Paris, autumn 2022. A hundred women in a maternity-ward waiting room, each with a baby on the way. Most want a childcare place for it. The town hall gives out the places in June.
You hold the fund's purse for four years. Each year you pick a tool, pick whom it goes to, and send it; the year runs, and the counter's board shows what came of it.
Four years and a purse: get the mothers without a degree a place as often as the mothers with one.
A beta, built from the paper. Builder's review only; the author finishes it.
Four years and a purse: get the mothers without a degree a place as often as the mothers with one.
The ward when the fund does nothing
mothers with a degreemothers without
Three steps up to the counter. Wants a place: asked while pregnant. Applied: a year later, sent in the forms. Got a place: the town hall said yes. Each bar is a share: of every hundred mothers in its colour, how many got this far. The darker band inside the top bars: the ones with a crèche place (a daycare centre, the kind most parents want) rather than a childminder.
The bars shrink as they climb, and the orange ones shrink faster. The board counts the gap between blue and orange, in mothers per hundred, at the applied step and at the place step. The second number is the one you are here to bring down.
Four years and a purse: get the mothers without a degree a place as often as the mothers with one.
Year
of four
Purse, €
to spend or keep
This order, €
nothing picked
mothers with a degreemothers without
Pick a tool, then whom it goes to. The bars stand where they would with nothing; after Send the year runs and they move. A dotted line shows which mothers an order reaches.
To whom
After four years
A beta, built from the paper. Builder's review only; the author finishes it.
The Crèche Counter
How to play
Four years and a purse: get the mothers without a degree a place as often as the mothers with one.
The ward
A hundred expectant mothers; blue bars are the mothers with a degree, orange the mothers without. Three steps up to the counter: Wants a place, Applied, Got a place. Each bar is a share: of every hundred mothers in its colour, how many got this far. The thin outline behind each bar is the ghost: where the bar stands when the fund does nothing. On the top step a darker band inside a bar is the mothers who got a crèche place (a daycare centre, the kind most parents want) rather than a childminder.
The board
Red numbers are the counter's. On the top board: the year, the purse in euros, and the cost of the order you are picking. Under the stairs: the gap between blue and orange at the applied step and at the place step, in mothers per hundred, with the no-help gap small beneath. The place gap is the aim.
Each year
Pick a tool. Videos: five text messages with short films and reminders, €5.50 a mother. Calls: the videos, then a helper on the phone who works out what fits and fills in the forms with her, €55 a mother.
Pick whom it goes to: everyone, the mothers without a degree, the mothers born abroad, the fifteen who know least about crèches, or the thirty not in work. A dotted line shows which bars the order reaches. The cost lights on the board; the purse is €3,000 a year, and what you do not spend stays for next year.
Send. The year runs: messages go out or phones ring (about half pick up), the bars move, the shutter opens in June, the places are given out, and the board counts. The game says what you sent, what the steps did, and what the paper found. Or send nothing and keep the purse.
The switch above the stairs shows the same ward by degree, by birthplace (blue born in France, orange born abroad), or the mothers you called against the rest.
The end
After year four the board shows the four years, and the verdict says what became of the two gaps on your path and what the paper found. There is no score.
The Crèche Counter
About
Built from Laudine Carbuccia, Arthur Heim, Carlo Barone and Coralie Chevallier, “A randomized controlled trial on the effect of administrative burden and information costs on social inequalities in early childcare access in France”, Nature Human Behaviour 10 (2026), 64–79, published online 11 September 2025. DOI 10.1038/s41562-025-02293-4. Preregistered as AEARCTR-0009901; data and code on OSF. Game built from this paper; the paper is the authority where the two disagree.
