Expanded Deference, Diminished Review, Relief Restricted

Here is a question worth sitting with: what happens to the right of appeal when the thing being appealed is a judgment call, and the reviewing court is told, in effect, not to second-guess judgment calls?

That is the practical center of Urias-Orellana v. Bondi, decided by a unanimous Supreme Court on March 4, 2026. The case is about asylum law specifically. But the doctrine it settles reaches far past immigration court, into every corner of American governance where a human or a system makes a call, and the question is who, if anyone, gets to check it. That question is about to matter more, not less, as agencies fold automated tools into the kind of fact-finding this case shields from close review — the immigration-law mechanics below are worth walking through first, because they show exactly how a deference doctrine locks in before anyone notices it's happened.

What the Court Actually Decided

The facts are stark and, unfortunately, not unusual. Urias-Orellana and his family fled El Salvador after death threats, a physical assault, and ongoing surveillance connected to a hitman. The immigration judge who heard his case found him credible. Believed him. And denied asylum anyway, ruling that what happened to him did not rise to the legal threshold of "persecution" under the Immigration and Nationality Act. The Board of Immigration Appeals affirmed. The U.S. Court of Appeals for the First Circuit affirmed too, applying the deferential substantial-evidence standard and concluding the record did not compel a different outcome.

The question the Supreme Court took up was narrower than it sounds, and that is exactly why it matters so much: when an immigration judge decides that a set of facts does or doesn't amount to persecution, is that a factual finding, entitled to heavy deference on appeal, or a legal conclusion, open to closer, de novo review? Courts of appeals had split on this for years. Some treated it as a "mixed question," law applied to fact, warranting real scrutiny of the legal reasoning even while deferring to the facts themselves. Others folded the whole thing into deference.

Justice Ketanji Brown Jackson, writing for a unanimous Court, resolved the split against the asylum seeker. The opinion holds that the substantial-evidence standard governs the entire persecution determination, not just the underlying facts but the application of the legal standard to those facts. Congress, the Court reasoned, meant for the whole mixed determination to receive deference under 8 U.S.C. § 1252(b)(4)(B), the provision that says administrative findings of fact are "conclusive unless any reasonable adjudicator would be compelled to conclude the contrary." That is an old standard, drawn from INS v. Elias-Zacarias in 1992, and it was already a hard bar to clear. Urias-Orellana extends it to a category of decisions that used to get a second, more searching look.

The practical effect is simple to state and hard to sit with: an immigration judge can believe you, can find your testimony truthful, and can still rule against you on a characterization question, an appellate court will now uphold that ruling unless the record leaves no other reasonable conclusion available. That is a much higher bar than "the judge got the legal question wrong."

The Doctrine Splitting Hairs: Fact, Law, and the Space Between

Administrative law spends a lot of energy on a distinction that sounds academic until you are the person on the losing end of it: the line between a fact and a legal conclusion drawn from that fact. If an immigration judge finds "the petitioner was beaten and threatened," that is a fact. If the judge then concludes "being beaten and threatened once does not constitute persecution," that is where things get interesting, because the word "persecution" is doing legal work. It has a threshold, a meaning shaped by decades of case law, and applying it to a specific set of events is an act of legal judgment, not just observation.

Urias-Orellana treats that entire act, fact-finding and legal characterization together, as a single deferential package. The table below lays out what changed.

Notice what that last row does. Resolving a circuit split usually reads as a win for predictability, and in one sense it is. But predictability in the direction of "no circuit will look closely" is a different outcome than predictability in the direction of "every circuit will look closely." The uniformity here is real. It is uniformity toward diminished review.

