Navy Federal mortgage underwriting

What this models

This example runs on the Credit-union-class semi-automated mortgage underwriting network: 12 components and 24 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 1 assumed · 9 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

• baseline

  D48-derived new org (Phase 6, lending-credit-collections). REGISTER FIRST, because it governs every value on this diagram. No court and no regulator has ever found that this credit union discriminated in mortgage underwriting. The quantity this board is built around is an OUTCOME MEASUREMENT over published filings, and it is never a mechanism finding: no biased variable, weight or rule has been identified by anyone, and the algorithm was never put on any record. The February 2026 appellate ruling is a procedural holding about when a district court may decide class certification, and the panel said in terms that discovery might show the complaint's allegations are false. The controlled figures from the December 2023 investigative analysis are journalism, always attributed, with the operator's dispute carried alongside. The commissioned reviewer's under-one-per-cent residual is a claim by the credit union's own defence counsel, resting on data, a model and a report that were never released. The eight-year denial series is primary federal disclosure data, it is UNCONDITIONAL, it controls for nothing whatsoever, and it is not evidence about causation.

• baseline

  TOPOLOGY. Twelve nodes, all documented, none decorative. ONE model because the record describes one at-least semi-automated process over one proprietary algorithm, and because whether that algorithm spans every product is CONTESTED — the appellate majority read the pleading as alleging a single algorithm applied to everyone and the dissent read the same pleading as using the plural elsewhere. TWO input sources because the two feeds the causal dispute actually runs on are separately documented and reach different readers: the consumer credit bureau file, whose single decisive field is excluded by rule from the public record of the same application; and the income and asset verification set, which is where the operator locates its own explanation of the residual and which the underwriters themselves judge complete. THREE record stores because one decision writes three records with three different readerships — the complete file the operator and its supervisors hold, the modified file published to everyone else, and the notice the applicant receives. TWO operator classes because the sources document two groups with different authority over the decision: the underwriters who hold it, and the fair-lending function nothing public describes. The credit union's disclosure reporting staff are a real third group, and their ordinary, entirely compliant work is what splits the record; but their whole documented conduct is producing the full submission and the modified public file from the same rows exactly as the rule requires, so that split is drawn once, as the loan file deriving the published file, and the function is named on that pathway and on both stores rather than drawn as a class that adopts or corrects anything. FOUR reviewers because the record documents four separately constituted external channels with three different data endowments and four different kinds of result, which is the whole case: a commissioned review with the best access and no published method, a supervisor with the complete file that named no institution, an outside channel with the redacted file and no compulsion, and a court channel with compulsion that was withdrawn before it read anything.

• baseline

  ABSENCES ARE DERIVED TOO, and four of them are load-bearing. There is NO model-to-model self-loop, and its absence is the most deliberate choice on this diagram: a monoculture self-loop would assert that one algorithm scores every product for every applicant, which is exactly the proposition the appellate majority and the partial dissent read out of the same pleading in opposite directions, and which discovery would have tested and never did. Drawing it would settle a question the record leaves open. There is NO enforcement node: a denial is a decision, and no downstream action system driven automatically by the record is documented anywhere here. There is NO worklist: the only queue in the record is an appellate dissent's hypothetical that a processing-time decision might turn on the length of a backlog, which is a judge's supposition about a harm nobody measured, and a node drawn from it would be a decoration. There is NO external boundary and no egress pathway: the published disclosure file is a legally mandated publication rather than data crossing a boundary without a guardrail, and it is drawn as the record store it is. There is also no guardrail and no retriever: no automated output screen and no retrieval component appears anywhere in this record.

