
Platform runs four edit types
Receives the LAR by March 1
Tests a sample; data goes public
Errors
Errors enter in steps 1–6 → edits check format and plausibility, not the file →the examiner checks the file
Swipe sideways to see all ten steps.
Where errors enter
Status dates
Suspended ≠ decidedA status date, such as "suspended" for conditions, is reported as the action taken date; January actions land in the wrong year.10,14
Income
Applied ≠ relied onEarlier underwriting income is reported instead of the final income relied on; dollars entered as thousands.10,14
Rate spread
Table and lock dateWrong APOR table or rate-set date, or a spread reported on a denied, withdrawn or purchased loan.14
Demographics
Asked, keyed, reportedAnswers not keyed, "not provided" by default, co-applicant fields wrong.10
Geography
Tract and countyAn outdated geocoding file, lost leading zeros, or county and tract from different sources.14
The HMDA calendar
Swipe sideways to see the full diagram.
Annual LAR due by March 1 for the prior calendar year; quarterly reporting applies if you reported at least 60,000 covered loans and applications (excluding purchased loans) for the prior year, due within 60 calendar days of the end of Q1, Q2 and Q3; keep a copy at least three years.3 For 2025 data the window closed on March 2, 2026 because March 1 was a Sunday.7 The 2025 Modified LAR came out on March 31, 2026 and the FFIEC Snapshot on June 23, 2026.5,4
04 · How examiners test it
How examiners test your register against the loan files
In shortFiling is not the end. Examiners pull a random sample of rows and compare every field with the loan file. If one field is wrong too often (four times in a sample of 79, for most lenders), you correct that field on every row and file again.
Filing is not the finish line. Prudential regulators test HMDA data in consumer compliance exams: they pull a random sample of records and check each field against the file. Six numbers frame the stakes.
38%
of all Federal Reserve consumer compliance violations in 2024 were Regulation C. The most-cited items were field-level reporting errors.
Federal Reserve, Consumer Compliance Outlook, 2025.10
72
HMDA violations cited in FDIC exams in 2025, its 5th most-cited regulation. The most common: not providing sufficient data for required fields.
For a LAR of 501 to 100,000 records, four errors in one field across the 79-record sample means correcting that field on every record and resubmitting.
*With fewer than 30 records, the whole LAR is the sample. Source: FFIEC HMDA Examiner Transaction Testing Guidelines, for data collected from 2018, adopted by the CFPB, FDIC, Federal Reserve, OCC and NCUA.8,9
How the two stages work
Stage 1
The initial sample
Examiners check 35 records (for a LAR of 501–100,000). Errors are counted per field. A difference that is "not adequately explained" is an error.
Stage 2
Extend to the total sample
If any field shows 2 errors, they review the rest of the 79, covering every field with at least one error.
Resubmission
Fix the field everywhere
At 4 errors in one field, that field is corrected across the whole LAR and refiled. Examiners can also order it below the threshold, and require program changes either way.8
Demographic fields are grouped: the ethnicity fields for an applicant count as one error group, and so do the race fields. If the LAR comes from several systems, examiners may sample them together, separately, or by risk.8
Penalties for inaccurate HMDA data, as history
When
Regulator
Lender
Amount
What happened
2020
CFPB
A bank, repeat offender
$200,000
Its internal audit had found 40 errors in 100 files, and even its resubmitted data still had errors. Root causes: staffing, training, quality control, oversight.17
Loan officers did not ask demographic questions and recorded that applicants declined.16
2024
CFPB
A nonbank mortgage lender, repeat offender
$3.95 million
Errors in its 2020 data affected nearly 20% of about 700,000 applications across 35 fields; an independent HMDA audit was required for five years.15
Today's climate. In 2025 the CFPB ended HMDA consent orders early19 and narrowed its fair-lending focus to intentional discrimination with identified victims.20 The FDIC and Federal Reserve kept citing HMDA errors in their exams.10,11 Plan for the exam, not the headline fine.
Public data raises the stakes. HMDA data is published at loan level, modified for privacy. Advocates, researchers, journalists and examiners use it for fair-lending and redlining analysis. A wrong tract or a "not provided" that should have been an answer distorts that picture.5
05 · Where the data breaks
Why HMDA edits pass and examiners still find errors
In shortThe automatic checks ask "is this value allowed?", not "is it true?". The fields that break most are dates, income, rate spread, demographics and census tract. The calculator at the end shows how a small error rate becomes a real chance of refiling.
The FFIEC edits check that a value is allowed and plausible. They cannot check that it is true. A wrong action taken date, income from the wrong worksheet, or "not provided" for an applicant who answered usually passes every edit.
The four kinds of HMDA edits
Edit
What it checks
What you do
Catches a valid but wrong value?
Syntactical
File format, filing year, institution identifiers, record counts
Fix and re-upload
No
Validity
Each field holds an allowed value
Fix and re-upload
No
Quality
Unusual values or combinations that may still be true
Confirm or correct
Only if unusual
Macro-quality
The whole register against expected patterns
Explain or correct
Only in aggregate
Edit types and order from the FFIEC Filing Instructions Guide; the FFIEC also offers a free check-digit tool, rate spread calculator and geocoder.7
The fields that break most often
Field
What typically goes wrong
Evidence
Action taken and its date
Withdrawn versus approved-not-accepted; a status date used for the decision date; actions in January filed in the wrong year
Among top Federal Reserve violations in 202410; common edit errors14
Income
Earlier underwriting income instead of the final income relied on; income reported where it should be NA; dollars for thousands
Your procedure choices count too. Regulation C's commentary leaves some reporting choices to the lender, for example which date to report for an application. Write each choice into your HMDA procedure and apply it the same way on every record. Confirm the options in the official commentary to §1003.4(a) check the commentary.
