Not Steffen’s: a gift for the paper’s authors, with the beta line in the footer and no signature or counter. No recipient named. No ideas given; every choice below is the engine’s.
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 (January 2026): 64–79. Published online 11 September 2025. DOI 10.1038/s41562-025-02293-4. Preregistered as AEARCTR-0009901. Data and replication code on OSF (DOI 10.17605/OSF.IO/RH7EB) and GitHub (LaudineC/RCT_RR_NHB).
What was available: the typeset PDF with the main text, Tables 1–2, Figures 1–6, Extended Data Figures 1–3 and Extended Data Tables 1–6, and the reporting summary. Not the supplementary information, not the data. Numbers from the Extended Data tables are typeset and exact; the subgroup effects of Figures 5 and 6 are read off the figures where the text does not give them, and are tagged as such in the config block.
Primary: an estimate. A three-arm RCT with 1,849 pregnant women recruited in eight Paris-region maternity wards, randomised to an active control, five text messages with short videos (information only, €5.50 a family), or the same videos plus phone calls from a trained helper who diagnosed the family’s needs, chose a childcare solution with them and filled in the forms (information + support, €55 a family, 15 minutes on average; about half of those offered the calls took them). One year later: did the household apply for early childcare, and did it get a place. Intention-to-treat effects overall, by the mother’s education (the paper’s SES measure), by whether she was born abroad, and for subgroups facing high information or psychological costs; average effects on those who actually took the calls, by instrumenting take-up with assignment.
Secondary: an argument. The multilevel framework of non-take-up: information costs, process costs (the administrative burden of the forms), psychological and social costs, and beneath them all structural barriers: too few places and town-hall allocation rules (“previousness”) that favour dual-earner families. The paper’s reading of its own null on access rests on this argument; the trial does not identify the allocation rules.
The game reconstructs the estimate as a data-generating process and uses the argument only in words, where the paper does.
A counterintuitive result, twice over. Policy and much of the public believe that families who do not use a service need to be told about it: cheap, scalable information. The paper shows that five videos and reminders change nothing it can measure, while a helper on the phone gets mothers without a degree and mothers born abroad to apply about as often as the others: for those who took the calls, the application gap nearly closes. Then the second turn: the applications go in, and the places do not follow. The gap in places between mothers with a degree and mothers without is as wide as before. Where the calls did win crèche places, they went to mothers with a degree, both parents in work, who swapped a childminder for a crèche. What stands between an application and a place is not in the mother’s hands.
The kind of result shapes the game: the player must be able to act on the belief (send information), meet the null, then act on the finding (send help), cheer at the applications, and meet the wall.
A hundred women are waiting at a maternity ward in Paris, each with a baby on the way, and most of them want a crèche place for it. You hold a purse of public money for four years. You can send everyone a few text messages with little films about how to apply, which is cheap. Or you can have a helper phone some of them and fill in the forms with them, which costs ten times more a family. The films change nothing. The phone calls get the mothers who were falling behind to send in their forms. But the town hall gives out the places the way it always did, and the mothers who were falling behind still get fewer. The help worked on the mothers; it could not work on the town hall.
| Game | Position | Format | Feedback device | Chassis |
|---|---|---|---|---|
| Leavitt’s Cloud | Experimenter | Discovery | The dial and fold, then the chart (the drop, the trajectory) | Austere / Instrument: a glass plate on a light box, velvet and lavender glass, violet ink; Spectral + Josefin Sans; pill buttons; dark |
| Jevons’s Noon | Controller | Challenge | The fill (squares of coal) and the trajectory (the curve) | Instrument: Jevons’s engraved plate, cream and soot, furnace orange; Bodoni Moda; engraver’s double-rule buttons; light |
| Julian’s Window | Subject | Dilemma | The strip (hers) | Drawn: shōjo manga, ink on white, blanket rose; Delius + Italiana; panel buttons; light |
What this game does differently. Position: the fund, a controller; this is the seat the paper plainly wants (the policy question is the Cnaf’s own, and the paper’s two treatments are two things a fund can buy), so the position is shared with Jevons’s Noon, and the other three dimensions differ instead. Format: strategic rather than a challenge against a stock: the player acts on the applicants, and a second party, the town hall’s commission, answers with the places by its own rule, which the player does not hold. Device: the counter (two glowing numbers in the aim’s units) and the before/after (the player’s cohort drawn over the ghost of the cohort that got nothing); not the fill, not the trajectory, not the drop, not a strip. Chassis: Techno, the ticket-number display of a French public counter, on a light institutional wall: after Leavitt’s dark glass, a light screen; after a fine serif, a Bodoni and a comic hand, a signage grotesque; after pills, double rules and panels, ticket stubs.