• baseline

  WHERE THE LAB SHAPE DIVERGES FROM THE PAN SHAPE, and nothing is asserted here that the PAN file does not already record. Five divergences. First, PAN folds the credit bureau file and the verification documents into its complete-application store; the Lab draws them as two input sources with their own pathways, because the dispute is about those two feeds specifically — one carries the field the public record omits and the other carries the operator's stated explanation of the residual. Second, PAN carries the outside analysis and the congressional letters in its governance block as actors with no dynamical authority; the Lab draws that channel as a reviewer, because it read a store and its read is the measurement the whole matter is about. Third, PAN has no edge kind for a check at all. One of the two checks here is a PAN peer edge REDRAWN as a check — fair-lending training reaching the deciding desk — because a channel that corrects a decision is inhibiting in the Lab's vocabulary and reinforcing in PAN's, and its width still comes from PAN on the stated mapping; the other, setting the two complete-file analyses against each other, has no PAN counterpart and is derived from the cited record. Fourth, PAN carries the court in its governance block; the Lab draws the discovery channel as a reviewer with two inbound pathways, both of which say what it actually reached. Fifth, PAN draws the disclosure reporting staff as a class of their own and draws several flows more than once — the full submission reaching the supervisor as a write, a peer route and a read; the officer's write into the notice beside the decision's; the appeal returning to the officer beside the appeal returning to the decision; the two review channels reporting to compliance beside their routes to the stack; and a thin, explicitly uncertain route from the underwriter into the stack. The Lab draws each of those flows once and narrates the rest on the pathway that carries it.

• baseline

  BASELINES, and exactly how far the PAN org carries them. The PAN entry for this deployment holds twenty-five edges. Fourteen of this network's twenty-four pathways have a one-to-one counterpart among them, and every one of those fourteen mirrors that edge's width on a single rung mapping (0.50 and above to 3, 0.30 to 0.49 to 2, 0.06 to 0.29 to 1, documented absent to 0) with no exceptions, including the peer edge redrawn here as a check. The other eleven PAN edges describe flows this board draws once rather than twice, and each is narrated on the pathway that absorbs it, which keeps its own width. The remaining ten pathways are derived from the cited record directly and each says so on its own line. Four contrasts are load-bearing. The widest read into a review channel runs into the redacted store, not the complete one: the outside analysis reads the published file at the top rung, which is the whole of its evidence, while the pathway that would let it check that file against its source runs at zero. The two channels with complete-file access also read at the top rung, and both of their outbound pathways to the stack run at the low rung, which is this record's shape in two numbers. The appeal pathway runs at the low rung, and the value is a finding rather than a guess: a correction route that costs the applicant a second hard credit inquiry is a route the one detailed account in the record describes being abandoned. And the three routes from the review and compliance channels that could act on the stack all sit at the low rung while the reads into the decision and into the complete-file channels sit at the top, which is what a governance loop that ran to completion without answering its own question looks like when it is drawn.

• baseline

  DRAWN AT THE COARSEST GRANULARITY THE RECORD SUPPORTS. This board was first derived at thirty-six pathways and has been re-derived at twenty-four without dropping a documented fact: every flow the record describes is still on it, drawn once. Where one flow had been drawn two or three times, the survivors now carry the rest in their own words. The full submission reaching the supervisor is one read of the complete file, and the supervisor's occasional use of the coarsened copy is narrated there. Decisions reaching the compliance function travel by way of the loan files they are written into. The underwriter's statement of a denial reason and handling of any complaint are narrated on the notice the decision writes. The appeal returning first to the people who decided is narrated on the one appeal pathway. The review channels reporting back to the credit union's compliance function are narrated on their routes to the stack. The compulsory channel, withdrawn before it read anything, is one pathway at zero rather than two. No width moved in the process.

• baseline

  DEMAND 3 / CAPACITY 3, and the pair is a reading rather than a default. Demand 3 rests on primary federal disclosure data queried directly for this institution: for filing year 2022, 43,765 originated and 19,635 denied applications from white applicants and 12,997 originated and 13,410 denied from Black applicants, across all reportable loan types and purposes — roughly ninety thousand decided applications from two cohorts alone in one year, against the operator's own annual-report figures of almost 50,000 closed mortgage loans totalling $16.5 billion and an $84.3 billion portfolio. Capacity 3 rests on three documented facts and on the deliberate absence of a fourth. The process is described on the record as at-least semi-automated, so the human comparator is not hypothetical: a human underwriter holds the decision by the operator's own account. The institution reports 25,200 employees, $197.1 billion in assets and $18.9 billion in equity as of 31 December 2025, with a dedicated real-estate lending division. And the applicant population is unusually well documented, because the field of membership covers the defence department, all uniformed services, veterans and their families, so incomes are documented and often federally guaranteed. What the record contains no trace of is a backlog figure, a staffing shortfall or an alert flood. The pair therefore says what the whole file supports: the failure here is not throughput and not resourcing.

