Bill-splitting tools compete on arithmetic. Divide by the right denominator, allocate tax and tip proportionally, minimize the number of payments. It is the part of the problem the category has spent years getting right, and a phone calculator got most of the way there first.
None of it addresses the moment the split actually fails. That moment is not the calculation — it is the two seconds after someone reads “you owe $34” and has to decide whether to believe it. Arithmetic produces a number. What the other person needs is a way to check one. Those are different problems, and the category has concentrated almost entirely on the first.
This is the argument for treating the receipt as something other than input to a calculator — and the research on verification points two ways at once. On one side, an unverifiable claim tends to get discounted: whatever you didn’t show is assumed to be the part you wouldn’t want shown. On the other, verification hardened into enforcement can reprice the very norm it was protecting, and the repricing doesn’t reverse. Push too little and you aren’t believed. Push too far and there is nothing left worth believing in.
Those two pressures leave a narrow band, and the band is this article’s actual claim: the right instrument is a record that is always available and almost never invoked. A receipt, shared with the ask, is close to the only everyday object that fits. It is checkable without being binding, which is the whole trick.
What kind of argument this is: none of the research below studied a restaurant table. It studied salesmen, quality disclosure in markets, laboratory contracts, and day-care pickups. The mechanisms are borrowed and the borrowing is flagged each time it happens. Treat what follows as a way to reason about splitting a bill, not as measured findings about people splitting bills.
Source: splitty first-party data — aggregates across splitty’s US-leaning restaurant receipts, July 2026 snapshot. Scoped to splitty’s own scans; not a national estimate.
Why do bill splits get disputed when the math is right?
The argument of this article is that a per-person total is a claim, and the person receiving it has no way to test it. No study cited below measures that — nobody has run the experiment on diners — so take it as a frame worth reasoning from rather than a demonstrated cause. The arithmetic being correct is invisible to the recipient. What they can see is one number, produced by someone with an interest in the outcome, covering an evening they only partly remember.
Picture the table this actually happens at. Five people, two shared appetizers, a bottle of wine three of them finished, one seat that ordered still water all night. The check lands face-down. Someone opens a calculator, types a total, divides, and sends four numbers to four phones. Nobody else at the table watched the arithmetic happen — they only watched the check arrive and the number leave. The person who paid knows exactly what the total is made of. Everyone else is holding a conclusion with the working stripped out.
The scale of what is being taken on faith is easy to underestimate. Across splitty’s US-leaning restaurant receipts, the average bill carries about nine line items. A single “you owe $34” is not one claim to accept — it is a compressed summary of nine of them, plus a tax allocation and a tip allocation nobody watched being made.
Nine of what, though, is the part no general rule covers. Bills vary enough in size alone that no standard basket exists to check a total against: about 48% run to $150 or more, while roughly one in seven lands under $50. Each bill is its own particular set of roughly nine claims, and the only document that records which nine is the receipt from that table.
That does not make the receipt sufficient. It records items, not people — it will not tell you who ordered the second glass of wine, and it knows nothing about what anyone agreed to share. Verifying a personal share takes both halves: the receipt for what was on the table, and the group for who had what. What a bare per-person total does is discard the first half and ask you to take the second on faith.
The distinction that matters: a fair split and an acceptable split are not the same output. A split can be fair and still be refused, because the person refusing has no way to see that it is. Acceptability depends on what they can check — which is a separate problem from getting the allocation right, and the one that gets far less attention.
What does a number without a receipt actually signal?
In the standard economic model of disclosure, it signals something mildly negative — not because the sender is lying, but because a rational receiver reads silence as unfavorable. Paul Milgrom formalized this in the Bell Journal of Economics in 1981. His setup: an interested party holds information a decision-maker wants, and can choose what to reveal.
“If communication between the parties is costless and if the decisionmaker can detect any withholding of information, then, at equilibrium, the decisionmaker adopts a strategy of extreme skepticism: he assumes that information is withheld only if it is very unfavorable.”
Paul R. Milgrom, “Good News and Bad News,” The Bell Journal of Economics (1981)
Faced with that skepticism, Milgrom shows, the interested party’s best strategy is full disclosure. Note the load-bearing assumption: disclosure in his model is costless. That matters later, because it is the assumption a restaurant table breaks.
Milgrom was modelling salesmen and share prices, not dinner — the application to a group check is an analogy, not a measured finding about diners. What it offers is a shape for something the fairness literature leaves alone: in his model, withholding is itself informative, and the inference runs one direction — whatever wasn’t shown probably wasn’t in my favor. Whether a friend reading “just send me $34” performs anything like that inference is untested. It is a hypothesis the model makes worth entertaining, not a result.
That suggests a reading of the awkwardness of asking, though the study tests none of it: requesting the receipt feels like an accusation, so the doubt mostly doesn’t surface as a dispute at all. It surfaces as a “I’ll get you tomorrow” that never resolves. That is an inference about how the mechanism would show up at a table, not a measured effect on repayment.
Why doesn’t everyone just show the receipt?