The aim, on its own line: Four years and a purse: get the mothers without a degree a place as often as the mothers with one.
The yardstick, in the aim’s units: the board shows two numbers, each the gap between the two groups in a ward of a hundred mothers: how many more mothers with a degree applied than without, and how many more got a place. The second is the aim.
The verdict leads with: You spent [the purse] on [the four orders]. The gap in applications [closed by a third in the years you called the mothers without a degree]; the gap in places [hardly moved: 33 to 30 in a hundred]. The paper found the same: the videos changed nothing it could measure; a helper on the phone got mothers without a degree, and mothers born abroad, to apply nearly as often as the others; places did not follow, except for the few who knew least about the system; and the crèche places the calls did win went mostly to mothers with a degree. What stands between an application and a place is not in the mother’s hands: too few places, and the town hall’s rules for giving them out. Then what the paper says about the rules. No score, no best path.
A. The Trial (Experimenter, Discovery). The player runs the RCT: splits the ward into arms, picks the arms (nothing, videos, calls), picks how many families, waits a year, and reads noisy results; a second trial lets them re-split. Controls: arm sizes. Deliberation: how big a trial to see a 4-point effect. Discovery: the videos’ null, then that the half who picked up the phone look different from the half who did not, and why assignment, not take-up, is the fair comparison. Fun: anticipation, discovery. Replay: a different split. Weak for the stranger: the action is sample sizes, and the discovery is the method’s, not the mothers’.
B. The Counter (Controller, Strategic) — the pick. The player is the fund. Each year a new ward of a hundred expectant mothers, a purse that carries over, two tools at the paper’s prices, five targets; the player gives one order, sends it, and the year runs: messages go out, phones ring, the June commission sits, the board lights up. Controls: tool × target, one order a year, four years. Deliberation: cheap and wide or dear and narrow; and whom. Discovery: the videos do nothing; the calls close the application gap for the targeted mothers; the places do not follow; the places the calls do win go to the front; the one target where places move is the mothers who knew least. Fun: anticipation (the year runs), agency (the application bars move for the mothers you chose), discovery (the board’s second number), mastery (year three is spent better than year one). Replay: a different order of orders, the birthplace cut, the mothers not in work.
C. The Commission (the reckless one: the structural barrier as the player). The player sits at the town hall in June with the Seine-Saint-Denis supply (35 places per 100 babies), a stack of applications, and the rule to set: first come, both parents in work, social priority, a lottery; the applications are those the calls produced. Discovery: no rule closes the gap at 35 per 100 unless it favours the back, and “both parents in work” hands the places to the mothers with a degree. Fun: agency, discovery. Replay: other supplies (Paris at 80). Why it is reckless: the paper does not model the allocation; every number in the commission would be ours, and the discovery would follow from our rule, which build.md names as a mechanic and not a finding. Its walkthrough is in section 9.
A fourth, written out and dropped: The Waiting Room (Subject): one mother, no degree, born abroad, walking the year: watch the video or not, pick up or not, fill in the form or not. Every outcome would be a probability drawn for one person from an average effect, which is the trap build.md names; and the player is told, not finding.
The pick, B. Minute 0: a pale wall, a black board with red digits reading 100, a title, three lines (a maternity ward, a purse, two tools), the aim on its own line, a ticket that says BEGIN. Tap. Minute 1: the ward with no help. A staircase of three steps up to a counter window: WANTS A PLACE, APPLIED, GOT A PLACE. Two bars on each step, blue (with a degree) and orange (without). The bars shrink as they climb, the orange faster; the board reads APPLIED 24, PLACES 33. A line says what the stairs show and that this is what happens when the fund does nothing. The player is wondering: where does the orange go? Tap NEXT. Minute 2: year one. The purse reads 3000. Two tools as ticket stubs: VIDEOS, €5.50 a mother, five text messages with short films on how to apply; CALLS, €55 a mother, the videos plus a helper who phones and fills in the forms. Five targets: everyone (100), without a degree (39), born abroad (46), who know least about crèches (15), not in work (30). The player taps VIDEOS and EVERYONE: the board shows the cost, 550, against the purse. They are weighing: that reaches everyone for the price of calling ten. They tap SEND. The year runs: messages go out across the ward, the board counts the months, the stairs fill, the shutter opens, the places bars fill, the shutter drops. The board reads APPLIED 24, PLACES 33. The game says: a hundred mothers got the five messages; about as many applied as with nothing, and about as many got a place; the paper found the same. Frown. Minute 3: year two, purse 5450. Calls, without a degree, 2145. SEND. The orange APPLIED bar grows past its ghost: the board reads APPLIED 15. Grin. The PLACES bar: 30. The game says: of the 39 you called, about half picked up; the mothers without a degree applied nearly as often as the others; the town hall gave out the places as before; the paper found the same, and says why. Minute 4: year three, purse 6305. The player wants to know whether anyone can be got a place. Calls to everyone, 5500? Or the 15 who know least, 825? They try the 15: the board’s PLACES number drops to 28, and the game says that of the fifteen, one in twenty had got a place before and now one in three; a small group, the paper says, and an imprecise estimate. Minute 5: year four, calls to everyone; the board’s places number barely moves and the game says where the crèche places went. The verdict.