• baseline

  EVIDENCE STATUS, labelled where it is used, because this record mixes tiers harder than most and the tiers are the case. PRIMARY GOVERNMENT DATA, queried directly: the eight-year denial-share series and the 2022 application volumes, from the federal disclosure browser for this institution's legal entity identifier — unconditional, all products, controlling for nothing. PRIMARY COURT RECORD: the published appellate opinion of 9 February 2026, cross-verified against a government-hosted copy, and the district docket including the 30 May 2024 order and the 25 August 2026 dismissal entries. PRIMARY GOVERNMENT: the supervisor's own economics research note, the 4 September 2025 disparate-impact removal and Letter 25-CU-04. CONGRESSIONAL PRIMARY: four letter campaigns, including a full-text ten-question demand with a deadline. JOURNALISM, ALWAYS ATTRIBUTED AND ALWAYS WITH THE OPERATOR'S DISPUTE ALONGSIDE: the controlled approval-rate analysis and its published caveats, both of whose articles return a legal-block status to the verifying environment and every figure of which is corroborated through the appellate opinion, the congressional letters, the pleading and a verified syndication. PLEADING, ALLEGATIONS ONLY: the proprietary-algorithm description, the proxy-for-race theory and the lead plaintiff's account of the appeal friction. OPERATOR TIER: the scale figures, the response of 18 December 2023, the statement of 21 March 2024 and the under-one-per-cent residual, which is a claim by defence counsel in the same case and rests on a report, a model and data that were never released. No parameter on this diagram is set by the operator's figure or by the journalist's figure; both are recorded description.

• baseline

  THE STRUCTURAL FACT THIS BOARD IS DRAWN AROUND, stated as a property of the record rather than as a measurement. Two records come out of one loan file and they are not the same record. The complete submission carries the applicant credit score, the debt-to-income ratio and the loan-to-value; the published modified file excludes the credit score by rule and coarsens other fields. That single suppression decides who could say what. The party with the redacted file published a controlled analysis and its own caveats. The party with the complete file and an interest in the answer published a conclusion without the analysis. The party with the complete file, the strongest model and the supervisory authority published an industry aggregate that names no institution. Congress supplied volume and no compulsion; the one request whose compliance is documented went unmet. And the only compulsory instrument opened in February 2026 and was withdrawn by the parties in August 2026. Nothing on this diagram computes a harm to any applicant, and nothing could: the reconciliation that would close the question is drawn at zero because no rule permits it, not because anyone declined to run it.

• assumed

  Applicants are not in the dynamics. No approval, denial, price, rate or household outcome for any person is computed from anything drawn here, and no score over any person is authored anywhere in this network. The denial shares, the approval rates, the odds ratios and the dollar figures are recorded external observations and they set no parameter. The credit union is member-owned with no outside shareholder, so its governing principals are the fifteen point three million members who are also its applicants; that is a real governance fact and it is not a reason to put a served person inside the dynamics, and this board does not. The operator's counter-framing — that it ranks first among large lenders in the share of mortgages made to Black borrowers and that roughly one in four members is Black — is operator-tier, is materially true as a volume statement, and is orthogonal to the approval-rate question; both are carried and neither erases the other.

