the open model · capacity
What the app does with the fact that you work fifty hours
This is the concession, and it is made in arithmetic before anybody is asked for anything. Long hours genuinely reduce how much of the doing you can take on. They do not reduce how much of the remembering you can take on — and in the research, they don't.
The app never says an even split is fair
It has no opinion about that and it is not entitled to one. What it does instead is derive an expectation from your household's own answers — paid hours, commute, other involuntary commitments — and then report the distance between that and what is actually happening.
A number you helped calculate is much harder to wave away than a number handed to you. That is the entire design.
capacity = 112 − paid work − commute − other obligations112 is waking hours in a week, the same constant for both of you, so it cancels and nobody has to argue about who sleeps more. Paid work is meant to include the invisible work-hours — lesson prep, marking, on-call, evening email, the invoicing a self-employed partner has never counted. If the question does not say so out loud, the teacher and the freelancer get systematically undercounted.
Raw capacity is not used raw
Left alone, it produces results nobody would defend. Someone working seventy hours has 42 capacity hours against a partner's 105 — a 71/29 expected split — and by eighty hours it approaches you owe nothing. So the ratio is compressed toward parity:
expected_execution = clamp(0.5 + 0.70 × (raw_share − 0.5), 0.28, 0.72)
expected_cognitive = clamp(0.5 + 0.20 × (raw_share − 0.5), 0.28, 0.72)Two different numbers, because they are two different claims
For execution, paid hours genuinely predict the split. A study of 7,535 dual-earner couples across ten European countries found a wife working more hours than her husband does 30.5 percentage points less housework. At the tails it is starker still: where she contributes under 35% of the couple's combined paid time she does about 85% of the housework; above 60%, he does about two thirds. So the coefficient here is high. The app concedes this almost entirely.
For cognition, paid hours predict essentially nothing. A survey of 2,133 US parents found that employment cuts a mother's physical labour by 17–22% and her cognitive labour by zero — and that out-earning her partner does not move it either. This is the part the app holds the line on, and it holds it because that is what the evidence says, not because it is the answer anybody wanted.
If those two were collapsed into one coefficient, it would be too low to be fair on the execution side and too high to be true on the cognitive one. Two numbers cost an extra sentence of explanation. That is the whole price.
The part of this that is a guess
Stated here rather than buried, because the page would be dishonest without it.
- The cognitive coefficient (0.20) is a guess inside an evidence-supported direction. The evidence says near zero. 0.20 is not zero, and the reason it is not zero is judgement: a coefficient of exactly 0 would assert that paid work has no bearing whatever on what a person can carry, which is a stronger claim than one survey supports and a worse thing to defend in a kitchen. The execution half rests on a large European dataset; the cognitive half rests on a single US survey with no European replication.
- The execution coefficient (0.70) and the 28–72% clamp are guesses, carried unchanged from the first draft of the model and labelled as such ever since. The clamp has some external cover — people asked to judge divisions of labour rate woman-heavy ones as less fair than the mirror image, so a pull toward parity matches a measured contemporary norm rather than inventing one — but the specific numbers are not measured.
Both get calibrated against real households before any of this is settled.
What the app does with the answer
It sets the mark for what your hours suggest on each axis, and reports how far what happens sits from it. Whatever the reading says is said about the household, never about a person.
And the question is only asked at the point where it buys something — not in a set-up questionnaire, and never as a gate. The carrying half of the reading needs none of it, which is why staging it this way is honest rather than a delay. When it is asked, it arrives with its reason and its consequence attached: what your answer changes, and what it does not.