# The Student Loan Payment on the Credit Report Is Probably Stale: Liabilities, DTI and What an Underwriting Agent Has to Reconcile After SAVE

*November 1, 2026 · 7 min read · Ramkumar Venkataraman*

> SAVE is gone, RAP has a $10 floor, and millions of credit reports still show a student loan payment that no longer exists. A technical account of how an underwriting agent reconciles every liability across the 1003, the tri-merge and the documents, why the same borrower carries a different DTI under Fannie Mae, Freddie Mac and FHA, and why we let the agent add a debt on its own but never remove one.

## Who This Is For

Underwriting managers, credit policy owners and Directors of Mortgage Ops who are seeing more student-loan conditions than they did a year ago, and anyone in model risk who has to sign off on an agent that touches the debt side of the ratio.

## The Debt Side of DTI Is a Reconciliation, and Nobody Treats It Like One

Income gets the attention. There are vendors for it, there are calculators for it, and every underwriter has an opinion about Schedule C add-backs. The debt side of the ratio gets far less, because it looks like a lookup: pull the tri-merge, sum the monthly payments, divide. That is roughly how a lot of files get worked, and it is wrong in a specific way, because the monthly debt figure is the output of a reconciliation across three sources that routinely disagree.

The first source is what the borrower declared in Section 2 of the URLA. The second is what the credit bureaus report, which is a snapshot of what each creditor furnished in its last cycle. The third is whatever documents are in the file: a student loan servicer statement, a payoff letter, a court order assigning a debt to an ex-spouse, twelve months of a business account paying the borrower's truck note. The qualifying payment for each liability comes from applying the investor's rule to whichever of those sources the rule says to trust, and the rule differs by investor.

Reg Z puts the obligation on the creditor, not on the bureau. The [ability-to-repay rule at 12 CFR 1026.43(c)(2)(vi)](https://www.consumerfinance.gov/rules-policy/regulations/1026/43/) requires the creditor to consider "the consumer's current debt obligations, alimony, and child support," and 1026.43(c)(3) requires it to verify that information using reasonably reliable third-party records. A credit report is a reasonably reliable record. It is not always a current one, and 2026 is the year that gap got wide.

## What Changed in 2026: SAVE Ended and the Credit Reports Did Not Catch Up

For most of the last three years, a large share of federal student loan borrowers sat in the SAVE plan, much of that time in an administrative forbearance with a $0 required payment. Credit reports reflected that. Then the plan ended. Litigation closed out SAVE in early 2026, and the new [Repayment Assistance Plan](https://www.nyc.gov/site/dca/talk-money/Student-Loans-Key-Changes.page) launched on July 1, 2026, with payments set at 1% to 10% of adjusted gross income and a $10 monthly minimum. Borrowers coming off SAVE were given a window to pick a new plan, and those who did not choose are moved to a standard plan. [Federal Student Aid's servicer notices](https://edfinancial.studentaid.gov/RAP) describe the same transition.

So in the second half of this year, an underwriter opening a file is looking at a credit report that may show $0, or the old SAVE figure, or nothing at all, for a borrower whose real payment is now somewhere between $10 and a fully amortizing standard payment that could be several hundred dollars higher. The reported number lags the servicer by at least one furnishing cycle, often more while servicers work through the plan changes. That lag is a DTI risk in both directions. Use the stale $0 and the ratio is understated, which is an ATR and a repurchase problem. Use a default proxy when the borrower has a documented lower payment and you may decline or misprice a borrower who qualifies.

## Same Borrower, Same Report, Three Different Ratios

Here is where a lookup breaks. Take one borrower with a $60,000 federal student loan balance and a credit report showing a $0 payment. This example is illustrative, but the rules are not.

Under Fannie Mae's [B3-6-05, Monthly Debt Obligations](https://selling-guide.fanniemae.com/sel/b3-6-05/monthly-debt-obligations), the lender uses the payment on the credit report, or a documented payment from the servicer. If the reported payment is $0 or missing and there is no documentation of the actual payment, the lender uses either 1% of the outstanding balance or a fully amortizing payment from the documented repayment terms. A documented $0 payment under an income-driven plan can be used as $0. On our example, the proxy is $600 a month.

Under Freddie Mac's Guide, when the payment reported is zero, the proxy is 0.5% of the outstanding balance, which is $300. A $0 income-driven payment does not carry over the way it does at Fannie.

Under FHA, [Mortgagee Letter 2021-13](https://archives.hud.gov/news/2021/2021-13hsgml.pdf) set the rule for case numbers assigned on or after August 16, 2021: use the payment on the credit report, or 0.5% of the outstanding balance when that payment is zero, and get written evidence from the servicer if you use anything lower than what the report shows.

That is $600, $300, or a documented number, from one borrower and one report. On a borrower with $7,000 of monthly income, the difference between $300 and $600 is more than four points of DTI, which is the difference between a clean DU or LPA finding and a file that needs a second look. And RAP shrinks the documented-$0 path, because a plan with a $10 floor does not produce a $0 payment, so a file relying on a $0 servicer letter dated before July has to be read against its date.

This is why the rule has to live in the [Rulebook](/how-it-works#rulebook) by investor and by effective date, which is how we [version every agency update](/blog/rulebook-change-management-agency-selling-guide-updates-ai-mortgage-effective-dates). An agent that "knows the student loan rule" in general is an agent that will apply the Fannie rule to an FHA file. The agent has to know which rule set the loan is being underwritten to, and has to re-run the ratio when a file moves from conventional to FHA mid-process, which happens more often than anyone likes.