Because disclosure is rarely as free as the model assumes, and where it isn’t, the theory’s tidy prediction breaks. Milgrom’s result — and Sanford Grossman’s, published the same year — predicts unraveling: the best-quality seller discloses to separate from the rest, the next-best follows, and disclosure cascades until only the worst stays quiet.
It frequently fails to happen. In their Journal of Economic Literature survey of quality disclosure, David Dranove and Ginger Zhe Jin are direct about it: “In reality, there are many markets in which voluntary disclosure is incomplete.” They list the assumptions the unraveling result depends on — among them that disclosure is costless, that quality sits on a single well-defined dimension, and that consumers hold rational expectations about what silence means. Violate any of them and the cascade stalls.
A restaurant table violates most of them at once. Photographing a receipt and circulating it is not costless — it costs the awkwardness of appearing to audit your friends. “Quality” is not single-dimensional: the shared appetizer, the bottle of wine two people finished, and the tip are three different arguments, not one. And the group’s expectations are anything but rational — among people who barely know each other, nobody is confidently inferring anything.
The practical read: Dranove and Jin study disclosure in markets, not diners — but the mechanism they identify suggests the likeliest explanation at a table is not that anyone is hiding something. It is that the act of sharing carries a social cost, and no one wants to be the person who introduced it.
Sources: Milgrom, The Bell Journal of Economics, 1981; Dranove & Jin, Journal of Economic Literature 48(4), 2010 (quoted from the NBER working-paper version)
What does the IRS require a restaurant receipt to prove?
Four things: the name and location of the restaurant, the date, the amount — and the number of people served. That is IRS Publication 463’s standard for a restaurant receipt to substantiate a business meal. The rules are written for business-expense substantiation, not for splitting a check with friends — but it is a striking place to find a version of this article’s argument already written down. A body that has spent decades deciding what makes a meal receipt credible does not accept the amount by itself. It also wants to know how many people the meal covered.
That last line is the IRS conceding, in its own guidance, that memory is unreliable: a contemporaneous record outranks a later statement because later there is “generally a lack of accurate recall.” The publication is administrative guidance rather than the tax code itself, and its bar is timeliness rather than instant capture — it accepts a weekly log. Every group that has tried to reconstruct a dinner three days afterward already knows the underlying point. The IRS wrote it down.
The $75 threshold is the other half of the lesson, and the transferable one. It is a rule about business-expense substantiation and says nothing about dinner — but the principle behind it does travel: verification costs something, so the paperwork is scaled to the stakes rather than demanded uniformly. Groups already apply a version of this instinct. Nobody itemizes a $12 coffee run. The judgment worth borrowing is only that the threshold should be deliberate, and that plenty of group bills sit well above wherever a table would set it.
Source: IRS Publication 463 (2025), Travel, Gift, and Car Expenses, chapter 5 (Recordkeeping)
Does making a split checkable make it colder?
It depends entirely on whether the record is binding. This is the sharpest finding in the trust literature for anyone designing around shared money, and it cuts against the instinct to formalize. Deepak Malhotra and J. Keith Murnighan ran two laboratory experiments, published in Administrative Science Quarterly in 2002, on what contracts do to interpersonal trust.
Their prediction, and their result: binding contracts lead people to attribute the other party’s cooperation “to the constraints imposed by the contract rather than to the individuals themselves, thus reducing the likelihood of trust developing.” You cooperated because you had to. That tells me nothing about you. And when binding contracts were subsequently removed, trust “dropped significantly.”
Malhotra and Murnighan tested contracts in a laboratory, not receipts at a table, and nothing in their design speaks to bill splitting. But the variable they isolated — whether cooperation is compelled or chosen — is the one that distinguishes a receipt from a rule. A shared receipt compels nobody. There is no enforcement mechanism and no penalty attached to it. It makes the claim checkable and leaves the choice where it was, which is a very different object from a ledger with a due date.
This is where the article’s own claim gets its second half. Milgrom says show everything; Malhotra and Murnighan say do not turn showing into compelling. Both hold only for a record that is always available and almost never invoked — present enough that nothing looks withheld, inert enough that nobody is being held to it. Their experiments say nothing about passive verification either way, so that band is mine to argue for rather than theirs to have proven. It is also the only design target that satisfies both findings at once.
Source: Malhotra & Murnighan, “The Effects of Contracts on Interpersonal Trust”, Administrative Science Quarterly 47(3), 2002
What happens when verification turns into enforcement?
The norm it was protecting can be replaced by a price — and the replacement does not reverse. Uri Gneezy and Aldo Rustichini demonstrated this across ten Israeli day-care centers in 1998, in a study published in The Journal of Legal Studies under the title “A Fine is a Price.” Parents were arriving late to collect their children. The researchers introduced a fine for lateness at six of the centers; the other four served as a control group.
The authors’ reading is that a penalty enters an incomplete contract — social or private — and changes the information people have about the situation. Before the fine, being late meant imposing on a teacher. After it, lateness had a posted price and could simply be bought. And once the arrangement had been repriced, withdrawing the fine did not restore what it displaced.
