The Disclosure Desk: Change of Circumstance, the Redisclosure Clock, and What an AI Agent Is Allowed to Reset

The Fee Moved. Now Someone Has to Prove Why.

A mortgage fee increases between the Loan Estimate and closing for one of two reasons. Either something happened that the rule recognizes as a valid basis to revise the disclosure, or nothing did and the lender absorbs the difference as a tolerance cure. The disclosure desk is the function that decides which of those it is, loan by loan and fee by fee, and it decides on a three-business-day clock that starts the moment the lender has enough information to know the fee changed.

That determination is small, repetitive, evidentiary, and expensive when it is wrong. A fee increase the lender cannot tie to a valid changed circumstance is a tolerance violation the lender cures with its own money at consummation, and a pattern of them is a finding. A changed circumstance the lender documents badly, without the date it learned the facts or the specific fees the reset touched, fails the same way under a CFPB TRID examination even when a valid trigger existed. The work is not hard. It is voluminous, and the cost of getting it wrong sits at the intersection of a disclosure rule and the lender's margin.

We build the agent that runs across mortgage origination on lender platforms, and the disclosure desk is where an agent earns its keep, because the redisclosure decision is structured enough to compute and consequential enough that a human should sign the borderline calls. What follows is where the agent operates, the baseline it is allowed to reset, and the boundary that separates a documented changed circumstance from a manufactured one.

The Baseline Is Set at Application, and the Rule Guards It

When intake becomes an application under the six-piece definition at 1026.2(a)(3), the lender owes a Loan Estimate within three business days, and that estimate sets a good-faith baseline the lender is measured against at closing. Regulation Z at 1026.19(e)(3) sorts every charge into a tolerance category and holds the lender to it. Lender charges, transfer taxes, and charges for services the borrower cannot shop for carry a zero tolerance, so the amount at closing cannot exceed the amount disclosed. Recording fees and charges for services the borrower could shop for but took from the lender's written list carry a ten percent cumulative tolerance. Prepaid interest, property insurance, escrow deposits, and services the borrower shopped and chose independently sit in the good-faith category with no fixed cap, measured instead against whether the estimate was reasonable.

The baseline matters because the revised Loan Estimate is the only instrument that moves it. The lender cannot raise a zero-tolerance fee and disclose the higher number on the Closing Disclosure and call it even. It has to reset the baseline first, and it can only reset the baseline when a triggering event lets it, on the clock the rule sets. So the disclosure desk's real question is never "did the fee change." It is "does this change reset the baseline, and can I prove it on the timeline the rule wants."

What Counts as a Reset, and the Clock That Governs It

Regulation Z at 1026.19(e)(3)(iv) lists the events that let a lender use a revised estimate to reset the tolerance baseline. A changed circumstance that affects settlement charges or the borrower's eligibility. A borrower-requested change. A points or lock event that changes the rate-dependent charges. A consumer who indicates intent to proceed more than ten business days after receiving the Loan Estimate, which lets the estimate expire and its good-faith amounts reset. New construction with a settlement date more than sixty days out where the lender disclosed that it might revise. Each is a defined event with a defined effect, and each starts the same clock: the lender must deliver the revised Loan Estimate within three business days of receiving the information sufficient to establish that the event occurred.

That "sufficient information received" date is the fact the whole determination turns on, because the three-day window runs from it and a tolerance cure at closing depends on whether the lender met it. A lender that learns on Monday that the appraisal came in requiring a second inspection, and issues the revised estimate the following week, has blown the window even though the changed circumstance was real. The reset is a compliance act with a timestamp, and the timestamp is the thing manual disclosure desks lose track of when volume climbs.

There is a second boundary the rule draws at 1026.19(e)(4): the creditor cannot issue a revised Loan Estimate on or after it has provided the Closing Disclosure. That does not turn every later increase into a lender cure, because a valid triggering event that arises too late for a revised Loan Estimate can be reflected on a corrected Closing Disclosure and still reset the baseline, which is the mechanism the commentary provides for a changed circumstance discovered close to consummation. What becomes a cure is an increase with no valid triggering event behind it. So a disclosure process has to track the transition into the CD and keep telling those two apart, the late-but-valid change that resets on a corrected CD and the unsupported increase the lender absorbs, because conflating them produces the redisclosure defects we see most: a revised estimate issued into a window the rule has already closed, or a valid late change mishandled as a cure the lender did not owe.