What this example does not show

• Litigation and regulatory posture, carried exactly as the dossier states it. Oliver v. Navy Federal Credit Union, No. 1:23-cv-01731-LMB-WEF (E.D. Va., filed 17 December 2023), on appeal as No. 24-1656 (4th Cir.). On 30 May 2024 the district court dismissed the disparate-treatment theory for failure to allege plausible evidence of discriminatory intent, preserved the disparate-impact theory, dismissed four other counts and struck all class allegations before any discovery. On 9 February 2026 the Fourth Circuit, two to one, affirmed the denial of a Rule 23(b)(3) damages class and vacated the denial of a Rule 23(b)(2) injunctive class, holding only that a pre-discovery strike was premature and expressly reserving the merits. On 25 August 2026 all nine named plaintiffs filed a notice of dismissal WITH PREJUDICE, so-ordered the same day; the 28 August status conference was cancelled. The docket states no reason and discloses no settlement. The correct current posture is: concluded by voluntary dismissal with prejudice, merits never reached, no class ever certified. Do not assert that the case settled, that money changed hands, or that either side conceded anything.
• NO FINDING OF ANY KIND EXISTS, and nothing on this board should be read as one. No court, regulator or agency has found that this credit union violated any fair lending law in its mortgage underwriting. The February 2026 appellate ruling is a Rule 23 procedural holding about when a district court may decide certification, and the panel wrote that discovery might show the complaint's allegations are false. The complaint's account of the algorithm, and its theory that some collected data can be proxies for race, are allegations that were never proved and never tested.
• The disparity on this board is an OUTCOME MEASUREMENT and never a mechanism finding. No biased variable, weight or rule has been identified by anybody, because the underwriting algorithm was never put on any public record. This is a reporting-based case, not an enforcement-based one, and the difference is the whole point of drawing it.
• The eight-year denial series computed from the federal disclosure browser is UNCONDITIONAL. It covers all loan types and purposes and controls for nothing whatsoever — not credit score, not income, not loan-to-value, not product mix. It is a legitimate primary-data anchor for volumes and for the persistence of the raw gap and it is not evidence about causation. The 2025 filing year is the most recent available and may be revised; treat that row as provisional. The controlled 2022 conventional-purchase figures are a separate object and are journalism, always attributed, always with the operator's dispute alongside.
• The under-one-per-cent residual is a claim by the credit union's commissioned reviewer, who is a partner at the firm defending it in the same litigation. It rests on data, a model and a report that were never published, and when a congressional body asked for even aggregate supporting numbers the letter records that none were provided. It should never be presented as a finding and never without the dual-role fact attached. Equally it should not be dismissed: that reviewer had access to underwriting variables the public data genuinely lacks.
• The supervisor's own odds ratios — Black applicants at about 1.60 times the denial odds of white applicants on conventional purchase, 1.54 on rate-and-term refinance and 1.81 on cash-out refinance — are INDUSTRY-WIDE across credit unions on 2020 and 2021 data. The note names no institution and separately estimates none, and it carries the caveat that such results should not be interpreted as evidence of discrimination because they may reflect unobserved factors. Attaching an industry figure to one deployment would be exactly the misattribution the supervisor took care to avoid.
• The causal ambiguity is unresolved and is carried unresolved. The record contains two live theories of where any disparity would come from: a single automated underwriting algorithm applied to everyone, which is the appellate majority's reading of the complaint, and dispersed loan-officer discretion exercised in differing ways, which is the partial dissent's alternative. Nothing on the public record settles which, and the discovery that would have is gone.
• Do not conflate the two supervisory threads. The consumer bureau's November 2024 order requiring more than $95 million and its July 2025 termination concern surprise overdraft fees, an entirely separate matter from mortgage underwriting. They belong here as evidence about a supervisor's posture and capacity, never as evidence about lending discrimination.
• Both articles of the originating investigation return a legal-block status to the verifying environment and could not be read directly. Every figure attributed to them is corroborated by at least one verified independent source that quotes them: the Fourth Circuit's opinion, the March 2024 congressional letter, the Senate Banking release, the individual-senator release, the pleading, or a verified syndication of the same wire story.
• The operator's counter-framing is carried and is not erased: it says it ranks first among large lenders in the share of mortgages made to Black borrowers, that roughly one in four of its members is Black, and that it made $3.5 billion in 2022 mortgages to Black borrowers. Those are operator-tier volume statements, they are materially true as such, and they are orthogonal to the approval-rate question. Carry both.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

