When the Agency Reviews Itself: How Internal Appeals Actually Work

Listen Now

Spotify | Apple Podcasts | Listen on our site

What We Cover

• Why the Appeals Council's "denied" letter functions as a case-ending event for unrepresented claimants — and why the same letter is a federal-court trigger for represented ones
• The structural asymmetry of credibility review: ALJs see the witness, reviewing bodies see only the transcript, and the deference runs the wrong way
• How APA § 557(b) gives agency review bodies "all the powers" of the original adjudicator — including the power to redo credibility findings on a cold record
• Why Social Security, immigration, and enforcement agencies have each written their own regulations pulling back from the APA default to different degrees
• How the four-part new-evidence test plus a good-cause requirement makes the Appeals Council a damage-limitation step rather than a fresh opportunity to fix a record
• The Social Security model: hundreds of thousands of appeals, fifteen to twenty percent reviewed, summary denials that double as federal-court triggers most unrepresented claimants miss
• The immigration model: BIA review constrained on facts by a clear-error standard, paired with the Attorney General's power to certify any case and announce binding doctrine through it
• How asylum eligibility for survivors of domestic violence has whipsawed across three administrations — Matter of A-R-C-G-, Matter of A-B-, Matter of S-S-F-M- — through a single certification mechanism, with no statute changing
• The enforcement model: automatic review by the same commissioners who oversee the enforcement division that brought the case, with reversal patterns that consistently favor enforcement
• The selection power that distinguishes agency review from court review — picking which cases to review and which to designate as precedent is itself a form of doctrinal control

Full Rough Transcript

Intro

Gwen: Hello and welcome to Administrative Remedies, because you can't fix what you don't understand. Brought to you in part by the University of Tulsa College of Law. I'm Gwendolyn Savitz, an associate professor here at TU and the associate dean of research and intellectual life.

Marc: And I'm Marc Roark, the dean of the law school.

Gwen: We'll be breaking down complex doctrines with real life analogies and examples, to demystify the world of administrative law for everyone trying to understand how government actually works.

Marc: Agencies are the main way that the federal government gets things done. It's not through Congress — for reasons we'll be addressing over the course of the series.

Opening hook

Gwen: Marc, last episode we spent thirty minutes on the hearing-level picture. Judge-level variation up to forty-six points on the same claim. Another ten or fifteen points of variation between hearing offices. The representation effect — roughly doubled approval odds for claimants with experienced counsel. And a research environment where sophisticated lawyers can navigate all of that because they've built up private knowledge of specific judges in specific offices, while unrepresented claimants are essentially operating blind.

Marc: And the point was that none of that is controlled by any top-down mechanism at the ALJ level. The agency has some binding interpretive rulings, but they cover a narrow band of questions. Everything else is private — ALJ judgment on one side, experienced counsel's ambient knowledge on the other.

Gwen: Right. And the natural move, once you've seen that picture, is to say — fine, maybe the hearing level doesn't produce consistency on its own. But there's a review layer above it. Every agency has one. The Appeals Council at Social Security, the Board of Immigration Appeals, the Commission at the SEC, the Board at the NLRB. The theory is that even when ALJs produce inconsistent decisions, the bodies above them correct errors, smooth out conflicts, and pull outcomes toward consistency. That's the mechanism that's supposed to fix what the hearing level doesn't.

Marc: And today's question is whether it does.

Gwen: Today's question is what the review layer actually does. Because the answer turns out to be — not one thing. It's three different things at three different kinds of agency. And none of the three is what the textbook description suggests.

Marc: So we're not going to land on 'internal review doesn't work.'

Gwen: We're going to land on something more interesting. Internal review does different work at different agencies, and the kind of work it does at each one tells us something about what it actually means for an agency to review itself.

Segment 1 — What the reviewing body gets, and what it can do

Marc: Let's start with what they're looking at. When the Appeals Council or the BIA or the SEC Commission opens a file, what's actually in front of them?

Gwen: A transcript. The written exhibits. Whatever briefs got filed on appeal. That's essentially it.

Marc: Not the hearing itself.

Gwen: Not the hearing. The APA makes the hearing record the exclusive basis for decision. Read 5 U.S.C. § 556(e).

Marc [reading]: 5 U.S.C. § 556(e): The transcript of testimony and exhibits, together with all papers and requests filed in the proceeding, constitutes the exclusive record for decision.

Gwen: Exclusive record. Which means the reviewing body is working from paper. They didn't see the witness testify. They didn't watch the claimant shift in her chair. They didn't see her wince when she stood up. Everything the ALJ observed in the room that informed the credibility finding — none of that is on the transcript.

