🤖HomeMeet the Developer

Accountability Series · Part 2 of 2

The Open Secret of Family Court: A System Designed Not to Be Measured

A class of actors with absolute immunity, no published outcome data, an undefined legal standard, and a federal funding stream that pays states to enforce.

May 4, 2026·18 min read

A Companion Piece

This article is the family court counterpart to The Open Secret of Government IT: A Decade of Failure, Sold as Service. The two systems share a single underlying pathology — institutional actors protected from market and political accountability, presiding over a high-failure regime, insulated from the people they’re supposed to serve by complexity, opacity, and professional capture. The IT version costs the country money. The family court version costs the country something harder to put a number on, and ultimately more important.

If you have not read the IT piece, the structural argument there will make this one land harder. If you only have time for one, this one is more urgent.

What “Failure” Means Here

No national database. No aggregation. No accountability.

In the IT version, the failure rate is roughly 70% — 13% for projects over $6 million, by Standish Group’s count. Failure is measurable: cost overruns, missed schedules, cancellations, abandoned modernizations.

Family court is harder to score because the system has been deliberately designed to make scoring impossible. There is no national database of custody outcomes. No national database of guardian ad litem (GAL) recommendations vs. ground-truth child welfare outcomes five years later. No national database of custody evaluator findings against any independent measure of accuracy. No appellate review for most factual determinations. Most family court records are sealed, fragmented across roughly 3,000 county-level jurisdictions, and actively resist aggregation.

This is not an accident. A system that does not collect data on its own outcomes cannot be held accountable for them. Compare that to federal IT, where GAO at least has the authority to audit, count, and publish — flawed as that process is. Family court has no GAO. It has bar associations and judicial conduct commissions, both staffed and run by the same professional class whose conduct is theoretically being reviewed.

What we can observe — through state-by-state data releases, academic studies, and the lived experience of millions of litigants — is the shape of the dysfunction:

• The federal child support enforcement program (Title IV-D) collected only 18.9% of total support obligations owed in FY2024 when arrears are included, despite consuming $6.6 billion in combined federal and state administrative spending. Roughly 40% of the caseload received no payment at all. That is the system at its most-measured corner.

• Custody evaluations routinely cost between $3,500 and $12,000 or more per case, take months to a year to complete, and produce recommendations that — as a matter of “pure law” — are not binding on the court but, as a matter of practice, are followed almost without exception.

• A parent seeking to overturn a custody evaluator’s recommendation faces a system in which that evaluator has, in most jurisdictions, absolute immunity from suit, and any sustained substantive critique of their methodology is interpreted as obstreperousness rather than legitimate cross-examination.

The stated purpose of the family court system is the welfare of children. By the welfare of children — measurable in mental health outcomes, educational outcomes, suicide rates, substance abuse, incarceration rates, and the simple question of whether a child has both parents in their life — the system is not just failing. It is generating the dysfunction it claims to remediate.

The Accountability Vacuum, in a More Concentrated Form

In the IT article I argued that “accountability” in government is a word people use, not a thing they do. Family court takes that further. In federal IT, when a $200 million project implodes, there is at least a hearing, a GAO report, a press release, and the theater of consequence — even if the actual consequences are minimal. In family court, when a custody decision destroys a child’s life, there is no equivalent layer at all.

The legal infrastructure that makes this possible is explicit:

• Judges have absolute immunity for judicial acts. This is not metaphorical — it is statutory and constitutional doctrine going back to Stump v. Sparkman (1978). A family court judge can issue a custody order based on no evidence, in defiance of every recommendation, in the face of documented bias, and the parent has no civil remedy.

• Guardians ad litem and custody evaluators have absolute or quasi-judicial immunity in nearly every state. That language — disappointed litigants — is doing a lot of work. It conflates “disappointed” with “harmed by professional negligence,” and the law treats them identically.

• Judicial conduct commissions are dominated by attorneys and former judges, whose professional networks and economic interests overlap heavily with the conduct they’re evaluating. The overwhelming majority of complaints are dismissed without investigation.

A protected class with no civil exposure, no real disciplinary exposure, no published performance data, and a deferential appellate posture has, predictably, behaved exactly the way you would expect such a class to behave.

The “Qualifications” of the Decision-Makers

Lawyers who took family law cases, now deciding them.