• Navy Federal Credit Union underwrites residential mortgages through what the pleadings and the Fourth Circuit both describe as an 'at-least semi-automated underwriting process' built on a 'proprietary underwriting algorithm' whose variables and weights are, in the complaint's words, 'entirely up to Navy Federal,' which 'maintains secrecy' over both; no third-party underwriting vendor is identified anywhere in the verified record. CNN reported on 14 December 2023 that for 2022 conventional home purchase mortgages the credit union approved 77 per cent of White applicants, 69 per cent of Asian applicants, 56 per cent of Latino applicants, and 48 per cent of Black applicants — a spread of nearly 29 percentage points, the widest of the fifty lenders that originated the most mortgages that year — and that holding more than a dozen variables constant Black applicants were more than twice as likely to be denied as White applicants and Latino applicants roughly 85 per cent more likely. Navy Federal disputes that analysis, saying the statistics 'do not appear to have considered several key credit criteria,' and adds that it ranks first among large lenders in the share of mortgages made to Black borrowers, made $3.5 billion in 2022 mortgages to Black borrowers, and counts roughly one in four members as Black. Separately, and unconditionally, the CFPB's own HMDA Data Browser queried directly for this institution (LEI 5493003GQDUH26DNNH17) across all reportable loan types and purposes gives denial shares of originated-plus-denied applications, White against Black, of 23.8 and 45.0 per cent in 2018, 31.0 and 50.8 in 2022, 34.5 and 56.6 in 2023 and 27.4 and 46.8 in 2025; those counts control for nothing whatsoever, and in particular not for credit score, which the public loan-level file excludes by rule. No court and no regulator has ever found that Navy Federal discriminated in mortgage underwriting.

  empirical
  • Government United States Court of Appeals for the Fourth Circuit (2026, February 9). Oliver v. Navy Federal Credit Union, No. 24-1656 (published opinion, 47 pages) https://www.ca4.uscourts.gov/opinions/241656.P.pdf
  • Investigative Tolan, C., Ash, A., & Marsh, R. (2023, December 14). The nation's largest credit union rejected more than half its Black conventional mortgage applicants. CNN (URL returns HTTP 451 to this environment; every figure corroborated through the Fourth Circuit opinion, the congressional letters and the pleading rather than read directly) https://www.cnn.com/2023/12/14/business/navy-federal-credit-union-black-applicants-invs/index.html
  • Government Consumer Financial Protection Bureau and Federal Financial Institutions Examination Council (2026). Home Mortgage Disclosure Act Data Browser aggregations for Navy Federal Credit Union (legal entity identifier 5493003GQDUH26DNNH17), filing years 2018 to 2025 https://ffiec.cfpb.gov/data-browser/
  • Vendor Navy Federal Credit Union (2023, December 18). Navy Federal Credit Union Responds to Allegations Concerning Its Home Lending Practices https://www.navyfederal.org/about/press-releases/2023/navy-federal-responds-to-home-lending-allegations.html

• One loan file generates two records with different contents, and the difference decides who could say what. The full Home Mortgage Disclosure Act submission that goes to the regulator carries the applicant credit score, the debt-to-income ratio, and the loan-to-value; the public modified loan-level file suppresses the applicant credit score by rule and coarsens other fields. CNN published that limit on its own instrument in the same article that carried its findings, and the Fourth Circuit's opinion recites it: applicant credit score, available cash deposits, and relationship history with the lender are not available in the public mortgage data. The NCUA Office of the Chief Economist's research note documents the same split from the other side, describing which credit-risk fields the regulatory file contains and applying a federal logit model to them. The consequence is structural rather than rhetorical: an analysis built on the published file can be answered, accurately and unfalsifiably, by pointing at the variable it could not observe, and that answer cannot itself be checked, because the file it rests on is not published either. When Congressional Black Caucus members asked Navy Federal in February 2024 for 'at minimum, aggregate data ... regarding credit scores or any other non-public variable that Navy Federal has suggested serves as an explanation' for the gap, the March 2024 congressional letter records that 'Navy Federal failed to provide this information.'