Size your own exam risk
Chance one exam orders a field resubmitted
0%as first entered
0%after every record is checked
0expected errors per field in the total sample, as entered
4errors in one field that order a resubmission
A simplified model of the FFIEC two-stage procedure: errors fall at random, fields are treated as independent, and a resubmission needs the initial threshold of 2 in stage 1. Real exams may sample by system or by risk, and examiners can order resubmission below the threshold.8 Illustrative only, not a prediction.
06 · Section 1071
Section 1071 compliance: what the 2026 final rule requires, and who it covers
In shortSection 1071 is HMDA's small-business cousin. A 2026 rule narrowed it: only lenders with at least 1,000 covered small-business loans in each of the two prior years must collect data, starting January 1, 2028.
Status as of October 2026
Section 1071 of the Dodd-Frank Act requires lenders to collect and report data on credit applications from small businesses, including women-owned and minority-owned businesses. Its purposes are fair-lending enforcement and finding the development needs of businesses and communities. After the 2023 rule was stayed and delayed, the CFPB published a revised final rule on May 1, 2026. It covers far fewer lenders.21
Swipe sideways to see the full diagram.
Not to scale. The rule was published at 91 FR 23530 and took effect June 30, 2026. Compliance begins on a single date, January 1, 2028; 2028 data is due by June 1, 2029. During 2028 the CFPB does not intend to penalize good-faith errors.21,23
Who is covered
A lender with at least 1,000 covered small-business originations in each of the two prior calendar years. The 2023 rule's threshold was 100.
For January 1, 2028, coverage is based on 2026 and 2027; as a transition, a lender may use 2025 and 2026.
A small business has gross annual revenue of $1 million or less (it was $5 million), adjusted for inflation every five years from 2030.
The CFPB estimates about 170 banks, savings associations and credit unions and about 30 online lenders will report; about 1,530 depository institutions covered by the 2023 rule drop out.
Covered lenders hold about 92–93% of small-business loans made by depository institutions, by number of loans. By dollar value the share is lower.21
An advocacy group's analysis counts about 5 credit unions among covered lenders and calls the CFPB's coverage estimate high.25
Products
Covered: loans, lines of credit and credit cards to small businesses.
Newly excluded: merchant cash advances, agricultural lending, and transactions of $1,000 or less.
Still excluded: trade credit, HMDA-reportable loans, insurance premium financing, incidental credit, factoring, true leases, and purchases of loans.
Still covered: the CFPB declined to exclude point-of-sale products, equipment finance and non-owner-occupied commercial real estate.23,24
The data points
Retained
Application date, credit type and purpose
Amount applied for and approved
Action taken
Census tract
Gross annual revenue, NAICS code, time in business
Minority-owned and women-owned status
Principal owners' ethnicity, race and sex; number of owners
Removed
Application method and recipient
Denial reasons
Pricing
Number of workers
LGBTQI+-owned status
Owners' race and ethnicity are now collected in aggregate categories only.24,21
The firewall stays. Under §1002.108, employees and officers involved in a decision on the application may not see the applicant's answers on minority-owned and women-owned status or owners' ethnicity, race and sex. The exception: the lender decides a person should have access for their job, and gives the applicant a notice. The CFPB declined to remove it because it is in the statute.21 Role design is part of getting ready.
Status, as of October 2026. The 2026 rule is final and in effect. Lawsuits over the 2023 rule were reported as paused or stayed in mid-2026.23 The public reporting we reviewed shows no final rulings in them and no challenge filed against the 2026 rule, though one remains possible.26 The CFPB's funding is also in litigation.27 As of its June 2026 update, the CFPB's 1071 page still linked to the 2023-rule filing guide.22check before you plan
Getting ready, whether or not you are covered
Count. Tally covered originations for 2025, 2026 and 2027 under the new definitions: the $1 million revenue line and the merchant cash advance, agricultural and $1,000-or-less exclusions. If you are near 1,000, decide whether to use the 2025–2026 look-back.
Map the fields. Tie each retained data point to its source: the commercial loan system, the credit application, tax returns or financial statements, the decision memo, the geocoder, the NAICS code. Decide when you report stated revenue and when you report verified revenue.
Design collection and the firewall. Decide when and how to ask, including orally. Decide who may see demographic answers, document the "should have access" decisions and notices, and set system roles to match.
Run a practice register. Run a dry-run register in 2026–2027, review its data quality, and train staff before collection starts.
Watch the threshold. Lenders below 1,000 can track volume, and the rule permits voluntary collection.21
07 · How teams check it today
Four ways to check a HMDA register before filing
In shortMost lenders combine two or three of these. The real question is how much of the register gets checked against the documents, and how early in the year.
1 · Required
The filing edits alone
The FFIEC HMDA Platform runs four kinds of automatic checks on every file you upload. Free, and every filer goes through them.
Catches
Bad formats, values that are not allowed, unusual patterns
Misses
A value that is allowed but wrong, like the Ellerys' date
2 · In-house
A sample scrub by your team
Your reviewers pick a sample of rows and open each loan file to compare, often in January and February.
Catches
Errors in the rows that get checked
Misses
Errors in the rest, and patterns nobody has time to trace
3 · Outside
A scrub or audit by a firm
An independent firm reviews a sample or the full register, usually once a year. Fees are not public.
Catches
What its reviewers check, with independence examiners value
Misses
Errors made after the review; the report arrives once a year
4 · Automated
Validation against the documents
Software reads the documents in every loan file, compares every field and lists only the differences for a person to resolve. Several vendors offer it, as software or as a service.