Marc: And last episode we talked about how credibility findings are essentially unreviewable on appeal because you weren't in the room.

Gwen: That's the principle federal courts use. The agency review bodies operate under a different rule. Read 5 U.S.C. § 557(b) — the one that governs them.

Marc [reading]: 5 U.S.C. § 557(b): On appeal from or review of the initial decision, the agency has all the powers which it would have in making the initial decision.

Marc: All the powers. Meaning they can redo the case from scratch, including the credibility findings?

Gwen: Including the credibility findings. On paper. From a cold record. The ALJ who sat three feet from the witness and watched her testify for forty-five minutes makes a credibility finding, and a Commission or a Board can reverse it based on reading a transcript. That's not an abuse. That's what the APA says they can do.

Marc: That seems kind of messed up. Federal courts reviewing a trial court's credibility finding — they defer. They can't redo it from a cold record. The trial judge saw the witness.

Gwen: Right.

Marc: But the agency review body can redo the ALJ's credibility finding. And then the federal court reviewing the agency defers — again, because the agency is the finder of fact.

Gwen: So the one party who actually saw the witness — the ALJ — is the one whose credibility finding gets replaced. You can't just swap in a finding that she did make eye contact when the ALJ who was there said she didn't. But you can change the outcome that flows from that finding. The asymmetry runs in the wrong direction.

Marc: Give me a concrete version.

Gwen: Take an SEC insider-trading case. The respondent goes through an evidentiary hearing in front of an SEC ALJ. The ALJ watches the respondent testify across multiple days, watches the government's cooperator witness testify, makes a series of credibility findings — the respondent's explanations of the trades are plausible, the cooperator is shading testimony to fit his cooperation agreement. The ALJ finds insufficient evidence of scienter and dismisses the proceeding. Under SEC rules, that initial decision doesn't become final on its own. The Enforcement Division petitions the Commission for review. The Commission — five politically appointed commissioners who oversee the Enforcement Division — takes the case. They read the transcript. They don't see the respondent testify. They don't see the cooperator. What they see is a paper record where the trades look suspicious and the respondent's explanations look constructed after the fact. They reweigh credibility on the cold record and reverse. Violation found. Sanctions imposed.

Marc: And the federal court reviewing the Commission —

Gwen: Applies substantial evidence review to the Commission's determination, not the ALJ's. The original credibility finding is gone. And substantial evidence is a deferential standard — the court asks whether a reasonable person, looking at the record, could reach the conclusion the agency reached. Not whether the agency reached the right conclusion. Just whether a reasonable person could have.

Marc: Which means if the record contains evidence pointing in either direction, the agency wins. Because it's usually possible to reach either conclusion from a record that has evidence on both sides.

Gwen: Usually possible. So the practical effect is: the ALJ who saw the witness gets overturned; the Commission or Board that didn't see the witness gets deferred to; and the claimant or respondent who lost in the room can't recover. The asymmetry survives all the way through federal court.

Marc: I want to flag something before we move on. You picked the SEC for that example. Is that because every agency works the same way, or because the SEC is the place where the 'all the powers' language really bites?

Gwen: Good question, and the answer matters for the rest of the episode. The APA default — the 'all the powers' language — is the statutory baseline. But individual agencies have written regulations that pull back from that baseline. Social Security has one. Immigration has one. And they pull back in different ways and to different degrees.

Marc: Give me the SSA regulation first since that's the one that affects the most people.

Gwen: Read 20 C.F.R. § 404.970(a).

Marc [reading]: 20 C.F.R. § 404.970(a): The Appeals Council will review a case at a party’s request or on its own motion if — there appears to be an abuse of discretion by the administrative law judge; there is an error of law; the action, findings, or conclusions in the hearing decision are not supported by substantial evidence; or there is a broad policy or procedural issue.

Marc: “Not supported by substantial evidence.” That's the same deferential standard the federal court applies later.

Gwen: Same standard, walked back to the review stage. SSA wrote its own regulation that constrains the Appeals Council below the APA default. The Council can't just decide it would have weighed the credibility differently. It has to find that the ALJ's conclusion isn't supported by substantial evidence in the record — meaning a reasonable person couldn't have reached that conclusion. That's a much higher bar than “we read the transcript differently.”

Marc: So the scenario where the Appeals Council just substitutes its own credibility call from a cold transcript — that's not really what's happening at SSA.