## The Other Liabilities That Get Waved Through

Student loans are the loudest item this year. They are not the only place the debt figure goes wrong. These are the treatments we see mis-applied most when we backtest an agent against historical files:

- Installment debts with ten or fewer payments remaining, which Fannie allows the lender to exclude, except that lease payments count regardless of the months left. A car lease with six months to go is still a debt.
- Revolving accounts with a balance but no payment on the report, where Fannie's fallback is 5% of the balance unless a statement documents the actual minimum.
- Debts paid by someone else, which can come out of the ratio only with evidence that the other party made the payments for the most recent twelve months, and never when the payer is an interested party to the transaction.
- Debts assigned by a divorce decree or separation agreement, which can be excluded with the court order in the file, not with the borrower's word that the ex-spouse pays it.
- Debts being paid off at closing, which only leave the ratio if the payoff actually appears on the Closing Disclosure.
- Debts that are not on the report at all: the 401(k) loan, the private student loan from a lender that does not furnish, the debt that shows up on a bank statement as a recurring ACH to a lender nobody declared.

The last item is where liabilities and fraud meet. An undisclosed debt that surfaces between application and closing is its own problem, and we wrote about the [fresh-credit and undisclosed-debt monitoring](/blog/fresh-credit-undisclosed-debt-monitoring-ai-mortgage-permissible-purpose-fcra) side separately. The reconciliation here is about the debts that are in the file and treated wrong.

## How the Agent Builds the Liability Ledger

Inside UMOS, the [Income Desk](/for-lenders/income-desk) and the [Conditions Manager](/for-lenders/conditions-manager) work from one liability ledger in the [Loan Graph](/platform/loan-graph). Each row is one obligation. It carries where the obligation appears (the 1003 line, the tradeline on each bureau, the page of each document that mentions it), the treatment the agent applied, the Rulebook citation for that treatment, and the qualifying payment that resulted. The DTI is a sum over that ledger, so any number in the ratio can be traced back to a page and a rule in one click.

Building the ledger means matching things that do not share an identifier. A tradeline reported as "DEPT OF ED/AIDVANTAGE" with a partial account number has to be matched to a servicer statement with a different account format, to a line on the 1003 where the borrower wrote "student loans" with a rounded number. Federal student loans are usually reported as several tradelines, one per disbursement, so the agent also has to decide whether the servicer statement's single payment covers all of them or only some. Mismatches here are the most common reason a liability row stays open, and they are exactly the kind of thing a person does by eye and an extraction tool does not do at all.

When the ledger changes, the ratio changes, and the conditions change with it. If a servicer letter arrives documenting a RAP payment of $187, the agent replaces the proxy on that row, cites the letter, re-runs the DTI, and re-plans the conditions list: the student-loan condition clears, and if the new ratio moves the file across an overlay threshold, a new condition appears. Nothing waits for an underwriter to notice that a document landed.

## Why the Agent Can Add a Debt but Never Remove One on Its Own

The design decision we spent the most time on was an asymmetry. Every treatment that raises the debt figure (using a proxy instead of a $0, counting a lease with three months left, adding an undeclared obligation found on a bank statement) the agent can apply on its own, cited to the rule, because the error mode is conservative. Every treatment that lowers the debt figure (excluding a debt paid by others, dropping a court-assigned obligation, using a documented payment below what the bureau shows) requires a document in the file that the agent cites to the page, has to pass the independent check in [Triple Check](/how-it-works#triple-check), and has to clear the confidence floor the lender set for exclusions. If the only evidence is a borrower explanation, the row goes to the underwriter as a named exception with the question that would settle it.

We made this choice because an exclusion that should not have happened is the failure that turns into an ability-to-repay finding, and the one a QC reverification catches months later. Every lender we work with already holds exclusions to a higher bar in their own underwriting; the agent just enforces that bar on every file instead of most of them. It also makes the exception queue honest. When we backtest against an institution's historical book, the exclusions are where agent and underwriter disagree most, and that disagreement is useful, because it shows which exclusions the institution was accepting on thinner documentation than its own policy requires.

## What We Would Check First

If you want to know where your own operation stands, pull the funded files from the last ninety days with a federal student loan on the report and compare the qualifying payment used against the servicer's current figure. On files closed after July 1, look specifically for a $0 payment supported by a pre-July servicer letter. That one query will tell you whether your liability rules have caught up with the plan changes, and it is the same test we run when we [backtest an agent](/blog/testing-long-horizon-mortgage-ai-agent-eval-harness-golden-files-sr-11-7) against a lender's book before it touches a live file.

The debt side of the ratio will keep moving, because student loan policy is not finished changing and investors will keep adjusting their proxies. A lookup cannot keep up with that. A ledger where every row names its source and its rule, and a Rulebook that knows which version applied on which date, can.

---

_Source: [https://www.seiright.com/blog/student-loan-liabilities-dti-reconciliation-ai-underwriting-save-rap-fannie-b3-6-05](https://www.seiright.com/blog/student-loan-liabilities-dti-reconciliation-ai-underwriting-save-rap-fannie-b3-6-05) · Sei AI_