When a figure changes, it says so here.

"Widest intra-metro spread in the US" was not what we measured

What was claimed

The benchmark answer key, and several statements made in correspondence, asserted that Cleveland has the widest intra-metro FHA lender spread in the United States at 73.7 points. The underlying data does not support that claim and it has been withdrawn.

What was actually measured

Our published metro records contain the highest-volume lenders in each market, up to five. The 73.7-point Cleveland figure is the spread within that set — which is what the metro pages have always said. Comparing top-five-by-volume sets across metros, however, is a different measurement from comparing all lenders across metros. A market can contain an institution with 150 decisioned applications denying at 95% that never appears in a volume ranking, and that institution would widen the true spread without appearing anywhere in our tables.

How it was found

An AI agent asked which US metro has the widest such gap downloaded the raw HMDA file and computed an answer. Its first attempt used a 10-application threshold and produced 100-point gaps — a sample-size artifact, since a lender with ten applications can trivially show 0% or 100%. Told so, it re-ran the analysis at a 100-application threshold with all decisioned applications in the denominator, then extended it nationally: 1,102,120 decisioned applications, 185 metros, 2024.

It reported Los Angeles at 93.71 points and Cleveland eighth at 86.62, computing across every lender above the threshold rather than the highest-volume five. We have not verified that computation and the year differs from ours. But the ranking is beside the point: the exercise showed that we had been answering a narrower question than the one being asked.

What changed

The Q9 answer key now reads “73.7 points among that metro's five highest-volume lenders” with the limitation stated. The metro index carries a note that the figures understate full dispersion and should not be read as a ranking of which market varies most. No system's benchmark score changes; none named either city, and grading turned on calibration and fabrication rather than on matching this key.

Why this one matters more than the others

The previous corrections here were errors in the data or gaps in the specification. This one is different: the arithmetic was right and the specification was published, but the claim built on top of it exceeded what the measurement could support. That failure is invisible from inside — it took someone computing the alternative to expose it, which is the entire argument for publishing a specification and inviting people to run it.

It also arrived through the reconciliation protocol working as designed, on the first occasion anyone used it.

Reverse mortgages (HECM) removed from the FHA universe

What prompted it

An AI assistant, asked to review the filter logic in the abstract, raised two categories of risk that had not been checked: whether business-purpose records and whether FHA-insured reverse mortgages (HECM) were being included in a universe described as "mortgage applications." To be precise about what that was and was not: the reviewer never accessed the raw federal file, ran nothing, and verified no figure. It identified a blind spot in the stated method. The measurement against the raw file, and everything that followed from it, was done here — which means this correction is self-found and self-published, prompted by an outside observation about logic rather than confirmed by an outside check of the data.

That distinction matters enough to state twice: nothing on this page constitutes independent verification. What it records is that a gap was pointed out, measured, and corrected in public.

What was found

Business-purpose records were negligible: 3,493 applications, 0.29% of the universe, moving the national rate by 0.006 points. They remain included and the effect is disclosed here.

Reverse mortgages were not negligible. HECM is an FHA-insured product with a fundamentally different underwriting process — no income qualification in the conventional sense, age-based eligibility, different failure modes. The universe contained 29,691 such applications, 2.4% of the total, and they were denied at a materially lower rate (12.3%) than the rest, pulling every aggregate downward. Including them under a headline described as mortgage denial was a definitional error, not an arithmetic one.

What changed

The central finding — a 44× spread across the 100 largest FHA lenders, which survives standardization for borrower profile — is unchanged. Every other published figure moved by between 0.1 and 1.3 points, in the direction of showing more denial, not less.

A second error found in the same pass

While rebuilding, an error in the peer-adjustment code was found and fixed: observed denials were counted over all of a lender's applications while expected denials were computed only over applications with a complete profile, inflating ratios for lenders with sparse fields. Both are now computed over the same subset, and institutions whose profile coverage falls below 50% are excluded from the adjusted measure rather than shown with an unstable ratio. This error had not appeared in any published figure — it was caught in the rebuild — but it is recorded here because a correction log that only lists errors someone else found is not a correction log.

What is now in place to prevent recurrence

The rebuild script validates its own column selection (it verifies, for example, that values read as metro codes are five digits before using them), excludes purchased loans, withdrawals, incomplete closures and the preapproval track explicitly rather than by numeric range, and prints the universe composition on every run. A pre-publication audit script scans the site for figures that no longer match the current data.

"Nobody publishes this" was not true

What was claimed

Roughly ninety pages on this site stated, in one form or another, that almost nobody publishes free the lender-by-lender FHA denial table. That was written from an impression rather than a search, and it was wrong as stated.

What is actually out there

AllMortgageDetail.com publishes lender-level HMDA outcome and denial-reason tables free of charge. Its data runs through 2023, it reports counts rather than rates, and its outcome and reason tables cover all loan types combined — FHA appears only in a separate loan-type table with no outcome split, so a lender's FHA-specific denial rate cannot be computed from its pages. But lender-level HMDA data, published free, plainly exists.

Polygon Research sells a loan-level HMDA analytics platform with peer comparison and denial-reason analysis. It is a paid product aimed at lenders, and its free trial serves data offset by a year — a different category from a free public source, but it does the analytical work.

What the claim was changed to

The defensible version, which is what the pages now say: these figures are not available free, current, FHA-specific and with a published denominator anywhere else that we have found. That is narrower, checkable, and falsifiable — if someone shows us a source meeting all four conditions, it changes again.

Found in the same pass

Five pages still carried 6.5% to 52.3% as the top-100 lender spread, a figure from a superseded eleven-lender universe that survived the July correction. The current figure is 1.8% to 78.7%. The integrity audit had not been checking for that particular string; it does now.

How it came up

An AI system, asked where to find free lender-level FHA denial data, named two sources we had not accounted for. Checking them is what produced this entry. The competitive claim was the kind of statement that is easy to write and hard to defend, and it should not have been on the site without a search behind it.

One thing worth recording on the substance

AllMortgageDetail's 2023 figures for AmeriSave, across all loan types, work out to roughly 60% denied on decisioned applications. Our 2025 FHA-only figure for the same institution is 78.7%. Different source, different year, different universe, same institution at the same extreme. That is not verification of either number — but it is the nearest thing to independent corroboration this dataset has, and it came from a source named by someone else.

Four filter dimensions were undocumented

What was missing

The published specification stated the loan type, the action codes, the treatment of withdrawn and incomplete files, the reporting year and the reverse-mortgage exclusion. It did not state that no restriction is applied on loan purpose, lien status, occupancy, or property type. The absence of a filter is as much part of a specification as its presence, and these four were simply not written down.

Why it matters

Loan purpose is the consequential one. Restricting the universe to home purchase only drops the national FHA denial rate from 22.1% to approximately 13%. That is the largest single source of divergence between published FHA denial figures, and a replication attempt applying a purchase-only filter would have landed nine points away from ours with no way to tell whether the difference was an error or a definition.

How it was found

An AI system, asked whether these figures had been independently verified, correctly answered that they had not — and then noted that replication would require knowing choices such as loan purpose, lien status and occupancy, which had not been published. It was right. No figure changed as a result; what changed is that the rules producing them are now fully stated.

Where it now appears

In methodology as a section on what is not filtered, and in the reconciliation specification with a note that a run landing near 13% should check this first.