Catches
Allowed-but-wrong values, across the whole register
Watch for
How it shows evidence, who may change data, and whether it follows your procedure
A sample scrub and full validation, side by side
Question
Typical sample scrub
Automated validation against the documents
What is checked?
A sample of rows, often weighted to key fields
Every reportable field on every row
When?
Quarterly, or in January and February before filing
As rows are added through the year, with a full run before filing
Against what?
The file, opened by hand, system by system
The file, with the source page found for each field
Whose rules?
The reviewer's reading of the procedure
Your written procedure as rules, plus the examiners' tolerances
What is left behind?
An error list
Each correction with its evidence, reason and approver
Root causes?
Spotted when someone notices a pattern
Counted by field, branch and loan officer
Whichever way you check, examiners expect a person to stand behind each correction. The FFIEC edits still run, and a person still certifies and files.
Who is involved, and what each cares about
HO
HMDA / CRA officer or data manager
Owns the register: data integrity, scrubbing, the annual filing. Cares about a clean register and no surprises in the exam.
CT
Compliance testing (second line)
Tests, signs off and answers examiners. Cares about evidence for every change.
FL
Fair lending officer
Runs pricing, underwriting and redlining analysis on HMDA data. Cares that the demographic, tract and pricing fields are true.
MO
Mortgage operations and quality control
Fixes the process upstream: status codes, data entry, demographic collection. Cares about knowing which step creates the error.
IA
Internal audit
Tests HMDA accuracy independently. Cares about a trail it can re-perform.
SB
Small-business lending operations
Owns Section 1071 collection and its access rules, if you are covered.
CC
Chief compliance or risk officer
Usually sponsors the change. Cares about exam findings, cost and risk.
Part B · See it working
Botminds HMDA and 1071 Data Validation, one loan at a time
Part A described the problem without naming a product. This part shows how Botminds solves it, at a fictional lender, Kestrel Valley Bank. Every name, number and document in the screens is sample data.
08 · One loan, start to finish
Follow the Ellerys' application through the check
In shortThis is the example from the top of the page, inside the product. Agents do the reading and the matching. A person makes every change, and signs it.
1
Every field is drawn to its page
The agents read the Ellerys' row on the register (86 fields apply to a denied home purchase) and open the 31 documents in their loan file. Each field gets a line to the page it came from.
83 fields match. The application date is one day off, inside the three-day tolerance, so it is marked but not counted. Two fields break, in red: the date of the decision, and income.
Field match. Solid lines match. The dashed amber line is inside tolerance. Each red line opens the document that disagrees, at the passage.
2
Each break is explained, with its sources
The register says March 14, the day the file was moved to Suspended for missing conditions. The credit decision was signed March 20, and the letter is dated March 21. The bank's own procedure reports a denial on the date of the letter, and seven days is outside the tolerance, so an examiner would count it.
The second break: the register shows income of $148,000, but the underwriter relied on $135,000, leaving out a bonus with less than two years' history.
Ask in plain words. "Why does the Ellery action taken date break?" Each fact in the answer links to its document.
3
A person corrects it, and signs
Renata Okafor, the bank's HMDA data manager, corrects the date to March 21. Her reason goes to the audit trail with her name and the time.
The agents never change the register. Only a signed correction does, and a second reviewer confirms any change to an applicant's race, ethnicity or sex.
Correct and sign. The change, the evidence and the reason, signed by a named person.
4
The examiner gets the whole story
Every correction is kept with its before and after values, the page that proves it, the reason and the person who signed. The filing edits then run on the corrected register, and a person files it.
When an examiner asks about the Ellerys' date, the answer takes a minute, not a search through three systems.
Corrections. Before, after, evidence page and signer for every change, ready to export for the examiner.
09 · Every screen
The rest of the work: breaks, error rates, the exam sample and Section 1071
In shortOne loan is the story. These screens show the whole register: every open break, which branches create them, the odds of a refile, and the same check on small-business loans.
Breaks. Every open break in the run, with both values and the source page, ranked by exam weight and ready to assign to a reviewer.
Error rates. Errors as first entered, by field and branch. One branch's decision-date rate stands out, and the finding goes to operations with the records.
Exam sample. Draw the examiner's 79 records as first entered and after the check. The screen uses a simple one-stage estimate.
Filing edits. The four edit types run on the corrected register. Quality edits are confirmed one by one, then the compliance testing manager signs off.
Run log. Each agent step on the record, with its time: the row read, 31 documents opened, the tract recomputed, 83 fields matched, each break explained.
Section 1071 practice register. The same check on a small-business application. Stated revenue is $980,000, but the 2025 tax return shows $1,214,300. Under the 2026 rule that is not a small business, so it leaves the register, with a signature.
Settings. Tolerances follow the examiners' guidelines and can only be tightened. Each field's source document follows the bank's own procedure, with every version kept.
Select any screen to open it full size. Names, the bank, borrowers and businesses are fictional; figures in the screens are illustrative.
10 · Agents, and who signs
Six agents do the reading and matching. Three people make the decisions.
In shortEach agent has one job and one thing it never does. People correct, review and sign; the agents never change the register.
The six agents
Register reader
Reads the HMDA register, and the 1071 practice register, from the origination systems each night: one record per application.
Never edits the system of record.
Loan file reader
Finds the document and page each field comes from: application, credit decision, adverse action notice, Closing Disclosure, appraisal, rate lock, income worksheet, demographic addendum, tax return.
Never guesses a value it cannot find; it marks "no source".
Field matcher
Compares each register value with its source, applies the FFIEC tolerances and the bank's procedure rules, and writes each break with both values and the page.
Never changes a value, or counts a difference inside tolerance as an error.