  empirical
  • Government United States Court of Appeals for the Fourth Circuit (2026, February 9). Oliver v. Navy Federal Credit Union, No. 24-1656 (published opinion, 47 pages) https://www.ca4.uscourts.gov/opinions/241656.P.pdf
  • Government National Credit Union Administration, Office of the Chief Economist (2022). Observations on Credit Unions' Mortgage Lending to Minority Borrowers (research note) https://ncua.gov/files/publications/analysis/observations-credit-unions-mortgage-lending-minority-borrowers.pdf
  • Government Cleaver, E., Horsford, S., & Kamlager-Dove, S. (2024, March 1). Letter to the Chairman of the National Credit Union Administration and the Director of the Consumer Financial Protection Bureau https://horsford.house.gov/sites/evo-subsites/horsford.house.gov/files/evo-media-document/Navy_Federal_Regulator_Signature.pdf
  • Government Consumer Financial Protection Bureau and Federal Financial Institutions Examination Council (2026). Home Mortgage Disclosure Act Data Browser aggregations for Navy Federal Credit Union (legal entity identifier 5493003GQDUH26DNNH17), filing years 2018 to 2025 https://ffiec.cfpb.gov/data-browser/

• Three analyses of this lending exist, each run from a different slice of the data, and no forum ever set one against another. CNN's was built on the public file with the credit score removed by rule, and published its own caveats. Navy Federal's was commissioned from Debo P. Adegbile and reported complete on 21 March 2024: the credit union announced 'no race-based decision making' once 'all non-public underwriting factors are accounted for — including credit score, income verification, debt-to-income ratio, and incomplete credit applications,' and Adegbile stated that 'when all relevant factors are controlled for, which CNN did not do, the difference in approval rates between Black and White borrowers falls to less than 1%,' with the remainder 'explained by legitimate, non-race factors like income verification and incomplete credit applications.' The report, the model, and the underlying data have never been released; Adegbile is a partner at WilmerHale, which is Navy Federal's defense counsel in the same class action and argued the appeal for it; plaintiffs' counsel called the arrangement 'a classic conflict of interest'; and the credit union nonetheless described the work as an 'external review.' The third analysis is the NCUA Office of the Chief Economist's, applying the FDIC's Popick logit model to 2020 and 2021 HMDA data including the credit-score, debt-to-income, and loan-to-value fields the public file suppresses, and reporting credit union denial odds against White applicants of about 1.60 for Black applicants on conventional purchase, 1.55 for Hispanic applicants and 1.60 for Asian applicants, 1.54 on rate-and-term refinance and 1.81 on cash-out refinance, with average marginal effects of two to four percentage points and contract-rate premiums of eight to thirteen basis points — published as an INDUSTRY-WIDE aggregate that names no institution and separately estimates none, and carrying the caveat that such differences 'should not be interpreted as evidence of discrimination ... as such results may reflect unobserved factors.' Congress supplied volume and no compulsion: at least four letter campaigns, a chief-executive meeting requested by forty members, and a ten-question demand of 1 March 2024 asking what matters requiring attention had issued in five years, how fair lending findings had affected ratings, whether a referral to the Department of Justice had been made, and how each agency ensures a lender searches for a less discriminatory alternative — a test Consumer Reports had put to the Bureau ten weeks earlier — with answers requested by 5 April 2024. No public response has been located.

  empirical
  • Vendor Navy Federal Credit Union (2024). Statement on Conclusion of External Review, submitted for the record, hearing of the U.S. House Committee on Veterans' Affairs, June 12, 2024 (operator statement hosted in the congressional record) https://docs.house.gov/meetings/VR/VR10/20240612/117409/HHRG-118-VR10-20240612-SD011.pdf
  • Vendor Navy Federal Credit Union (2024, March 21). Statement from Navy Federal Credit Union on Conclusion of External Review https://www.navyfederal.org/about/press-releases/2024/navy-federal-statement-conclusion-of-external-review.html
  • Investigative WRAL, carrying the CNN wire (2024, March 23). Navy Federal says external review finds 'non-race factors' explained mortgage approval disparities https://wral.com/story/navy-federal-says-external-review-finds-non-race-factors-explained-mortgage-approval-disparities/21344125
  • Government National Credit Union Administration, Office of the Chief Economist (2022). Observations on Credit Unions' Mortgage Lending to Minority Borrowers (research note) https://ncua.gov/files/publications/analysis/observations-credit-unions-mortgage-lending-minority-borrowers.pdf
  • Government Cleaver, E., Horsford, S., & Kamlager-Dove, S. (2024, March 1). Letter to the Chairman of the National Credit Union Administration and the Director of the Consumer Financial Protection Bureau https://horsford.house.gov/sites/evo-subsites/horsford.house.gov/files/evo-media-document/Navy_Federal_Regulator_Signature.pdf
  • Advocacy Consumer Reports (2023, December 20). CR urges CFPB to investigate Navy Federal Credit Union for unfair mortgage lending and hold lenders using algorithmic scoring accountable for preventing discrimination https://advocacy.consumerreports.org/press_release/cr-urges-cfpb-to-investigate-navy-federal-credit-union-for-unfair-mortgage-lending-and-hold-lenders-using-algorithmic-scoring-accountable-for-preventing-discrimination/