One intervention at ten day-care centers does not establish that every informal arrangement behaves this way, and the authors do not claim it does. What it does establish is that the failure mode exists and that reversal is not guaranteed — which is enough to be careful. Note what they actually tested, though: a monetary fine. A late fee between friends, or interest on an unpaid share, is the same shape. Merely expecting people to settle tonight is not, and nothing here says a shared record carries that risk. The warning is specific to attaching a price to the failure, and it says you should not assume you can take the price back off.
Source: Gneezy & Rustichini, “A Fine is a Price”, The Journal of Legal Studies 29(1), 2000
Does being checkable change what people do, or just what they say?
It is a live question in the research, not a settled one, and it is worth stating honestly rather than overselling. Jennifer Lerner and Philip Tetlock’s review in Psychological Bulletin surveys what they describe as a now-extensive literature on how accountability shapes social judgments and choices — and one of the four questions the review organizes itself around is precisely whether accountability “alters how people think or merely what people say they think.”
One thing follows directly, with a caveat about scope. What Lerner and Tetlock review is accountability — expecting to justify a judgment to an audience — which is a stronger condition than a record merely being available to look at. Even so, their headline is that the effect is conditional rather than automatic: accountability can attenuate, have no effect on, or amplify a given bias depending on the ground rules. Anyone promising that a shared record simply makes people behave better is overselling a literature that says it depends.
The design conclusion is mine rather than theirs, and it is the one thing in this article I would defend without a citation: if the record only appears once someone is defending a position, it has already changed character. A receipt produced during an argument is evidence in a dispute. The same receipt shared when the check lands is just how the group sees the bill.
Source: Lerner & Tetlock, “Accounting for the Effects of Accountability”, Psychological Bulletin 125(2), 1999 (abstract)
What makes a receipt a good trust instrument?
Always available, almost never invoked — that is the target, and it cashes out as four criteria. These are mine, not the literature’s: a receipt only does the work described above when it is shared before settlement, itemized rather than totaled, visible to everyone in the same form, and carrying no enforcement. Most of the ways a split turns sour are a failure of one of the four.
This is the design splitty is built around, and the reason the camera comes first. It scans the receipt and reads the line items, each item is assigned to the people who actually shared it, and tax and tip are distributed proportionally to each person’s share rather than spread evenly across the table. What each person receives is a pre-filled request in their own payment app — a number assembled item by item, rather than a total divided by headcount.
The part that matters most for the argument here is the Share Split link. Recipients open a web page with their share and the items behind it, and settle through their own preferred payment method — only one person needs the app, and nobody else has to create an account to see what they are being asked to pay for. Everyone sees the same record. It binds no one.
Evidence offered up front is context. The same evidence produced after a challenge is a rebuttal.
An itemized record can be checked line by line. A single total can only be believed or doubted.
One shared link, not a screenshot forwarded to whoever complained. Everyone sees the same record.
No ledger, no running balance, no penalty. splitty settles tonight’s bill and stops.
What this changes about how to ask for money
Attach the evidence to the ask, every time, and stop treating that as passive-aggression. The instinct that sending an itemized breakdown implies distrust probably has it backwards. On the disclosure model, it is the bare number that carries the unflattering signal, because a claim with nothing behind it is the one a careful reader has reason to discount.
None of which requires anybody to audit anybody. The receipt does its work by being available — a shared record, produced at the table, that nobody has to invoke. The math was never the hard part, and neither is the trust. What was missing was the artifact that connects one to the other.
FAQ
Questions & Answers
01 Should you show the receipt when splitting a bill?
Yes, and send it with the request rather than waiting to be asked. In the standard economic model of disclosure — Milgrom's 1981 result, which studied salesmen rather than diners — a receiver treats withheld information as unfavorable, so a bare per-person total plausibly carries a worse signal than an itemized breakdown does. Sharing proactively also removes the awkwardness of someone having to ask, which is a likelier explanation for unshared receipts than anyone hiding anything.
02 Is asking for the receipt rude?
It can read that way, which is likely why the question so often goes unasked. No study measures how diners react to being asked for a receipt, so treat that as a reading of the situation rather than a finding. Either way the fix is structural rather than social: if the itemization is shared by default when the check lands, nobody ever has to be the person who requested it.
03 How many items are on a typical restaurant bill?
Across splitty's US-leaning restaurant receipts, the average bill carries about nine line items, and they vary widely in size — about 48% total $150 or more, while roughly one in seven comes in under $50. So a single per-person total is compressing about nine separate claims into one number the other person is asked to accept on faith, with no standard basket to check it against.
04 What does the IRS require on a restaurant receipt?
IRS Publication 463 states that a restaurant receipt substantiates a business meal if it shows the name and location of the restaurant, the number of people served, and the date and amount of the expense. Documentary evidence is generally required, with an exception for expenses other than lodging under $75.
05 Does tracking who owes what damage friendships?
Formal enforcement may. In laboratory experiments — on contracts, not on bill splitting — Malhotra and Murnighan found that binding contracts led people to attribute cooperation to the constraint rather than to the person, interfering with trust, while non-binding arrangements did not carry that cost. Applying that to a table is an extension of their work rather than a result of it, but the distinction it turns on is real: a shared receipt makes a claim checkable without compelling anyone, which makes it a record rather than a rule.
