How Long to Keep a Split Record After the Bill Is Paid

How long do you have to dispute a restaurant charge on a credit card?

Sixty days — but counted from a date most people guess wrong. Under Regulation Z, the rule implementing the Fair Credit Billing Act, a billing error notice only counts if it is received by the creditor “no later than 60 days after the creditor transmitted the first periodic statement that reflects the alleged billing error.” The statute uses the same trigger: the clock runs from when the creditor transmitted “a statement of the obligor’s account” — not from when the statement reached you, and not from when you ate.

The distance from the meal still cuts in your favor. Dinner on the second day of a billing cycle is not on a transmitted statement for roughly a month, and only then do the sixty days begin. So the practical outside window for catching a wrong restaurant charge is closer to three months after the meal than to two — the exact figure depends entirely on where in your cycle the dinner fell. That is arithmetic on the rule, not a published number, and it is worth doing once so you stop assuming the window is shorter than it is.

The catch is what the rule asks of you in return. A billing error notice must, “to the extent possible,” indicate your belief and reasons that an error exists and “the type, date, and amount of the error.” That is a low bar written generously — but it is a bar about specifics. Sixty days after a statement transmitted a month after a seven-person dinner, “the total looks too high” is hard to write. “We were charged for four espresso martinis and the table ordered two” is not.

Sources: 12 CFR § 1026.13, Billing error resolution, Consumer Financial Protection Bureau; 15 U.S.C. § 1666, Cornell Law School LII

Why the number you approved is not the number that posts

Because a card can be approved before the final amount is known. Visa’s guidance to merchants sets out the mechanism: an estimated authorization “must be a genuine estimate” and, to prevent over-authorization, “must not contain incidental spend amounts such as tips or a buffer for damage.” A tip, in other words, is added after. If the final figure comes in higher, the merchant may raise the total with an incremental authorization; if it comes in lower, Visa requires the difference be reversed within 24 hours of the transaction completing. A merchant using an estimated authorization is also required to tell the cardholder that the authorization is an estimate and not final.

Two honest caveats, because this is the part most easily overstated. Visa describes the mechanism generically and never names restaurants: nothing in the document says an ordinary dinner check is authorized this way. And not every check is. What the mechanism establishes is that a card approval is not always a final number — which is enough, because you do not get told which kind yours was.

None of that is instant. Visa sets a maximum processing timeframe from a valid estimated authorization to processing, and for card-present transactions — which is what a check settled at the table is — it is 5 days from the day the estimated authorization is approved. That is an outside deadline rather than a typical duration, and it applies only to transactions actually begun as estimates. But it sets the boundary that matters here: the rules allow a gap of up to five days between the number your group agreed on and the number that reaches your statement.

Visa also names the failure mode directly. When authorization messages cannot be matched to the clearing record, it warns, “funds remain held for a longer period or in duplicate,” and “cardholders may interpret this as an extra charge by the merchant.” A receipt will not tell you whether two entries are a stale hold or a genuine double charge — only the issuer can settle that. What it tells you is the one number the entries should reconcile to, which is the question you cannot answer from memory three weeks on.

This is a different question from the one answered in the comparison of splitting at the swipe versus at the receipt, which is about the moment a split gets computed. This is about what happens to the record in the days after that moment, while the transaction is still finishing.

Source: Visa Authorization and Reversal Processing: Performing estimated and incremental authorizations, Visa, 2024

What does a dispute actually ask you to produce?

Specifics, in writing. A billing error notice has to reach the creditor at the address it disclosed for the purpose, identify you and the account, and — “to the extent possible” — give your belief and reasons that an error exists plus the type, date and amount of it. A receipt is nowhere on that list. What it buys you is not standing; it is the ability to be specific on demand, months later.

Visa’s guidance to merchants shows what the other side of the table brings. It lists the dispute conditions and the evidence each one turns on, and the transaction receipt is a recurring answer among several — a merchant may also supply authorization evidence, document a credit already processed, or simply accept the dispute. This is the merchant’s playbook, not a requirement placed on you — but it tells you what the argument will be about.

Note what those three have in common. Every one is a fight about an amount and a date, and a document listing both is the merchant’s most direct way out of each. The merchant keeps that document because its own money depends on it. Nobody makes you keep yours.

The Federal Trade Commission puts the practical half plainly: “Keep your receipts: Having the details of your transactions readily available can help you get inaccurate charges fixed.” Its how-to for a written dispute says to “include copies (not originals) of receipts or other documents that support your position.” That is advice, not a legal precondition — advice written for a person who paid their own bill, not for six people who divided one.

Sources: Dispute Management Guidelines for Visa Merchants, Visa, June 2024; Using Credit Cards and Disputing Charges, Federal Trade Commission

The clocks, in order

Laid end to end, the timeline is the argument. Every stage after the first is one that a record living only in the moment cannot reach.

The last line is the one people skip, and its baseline is different from every line above it: the IRS clock runs from the return the deduction was claimed on, not from the meal, so a January dinner can sit inside that window for well over three years. Whether a given meal is deductible at all is a separate question this page does not answer. What an itemized receipt has to prove for that purpose is a separate argument again, made in the case for showing the receipt rather than the total. This piece is only about how long the thing has to exist.

Source: How long should I keep records?, Internal Revenue Service

Is a self-destructing split record a privacy win?

Partly, and the case for it deserves to be made properly rather than waved off. Every copy you stop holding is one fewer copy that can leak or be handed over — a genuine reduction, even if deletion from one server says nothing about backups, logs, or the copies other people made. A split contains a surprising amount: where you were, when, with whom, and what you each ordered. Minimizing how long that sits on a server is a real and defensible goal, and an app that ships expiry by default is not doing something cynical.

Resplit, a bill-splitting app, makes it an explicit selling point. Its homepage lists under “Private by design” that “the link shares one trip — nothing else travels,” that “first names and items go out — never phone numbers,” and that “live splits self-destruct from our server in ten minutes.” All three are defensible privacy choices, and the first two are straightforwardly good design.

Read the page carefully, though, and the ten minutes is not the whole story. The same section says “a link carries the split. Your history stays in your iCloud.” So the record does not vanish — it relocates. What expires in ten minutes is the copy everyone can see. What persists is the copy belonging to whoever ran the split.

That asymmetry is the fairness problem, and it is sharper than plain deletion would be. The page names one place the split still lives after ten minutes, and it is whoever ran it; everyone else is left with whatever they thought to screenshot. It is the same dynamic that makes parking-lot math go badly, and it is why a group’s record is worth judging by who holds it, not only by how long it lives.

Source: Resplit homepage, fetched 1 August 2026

What should a split record survive?

Four things, in increasing order of how long they take. Use them as the spec, and any tool’s retention behavior becomes easy to judge.

Where splitty lands on this is narrower than the full spec, and worth stating plainly. Scanning the receipt turns the itemization into a document rather than an agreement six people remember differently, and each person’s request is built from their own items rather than from a division of the total. That settles the argument at the table, and it means the line-item detail exists at all. It does not file your paperwork — the later clocks are still yours to cover, which is the whole reason to put the receipt image somewhere you will still have it when the statement arrives.

Catching the error at the table is still better than catching it on a statement, which is the whole subject of what to do when the bill has errors. This piece is the fallback for the errors nobody caught — and for the ones the restaurant made after everyone left.