The reckless, C. Minute 0: the same board, the commission’s table. Minute 1: a stack of applications, orange and blue; 35 places. The rule as four ticket stubs. The player picks “both parents in work” because it sounds fair; the places go blue. Picks the lottery; the gap closes to the application gap. Picks social priority; the gap reverses. By minute 3 the player has found that the rule decides, which is true and not the paper’s; nothing in the paper says what a lottery would do in Paris. “And I don’t know what I’m supposed to feel”: the player operates a thing the authors did not measure. Dropped; what it gives is kept in words at the shutter.
B, The Counter. The player is in the paper’s own seat with the paper’s own two purchases at the paper’s own prices, and the two reversals land inside the play, in the order the stranger will meet them: the cheap thing first, then the helper, then the wall. The sentence afterwards has no “guess”: I sent everyone the videos and nothing happened; I had the mothers without a degree phoned and they applied like everyone else; and the town hall gave out the places exactly as before. Originality: no game in the table has had the player buy a policy for a population and watch an RCT’s estimate answer, nor made a null result the first thing the player meets.
Author test. The authors would recognise the seat (the Cnaf’s), the prices, the half who picked up, the application gap closing for the targeted mothers, the places not following, the crèche places going to the dual-earner mothers with a degree, and the mothers who knew least as the one place the support moved access. What they will enjoy is watching a player do what policy does: buy the videos first.
Runner-up: A, The Trial. Switch if the author wants the identification lesson (the half who picked up against the half who did not) as the spine, for a methods class; it costs the stranger the mothers.
What the reckless concept would have given. The structural barrier under the player’s hand. Kept: the counter’s shutter, which opens in June and drops with tickets still held, and the line at the end about the town hall’s rules. Given up: setting the rule.
Register: the ticket display of a French public counter (Techno family). From the setting of the data: the mairie’s crèche desk, the Cnaf’s guichet, the maternity ward’s waiting room: a pale institutional wall, a black board with red seven-segment digits, numbered paper tickets, a shutter. Nothing of the journal.
#EDE7DC (the page; a pale institutional cream); ink #1B1C20 (type, the board’s body, the shutter); LED red #FF3B2E on the board only, with its unlit segments in #3A2422; République blue #2242A6 (mothers with a degree, or born in France); tangerine #E8792B (mothers without a degree, or born abroad); ticket #FFFDF7 for the stubs.The information-only arm’s subgroup estimates (all nulls; the game uses the overall null for every group); the minority-norms, never-used and low-trust subgroups (two exploratory targets are enough); the mothers’ labour supply (one line in the verdict); the Paris / Val-de-Marne / Seine-Saint-Denis supply differences (one ward); the post-lasso and logistic robustness; the balance table; attrition; the timing of allocations within the year; the paper’s comparison with the German trial.
Kind: reconstructed. mechanism.js runs a data-generating process built from the paper’s estimates: for each group, the untreated share at each step (wants, applied, got a place, with crèche places inside places), plus the paper’s intention-to-treat effect of the chosen tool for that group, times the share of the group the order covers; access is capped at application and crèche places at both. The information-only arm is played as zero: its point estimates (−0.01, −0.02, +0.01, −0.01) sit inside simultaneous confidence intervals that include zero, and applying a non-zero null to one group would widen a gap by rounding.
