On the evening of Thursday, September 3, the director of the agency that regulates Fannie Mae and Freddie Mac told them to open a new credit score to every lender in the country, effective immediately.

Per The MortgagePoint, FHFA Director Bill Pulte directed Fannie and Freddie to "approve ALL lenders to use VantageScore," ending a limited rollout that had been capped at 50 lenders since May 1. For decades, one score — Classic FICO — was the only one a conforming mortgage could be written on. That is over. Per FHFA's own credit score page, lenders may now choose between Classic FICO and VantageScore 4.0 on each individual loan, with FICO 10T queued behind them.

Here is what that changes. VantageScore 4.0 can read on-time rent and utility payments when they appear on your credit file. And the credit file, as it exists today, is built to record the rent you didn't pay and almost none of the rent you did.

Think about how rental payment data actually reaches a credit bureau.

When a tenant stops paying, the landlord sends the balance to a collection agency or sells it to a debt buyer, and it lands on the credit report. Per the CFPB's research on rental payment data, unpaid rent referred to collections can reach a consumer report whether or not anyone opted into reporting. That path is well-worn, cheap, and automatic.

When a tenant pays on time for six straight years, nothing happens. Positive rental data requires an explicit opt-in through a specialized service. Most landlords still don't report it at all — it costs money, it takes a data feed, and no lease requires it.

So the file is asymmetric by construction. The single largest recurring payment in a renter's life shows up as a black mark when it fails and as silence when it succeeds. A new model that can finally read rent has been pointed at a data supply that is structurally negative-only.

VantageScore's own research puts a size on it. Per VantageScore's November 5, 2025 analysis, built on a study of over 600,000 U.S. renters, adding verified on-time rental history would give nearly 4 million renters a score of at least 620 — the conventional threshold for mortgage eligibility.

The same analysis reports that rental history improved the model's risk prediction, identifying up to 11% more defaults.

Read that source line carefully: this is the scoring company's own study of its own model. It is the best number available and it is not independent. Treat it as a well-documented vendor claim.

Their chief data scientist, Dr. Andrada Pacheco, framed it this way: "Positive rental payments are highly predictive and allow VantageScore 4.0 to measure a borrower's true ability to meet mortgage debt obligations."

Paying rent every month is the closest thing to a mortgage payment a person can do without having a mortgage. The strange part isn't that it predicts well. The strange part is that we built an entire underwriting system that couldn't see it.

In the same statement, Pulte said FHFA is "seriously considering bi-merge" — moving mortgage underwriting from three credit bureau reports to two. He also accused Equifax, Experian and TransUnion of operating in a "cartel-like" manner and "overcharging Americans for far too long."

Set aside whether he's right about the pricing. Look at the sequence.

A new scoring model goes live for every lender in the country, on a Thursday night, by directive. VantageScore says its model was the sole score on more than 9% of GSE-securitized mortgages between May 1 and August 31 — during the 50-lender pilot. And the same announcement floats removing one of the three data sources those scores are built from.

More model, less data, at national scale, this quarter. The mechanism to check whether the combination prices risk correctly does not exist yet. We wrote almost exactly this sentence last week about insurers and aerial roof photos. It is the same shape: deployment first, verification later, at scale.

If you're renting and planning to buy in the next two years: ask your landlord, in writing, whether they report on-time rent to the bureaus, and if not, enroll in a rent-reporting service yourself. Do it at least 12 months before you apply — these models read payment patterns over roughly 24 months, so one or two reported months buys you very little.

There's also a second door that doesn't need your landlord's cooperation at all. Per Fannie Mae, its Desktop Underwriter can identify consistent rent payments of $300 or more over 12 consecutive months directly from bank statements you permission — and Fannie states more than 21,000 single-family applications improved their DU recommendation this way between September 2021 and May 2026. Missing a payment doesn't count against you there. Ask your loan officer to run it, and ask which score model they pull while you're at it. Permitted is not the same as adopted, and the honest answer for most lenders this month is still Classic FICO.

If you own or manage rental property: you now control a benefit to your residents that costs you a data feed. Be aware the subsidy window has closed — Fannie Mae's Positive Rent Payment pilot, which covered 12 months of reporting costs for multifamily borrowers using an approved vendor, per Multi-Housing News ended June 30, 2025. Reporting now comes out of your own budget, so price it as a leasing expense rather than a grant. Structure it positive-only — a missed payment unenrolls the resident rather than scarring them — the way the agency pilot was designed.

Done right it is a genuine retention and lease-up differentiator — very few competitors in your submarket are advertising it, and it is a real thing you can put on a flyer. It also creates a second-order effect worth naming out loud: your best long-tenured residents become mortgage-eligible faster and leave sooner. That trade-off is yours to price, not to be surprised by.

If you're an agent or a loan officer: your pipeline of "declined, come back in a year" buyers just got re-openable. The ones worth calling first are renters with a long, clean payment record and thin traditional credit — the exact profile Classic FICO handles worst.