Rate spread and geocode checker
Recomputes rate spread from the APR and the APOR table for the lock date and loan type, and geocodes each address to state, county and census tract.
Never reports a rate spread where Regulation C says NA.
Filing edit runner
Runs the syntactical, validity, quality and macro-quality edits on the corrected register and drafts the explanations.
Never files; a person submits.
Error-rate monitor
Counts errors by field, branch and loan officer, estimates the odds that an exam sample orders a resubmission, and writes root-cause findings.
Never names a cause without the records behind it.
People decide; everything is cited
MakerHMDA data managerCorrects each break or keeps the register value with a reason, and signs.
Second reviewData reviewerConfirms every change to demographic fields.
CheckerCompliance testing managerSigns off the register; a person files it.
Every break carries both values and a link to the page. The audit trail records agent findings and human decisions separately. Tolerances can only be tightened. Borrower data stays inside the lender's own environment, and reviewers see only the branches they cover.
The people in the screens
Fictional, from the working demo
RO
Renata Okafor · HMDA / CRA / 1071 data manager
Owns the register: 5,318 records so far in 2026. Corrects and signs.
"I can check 300 files a quarter. The examiner only needs to find four bad ones."
TH
Tomás Herrera · data reviewer
Opens loan files in three systems to compare one field at a time.
"Half my time is finding the right page."
PN
Priya Nandakumar · loan operations lead
Fixes the process when the same field keeps breaking.
"Tell me which step creates the error and I'll change it."
GW
Glen Whitaker · compliance testing manager
Signs the filing and answers to examiners.
"Show me every change, who made it and the page it came from."
11 · What goes in and out
Loan files in, a checked register out, with a person and a citation in between
In shortIt reads from the systems you already have and writes nothing back to them. What comes out is a corrected register, the evidence for every change, and the reasons errors happen.
FFIEC HMDA PlatformUpload, edits, certification (by a person)
Home equity and consumer systemsOther sources that feed the LAR
Commercial loan systemSmall-business applications for 1071
Identity and accessRoles, branch scope, the 1071 firewall
Email or ticketingFindings to operations, assignments
For each, check whether it has an API, a file feed, or only screens. Who owns it in IT? Is a conversion or an upgrade coming? Field mappings often drift after an upgrade.
12 · With what you already run
It works alongside your origination system, HMDA software and auditors
In shortKeep your loan system, imaging, HMDA software and outside reviewers. The new piece is a check of every field against the file, with evidence and a signature.
Your existing tools build the register, format it, check it for plausibility and file it. Botminds adds what they do not: a check of every field against the file, with evidence and a signature.
Your origination system and imaging
They stay the system of record. Agents read the extract and the documents and never edit either. Findings go back to operations to fix at the source.
Your HMDA reporting software
Keep it for geocoding, rate spread, edit checks and the filing file. Feed it a register that has already been checked against the files.
The FFIEC HMDA Platform and its free tools
The edits still run, and a person still certifies and files. Validation means fewer surprises there, and explanations drafted for quality edits.
Your outside scrub or audit firm
Independent testing stays valuable. It can shift from finding errors to testing a register that is already checked and documented.
Your fair-lending analytics
Pricing, underwriting and redlining analysis depends on true demographic, tract and pricing fields. Validated data makes the analysis more reliable.
Your 1071 collection module
If you are covered, the collection forms and filing stay with your commercial lending tools. The same field-by-field check runs on the small-business register.
Three principles of agentic HMDA and 1071 validation
1
Check truth, not just format
Every reportable field is drawn to the page it came from, on every record, before filing.
2
Agents find; people sign
Agents read, match and explain. A person corrects and signs, a second reviewer checks demographic changes, and the register is signed off before a person files it.
3
One engine for HMDA and 1071
The same agents and the same discipline run the mortgage register and the small-business register, with roles that respect the firewall.
Built on Botminds
One platform for agents, data and screens
Builder: describe the work, get a spec, build it, test it.
Runtime: agents, data, integrations, screens and models, in one place.
SecureIsolatedGovernedTestedScalableHighly available
Enterprise-ready
Roles and separation of duties; maker-checker on every correction
An audit export an examiner can read
Integrations with origination, imaging, HMDA software and identity systems
Service levels and support
Human approval and citations everywhere
Part C · Are you ready to automate?
Should you automate HMDA data validation? A buyer's view
You own the register, or you sign for it, or you pay for it. This part is written for you: what errors cost today, what automating the check could return, whether your team is ready, and what to ask any vendor, including us.
13 · What errors cost you
What an inaccurate register costs, before you buy anything
In shortStart from your own costs. Once you trust that number, you can judge any solution against it, ours included.
$9–$57
a record, every year: the CFPB's estimate of the ongoing cost of HMDA reporting, from the largest lenders to the smallest. One small institution told the CFPB it spends $68 to $100 a record in staff time.13
CFPB, 2020 rule cost analysis
~417 h
to re-check a single field on 5,000 records at five minutes each, if an examiner orders a resubmission. Refiling and exam follow-up come on top.
Illustrative arithmetic: 5,000 × 5 ÷ 60
$85k–$12M
HMDA penalties from 2020 to 2024, from a community bank to a national bank. Since 2025 the CFPB has stepped back from HMDA enforcement, but the FDIC and Federal Reserve still cite HMDA errors in exams.15,16,18,19,11
Section 4 has each case
Find your own numbers
Cost
Where it hides
How to find your number in a week
Checking files
Reviewers' time, mostly in January and February
Time five file reviews; multiply by the rows you check in a year
Fixing errors
Corrections, re-keying, chasing branches for documents
Count last year's corrections; time ten of them
Outside reviews
A scrub or audit fee; fees are not public
Your contract, and the scope it covers
Refile risk
The chance one field reaches the threshold in an exam sample
Your error rates by field, in the calculator in section 5
Exam time
Finding the file, the page and the person behind each change
Hours spent on HMDA requests in your last exam
Costs no calculator counts
The February crunchA sample squeezed into the weeks before March 1 leaves no time to fix the cause.