Why "Refugee Roulette" Doesn't Go Away, It Gets Insulated

The wide variation in asylum outcomes depending on which immigration judge hears the case has long been described as "refugee roulette." Grant rates for similarly situated applicants can differ dramatically from one judge to the next: the 2007 Stanford study found that Colombian asylum seekers had a 5 percent chance of winning before one Miami immigration judge and an 88 percent chance of winning before another judge in the very same courthouse. The underlying claim wasn't that immigration judges are unusually biased people. It was that a system built on individual, largely unreviewable judgment calls will produce wildly inconsistent outcomes almost by construction, because two reasonable people can look at the same facts and land in different places on a fuzzy legal threshold like "persecution."

Sarah Paoletti's essay in The Regulatory Review that prompted this piece makes the point directly: expanded deference doesn't fix refugee roulette, it protects it. If de novo review of the legal characterization was ever going to smooth out inconsistency between judges, one venue, and one applicant's outcome, at a time, that avenue just narrowed. The disparity documented in 2007 doesn't disappear under Urias-Orellana. It becomes harder for a court to correct once it shows up in an individual case.

I find it worth pausing on the timing here, because it cuts against the direction administrative law has been moving on the other side of the ledger. In 2024, in Loper Bright Enterprises v. Raimondo, the Supreme Court ended Chevron deference, the decades-old doctrine under which courts deferred to an agency's reasonable interpretation of an ambiguous statute. That decision was framed, correctly, as courts reclaiming their role as the final word on what the law means. Two years later, in a different corner of administrative law, the same Court is expanding deference to agency fact-finders on a different kind of mixed question. Less deference to agencies interpreting statutes. More deference to agencies applying statutes to facts. Read together, the two decisions don't describe a simple story about courts getting more or less deferential. They describe courts drawing a new line about which kinds of institutional judgment get checked and which get trusted, and the line isn't obviously about which choice serves the person the decision is actually made about.

The Pattern Behind the Pattern

Here is where I think this case earns a place on a site about AI and institutions rather than staying inside immigration law journals, and it's not a stretch, it's the same structural question wearing a different coat.

Every adjudicative system, whether it decides asylum claims, disability benefits, unemployment appeals, or loan denials, has to answer the same design question: who makes the initial call, and who gets to check it, and under what standard. Urias-Orellana answers that question for one system by shrinking the checking function. What makes this relevant well beyond immigration is that the same fact-finder whose characterization now gets heavy deference is a human being applying human judgment to a human story.

That won't stay true by default. Immigration enforcement already uses algorithmic tools next door to this process — U.S. Immigration and Customs Enforcement's Risk Classification Assessment tool, for instance, scores detention and bond decisions, not the persecution determination itself. But nothing in Urias-Orellana's reasoning depends on that boundary holding, and other adjudicative agencies are already experimenting with automated document review and pattern-matching tools closer to the fact-finding role itself. If a tool like that moves into the fact-finding process for asylum claims, the "finding" an appellate court is deferring to won't be just one judge's read of one witness anymore. It will be a judgment shaped, in part, by a tool whose internal reasoning the judge may not fully understand and the reviewing court will almost certainly never see.

That is the deeper problem Urias-Orellana previews. Deference doctrines were built for a world where the fact-finder was a person you could cross-examine about their reasoning. Extend the same deferential posture to a decision pipeline that includes an opaque scoring tool, and the practical reviewability doesn't just narrow, it can collapse, because the artifact a court reviews is the judge's stated rationale, not the tool's actual logic. Nobody has to intend that outcome for it to happen. It falls out of applying an old, reasonable-sounding standard, substantial evidence, to a new kind of decision process that the standard was never designed to test.

I've come to think this is the pattern worth watching across every domain where AI is entering adjudication: institutions get very good at explaining why a human decision-maker deserves deference, and that explanation quietly extends to cover whatever the human decision-maker relied on to reach the call, tool included. Nobody votes on that extension. It just happens, one substantial-evidence ruling at a time, in cases that look narrow and technical until you notice how many other decision systems share the same architecture.

What This Means for Anyone Tracking Institutional Oversight

If you follow AI governance, algorithmic accountability, or administrative law as a discipline rather than a specialty, Urias-Orellana is worth filing under a broader question: as institutions build more decision pipelines that blend human judgment with automated tools, what does meaningful review of those pipelines actually require?

