Veröffentlicht 15. August 2026

Everyone thinks they do more than half

Ross and Sicoly asked married couples, separately, what proportion of the housework each of them did. Then they added the two answers together. The totals came out well over 100%.

Nobody was lying. Your own contributions are simply more available to your memory than anyone else's — you remember scrubbing the shower because you were there, with wet knees; you did not witness your flatmate doing it on Tuesday while you were out. Both people report honestly from the evidence they have, and both are wrong in the same direction.

That single finding reframes most chore conflict from a dispute about character into a dispute about information — and it is the reason the chore page shipped today looks the way it does.

What was there before

A bar chart of completed tasks per resident, all time, scaled to the highest number.

Problem	What it actually did
Counted rows, not time	A 2-minute bin run scored the same as a 25-minute shower scrub
Scaled to the maximum	Always paints someone's bar full — a podium, manufactured from nothing
All-time totals	Long residents coast on last year's credit; a newcomer looks permanently delinquent

The middle one is the dangerous one. A public ranking is a punishment mechanic whether or not you call it one, and Herrmann, Thöni and Gächter (Science, 2008) found substantial antisocial punishment across many societies — people punishing the high contributors, not the low ones. Where it was common, the ability to punish destroyed cooperation instead of sustaining it. In a product that places people from many countries into one flat, that is close to a disqualifying result for a leaderboard.

What shipped

A balance, not a ranking. Total the minutes, split evenly, and show each person their signed distance from their own share — a diverging bar with the centre line at exactly your share, scaled to the largest imbalance rather than the largest contribution. A near-even month now reads as near-even. Balances sum to exactly zero, and a test pins that: a fairness number that does not close is not a fairness number.

Weighting is logged duration → the task's estimate → a deliberately small default, so an unestimated chore can never outweigh a measured one. Minutes are also the unit Swiss federal statistics use to value unpaid household work, which makes it a number an institution can read without translation.

One deliberate asymmetry with the money ledger it borrows from: an expense balance is settled by a transfer, but a chore balance can only be settled by doing the next thing. So there is no simplifyDebts counterpart, on purpose, and the UI says so — ein Saldo wird nicht zurückgezahlt, sondern mit der nächsten Aufgabe ausgeglichen.

A definition of done made of actions, not outcomes. The most common shape of a chore dispute is not "you didn't do it" — it is "that's not clean". That argument is unwinnable, because "clean" is a gradient and two people with different standards can both be honestly right. So a checklist item is a binary, observable action:

☐ Boden gewischt
☐ Abfalleimer geleert
☐ Spiegel gewischt

Never "Bad ist sauber". Flatmates never have to agree what sauber means, only whether the floor got wiped. Partial ticks are allowed and stay visibly partial in the history — forcing all-or-nothing pushes people to either overclaim or log nothing, and both corrupt the record. Where a task has no agreed checklist, the page says so out loud rather than hiding the section; in most real disputes, "we never agreed what counts" is the actual finding.

A rota that is a default, not a lock. Whose turn it is, is derived from the rotation order and the completion count — never stored, so it cannot drift from the log. Completing a task advances the rota regardless of who did it, because covering for a flatmate must not cost you your own place in the queue. Anyone may do anything at any time.

The honest case for a rota is not allocation — an honour system allocates work fine. It is that the rota removes the need to ask, and the asking is the expensive part: that is the moment a household conversation becomes a household conflict, and eventually a staff mediation hour.

What we deliberately did not build

No score, no quota, no streak, no karma, no rank, no badge, no peer-punishment mechanic.

A monthly minutes target was the first thing designed and the first thing cut. Extrinsic rewards can crowd out the norm they were meant to support (Frey & Jegen, motivation crowding): a quota converts a shared obligation into an individual one that people then discharge — exactly N minutes, nothing off-list. Worse, it introduces the possibility of failing, and the residents who would fail it most often are the ones having the hardest month. An automatic system that names them delinquent is not a fairness feature.

The shipped version can tell you that you are 40 minutes behind. It cannot tell you that you have failed, and there is nothing for it to escalate to.

Equal time is also not the same as fair. Fair division defines fairness over each person's own disutility, and twenty minutes of bathroom is not the same cost to two people. We do not elicit that — it is expensive to ask and trivial to game — so swapping a turn stays a first-class action rather than a violation. Voluntary trade is the cheap decentralised approximation.

The sentence above the numbers

The heaviest-lifting element on the screen is not a number. It is one line explaining that almost everyone overestimates their own share, because we remember our own work better than other people's.

Without it, a negative balance reads as an accusation, and people argue with accusations. With it, the same number arrives as a correction that applies to everyone in the room — including whoever is currently feeling smug about being in the green.

What would count as evidence

None of the research above was conducted on this population, in this housing, with this software. It shaped our priors about what could go wrong, which is a different thing from evidence that the design works.

What would count is already being tracked, and already has a baseline: incidents per month, relocations caused by incompatibility, and staff hours spent on mediation. If this is doing anything, those move. If they do not, this is decoration and should be called decoration.