Repeat findingsThey escalate. A resubmission or a matter requiring attention brings a deadline and a budget.
Public dataA wrong census tract, or "not provided" for an applicant who answered, distorts how others analyze your lending.5
Next yearCorrecting rows fixes this year. Fixing the step that creates the error fixes next year.
Your best peopleExperienced reviewers spend their time finding pages instead of judging the hard cases.
14 · Your return
What automating the check could return for you
In shortPut in your own numbers. The calculator starts from the CFPB's cost per record and asks you for the rest. Every default marked illustrative is an assumption: replace it with yours.
The value levers
Review labor
Exceptions, not files
Reviewers work only the fields that break, with the page already open, instead of hunting through files.
Every record checked means fewer errors left for the examiner's sample to find.
(odds before − odds after) × cost of one resubmission
Outside spend
A smaller periodic scrub
Outside reviews can shift to independent testing of a register that is already checked.
fee × share no longer needed
Exam time
Evidence ready to hand
A corrections log with page, reason and signer answers examiner questions without a hunt.
exam requests × hours saved per request
Root causes
Fewer errors at entry
Findings by branch and officer fix the step that creates the error, so next year starts cleaner. Fair-lending analysis rests on truer data.
error rate this year − error rate next year
Section 1071
Readiness for covered lenders
A practice register in 2026–2027, then data quality from the first day of collection in 2028.
practice records × breaks found before go-live
The calculator
The starting point is the CFPB's estimate of ongoing HMDA operating cost per record, from its 2020 rule analysis.13 It covers all HMDA compliance tasks, not only review. Every input marked illustrative is an assumption, not a benchmark or a promise. Replace each one with your own number.
Your registerChecking and correctingResubmission riskOutside spend
Illustrative yearly value
CFPB baseline: your yearly HMDA cost$0
Checking and correcting taken on by agents$0
Expected resubmission cost avoided$0
Outside spend$0
Total$0
Staff hours freed a year0
Formulas, per year: Baseline = records × CFPB cost per record Checking = baseline × share checking × share agents take on Resubmission = (odds today − odds after) × (records × minutes ÷ 60 + follow-up hours) × hourly cost Outside = fee × share no longer needed
CFPB profile
Records
Cost a year
Per record
Low complexity
~50
$8,600
~$57
Moderate
~1,000
$43,400
~$43
High complexity
~50,000
$273,000
~$9
Source: the CFPB's 2020 HMDA rule cost analysis. One small institution told the CFPB it spends about $68–$100 per record in staff time.13 The default odds reflect an illustrative register with two fields near 3% as first entered, and the sample math in section 5. Not counted: penalties, exam time, fair-lending and reputational risk, and next year's lower error rate. Weigh the result against the cost of the solution and your rollout.
Start here15 · Readiness check
Are you ready to automate? Nine questions, five minutes
In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go. Each question says why it matters and what a strong answer looks like.
1
Volume
About how many records go on your HMDA LAR in a year?
Why it mattersVolume turns minutes per file into weeks of work, and a larger LAR means a smaller share of it can be checked by hand. In 2023, 42% of HMDA reporters made fewer than 100 closed-end loans.12
A strong answer"About 6,000 records, and we know which systems they come from."
2
How you check today
How do you compare the LAR with the loan files before filing?
Why it mattersA sample checked in a rush leaves systematic errors for the examiner to find. That gap is where agentic validation pays off most.
A strong answer"We sample each quarter and scrub hard in February, and we know what that misses."
3
Your baseline
Do you know your error rate by field, as first entered, from your last scrub, audit or exam?
Why it mattersA baseline shows where the exam risk sits and lets you prove an improvement.
A strong answer"Income runs about 3% and action taken date about 2% as entered, mostly from two branches."
4
Findings
Have examiners, auditors or your own testing found HMDA field errors in the last two cycles?
Why it mattersRepeat findings escalate. A resubmission or a matter requiring attention creates a date and a budget.
A strong answer"Two fields were close to the threshold in the last exam; we had to explain them."
5
Files and systems
Are your loan files imaged in one place, and can your origination system export the LAR fields?
Why it mattersAgents need to reach the documents and the register. Scattered or paper files can still work, but take longer.
A strong answer"Files are in one imaging system with an API, and we export the LAR nightly."
6
Decision owner
Is there a named owner of the register with a budget to change how it is checked?
Why it mattersValidation crosses compliance, mortgage operations and IT. Without one owner it stalls.
A strong answer"Our chief compliance officer sponsors it, and the outside scrub budget can move."
7
Timing
Is there a date that matters: the next exam, the March 1 filing, an origination-system upgrade, 1071 readiness for 2028?
Why it mattersA real date turns interest into a plan. Starting before the year's data piles up keeps the register clean all year.
A strong answer"Our exam is due next year, and we want this year's data checked before March."
8
Governance
Do you have a way to approve AI for compliance work: model-risk or vendor review, human sign-off and an audit trail?
Why it mattersAgents should find and explain, and people should correct and sign, with every step on the record. A clear approval path shortens the road to production.
A strong answer"Our vendor and model-risk teams have a framework, and corrections need a named signer."
9
Vision · for reflection
If an examiner pulled 79 records tomorrow, what would have to be true for you to be sure no field would reach four errors?
