Industries / Government, education and community / Social services / Foster care and adoption
Placing a child
Placement is the work of finding a child who must leave home somewhere to live that night, then the right home after that, whether with relatives, a foster family, a group home or an institution, and keeping that home stable.
6
Milestones achieved, from 1868 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1868
Families paid to board children
The Massachusetts Board of State Charities began paying private families to board children in their homes, and in 1869 it appointed an agent to visit the children there. The Adoption History Project at the University of Oregon calls this the beginning of placing-out, the movement to care for children in families instead of institutions; the same timeline dates the orphan trains of New York's Children's Aid Society, which also placed children with families, from 1854.
Credit
Massachusetts Board of State Charities
2007
Foster care beats institutions
In the Bucharest Early Intervention Project, 136 abandoned children aged 6 to 30 months who lived in institutions in Bucharest, Romania, were split by lot: 68 moved to foster families that the project recruited, paid and supported, and 68 stayed in institutional care. At 42 and 54 months the children placed in families scored markedly higher on tests of thinking and learning, and the gain was largest for those placed youngest. A 2026 follow-up calls it the only randomized controlled trial (a study in which chance decides who gets which care) of foster care as an alternative to institutions; at age 22 it found higher body mass index and waist size in the foster care group than in those who stayed.
LessonThe age at which a child leaves an institution matters, and a family placement does not undo every early harm.
Credit
Charles A. Nelson, Charles H. Zeanah, Nathan A. Fox and colleagues, with the Institute of Maternal and Child Health of Romania's Ministry of Health
2009
The UN puts families first
On 18 December 2009 the UN General Assembly adopted the Guidelines for the Alternative Care of Children (resolution 64/142). They say care for young children, especially those under 3, should be in family settings, and that where large institutions remain, states should plan their "progressive elimination" under a deinstitutionalization strategy with set goals.
Credit
United Nations General Assembly
2017
Japan's family target lags
A 2016 amendment to Japan's Child Welfare Act made care in a family setting the first choice, and a 2017 government vision set targets: 75% of children under 3 in foster or family homes by the end of fiscal 2024, and 50% of school-age children by the end of fiscal 2029. A June 2024 report by the Ministry of Internal Affairs and Communications put the overall rate at about 20%; at the end of fiscal 2021 it was 25.3% for children under 3, and 69.4% of about 33,000 children in care lived in children's homes.
LessonA target for family care moves slowly while subsidies still pay institutions for each child they house, which is how Human Rights Watch explains the pace; a 2022 law began funding institutions that turn themselves into foster care support centers.
Credit
Ministry of Health, Labour and Welfare (now the Children and Families Agency); review by the ministry's Administrative Evaluation Bureau
2018
Federal limits on group care
The US Family First Prevention Services Act, part of a law signed on 9 February 2018, limited federal Title IV-E foster care funds for most group care, called congregate care, to 14 days from 1 October 2021, unless the setting meets set standards, such as those for a qualified residential treatment program. In a survey published in March 2026, the Government Accountability Office (GAO) found that 26 of 49 states had not reduced their use of congregate care, and all 49 reported trouble finding suitable placements. One state's share of Title IV-E funds spent on congregate care fell from 17.9% in 2018 to 1.6% in 2024, with states paying more themselves.
LessonLimiting federal money for group care moved much of its cost onto states, and in about half of them did not reduce group care while foster homes were short.
Credit
US Congress (Public Law 115-123, title VII); review by GAO (GAO-26-107592)
2021
Children sleep in offices
In 2021 Texas had months with more than 400 children in its long-term custody and no placement, sleeping in hotels, state offices or state-leased houses staffed by caseworkers, by the count of monitors appointed in a federal lawsuit that was then 11 years old. An expert told the court that most states have around 10 such children at a time. The monitors found that 61% of the children without placements were involved in a serious incident between July and September 2021, and Texas had lost more than 1,000 treatment beds as facilities closed.
LessonPlaces lost faster than new ones open are filled with offices and hotels, where the monitors recorded more harm.