• Oliver v. Navy Federal Credit Union, No. 1:23-cv-01731-LMB-WEF, was filed in the Eastern District of Virginia on 17 December 2023; related suits were consolidated and the operative complaint of 20 February 2024 named nine plaintiffs pleading the Fair Housing Act, the Equal Credit Opportunity Act, 42 U.S.C. section 1981 and state analogues. On 30 May 2024 Judge Leonie Brinkema dismissed the disparate-treatment theory because 'the Complaint has failed to allege plausible direct or circumstantial evidence of discriminatory intent,' dismissed four further counts, preserved the disparate-impact theory on the ground that at the motion-to-dismiss stage 'the statistical disparities reveal a disparate impact among non-white loan applicants and the underwriting algorithm and process is alleged to have caused the disparity,' and struck all class allegations before any discovery. On 9 February 2026 the Fourth Circuit decided No. 24-1656 in a published two-to-one opinion: Rule 23(c)(1)(A) rather than Rule 12(f) or Rule 23(d)(1)(D) is the source of a district court's authority to decide certification, and before discovery a court may deny certification only if the class allegations fail as a matter of law on their face. It affirmed the denial of a Rule 23(b)(3) damages class and vacated the denial of a Rule 23(b)(2) injunctive class, holding the complaint made 'a sufficient prima facie showing' of commonality and that the district court 'acted prematurely,' and it reserved the merits expressly: 'Of course, discovery in this case might show that the [complaint's] allegations ... are false.' Judge Richardson, concurring in part and dissenting in part, would have affirmed the strike, arguing the complaint uses the plural elsewhere, says nothing about how the algorithm and loan officers interact, and that the disparity 'may be caused not by any underwriting algorithm, but by the individual loan officers exercising their discretion in differing ways.' On 25 August 2026 all nine named plaintiffs filed a notice of dismissal with prejudice, which the district court so-ordered the same day; the 28 August status conference was cancelled. The docket text states no reason and discloses no settlement. The correct posture is: concluded by voluntary dismissal with prejudice, merits never reached, no class ever certified, and no finding of discrimination by any court.

  empirical
  • Government United States Court of Appeals for the Fourth Circuit (2026, February 9). Oliver v. Navy Federal Credit Union, No. 24-1656 (published opinion, 47 pages) https://www.ca4.uscourts.gov/opinions/241656.P.pdf
  • Government U.S. Government Publishing Office (2026). USCOURTS-ca4-24-01656-0: Laquita Oliver v. Navy Federal Credit Union (slip opinion) https://www.govinfo.gov/content/pkg/USCOURTS-ca4-24-01656/pdf/USCOURTS-ca4-24-01656-0.pdf
  • Government Oliver v. Navy Federal Credit Union, No. 1:23-cv-01731 (E.D. Va.), docket, via CourtListener and the RECAP Archive (HTML docket page returns HTTP 403 to automated clients; verified through the public search interface) https://www.courtlistener.com/docket/68095879/oliver-v-navy-federal-credit-union/
  • Advocacy Oliver and Jacob v. Navy Federal Credit Union (E.D. Va., filed December 17, 2023). Class Action Complaint (pleading; allegations only) https://bencrump.com/wp-content/uploads/2023/12/231217-Oliver-et-al-v-NFCU-Complaint.pdf
  • Trade press ABA Banking Journal (2026, March). Fourth Circuit revives class action challenging Navy Federal's mortgage lending practices https://bankingjournal.aba.com/2026/03/fourth-circuit-revives-class-action-challenging-navy-federals-mortgage-lending-practices/

