# The Three Tolerance Buckets: Balancing the Closing Disclosure Against the Loan Estimate With an AI Agent

*August 26, 2026 · 6 min read · Ramkumar Venkataraman*

> Between the Loan Estimate and the Closing Disclosure sits the fee comparison that decides whether a lender owes the borrower a refund. Zero tolerance, ten percent aggregate, and no tolerance are three different rules on three different sets of fees, and a changed-circumstance re-disclosure can move a fee from one bucket to another. Where an AI agent tracks every fee from LE to CD, catches the tolerance breach before consummation, and computes the cure the rule requires.

## The Comparison That Decides Who Pays

Every closed-end mortgage carries a promise the lender makes at the Loan Estimate and has to keep at the Closing Disclosure. The fees quoted early cannot drift upward past the limits the rule sets, and when they do, the lender owes the borrower the difference. This is the tolerance comparison, and it is where a loan that looked clean all the way through underwriting produces a compliance finding at the closing table, because someone entered a fee on the CD that was higher than the LE allowed and no one balanced the two documents before the borrower signed.

We build the agent that runs across mortgage origination on lender platforms, and the LE-to-CD tolerance comparison is a task that is pure arithmetic and pure discipline at the same time. The math is not hard. The hard part is doing it for every fee, on every loan, tracking each fee through every re-disclosure, and knowing which of three tolerance rules governs each line. Miss one, and the lender either eats a refund it did not budget for or, worse, closes a loan with an uncured tolerance violation that a repurchase or an exam will find.

## Three Rules on Three Sets of Fees

The tolerance framework under [Regulation Z at 12 CFR 1026.19(e)(3)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/) sorts every closing cost into one of three buckets, and the bucket decides how much the fee is allowed to increase from the Loan Estimate to the Closing Disclosure.

The zero-tolerance bucket, under [1026.19(e)(3)(i)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/), covers the fees the lender controls or has the information to know. Charges paid to the lender, a mortgage broker, or an affiliate of either, the origination fee and points among them, sit here, and so do transfer taxes and any charge for a required service the borrower was not permitted to shop for, including one performed by an unaffiliated third party the lender selected. None of these can increase at all. A dollar over the LE on any of them is a dollar the lender owes back, unless a valid changed circumstance re-set the baseline first.

The ten-percent aggregate bucket, under [1026.19(e)(3)(ii)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/), covers recording fees and charges for third-party services where the borrower was permitted to shop and chose a provider from the lender's written list. These fees are allowed to move, but only in the aggregate: the sum of the fees in this bucket on the CD cannot exceed the sum on the LE by more than ten percent. An individual fee can rise more than ten percent as long as the bucket total stays within the limit, which means the agent cannot evaluate these fees one at a time, it has to sum the bucket.

The no-tolerance bucket, under [1026.19(e)(3)(iii)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/), covers fees genuinely outside the lender's control at the LE: prepaid interest, property insurance premiums, amounts placed into an escrow account, and charges for third-party services the borrower shopped for and selected a provider not on the lender's list. These can increase without a tolerance cure, provided the original estimate was made in good faith and consistent with the best information reasonably available. No tolerance is not no rule, the good-faith standard still applies, and an estimate the lender knew was too low is not saved by putting the fee in this bucket.

## Why a Fee Does Not Stay in Its Bucket

The reason a static checklist fails at this is that fees move between buckets when the file changes, and the framework has a mechanism for it. A valid changed circumstance under [1026.19(e)(3)(iv)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/), an event beyond anyone's control, information the lender relied on that turned out to be inaccurate, a borrower-requested change, new information specific to the transaction, permits the lender to issue a revised Loan Estimate and re-set the tolerance baseline for the affected fees. Once a revised LE is issued for a valid reason and delivered inside the rule's timing, the CD is compared against the revised LE for those fees, not the original.

This is where the tracking gets real. A single loan can have an original LE, a revised LE after the appraisal came in and triggered a new inspection, and a second revision after the borrower changed the loan amount, and the tolerance comparison at the CD has to measure each fee against the correct baseline given its history. A fee that was in the ten-percent bucket can move because the borrower's choice of provider changed. A changed circumstance that is documented and timed correctly resets the baseline; one that is asserted after the fact to paper over a fee increase is a violation dressed as a reset. The agent has to know not just the current fee, but the entire history of what the fee was disclosed at, when, and under what justification.