Credit
court monitors Deb Fowler (Texas Appleseed) and Kevin Ryan (Covenant House International), as reported by Texas Public Radio
Watch for
Kin found by software
Maryland's Department of Human Services says that from September 2025 its caseworkers ran more than 4,500 searches with Binti's family-finding software, which searches several record databases for relatives and family friends, and found more than 4,300 possible connections. It says the share of children in care living with kin rose 33% by December 2025, or 330 more children, compared with before its kinship law took effect on 1 October 2024. Binti puts the average at 26 connections per search.
Achieved whenfederal AFCARS data or an outside evaluation confirms the rise in kin placements and shows how many came from the searches.
Claimed by
Maryland Department of Human Services and Binti, February 2026
AI matching ordered
US Executive Order 14359, signed on 13 November 2025, told the Department of Health and Human Services to expand states' use of predictive analytics and AI tools to raise carer recruitment and retention and to "improve caregiver and child matching", and to publish a yearly state scorecard that includes placement disruptions, placements that end before they were meant to.
Achieved whenan independent evaluation shows that AI-assisted matching in at least one state reduced placement disruptions against a comparison group.
Claimed by
The White House, November 2025
Open
01
A test set for matching
An open set of 300 made-up children's referrals and 150 made-up foster and adoptive family profiles, written from published practice guidance, with each child-and-family pair rated for fit by at least 10 experienced placement and adoption workers working blind. Use it to score matching software and general AI models on whether they rank families the way the panel does.
The briefCredit · unclaimed02
A national count of nights
A public tracker of how many children in care spent nights in offices, hotels or other unlicensed settings, by US state and month, built from court monitor reports, ombuds dashboards, legislative testimony and public-records requests. It holds aggregate counts only, each linked to its source document.
The briefCredit · unclaimed
Open 01 · the brief
A test set for matching
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of 300 made-up children's referrals and 150 made-up foster and adoptive family profiles, written from published practice guidance, with each child-and-family pair rated for fit by at least 10 experienced placement and adoption workers working blind. Use it to score matching software and general AI models on whether they rank families the way the panel does.
Why it is still open
The November 2025 executive order calls for AI tools to improve caregiver and child matching, and ACF's 2026 funding notice for regional matching hubs says most agencies still match children with families through "static photo listing services, fragmented data systems, and manual caseworker knowledge". The Associated Press found in 2023 that states had tried the Family-Match adoption algorithm without being able to see how it ranked families. Published research on matching algorithms, such as Terence Highsmith II's paper (first posted November 2024, revised August 2025), works from records of real placements. Searches of the web and GitHub for a synthetic or public benchmark for foster or adoptive matching found none, and found no independent evaluation of a matching tool's results.
Built with what exists
An open-weights language model such as Mistral Large 4 (see the Note "A trillion open parameters") to vary the profiles from a template; LimeSurvey or a similar survey tool for blind ratings; Python for agreement statistics such as Krippendorff's alpha. No data about real children or families is used.
First step
Write 20 child and 10 family profiles from a placement referral template and pilot the rating form with three practitioners.
Done looks like
The set published under an open licence with the panel's agreement score, and at least three tools or models scored on how often the panel's first-choice family appears in their top three.
Open 02 · the brief
A national count of nights
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public tracker of how many children in care spent nights in offices, hotels or other unlicensed settings, by US state and month, built from court monitor reports, ombuds dashboards, legislative testimony and public-records requests. It holds aggregate counts only, each linked to its source document.
Why it is still open
KFF Health News reported in June 2022 that there is no nationwide count of foster children sleeping in hotels or offices. A November 2026 search found state sources, such as the monthly count of these "placement exceptions" published by Washington's Office of the Family and Children's Ombuds, and Texas court monitor reports, but no national tracker.
Built with what exists
Public PDFs and pdftotext; a small model to pull counts out of tables, such as Meta's Muse Ember (see the Note "The jobs agents repeat"); a spreadsheet and a public Git repository; public-records request templates.
First step
List every state with a court monitor, ombuds or audit report that mentions office or hotel stays, and file records requests in five states that publish nothing.
Done looks like
Monthly counts for at least 30 states over at least 12 months, each figure linked to a public document, with states that hold no data marked as such.