The IT article observed that many senior government IT leaders couldn’t pass a technical screen at a mid-sized SaaS company. The family court parallel is more uncomfortable to state plainly, but the data is what it is.

Family court judges in most states are former family law attorneys who ascended through bench appointment or low-information judicial elections. There is no requirement that they have any background in child development, attachment theory, family systems psychology, trauma response, or the actual research on outcomes for children of separated parents. They are lawyers who took family law cases, and now they decide them.

Custody evaluators are typically licensed psychologists, marriage and family therapists, or social workers. Their training in custody-specific evaluation methodology is, in many jurisdictions, a weekend course. The actual peer-reviewed research on the predictive validity of standard custody evaluation methodology is, to put it kindly, contested. To put it less kindly: many of the instruments evaluators rely on were never validated for the population or the question they are being used to answer, and the research literature has been saying so for two decades.

Guardians ad litem, in many states, are simply attorneys who took a CLE course. In some jurisdictions they are volunteers with even less formal training. Their recommendations carry near-dispositive weight in court.

The point is structural: a system that grants near-final decision-making authority over children’s lives to a class of actors with no standardized competency requirements, no outcome tracking, no peer review of recommendations against ground-truth outcomes, and no civil exposure will produce the distribution of quality you’d expect from those incentives.

The Bar’s Capture of the Court

The same small group generates, reviews, and polices each other’s recommendations.

In the IT article, the villain was the Big Integrator: large vendors who have spent decades shaping procurement around their own business model. In family court, the equivalent capture is by the family law bar itself.

Family courts run on professional referrals. A judge knows the GALs and evaluators by name; the GALs and evaluators know the judges by name; the lawyers know all of them. The same evaluator gets appointed across hundreds of cases over their career. The same GALs cycle through the same dockets. Disagreement with this professional in-group is treated, by the in-group, as a sign of unreasonableness in the litigant — which is also the thing the in-group is being asked to evaluate.

This produces a structural problem the IT version doesn’t quite have. The same small group of professionals are simultaneously: (a) generating the recommendations, (b) reviewing each other’s recommendations, (c) sitting on the bar association committees that govern the field, (d) advising the legislatures that write the statutes, and (e) populating the judicial appointment pipeline. There is no exit. There is no outsider. The field reviews itself.

The economic incentive that holds this together is straightforward: conflict is the revenue model. A custody case that settles quickly generates modest fees. A custody case that runs for two years with a $10,000 evaluation, GAL fees, parenting coordinator fees, multiple status hearings, and a contested trial generates an order of magnitude more revenue across the same set of professionals. The professionals are not deliberately prolonging conflict — they don’t have to. The system they built rewards prolongation by default.

The “Best Interests of the Child” — Undefined by Design

In the IT article, I argued that requirements failure is the original sin: agencies cannot define what they want, vendors build for the ambiguity, and projects fail predictably.

The family court analogue is the best interests of the child standard — the legal criterion that governs almost every custody, visitation, and parenting decision. It is, by design, undefined. Each state lists factors a court “may consider,” but no factor is dispositive, no weighting is mandated, and the ultimate determination is left to the discretion of the trial judge. Appellate review is heavily deferential.

The parallel

This is functionally identical to a federal IT contract written without acceptance criteria. Whatever the contractor delivers is by definition the deliverable. Whatever the family court judge decides is, by definition, in the best interests of the child — because there is no measurable standard against which the decision can be tested.

If you proposed running federal IT procurement this way — “the agency will pay whatever the vendor delivers, and the vendor’s good faith is presumed, and there is no objective acceptance test” — you would be laughed out of OMB. We run the most consequential decisions in family life this way as a matter of doctrine.

The Federal Money Pipe: Title IV-D

$25.8 billion paid to families in FY2024 against $6.6 billion in costs; the program counts every dollar as its own.

The IT article noted that federal IT spending exceeds $100 billion a year, with the vast majority going to operations and maintenance of legacy systems. The family court parallel is Title IV-D of the Social Security Act — the federal child support enforcement program — and the perverse incentives it creates at the state and county level.

Here is how it works:

• The federal government reimburses states 66 cents on every dollar they spend on child support enforcement administrative costs. The reimbursement is open-ended — there is no ceiling.

• On top of that, states receive federal incentive payments based on performance metrics — specifically, paternity establishment rates, order establishment rates, and current support collected. These metrics reward enforcement activity volume, not enforcement accuracy or outcomes.

