There is no manual for what happens when Child Protective Services knocks on your door.
Not for the mother who has never broken a law in her life but answered the door on the wrong day. Not for the father who doesn’t know whether he’s required to let them inside. Not for the grandmother who is watching her grandchild disappear into a system she doesn’t understand and can’t afford to fight.
The law exists. It’s written in statutes. It’s been interpreted by courts. It’s been shaped by federal funding requirements that most families will never know about.
But it’s scattered across hundreds of state codes, thousands of court decisions, federal regulations, administrative rules, and agency policy manuals. No single resource has ever mapped the entire system in a way that a person — or a machine — can actually navigate.
Today, we’re changing that.
Project Milk Carton is releasing the US Child Welfare Decision Chain Dataset — the first open-source, structured, machine-readable legal knowledge graph of the entire child protective services system in the United States.
It covers all 50 states and the District of Columbia. It maps decision points from the moment a report is made to the final outcome. And it’s free. Forever.
The repository is live now here!
https://github.com/SpartanAltsoba/child-welfare-decision-chains
What’s In It
The dataset contains 2,854 files organized across 2,142 decision nodes — 42 per jurisdiction.
Every node represents a specific point in the CPS process where a decision is made, an action is taken, or an outcome is determined. At those points, the dataset includes:
The statute that governs it — with the actual citation and a URL to the official text
The constitutional constraints — what the 4th and 14th Amendments may require at this step under applicable law
The burden of proof — what standard of evidence is required, and who bears it
The timeline — how long the agency has to act, and what may happen if they miss the deadline
The consequences of a violation — potential legal remedies and procedural responses available under applicable federal and state law, including, where appropriate, Section 1983 claims
The funding incentives — where federal money flows at this decision point, and whether the funding structure may create incentives against keeping families together
What happens when CPS shows up — what authority they have to enter your home, what rights you have at that moment, and what you should and shouldn’t say
That last one matters more than most people realize.
The Two Axes
The dataset is built on two axes that intersect at every node.
The Vertical Axis: Who Governs
This is the hierarchy of legal authority — from the Constitution down to a caseworker’s desk.
U.S. CONSTITUTION: 4th Amendment: Government removal of a child can constitute a seizure, subject to jurisdiction-specific standards governing reasonableness, judicial authorization, and exigent circumstances. 14th Amendment: Parents have a fundamental liberty interest in the care, custody, and management of their children. 42 U.S.C. § 1983 may provide a remedy for certain violations of federally protected rights.
FEDERAL STATUTES: CAPTA, ASFA, Title IV-E, ICWA, FFPSA. These laws establish a combination of federal requirements, minimum standards, and conditions on federal funding. States must comply with the requirements applicable to them. Consequences for noncompliance vary by statute and may include corrective action, funding consequences, or judicial remedies where authorized.
STATE CONSTITUTION: May provide GREATER protection than the federal Constitution. Example: Texas protects the right to a jury trial in TPR proceedings when properly demanded.
STATE STATUTES: The specific child welfare laws of each state.
ADMINISTRATIVE RULES: Agency regulations and standard operating procedures.
CASE LAW: Court decisions interpreting all of the above.
Every node in the dataset is anchored to this hierarchy. If a state procedure conflicts with an identified constitutional constraint, the dataset flags it.
The Horizontal Axis: What Happens
This is the decision flow — the process a case follows from entry to exit.
REPORT → SCREEN → INVESTIGATE → ACT → OUTCOME → OVERSIGHT (with FAILURE branching to LAWSUIT)
Twelve ways a case enters the system. Six decision points. Six possible actions. Six outcomes. Six failure modes. Six oversight mechanisms. That’s 42 nodes per state, and we have all 51 jurisdictions.
Follow the Money
One of the most important features of this dataset is funding incentive tracking.
At every decision node, the schema tracks:
Whether Title IV-E reimbursement is triggered
The federal match rate
Whether adoption and legal guardianship incentive payments apply
Which CFSR performance items are affected
Whether the funding structure may create a perverse incentive against family preservation
This matters because the child welfare system operates in significant part on federal funding. The federal government reimburses states for a share of eligible Title IV-E foster care expenditures. It also provides Adoption and Legal Guardianship Incentive Payments to states based on improvements in adoption and guardianship performance relative to established baselines. The financial architecture of the system can create measurable incentives at specific decision points.
We’re not making accusations. We’re showing the structure. The data speaks for itself.
Why Open Source
We’re a 501(c)(3) nonprofit. Our mission is child welfare transparency. Keeping this behind closed doors would contradict everything we stand for.
But there’s a practical reason too.
This dataset is incomplete. We know that. We’re being honest about it.
The federal baseline is done. All 51 state chains are populated with real statute citations. The constitutional plane is complete. But there are gaps:
49 of 51 state constitutional planes are empty templates. Only Texas and California are filled.
Case law needs expansion. We have CourtListener data for every state, but the nodes need more specific controlling cases.
Administrative rules vary wildly by state and are often buried in agency policy manuals.
We can’t fill these gaps alone. But the legal community can.
If you’re a law student, pick a state and fill in the constitutional template. One state is one pull request. Your contribution will be used by families, attorneys, and researchers across the country, and your name will be on it.
If you’re a family law attorney, add the controlling cases you know by heart. The ones that actually win motions.
If you’re a researcher, use the data. Build on it. Publish with it. The BibTeX citation is in the README.
The license is CC BY-SA 4.0 — use it for anything, including commercial purposes, as long as you credit Project Milk Carton and share any improvements under the same license.
Who This Is For
For families: This is a map of the system you’re in, including rights, requirements, timelines, and authorities mapped at each stage of the process. What the agency must do, what they can’t do, and what may happen when they break the rules.
For attorneys: Structured cross-state comparison. If you practice in multiple jurisdictions, or if your client’s case touches an interstate compact, the data is already organized for side-by-side analysis.
For journalists: Follow the money. The funding incentive fields at each node show where federal dollars flow and where the structure may create pressure to remove children rather than preserve families.
For researchers: 2,142 nodes of structured data with provenance chains. Every claim traceable to a source URL. Ready for quantitative analysis.
For AI developers: The LEAF tool call hooks in the schema are designed for LLM integration. Build tools that reference real law instead of hallucinating it.
For legislators: See exactly how your state compares to every other state and applicable federal requirements. Identify compliance gaps before the next CFSR review finds them for you.
What Comes Next
This release is the beginning, not the end.
We will continue populating the gaps — state constitutions, case law, administrative rules. Every update will be pushed to the public repository.
We are applying for grants to fund the next phase: connecting this static dataset to live data sources. Real-time AFCARS outcome data. Live legislative tracking from OpenStates. Automated case law updates from CourtListener.
And we will continue building ARIA — our AI system that traverses this dataset to answer questions, generate compliance reports, and help families understand their rights. ARIA reads from this dataset. She doesn’t guess. She cites.
The dataset is the foundation. Everything else builds on top of it.
Get Involved
Use it: https://github.com/SpartanAltsoba/child-welfare-decision-chains
Contribute: Pick a state, fill a gap, submit a pull request.
Share it: Send this to family law attorneys, law school clinics, child welfare researchers, and journalists.
CPS overreach and how to counteract it is an astonishingly good use case for AI: the laws are often confusing, contradictory, opaque and heavily tilted away from families. Children without families make much easier trafficking victims: they've already been turned into a commodity via various government funding mechanisms, have no true advocates and can easily be disappeared. Beyond just citing case law though I wonder if judicial and investigator bias can be uncovered: there might be certain people in key nodes of power who always rule against families while others try their best in a corrupt system. Usually though it's about following the incentive structure...