The Points-and-Fees Test Runs at Pricing, Not at Closing: HOEPA High-Cost Coverage and the QM Cap With an AI Agent

A Line You Do Not Want to Cross by Accident

There is a category of mortgage most lenders will not originate on purpose. A high-cost mortgage under the Home Ownership and Equity Protection Act carries disclosure obligations, prohibited terms, and a liability exposure that makes it a loan most institutions have decided, as a matter of policy, not to make. The problem is that a loan can become a high-cost mortgage without anyone deciding to make one, because coverage turns on a fee calculation, and fees get added to a file late and by people who are not watching the coverage line. A single fee entered in the last week before closing can push a loan over the threshold, and a lender that intended to originate a standard loan discovers it originated a high-cost one it has no program to support.

The points-and-fees test is the calculation that decides this, and the same fee total drives a second test that matters just as much: whether the loan keeps its Qualified Mortgage status. We work with lenders on where these tests belong in the origination workflow, and the answer we keep arriving at is that they belong at pricing, run continuously, not at closing as a final check. A test you run only at the end tells you the loan already crossed the line. A test you run at pricing tells you in time to reprice.

Two Tests, One Fee Total

The high-cost coverage test lives in Regulation Z at 12 CFR 1026.32. A closed-end mortgage is a high-cost mortgage if it crosses any one of three coverage triggers: an APR that exceeds the average prime offer rate by more than the specified margin, a prepayment penalty that runs too long or too high, or points and fees that exceed the coverage threshold. The points-and-fees trigger under 1026.32(a)(1)(ii) is five percent of the total loan amount for a larger loan, and for a smaller loan below the annually adjusted size threshold it is the lesser of eight percent of the total loan amount or a fixed dollar figure, so the dollar figure caps the trigger rather than setting a floor under it. The CFPB adjusts the size threshold and the dollar figure every year for inflation. Cross the trigger and the loan is high-cost, with everything that follows.

The second test uses the same points-and-fees figure for a different purpose. To be a Qualified Mortgage under 1026.43(e)(3), a loan's points and fees cannot exceed three percent of the total loan amount for larger loans, with higher tiered caps for smaller balances, again adjusted annually. A loan that exceeds the QM points-and-fees cap loses the QM designation and the ability-to-repay presumption that comes with it, which changes the loan's liability profile even if it never comes close to the higher HOEPA high-cost line. So one fee total is measured against two ceilings, a lower one that decides QM status and a higher one that decides high-cost coverage, and a fee increase can breach the QM cap while leaving the loan well short of high-cost.

What Actually Counts as a Point or a Fee

The reason this is not a simple sum is that the definition of points and fees is specific about what is in and what is out, and getting the inclusion wrong is how the calculation goes wrong. Under 1026.32(b)(1), points and fees for a closed-end loan include the items in the finance charge with certain exceptions, the loan originator compensation the rule specifies, and real-estate-related fees that are not bona fide and reasonable, among other categories. Some third-party charges are excluded when they meet the bona-fide-and-reasonable test and the lender does not retain the charge. Loan originator compensation is included in a way that intersects with the Reg Z 1026.36 compensation rules, and double-counting or miscategorizing it is a common source of an inaccurate total.

This is where a test that looks like adding up fees becomes a classification problem. Every fee on the loan has to be evaluated for whether it counts toward points and fees, and the answer depends on what the fee is, who receives it, whether it is retained by the lender or an affiliate, and whether it meets the bona-fide-and-reasonable standard. A fee that is correctly excluded keeps the loan under the cap, and the same fee misclassified as included can push it over, or the reverse, an included fee left out gives false comfort that the loan is compliant when it is not. The test is only as good as the classification underneath it.

Why the Test Belongs at Pricing

The decision we keep coming back to with lenders is timing. If the points-and-fees test runs only as a pre-closing compliance check, it can only tell you what already happened, and by then the fees are baked into a file that is days from funding, the borrower has an interest rate lock, and the options for bringing a loan back under the cap are narrow and expensive. Reducing the fees means the lender absorbs the difference. Repricing means going back to a borrower who thought the terms were set.

Run the same test at pricing and keep it running as the file changes, and the breach shows up while the loan can still be structured to stay under the line. A loan that is approaching the QM cap can be repriced, a fee that is pushing it over can be examined for whether it is really includable, and the lender can make the decision with room to act instead of discovering the problem at the closing table. The agent we help lenders put on this runs the points-and-fees calculation continuously against both the QM cap and the high-cost threshold, and when a new fee moves the total toward either line, it surfaces the loan with the current total, the distance to each ceiling, and the specific fees driving it. A lender that sees the loan at three percent minus a small margin has a decision to make with time to make it. A lender that sees it at three percent plus a small margin on the closing CD has a problem.

We built the continuous version of this test after seeing the failure mode it prevents: a loan that was comfortably a QM at application picked up a fee late in processing, no one re-ran the points-and-fees calculation against the new total, and the loan closed over the QM cap. Nothing in the file was fraudulent and no single person made an obvious mistake. The fee was legitimate, it was just added by someone who was not looking at the cap, at a stage where no one re-ran the test. That is the entire argument for running the test continuously rather than once: the breach is almost never a bad decision, it is an unwatched total.

The Line Between Calculation and Judgment

The agent runs the calculation and flags the coverage question. It does not make the final determination that a particular fee is or is not bona fide and reasonable in a close case, because that is a judgment about the fee and the market that sits with the lender's compliance function. What the agent does is apply the classification consistently, surface the fees whose treatment actually swings the result, and show its work, so the compliance reviewer is deciding the two or three fees that matter rather than re-adding the whole loan. When a fee's classification is genuinely uncertain and it is the fee that decides whether the loan is over the cap, the agent flags exactly that, the fee, its amount, why its treatment is uncertain, and how the total lands each way, so the human owns the call that changes the outcome.

The value is not that the arithmetic is hard. It is that the arithmetic has to be right on every loan, against a definition with real inclusions and exclusions, run early enough to matter, and re-run every time a fee changes. That is consistency at a point in the workflow where the cost of missing it is a loan the lender either cannot sell as a QM or, at the far end, cannot originate at all under its own policy.

The Honest Read

Whether a mortgage is a high-cost loan under HOEPA and whether it keeps its Qualified Mortgage status both come down to a points-and-fees calculation, run against two different ceilings, off a fee total that a single late charge can move across a line. The calculation is a classification problem, because the definition of points and fees at Reg Z 1026.32 includes some fees and excludes others depending on what they are and who keeps them, and the test is only as good as the classification underneath it.

The decision that matters most is when the test runs. A points-and-fees check at closing tells a lender it already originated a loan it did not intend to. The same check at pricing, run continuously as fees change, catches the breach while the loan can still be repriced. At Sei, we help lenders put the test at pricing, run it against both the QM cap and the high-cost threshold, and surface the specific fees driving the result so compliance decides the handful that swing it. The line is one you do not want to cross by accident, and the way you avoid crossing it by accident is to watch the total the whole way, not just at the end.