• States also retain a portion of collections from TANF cases to reimburse themselves and the federal government for assistance paid to the family.

The behavioral consequence is straightforward and well-documented by litigants who’ve lived it: state child support enforcement agencies are economically incentivized to open IV-D enforcement cases, even when no enforcement is needed. Cases get converted from private agreements to state-enforced orders without the obligor missing a payment. Orders get established in default when the obligor wasn’t properly served. Modifications get denied or delayed because reduced obligations would reduce the state’s collection metric.

The numbers tell the story: in FY2024, the program paid $25.8 billion to families against $6.6 billion in administrative costs. On its face, $4.24 collected per dollar spent looks like a functional program. Strip out the cases that would have paid voluntarily without state involvement — which the program does not separately measure — and the marginal economics of state enforcement collapse. The program counts every dollar collected, whether the state’s involvement was the cause of the collection or not.

The Worker Disincentive Problem, Inverted

In the IT version, the line workers protected by union contracts and pensions had every incentive to obscure the actual workflow because streamlining would eliminate their headcount.

In family court, the equivalent dynamic is more direct: the entire professional class downstream of the bench is paid by the conflict. A streamlined family court — one with mandatory early mediation, presumptive equal parenting, evidence-based outcome tracking, and meaningful immunity reform — would dramatically reduce the work available to GALs, custody evaluators, parenting coordinators, family law litigators, and the judges who manage their dockets. None of these professionals individually want bad outcomes for children. All of them collectively benefit, professionally and economically, from a system that produces protracted conflict and treats the resulting fee generation as evidence of thorough process rather than systemic failure.

This isn’t a conspiracy. Perverse incentive structures, sustained over decades, produce coordinated-looking outcomes without coordination. Each individual actor is doing what their position rationally rewards. The aggregate is a system that fails its stated purpose almost every day, and is structurally incapable of acknowledging it.

Why AI Changes the Calculus

Profile the pattern — not the person.

This is where the family court argument becomes more potent than the IT argument, not less.

The IT version of the AI thesis is: AI can ingest process logs and produce honest maps of where the work actually flows, exposing waste. The family court version is: AI can ingest court records, GAL reports, custody evaluations, and IV-D enforcement actions and produce, for the first time, an honest empirical picture of who decides what, on what basis, with what consistency, and to what effect.

A modern model, applied to even partially-released family court data, can:

• Profile individual judges by ruling pattern: custody-to-mother rate, custody-to-father rate, joint vs. sole custody distribution, modifications granted, sanctions issued, IV-D enforcement orders entered, ratio of custodial parent gender against the population baseline.

• Profile individual GALs and evaluators by recommendation pattern, alignment with judicial outcome, reversal rate on appeal where appellate review exists, and — crucially — by outcomes for the children they were appointed to protect where any longitudinal data is available.

• Identify “Forced IV-D” patterns — cases where state enforcement was opened in the absence of any missed payment, private agreements converted to state enforcement without the obligor’s consent, per-judge IV-D conversion rates, county-level patterns. The Family Court Record prototype at familycourtrecord.org is doing exactly this kind of analysis with synthetic data, demonstrating what real data would expose.

• Generate ruling-pattern comparisons across jurisdictions, surfacing places where similarly-situated cases produce systematically different outcomes — which is the empirical signature of bias, by any honest definition.

• Read briefs, evaluations, and orders side by side and surface the gaps between cited evidence and stated conclusion that human appellate review almost never catches, because no human appellate panel has the time.

None of this is technically difficult. The capability exists today on commodity hardware. The reason it has not been done at scale is the same reason the IT version hasn’t: the people who would be exposed are the people who control the data.

The Resistance Will Look Different Here

In federal IT, the resistance to AI-driven accountability comes wrapped in security reviews, FedRAMP processes, and procurement preferences for incumbent vendors. It will be slow, technical, and bureaucratic.

In family court, the resistance will be constitutional and ethical in framing, and territorial in substance. Expect:

• “Privacy of minors.” Real concern, real cover. Aggregated, anonymized analysis does not require disclosure of any minor’s identity, and the bar knows this. The privacy framing will be deployed to block any analysis that would expose ruling patterns.