A few things follow from this case that apply outside immigration court entirely.

• Old standards meet new pipelines. Deference standards written for a pre-automation world do not automatically produce good outcomes when the fact-finding process changes underneath them. The statute at issue here, 8 U.S.C. § 1252(b)(4)(B), was written in 1996. Nobody drafting it was contemplating a decision pipeline that includes a scoring algorithm.
• Uniform review isn't the same as sound review. Resolving a circuit split toward uniformity is not the same thing as resolving it well. Predictability is valuable, but predictable diminished review is still diminished review, just with better branding.
• Explained reasoning isn't testable reasoning. This is the one I keep returning to: the gap between "the decision-maker explained their reasoning" and "a reviewer can actually test that reasoning" is where oversight quietly fails, and it fails in ways that look, from the outside, like the system working exactly as designed.

None of this means deference is inherently wrong. Courts genuinely can't relitigate every immigration case from scratch, and there are good reasons to trust the judge who actually watched the witness testify over a panel of appellate judges reading a transcript. But deference is a tool with a cost, and the cost falls on the person the decision is about. When the fact-finding process itself becomes more automated, more opaque, or more layered, the honest response is to ask whether the old deference calculus still holds, not to assume it does because the label on the standard hasn't changed. That's a live question in AI governance right now, and it's worth reading the effective oversight literature on it, including the case for proof drills over paper compliance as one way institutions can actually test whether their review function still does anything. It's also worth watching for the pattern Paoletti names directly in that same essay, where the entities being reviewed have real influence over how the review standard gets written in the first place, a dynamic covered in more depth in the piece on regulatory capture in AI governance.

The blunt version: a system that tells you it takes oversight seriously, and a system that has actually preserved a path for a mistaken decision to get corrected, are not the same system. Urias-Orellana is a reminder that you have to check which one you're looking at, case by case, statute by statute, because the label rarely tells you.

Frequently Asked Questions

What did the Supreme Court decide in Urias-Orellana v. Bondi? On March 4, 2026, the Court ruled unanimously, in an opinion by Justice Ketanji Brown Jackson, that the substantial-evidence standard governs an immigration judge's entire persecution determination in an asylum case, including the legal conclusion of whether facts amount to "persecution," not just the underlying factual findings.

What is the "substantial evidence" standard, and where does it come from? It is codified at 8 U.S.C. § 1252(b)(4)(B), which states that administrative findings of fact are conclusive "unless any reasonable adjudicator would be compelled to conclude the contrary." The Supreme Court applied this "compelled" standard to legal characterization questions as well in INS v. Elias-Zacarias (1992) and reaffirmed and extended its reach in Urias-Orellana.

How does this relate to the Supreme Court ending Chevron deference in 2024? Loper Bright Enterprises v. Raimondo (2024) ended judicial deference to agency interpretations of ambiguous statutes. Urias-Orellana moves in a different direction on a different question, expanding judicial deference to agency fact-finders applying settled legal standards to specific facts. The two decisions together show courts narrowing deference in one area of administrative law while expanding it in another.

What is "refugee roulette," and why does this ruling matter for it? The term comes from a 2007 Stanford Law Review study by Jaya Ramji-Nogales, Andrew Schoenholtz, and Philip Schrag, documenting large disparities in asylum grant rates for similar cases depending on the assigned judge. Urias-Orellana doesn't cause that inconsistency, but by narrowing appellate review of persecution determinations, it removes one of the checks that could otherwise correct it in an individual case.

Does this decision only apply to immigration law? Directly, yes. But the underlying question, how much independent review a fact-finder's judgment call deserves, applies to any adjudicative system, including ones that increasingly rely on automated tools as part of the fact-finding process. The doctrine here didn't anticipate that shift, and it's worth asking whether it should still apply the same way once it does.

Last updated: 2026-08-22