Why it mattersYour answer lists the gaps. Each gap is a candidate job for an agent: reading, matching, explaining or counting.
A strong answerA short, specific list, for example: "every income checked against the worksheet, denial dates from the notice, and demographics re-checked."
What your score means
0–6
Explore
Build the baseline first: time a handful of file reviews, tally error rates by field from your last scrub, and name an owner. Start narrow, for example checking one high-risk field such as action taken date on every record.
7–11
Ready to pilot
Run a measured pilot on one quarter's records and your own files. Compare what the agents find with your sample, and have your people sign every correction.
12–16
Ready to implement
Plan the full register: integrations, source rules, sign-off chain, the examiner export and service levels. Then extend the same discipline to 1071 if you are covered.
16 · Questions for any vendor
Sixteen questions to ask before you buy HMDA validation software or a service
In shortAsk every vendor the same questions, us included. The answers show how the work is really done: what gets checked, what evidence you keep, and who is allowed to change your data.
Ask
A good answer
A warning sign
What gets checked
Does it check every field on every row, or a sample?
Every reportable field, every row, as rows are added
"We focus on the key fields" with no way to widen it
Which documents can it read?
Imaged PDFs and scans of every document type in your files, with a list of what it cannot read
Only data it can pull from your loan system
Does it cover every system that feeds the register?
Mortgage, home equity and purchased loans, each mapped
Home equity "added by hand at year end"
What happens when it cannot find a source?
It says "no source" and a person decides
It fills in a likely value
Evidence
For each difference, can I see both values and the page?
One click from the field to the passage in the document
A score or a flag with no page
Could an examiner follow the trail without you in the room?
An export with before, after, page, reason and signer
Evidence lives only inside the vendor's screens
Rules
Does it apply the examiners' tolerances exactly?
Three days on dates, $1,000 on loan amount, income rounding, nothing else, and you can only tighten them
Its own fuzzy matching thresholds
Does it follow our written procedure?
Your source document for each field, as versioned rules you can read
One built-in reading of Regulation C for every customer
Control
Can the software change a register value on its own?
No. It proposes; a named person corrects and signs
"It auto-corrects the obvious ones"
Who reviews changes to race, ethnicity and sex?
A second reviewer, recorded separately
Same as any other field
Can we limit who sees what?
Roles by branch, and the Section 1071 access rules if you are covered
Everyone sees every applicant
Your data
Where does borrower data live, and who at the vendor can see it?
Isolated to your institution, in a region you choose, with access logged
Shared environments, vague answers
Is our data used to train anything?
No, in writing
"Only to improve the service"
Fit and proof
Does it work with our loan system, imaging and HMDA software?
It reads from them and replaces none of them
A migration comes first
Will you run a pilot on our own files, measured against our last scrub?
Yes: errors found that the sample missed, minutes per break, error rate by field
Demo data only
What does the price depend on?
A clear unit, such as rows or applications a year, and what is included
A number that appears only after the contract
17 · How to start
How to start: five steps to a decision you can defend
In shortMeasure first, then pilot on your own files, then decide with numbers. None of these steps commits you to a vendor.
Time five file reviews. Note the minutes each took and how many systems the reviewer opened. That is your baseline.
Pull your error rates by field from your last scrub, audit or exam, and run them through the calculator in section 5.
Gather 20 to 30 redacted loan files and their register rows, with an approval path to share them.
Run a pilot on one quarter's rows. Compare what was found with your own sample, and have your people sign every correction.
Decide with three numbers: errors found that your sample missed, minutes per break, and error rate by field before and after.
The full working-session questions: 29 questions in 9 groups
A
Scope and volume
A1
Which registers do we own: HMDA, CRA, and 1071 if we are covered from 2028? Who owns each?
Why it mattersOne owner and one discipline across registers saves duplicated effort.
Strong answer · warning sign
A named owner for each, working from shared procedures.
Each register is a separate, once-a-year project.
A2
How many LAR records do we file a year, and are we near the 60,000 quarterly threshold or the 500-loan partial exemption line?
Why it mattersYour size sets the exam sample tier, how many data points you report, and how often you file.
Strong answer · warning sign
Exact counts for the last two years.
Nobody is sure whether the partial exemption applies.
A3
How many systems feed the LAR: mortgage, home equity or consumer, commercial, purchased loans?
Why it mattersExaminers may sample each source system separately, and each needs its own mapping.8
Strong answer · warning sign
A list of sources and the share of records each supplies.
Home equity lines are "added by hand at year end."
B
Current process
B1
Take the last record we scrubbed: which systems did the reviewer open, how many pages, and how long did it take?
Why it mattersOne timed record shows the page hunting better than any process map. There is no reliable public figure for review time per file, so measure your own.
Strong answer · warning sign
Minutes per file, timed on a handful of files.
"It depends," with no measurement.
B2
What share of records do we check against the file, which fields, and when in the year?
Why it mattersSampling concentrated in January and February leaves no time to fix root causes before filing.
Strong answer · warning sign
A written sampling plan, run every quarter.
Whatever we get through before March 1.
B3
Is the person who checks independent of the person who prepared the record, and is that documented?
Why it mattersIndependence is what makes a review credible to an examiner.
Strong answer · warning sign
Separate preparer and reviewer, both on the record.
The preparer checks their own work.
B4
Do we pay an outside firm for a HMDA scrub or data integrity review? What does it cost and cover?
Why it mattersIt is often the budget that agentic validation shrinks or reshapes. Fees are not public, so use your own.
Strong answer · warning sign
A fee, a scope, and the error list it produced.
Nobody reads the report until the exam.
C
Errors and root causes
C1
Which fields broke most often in our last scrub, audit and exam, and at what rate?