• The supervisory test moved while the dispute was live. Executive Order 14281 of 23 April 2025 directed federal agencies to eliminate the use of disparate-impact liability, and on 4 September 2025 the National Credit Union Administration issued Letter to Credit Unions 25-CU-04 removing every reference to disparate impact from its Fair Lending Guide and other issuances and stating that its 'examination and supervision processes will no longer include reviews for disparate impact,' while continuing HMDA analysis and examinations for disparate treatment. The theory the district court dismissed in Oliver is the one the supervisor kept; the theory that survived dismissal and was revived for class treatment on appeal is the one the supervisor stopped examining for. In the same window the consumer-compliance supervisor's posture toward this institution reversed on an entirely separate matter: the Consumer Financial Protection Bureau ordered Navy Federal on 7 November 2024 to pay more than $95 million — $80.6 million in redress and a $15 million penalty — over surprise overdraft fees charged between 2017 and 2022, and on 1 July 2025 the Bureau terminated that order and waived any alleged non-compliance, a decision House Financial Services Democrats questioned in an August 2025 letter to the credit union's chief executive, which records Navy Federal saying it 'firmly believe[s] the CFPB's decision to terminate the order was appropriate.' The overdraft matter concerns deposit-account fees and is not evidence about mortgage lending; it is recorded here as evidence about a supervisor's posture and capacity.

  empirical
  • Government National Credit Union Administration (2025, September 4). NCUA Disparate Impact References From Fair Lending Guide and Other Materials https://ncua.gov/newsroom/press-release/2025/ncua-disparate-impact-references-fair-lending-guide-and-other-materials
  • Government National Credit Union Administration (2025, September 4). Removal of Disparate Impact (Letter to Credit Unions 25-CU-04) https://ncua.gov/regulation-supervision/letters-credit-unions-other-guidance/removal-disparate-impact
  • Government Consumer Financial Protection Bureau (2024, November 7). Enforcement action page: Navy Federal Credit Union (overdraft), order terminated July 1, 2025 https://www.consumerfinance.gov/enforcement/actions/navy-federal-credit-union-overdraft-2024/
  • Government Democratic staff, U.S. House Committee on Financial Services (2025, August 28). Letter to the President and Chief Executive Officer of Navy Federal Credit Union regarding the termination of the 2024 consent order https://democrats-financialservices.house.gov/uploadedfiles/ltr-2-nfcu-08.28.2025.pdf

Where this connects

Institutional pressures in this domain

• Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
• Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.
• Data & policy drift — The world, the intake process, and the rules change under a system trained on how things used to be — two mechanisms with different remedies: the statistical properties of what the system processes move (concept drift), or the mixture of inputs arriving in deployment differs from the mixture it was trained on (covariate shift).
• Austerity & recovery incentives — Cost-cutting and overpayment-recovery targets tilt the system toward denial and enforcement errors.
• Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.

All of them in context on the Lending & credit collections AI domain page.

Levers available here and the patterns behind them

• Upgrade model — Improve the model
• Check with a second model — Cross-model verification
• Check copied records — Reconcile copied records
• Review on schedule — Oversight cadence & retrospectives
• Understand the system — Understand the system
• Peer sharing rules — Peer-edge governance
• Mark AI-written records — Provenance labeling

Documented case histories

• Navy Federal mortgage underwriting & three readings of one gap
• Automated underwriting with its fair-lending testing on the record
• Cleared on the numbers but faulted on the explanation
• The governance an enforcement action had to write
• M-Shwari & Kenya's Digital Credit Market
• Citi Retail Services Judgmental Review & the Armenian surname screen
• Santander Consumer USA subprime vehicle loan scoring
• Credit Acceptance Corporation's net-collections score
• Wells Fargo refinance underwriting & the bridge nobody could build
• Enova International servicing defects & the debits nobody authorised
• Equifax Online Model Server coding error (2022)
• TransUnion's OFAC Name Screen & the people who could not sue
• Dave ExtraCash: an advertised ceiling, an automated amount, and a case that never asks how the amount is set
• Hello Digit's automated-savings algorithm
• Oportun's legal-collections filing pipeline