Where the Agent Operates on the File

The agent watches the events in the loan file that change a fee, because a triggering event is not an abstraction, it is a concrete change in the record: an appraisal value that requires a repair set, a title fee that came back higher than quoted, a borrower who switched from a thirty-year to a fifteen-year term, a rate lock that moved the discount points, a flood determination that added required insurance. When one of those changes lands, the agent identifies which charges it touches, maps each charge to its tolerance category, and computes whether the change is one the rule recognizes as a reset.

For the changes that reset, the agent assembles the redisclosure package the way an examiner would want to read it back. It records the triggering event and ties it to the specific fees whose baseline moves, not the whole estimate. It captures the date the lender received information sufficient to establish the event, because that date starts the three-day clock and the revised estimate has to cite it. It runs the tolerance math on both sides of the change, the disclosed baseline and the revised amount, so the good-faith comparison at closing is reconstructable rather than asserted. Then it drafts the revised Loan Estimate with the changed-circumstance reason populated and routes it for issuance inside the window.

For the changes that do not reset, the agent does the more useful thing, which is to say so. A fee that increased with no valid triggering event is not a redisclosure, it is a cure the lender owes, and the agent flags it as a cure rather than dressing it up as a changed circumstance. Surfacing the cure early, while the loan is open, lets the lender decide whether to absorb it or correct the underlying error, instead of discovering it in a post-closing quality-control review after the loan has funded and the cure has grown teeth.

The Line the Agent Does Not Cross

The tempting failure mode in an automated disclosure desk is the one that matters most, so it is worth being exact about it. The value of resetting a tolerance baseline is that it moves a fee increase from the lender's ledger to the borrower's. That incentive is precisely why the rule requires a valid triggering event and a documented basis, and it is why an agent must never generate a changed circumstance to fit a fee it would rather not cure. Manufacturing a changed circumstance is a Regulation Z violation on its face, and because it shifts cost to the borrower without a valid basis, it carries UDAAP and fair-lending exposure on top, especially if the pattern falls unevenly across a protected class.

So the agent proposes and documents, and it does not decide the borderline case on its own. When the triggering event is clean, an appraisal that plainly requires a repair inspection, a borrower who plainly requested a shorter term, the agent drafts the revised estimate and the desk confirms it. When the event is ambiguous, a fee that drifted for reasons the record does not cleanly explain, the agent routes it to a human with the tolerance math and the timeline attached, because the good-faith determination on a marginal changed circumstance is a judgment the lender should own and be able to defend. The agent's job is to make sure the judgment is never made on missing information, and to make sure the clean cases and the marginal cases are told apart rather than run through the same reset by default.

What This Buys the Lender

Two things, both measurable. The first is that the three-day clock stops getting missed, because the agent starts it from the moment the triggering information enters the file rather than from whenever a processor gets to the fee. In our deployments the redisclosure timeline is the metric we watch first, and holding revised estimates well inside the window is the single change that removes the most tolerance-cure exposure, because a valid changed circumstance disclosed late fails the same way as no changed circumstance at all. The specific timing target we hold is illustrative and set with each lender, but the shape is the same everywhere: the clock is a control, and the agent's job is to make missing it rare.

The second is that the paper trail is built at the moment the decision is made, not reconstructed under examination. Every baseline reset the agent proposes carries the triggering event, the information-received date, the affected fees, and the before-and-after tolerance math, so a quality-control reviewer or a CFPB examiner can read why a fee moved without asking the loan officer to remember. At Sei we build the disclosure agent to make the redisclosure decision fast where it is clean, honest where it is a cure, and reviewable in every case, because the disclosure desk is one of the few places in origination where a small, repetitive judgment made a few thousand times a year decides whether the lender's good-faith record survives a look. The agent does the volume. The desk owns the calls that deserve a human. The record holds up either way.