Where each discovery comes from. The videos’ null, the application effect for the targeted mothers, the absence of a place effect for them, and the crèche places for mothers with a degree are the paper’s estimates (Table 2, Extended Data Tables 3 and 4). The low-knowledge place effect is the paper’s exploratory estimate (text and Fig. 5), converted from ATT to ITT by the first stage. The purse and the four years are design choices and produce no finding: they decide only how many orders the player gets and which combinations are affordable. The calibration check runs every order under alternative cross-shares and confirms that no order brings the place gap by degree under 25 in a hundred (floor 26), that only calls narrow the application gap by five or more, and that videos move nothing; the upper bound of the paper’s interval for places among mothers without a degree (+0.15) would leave a gap of 18, which the result screen names as the loose end.
calibration.js checks, against the paper’s typeset numbers: the two cuts’ weighted control means reproduce Table 2 within 0.02 (five outcomes, two cuts); the information-only point estimates sit inside their intervals, which include zero and rule out effects beyond ±0.08; the subgroup ITTs for calls average to Table 2’s overall ITT within 0.02 (four outcomes, two cuts); ATT = ITT / 0.52 within 0.015 of Extended Data Table 2 (four outcomes), and the text’s subgroup ATTs (0.17, 0.15) scale the same way; the derived low-knowledge and inactive ITTs land where Figures 5 and 6 put them; the headline order reproduces Extended Data Table 3 exactly (0.69, 0.39; gaps 24→15 and 33→30) and the complier gap is under 0.10; crèche places under calls to everyone reproduce Extended Data Table 4; cross-shares are consistent; the rest-of-ward ladders reconstruct Table 2; all 11 orders × 7 groups keep crèche ≤ place ≤ applied ≤ 1; costs are the paper’s prices times the counts; the purse rules; the robustness runs of section 16. Result: 72 checks, all pass (calibration-result.txt).
No score, but the obvious ways of playing were run against the mechanism before any screen existed. The gap by degree (mothers per hundred), applications / places, after each year; start 24 / 33:
| Strategy | Year 1 | Year 2 | Year 3 | Year 4 | Spent |
|---|---|---|---|---|---|
| videos every year | 24 / 33 | 24 / 33 | 24 / 33 | 24 / 33 | €2,200 |
| calls to no-degree every year | 15 / 30 | 15 / 30 | 15 / 30 | 15 / 30 | €8,580 |
| calls to abroad every year | 21 / 31 | 21 / 31 | 21 / 31 | 21 / 31 | €10,120 |
| save, then calls to all, twice | 24 / 33 | 16 / 30 | 24 / 33 | 16 / 30 | €11,000 |
| calls to those who know least, every year | 20 / 29 | 20 / 29 | 20 / 29 | 20 / 29 | €3,300 |
| calls to mothers not in work, every year | 20 / 32 | 20 / 32 | 20 / 32 | 20 / 32 | €6,600 |
| videos, no-degree, know least, all | 24 / 33 | 15 / 30 | 20 / 29 | 16 / 30 | €9,020 |
Reading: the decisions matter for the application gap (24 down to 15) and for the purse, and nothing the player can buy moves the place gap by more than four. That is the paper’s sentence, and the game lets the player find it rather than telling it: the one simple rule that wins everywhere on applications (call the mothers without a degree) loses on places like every other, and the cheapest order that moves places at all (the fifteen who knew least) is the exploratory one, which the result screen says.
mechanism.js written headless with every number tagged; calibration.js run; one check failed on the first run (videos targeted at one group widened the gap by rounding a null point estimate), fixed by playing the information-only arm as zero and keeping the point estimates and intervals in the config for the check and the About screen.loop.js played the four-year loop on placeholder bars: the printout showed something visible changing after every order, the null visible as bars on their ghosts, and two wording slips (“the 100 everyone”, “one in 3”), fixed.template.html built on the declared chassis: seven-segment digits drawn on canvas (no LED font), Barlow and Barlow Condensed embedded, ticket stubs with perforated edges, the staircase with the shutter; about.html; build.py verifies the embedded mechanism byte for byte.session-test.js drove a full session at 380 px in headless Chromium: an unaffordable order refused, four orders, help opened mid-play and closed, three views, the verdict, About, a retry, and a second run that saves and then calls everyone. No errors, no horizontal scroll.Steffen asked for a title that reflects the substance rather than alluding to an author. The Crèche Counter: the town hall’s counter where the places are given out, which the player cannot reach, and the board that counts the gap. The file is now the-creche-counter.html; the folder stays carbuccia/.