• “Judicial independence.” Translated: judges should not be subject to empirical review of their decisions because that would chill their discretion. This is the same argument every protected class has ever made for being exempt from measurement, dressed in robes.

• “AI cannot understand the nuance of family situations.” True, and irrelevant. AI does not need to make the custody decision. It needs to surface the patterns in who is making which decisions, with what consistency, and to what effect.

• “This will lead to harassment of judicial officers.” The data published by the Family Court Record prototype is the same kind of data published about any other public official’s decision-making. The bar’s discomfort with judges being publicly evaluated is not a privacy concern — it is a status concern.

Each of these objections contains a kernel of legitimate concern, and each is being weaponized to preserve a status quo whose failure is measurable, and whose human costs are catastrophic, and which has been left unmeasured for a century because the people inside the system have had every reason to keep it that way.

What Has to Happen

The reform agenda for family court is not mysterious. The reform community has been articulating it for thirty years. The political will and the diagnostic capability to act on it have been the missing pieces. AI removes the second one. The first is downstream of the second:

Mandatory outcome tracking, published quarterly, by judge, GAL, and evaluator. Custody outcomes, modification outcomes, IV-D enforcement actions, longitudinal child outcomes where available. The bar will fight this harder than anything else, which is how you know it matters most.

Reform of GAL and custody evaluator immunity. Move from absolute immunity to qualified immunity, with a meaningful negligence standard. The IT equivalent — moving federal contracting officers from total to partial protection — would be considered moderate reform. In family court it would be revolutionary.

Independent evidence review in contested custody. The custody evaluator should not be the same person whose recommendation the judge defers to without challenge. Build in an adversarial review layer.

Title IV-D reform: separate enforcement metrics from collection metrics. Pay states for accuracy of orders established and successful resolution of arrearages, not for raw volume of enforcement actions opened. Separately track and publish the rate of “Forced IV-D” — cases where state enforcement was opened without a missed payment.

Open the data. Publish anonymized custody outcome records the way every other state administrative dataset is published. Family court’s exemption from the open-data norms that govern the rest of state government is anomalous and indefensible.

Default to mediation and presumptive equal parenting. The empirical research on outcomes for children of separated parents is broadly consistent: where two fit parents are available, equal involvement produces better outcomes than sole custody arrangements. The legal system has been slow to adopt this finding because the legal system economically benefits from contested custody. AI will make that misalignment between research and practice impossible to hide.

The Bottom Line

The family court system has been allowed to operate for a century without the kind of empirical scrutiny that every other consequential institution in American life is subjected to. It has produced — predictably — a regime in which the actors with the most power face the least accountability, the actors with the most expertise face no obligation to demonstrate it, and the actors most affected by the decisions have no meaningful voice in their review.

This is not the result of bad people. As with the IT problem, it is the result of incentive structures sustained for so long that the people inside them no longer notice that the building is on fire. The clearest sign is how the bar reacts when an outsider points at the smoke: not with curiosity, but with concern about who’s holding the camera.

AI did not create this problem. AI will, however, make it impossible to keep hiding. The same way it will in the IT version. The same way it will, eventually, everywhere institutional power has organized itself around the absence of measurement.

The fight that’s coming is over whether honest measurement gets to happen at all. The people whose careers depend on the answer being “no” will fight it the same way they have fought every other accountability mechanism: with procedure, with delay, with framing, and with the moral authority that the existing system grants them by virtue of their proximity to children.

They will lose. The math has changed. The cost of an honest look at the data has fallen to nearly zero, and the public’s patience with a system that produces these outcomes has fallen with it.

The only question is how many more children pay the bill before the lights come on.

Companion piece: The Open Secret of Government IT: A Decade of Failure, Sold as Service. The structural argument is the same; only the domain changes.

Sources for the figures cited above: Congressional Research Service report RS22380 (Jan 2026); Office of Child Support Services FY2024 program statistics; Cooney v. Rossiter (7th Cir. 2009) and progeny on GAL/evaluator immunity; Stump v. Sparkman, 435 U.S. 349 (1978) on judicial immunity; Utah Code § 78A-2-228 as representative state immunity statute; Family Court Record prototype project at family-court-watch-dogs.netlify.app for synthetic-data accountability dashboard methodology.

← Back to all articles

© 2026 AI Chat Insights. Exploring conversational AI.

https://achatwithai.com/blog/family-court-failure-accountability