What the paper found
In 2022 the authors met 1,849 pregnant women in eight maternity wards around Paris and split them at random into three groups. One got nothing but unrelated health messages. One got five text messages linking to short videos about the kinds of early childcare, who qualifies, what it costs, where to apply and when, plus reminders: about €5.50 a household. One got the same videos and then phone calls from a trained helper who worked out with the mother what would fit, and helped with the forms: about €55 a household, a quarter of an hour on average; about half of those offered the calls took them. A year later the authors phoned everyone and asked two things: did you apply for early childcare, and did you get a place.
The videos changed nothing the trial could measure. The calls raised applications a little on average, and that average hid the point: for mothers without a post-secondary degree and for mothers born abroad, who in the untreated group applied far less often than the others despite wanting a place nearly as often, the calls raised applications by about nine in a hundred; for those who took the calls, the application gap nearly closed. Places did not follow. Among mothers without a degree the trial could see no rise in places, and among mothers with a degree, both parents in work, the calls won more crèche places in place of childminders. The one group whose chance of a place rose a great deal was the small group who knew least about the system at the start. The authors read the wall between application and place as structural: too few places, and town-hall allocation rules that favour dual-earner families.
What the game does
You are the fund, four years running, with the paper’s two tools at the paper’s prices and a ward of a hundred expectant mothers each year. Each year one order: videos or calls, to everyone or to one group. The stairs show, for the mothers with and without a degree (or born in France and abroad), how many in a hundred want a place, applied, and got one, over the ghost of the ward that got nothing. The board counts the two gaps. After four years the verdict says what became of them on your path, and what the paper found.
The mechanism, and what it claims
Kind: reconstructed. The game runs the paper’s intention-to-treat estimates as a data-generating process: the untreated group’s shares at each step, by group, plus the paper’s estimated effect of each tool for that group, times the share of the group an order reaches.
The game demonstrates the paper’s estimated effects: the videos’ null, the calls’ effect on applications for mothers without a degree and mothers born abroad, the absence of a measurable effect on places for them, the crèche places won by mothers with a degree, and the large effect on places for the mothers who knew least.
The game assumes everything the trial did not do. The trial treated whole arms; a targeted order in the game applies the paper’s estimate for that group to that group alone and nothing to the rest, with no spillover between mothers and nothing carried from one year to the next. The share of the low-knowledge and not-in-work mothers who are without a degree or born abroad, which the paper does not report, is ours; so are the ward of a hundred, the four years, the purse of €3,000 a year and its carry-over. The bars are expected shares, not draws, so the trial’s sampling noise is not played; where it matters, the game says it in words. The videos are played as exactly nothing, where the paper’s point estimates are a point or two below zero and indistinguishable from it.
The paper establishes the intention-to-treat effects of the two arms, overall and by the mother’s education and birthplace, in the Paris region in 2022–23, with their confidence intervals; the effects on those who took the calls, by instrumenting take-up with assignment; and, as exploratory, not preregistered, analyses with small groups and imprecise estimates, the effects for mothers with little knowledge of the system, mothers not in work and some others. It does not establish what a town hall would do under other rules, what targeting one group would have done to another, or what the same calls would do outside Paris.
The numbers
Every number sits in the config block at the top of the mechanism, tagged with its table or figure. The untreated shares and the effects by group are from Extended Data Tables 3 and 4; the overall ones from Table 2; take-up from Extended Data Table 2; the low-knowledge and not-in-work effects from the text and read off Figures 5 and 6, converted from effects on the treated to intention-to-treat by the take-up rate. Intention at baseline is from the text and Figure 2. On screen, the only numbers are the counter’s: mothers per hundred and euros.
What is left out
The videos’ subgroup estimates (all nulls); the subgroups by trust, by norms and by past use; mothers’ hours of work (the calls let mothers with a degree keep working more); the three départements and their very different supplies of places; the robustness checks; the comparison with a German trial where support was given in person and places did follow.
Credits
Built with Claude under the paper-to-game skill. Type: Barlow and Barlow Condensed (Jeremy Tribby), SIL Open Font License, embedded. Source, design memo and calibration check: github.com/steffenhuck/paper-to-game.