## What the Agent Tracks From LE to CD

The agent's job is to hold the fee ledger across the life of the loan. It captures every fee at the original LE, tags each into its tolerance bucket, and records the baseline. When a revised LE issues, it checks that the revision has a documented changed circumstance that actually supports the fees being changed, that the revision was delivered inside the timing the rule allows, and it updates the baseline only for the fees the changed circumstance justifies, because a revised LE does not reset fees unrelated to the triggering event. It carries that ledger forward to the Closing Disclosure and runs the three comparisons: every zero-tolerance fee against its baseline, the ten-percent bucket in aggregate against its baseline sum, and the no-tolerance fees against the good-faith standard.

When the comparison shows a breach, the agent does not just flag it, it computes the cure. A zero-tolerance fee that exceeds its baseline by eighty dollars is an eighty-dollar cure owed to the borrower. A ten-percent bucket that came in a hundred and twenty dollars over the allowed aggregate is a hundred-and-twenty-dollar cure. The rule under [1026.19(f)(2)(v)](https://www.consumerfinance.gov/rules-policy/regulations/1026/19/) lets the lender cure a tolerance violation by refunding the excess to the borrower and delivering a corrected CD within sixty days of consummation, so catching the breach before the borrower signs means the cure goes on the CD as a lender credit rather than becoming a post-closing refund the lender chases and an exam finds. The agent surfaces the breach and the exact cure amount while the CD can still be corrected, which is the difference between a fee entry someone fixes and a violation someone discovers.

## The Boundary the Agent Holds

The agent computes the comparison and the cure. It does not decide whether a changed circumstance is valid in the close cases, because that is a judgment about the facts of the transaction and the good-faith standard, and it sits with the lender's compliance and closing functions. When a fee increased and the justification offered is a changed circumstance the agent cannot tie to a real, documented, timely event, it does not quietly accept the reset and it does not quietly reject it. It flags the fee with the asserted justification and the reason the justification looks thin, and it routes the question to the human who owns the good-faith call. This is the same line we hold across [TRID timing](/blog/trid-loan-estimate-closing-disclosure-ai-agents-timing) and the rest of the disclosure architecture: the agent runs the math and enforces the comparison, and the human owns the judgment about whether a circumstance was genuinely changed.

The reason to keep the agent on the arithmetic and off the judgment is that the arithmetic is where the errors actually are. Tolerance violations at closing are almost never a dispute about whether a changed circumstance was valid. They are a fee that was entered on the CD higher than the LE allowed, in a bucket someone did not track, against a baseline someone did not update, and no one balanced the two documents before consummation. That is a consistency failure, and consistency at scale is exactly what the agent is for.

## The Honest Read

The Loan Estimate is a promise and the Closing Disclosure is the test of whether the lender kept it. Regulation Z sorts every fee into zero tolerance, ten percent aggregate, or no tolerance, and the comparison between the two documents decides whether the lender owes the borrower a refund. The reason this produces findings is not that the rule is obscure, it is that fees move between buckets when the file changes, revised Loan Estimates reset baselines for some fees and not others, and balancing every fee against its correct baseline on every loan is precisely the kind of every-line every-time discipline a manual review loses under closing pressure.

An AI agent earns its place by holding the fee ledger from the original LE through every revision to the CD, running the three tolerance comparisons, and computing the cure while the CD can still be corrected, so a breach becomes a lender credit at closing instead of a sixty-day refund an exam finds. At Sei, we build the tolerance agent to track the buckets, enforce the comparison, and compute the cure, and to route the good-faith judgment on a thin changed circumstance to the human who owns it. The math is not the hard part. Doing it for every fee, against the right baseline, before the borrower signs, is.

---

_Source: [https://www.seiright.com/blog/trid-fee-tolerances-closing-disclosure-cures-ai-mortgage](https://www.seiright.com/blog/trid-fee-tolerances-closing-disclosure-cures-ai-mortgage) · Sei AI_