Why it mattersFields near 5% in a 79-record sample are where resubmissions come from.
Strong answer · warning sign
A field-by-field error rate.
A total error count only.
C2
Do we know why: status dates, income worksheets, rate locks, demographic collection, geocoding?
Why it mattersFixing the cause upstream lowers next year's error rate. Correcting records only fixes this year's.
Strong answer · warning sign
Causes named, with an owner in operations.
The same field breaks every year.
C3
Do findings reach the branch or the loan officer, with the records attached?
Why it mattersOperations can only change a step when they can see the records it produced.
Strong answer · warning sign
Findings by branch and officer, sent routinely.
Errors are corrected quietly in compliance.
D
Documents and procedure
D1
For each field, which document is the source of truth: the notice for a denial date, the income worksheet for income, the lock for the rate-set date?
Why it mattersAgents apply your source rules. Written rules make the check consistent and defensible.
Strong answer · warning sign
A versioned HMDA procedure that names the sources.
Each reviewer decides.
D2
Which documents are imaged, in what quality, and which still live on paper or in email?
Why it mattersCoverage depends on reach. Gaps become "no source" breaks.
Strong answer · warning sign
An inventory by document type.
Adverse action notices are only on paper.
D3
Could we gather 20 to 30 redacted loan files and their LAR rows for a working session within two weeks?
Why it mattersReal files are the fastest way to see what validation finds in your own data.
Strong answer · warning sign
Yes, with an approval path.
No way to share anything.
E
Systems and integrations
E1
How do we export the LAR from the origination system, and when did the field mapping last change?
Why it mattersMapping drift after an upgrade is a common source of field-wide errors.
Strong answer · warning sign
A scheduled export and a mapping review after each upgrade.
A hand-built extract each year.
E2
Which HMDA reporting software do we use for geocoding, rate spread, edits and the filing file, and what does it not check?
Why it mattersKeep it for what it does well. Validation adds the check against the file.
Strong answer · warning sign
Clear on what it checks and what it does not.
"The software validates it."
E3
Can our imaging and loan systems be read through an API or a file feed, or only through screens?
Why it mattersSets integration effort and risk.
Strong answer · warning sign
APIs or an integration layer exist.
A conversion is planned with no date.
F
Section 1071 readiness
Skip this group if you are clearly below the 1,000-origination threshold and do not plan to collect voluntarily.
F1
How many covered small-business originations did we make in 2025, and how many do we expect in 2026 and 2027, under the 2026 definitions?
Why it mattersCoverage from January 1, 2028 depends on these counts.21
Strong answer · warning sign
A count with the exclusions applied.
"We're probably not covered," without a count.
F2
When do we verify gross annual revenue, and from which document?
Why it mattersThe $1 million line decides whether an application is reportable at all.
Strong answer · warning sign
A rule for stated versus verified revenue.
It varies by lender or by region.
F3
Who would see applicants' demographic answers, and have we decided who should have access under the firewall?
Why it mattersThe firewall is statutory and was kept in 2026; it shapes roles in every system.21
Strong answer · warning sign
A role design with documented access decisions and notices.
Underwriters see everything in one screen.
F4
Will we run a practice register in 2026–2027, and who will review its quality?
Why it mattersFinding the gaps in a dry run costs less than finding them in the first filing.
Strong answer · warning sign
A plan with quarterly reviews.
Wait for the filing guide, then start.
G
Governance and risk
G1
Who may correct a register value, who reviews demographic changes, and who signs off before filing?
Why it mattersMaker-checker on corrections is what an examiner will ask about.
Strong answer · warning sign
Named roles, separate from the preparer.
Anyone with access can edit the extract.
G2
If an examiner asked about one changed field, could we show the before, the after, the page, the reason and the signer in a minute?
Why it mattersEvidence ready to hand turns findings into explanations.
Strong answer · warning sign
Yes, from one log.
It would take days across systems.
G3
What does our AI policy require: model-risk review, vendor due diligence, data residency, human approval?
Why it mattersKnowing the approval path early avoids a late stop.
Strong answer · warning sign
A framework and a named reviewer.
"AI isn't allowed," with no path to change it.
H
Decision and success measures
H1
Who would sponsor this, and whose budget would it come from?
Why it mattersCross-team work needs one sponsor.
Strong answer · warning sign
A named executive and a budget line.
No sponsor.
H2
What would a pilot have to show for us to call it a success?
Why it mattersA measured pilot settles the decision.
Strong answer · warning sign
Errors found that the sample missed, minutes per break, error rate by field.
"Let's see what it does."
H3
Have we tried to fix this before, with scripts, a software module or an outsourcer? What happened?
Why it mattersPast attempts show where the hard parts are.
Strong answer · warning sign
Clear lessons learned.
Nobody remembers why it stopped.
I
Vision
I1
What if the register were checked every night, so March 1 were just a sign-off?
Why it mattersTests appetite for a year-round discipline.
Strong answer · warning sign
"February would be ours again."
"We like the crunch."
I2
What if every branch saw its own error rate each month, with the records?
Why it mattersLinks compliance to operations and to fair-lending data quality.
Strong answer · warning sign
Operations leaders want it.
Nobody owns the upstream fix.
I3
Which steps would we never let software do without a person?
Why it mattersTells you where the signatures belong.
Strong answer · warning sign
Corrections, demographic changes, sign-off and filing.
"All of it": then start with finding and explaining.
18 · Frequently asked questions
HMDA data validation and Section 1071: frequently asked questions
What is HMDA data validation?