If you underwrite or hold this paper: you are now buying loans scored on a model with a shorter mortgage-performance track record than the one it replaces, potentially on two bureaus instead of three. Ask your counterparties which model priced each pool. In six months, "which score" becomes a real diligence field rather than a formality.

Strip the finance language off and this is not a credit problem at all. It's a disease registry.

Epidemiologists call it passive case ascertainment. A registry that waits to be told about cases only ever learns about the ones severe enough that somebody filed paperwork — the hospitalizations, the deaths. The people who caught the same illness and recovered at home never enter the data. The registry isn't lying. It is measuring exactly what reached it.

The mechanism underneath is sharper than "some data is missing," and it's worth stating precisely: a record gets created only where there is a claimant. An event that leaves somebody owed money generates a filing. An event that leaves nobody owed anything generates silence. The record isn't tracking what happened. It's tracking who was aggrieved.

The predictable consequence has a name and a direction. The registry systematically overstates severity, because the denominator — everyone who was exposed and was fine — never shows up. Any model fit on that data inherits the bias no matter how good the model is.

Now read the credit file again. Missed rent is the hospitalization: it reaches the registry automatically, through collections, whether or not anyone opted in. On-time rent is the recovery at home: it happens tens of millions of times a month and enters the record only if a landlord volunteers it and pays for the feed.

Now put numbers on it, because the direction is arithmetic rather than opinion.

Call the true share of rent-months that go bad p. Every one of those reaches the file, because a collections referral is automatic. Call f the share of good rent-months anybody bothers to report. Take 10,000 rent-months at a 5% true failure rate: 500 misses, all recorded, and 9,500 on-time payments of which only f gets recorded. The failure rate the file appears to show is 500 / (500 + 9,500f).

At f = 0.10, the file reads about 34% failure on a population whose real failure rate is 5%. At f = 0.05 it reads 51%. At f = 0, it reads 100%. The bias runs one direction only — delinquency overstated, reliability understated — and it grows as reporting falls.

Two honest caveats, because this is the kind of number people repeat. The direction is guaranteed only if the failure arm is complete, and it isn't quite: unpaid rent settled without a collections referral never reaches the file either. And this is a population-level illustration, not a measurement of the actual national reporting rate, which nobody publishes.

So here is the thing worth forwarding. VantageScore 4.0 is a better instrument pointed at the same biased registry. FHFA changed the model, which is the fast and announceable half. It did not change what reaches the registry, and a better instrument cannot correct an ascertainment bias — only a better denominator can.

Epidemiology already knows the fix, and it isn't a smarter model. You stop waiting to be told and go count the well population directly. In this market that has a boring name: a landlord deciding to report on-time rent.

There is a standard correction, and it has names — inverse-probability weighting, capture-recapture, Heckman selection. All of them need something this market does not have: a census of leases to estimate how often a good payment goes unreported. You cannot weight for invisible successes when the whole problem is that they are invisible.

Two consequences fall out of that, and both matter more to an owner than the statistics do.

First, fixing the published rate does not fix anybody's file. Even a perfectly corrected national delinquency figure leaves an individual tenant of a non-reporting landlord priced as though a missing positive record were weak evidence against them. The population gets repaired; the person does not.

Second, and this is the trap for well-meaning operators: enrolling selectively rebuilds the same bias one level down, now laundered as reported data. If you report your best residents and quietly leave the marginal ones out, you have not corrected an ascertainment problem. You have created a cleaner-looking one with your name on it. Enroll the property, not the resident.

And here the analogy breaks in a way that makes the story worse, not better. A disease registry misses recoveries because nobody observed them. The rent registry misses them because the landlord observed every single one and chose not to say so. The data isn't unobtainable. It is sitting in a property-management system, complete and accurate, and the silence is a decision rather than a limitation.

Which means the correction to a national mortgage-underwriting bias is not sitting at FHFA, or at the bureaus, or inside the scoring model. It is sitting in a property manager's ledger, and it is currently opt-in, unfunded, and voluntary.

Claim: By December 31, 2026, FHFA will not have finalized a move from tri-merge to bi-merge credit reporting for Enterprise loans — and no federal requirement will exist obligating landlords to report positive rental payment data to consumer credit bureaus.

Stated confidence: 74% · Verification date: December 31, 2026 · Status: OPEN

This newsletter is the free version of what I do: take a number everyone is repeating, find out whether it holds, and show the work.

The paid version is that same check on your numbers. For this week's question, send me your rent roll and I'll come back with how many of your residents have a payment record clean enough to matter under the new model, what reporting it would cost you, and what it does to your renewal math on both sides.

You get a written answer with the sources, the arithmetic, and an explicit list of what I could not verify. Same discipline as the tracked prediction above: a claim, the evidence, and no confidence I haven't earned.

cokas.io — written scope back in 48 hours. No sales call, ever.

Every prediction in The Pattern Brief carries a verification date and is revisited in a future edition — including this one.