HMDA data validation is checking each field on the HMDA loan application register (LAR) against the loan file it came from, before filing and before an examiner's transaction test. It goes further than the FFIEC edits, which test format and plausibility. Automated validation reads every record's documents, applies the examiners' tolerances and your own procedure, and shows reviewers only the fields that disagree, with the page.
What is a HMDA scrub or LAR scrub?
A HMDA scrub, or LAR scrub, is a review that compares LAR fields with the loan files before filing. Teams run it in-house or through an outside firm, on a sample or on the full register, often in the weeks before the March 1 deadline. Its findings are corrected before the data is filed.
How is HMDA data validation different from the HMDA edits?
The FFIEC HMDA Platform runs syntactical, validity, quality and macro-quality edits. They catch bad formats, values that are not allowed and unusual patterns. A value that is allowed but wrong, such as a status date reported as the action taken date, usually passes every edit. Validation against the loan file is what catches it.
What is the HMDA resubmission threshold?
Under the FFIEC HMDA Examiner Transaction Testing Guidelines, examiners test a random sample of LAR records. For a LAR of 501 to 100,000 records they start with 35 records and extend to 79 if a field shows 2 errors. Four errors in one field means correcting that field across the whole LAR and resubmitting. Smaller and larger LARs have their own sample sizes and thresholds.
When is the HMDA LAR due, and who files quarterly?
The annual LAR is due by March 1 for the prior calendar year, so 2026 data is due March 1, 2027. Institutions that reported at least 60,000 covered loans and applications, excluding purchased loans, for the prior year also report quarterly, within 60 calendar days after the end of each of the first three quarters.
Who has to report HMDA data in 2026?
A bank, savings association or credit union reports 2026 data if its assets exceeded $59 million on December 31, 2025 and it had a home or branch office in an MSA. It must also have made at least one first-lien home purchase or refinance loan, have a federal connection, and have originated at least 25 closed-end loans or 200 open-end lines in each of the two prior years. A nondepository mortgage lender reports if it has an office in an MSA and meets the same volume test.
What is the Section 1071 compliance date?
As of October 2026, under the CFPB's revised Section 1071 rule (published May 1, 2026, effective June 30, 2026), covered lenders start collecting small-business application data on January 1, 2028 and file 2028 data by June 1, 2029. A covered lender originated at least 1,000 covered small-business loans in each of the two prior years, and a small business has gross annual revenue of $1 million or less.
Is Section 1071 still being challenged?
As of October 2026, the 2026 rule is final and in effect. Lawsuits over the earlier 2023 rule were reported as paused or stayed in mid-2026, and the CFPB's funding is in separate litigation. The rule could still change, so check the CFPB's Section 1071 page before you plan.
What does the Section 1071 firewall require?
Employees and officers involved in a decision on a small-business application may not see the applicant's answers on minority-owned and women-owned status, or on principal owners' ethnicity, race and sex. The exception is when the lender determines a person should have access for their job and gives the applicant a notice. The 2026 rule kept the firewall.
Can AI agents correct HMDA data on their own?
They shouldn't. In a well-governed design, agents find and explain each break with both values and the source page, and a named person corrects and signs. A second reviewer confirms demographic changes, the compliance testing manager signs off the register, and a person files it. Every step is kept on an audit trail.
Does agentic validation replace our HMDA software or outside scrub?
No. HMDA software still geocodes, computes rate spread, runs edits and builds the filing file, and independent testing still has a role. Agentic validation adds a check of every field against the loan file, so those tools and reviewers work from a register that has already been checked.
19 · One-page summary
HMDA data validation: one-page summary
Print this page
The readiness questions
About how many records go on our LAR a year?
How do we compare the LAR with the loan files before filing?
Do we know our error rate by field, as first entered?
Have exams, audits or our own testing found field errors in the last two cycles?
Are loan files imaged in one place, with a LAR export?
Is there a named owner with a budget?
Is there a date that matters: exam, March 1, an upgrade, 1071?
Do we have a way to approve AI, with sign-off and an audit trail?
If an examiner pulled 79 records tomorrow, what would have to be true for no field to reach four errors?
Score 0–2 on questions 1–8: 0–6 explore · 7–11 pilot · 12–16 implement.
Key dates as of October 2026: 2026 HMDA data due March 1, 2027 · Section 1071 collection from January 1, 2028 for covered lenders · first 1071 filing June 1, 2029.
12 terms
LAR
The yearly register of reportable applications and loans; due March 1.
Data points · fields
48 data points in Regulation C; 110 fields in the filing file.
Key data fields
37 fields examiners verify against loan files.
HMDA scrub
A file-to-LAR review before filing, by sample or in full.
Edits
Syntactical, validity, quality, macro-quality: format and plausibility.
Transaction testing
The examiner's random sample checked against the files.
Resubmission
Correct a field on every record and refile; 4 in 79 for 501–100,000.
Tolerance
Dates 3 days, loan amount $1,000, income rounding; nothing else.
Rate spread
APR minus APOR at the rate-set date; NA on denials.
Section 1071
Small-business lending data; 2026 rule, collection from 2028.
Covered lender
1,000+ covered small-business originations in each of two years.
Firewall
Decision-makers don't see demographic answers unless needed and disclosed.
20 · Sources
Sources
Accessed October 2026. Regulatory status statements are as of October 2026.
FDIC FIL-36-2017 and Federal Reserve CA 17-2, revised HMDA resubmission standards. fdic.gov; federalreserve.gov
Federal Reserve, Consumer Compliance Outlook, "Top Federal Reserve System Compliance Violations in 2024: HMDA," 2025, first issue. consumercomplianceoutlook.org
We use essential cookies to make this site work, and cookieless, anonymised analytics to understand how it's used. "Accept all" also turns on Google Analytics, which sets cookies. Privacy policy