Gwen: Not on the front end of review. The constraint is real. The SSA problem is different, and we'll get to it — it's mostly about review not happening at all, and about what the agency does with new evidence. Immigration walks back the APA default even further, through a different regulation that we'll read when we get to model two. So the picture is a spectrum. APA default at one end — agency heads can redo everything. SSA in the middle — substantial-evidence trigger for review. Immigration at the other end — clear-error review of facts, the strictest constraint. The enforcement agencies operate closest to the default. They're the place where “all the powers” isn't a theoretical baseline. It's the operating standard.

Marc: Each agency has effectively decided how much of the APA default it actually wants to use.

Gwen: And those choices weren't made by Congress. They were made by the agencies themselves, through their own rulemaking. That's worth holding onto for the rest of the episode.

Segment 2 — The narrowing record

Marc: And the record the reviewing body sees — can parties add to it? If new evidence comes in after the hearing, can it come in at review?

Gwen: Sometimes. But the window is narrow and it narrows further at every stage. The Social Security rule is representative.

Marc: That was the rule that required the evidence to be new, material, tied to the period before the hearing, and likely to actually change the outcome.

Gwen: Four requirements stacked on top of each other. And on top of that, good cause for why the evidence wasn't submitted earlier.

Marc: What qualifies as good cause?

Gwen: The regulation lists a few things. The agency misled you. You had a limitation that prevented you from submitting it. Some unusual circumstance — serious illness, destroyed records, evidence you actively sought but didn't receive in time.

Marc: Walk me through what this looks like when it goes wrong for someone.

Gwen: Here's the common version. A claimant's treating physician wrote an opinion a year before the hearing that would have directly addressed what the ALJ ended up hanging the denial on. Nobody submitted it — maybe the representative didn't know it existed, maybe the claimant didn't mention it, maybe it got misfiled. The hearing happens without it. The ALJ denies.

Marc: And the representative finds it after the hearing.

Gwen: Submits it to the Appeals Council. Too late. No good cause for why it wasn't in the hearing record. Appeals Council declines to consider it. The case goes to federal court.

Marc: And the federal court reviews the record the agency had.

Gwen: Which doesn't include the opinion. The court doesn't ask whether the opinion was persuasive. It asks whether it's in the record. It isn't. She loses — not on the merits of the opinion, but on the procedural failure to get it there.

Marc: Could she refile?

Gwen: She could file a new application. But the onset date resets. If her original application was 2020 and she refiles in 2026, she loses years of back benefits. Potentially tens of thousands of dollars.

Marc: So every stage after the hearing is damage limitation, not a fresh opportunity.

Gwen: That's the right way to put it. And the implication for review is important. If the record at the hearing is what it is — if the evidence that would have changed the outcome wasn't in the file — then even a well-intentioned review body can't fix it. They're looking at the wrong record. The hearing is the only real opportunity to shape what every subsequent reviewer will see.

Segment 3 — Three models of review

Marc: Alright, let's get to the question we opened with. What does the review layer actually do? Because I'd guess — from the volume issues alone — the different agencies handle this very differently.

Gwen: Dramatically differently. Let me give you three models, because each one represents a genuinely different answer to what 'review' means at an agency.

Gwen: Model one — Social Security. The Appeals Council receives hundreds of thousands of requests for review every year. It grants review in roughly fifteen to twenty percent of cases. The rest get a one-paragraph denial that says the Appeals Council has denied your request for review. No analysis, no engagement with the arguments, no explanation of what they looked at.

Marc: So most Social Security claimants who appeal to the Appeals Council don't actually get reviewed.

Gwen: They don't. And it's not just that — the design of the denial letter itself is one of the cleanest examples of procedural-design consequences in the whole adjudicative system. Unrepresented claimants misread it, and when they misread it they lose their case. We talked about this a little a couple episodes ago, but there's more to say. This isn't something anyone designed to produce this effect. It's a procedural artifact. But the effect is real.

Marc: Walk me through this.

Gwen: The Appeals Council's denial of review is the procedural event that triggers the claimant's right to file in federal district court. The denial letter is the gate being opened. The claimant has sixty days from the date of the letter to file a civil action in federal court. Miss that window, the right is gone.

Marc: And the letter itself —

Gwen: Is a paragraph. It tells the claimant the Appeals Council has denied their request for review. It doesn't analyze their case. It doesn't engage with their arguments. It does not explain in plain terms that they have the right to sue the Commissioner of Social Security in federal district court within sixty days. That information is somewhere in the boilerplate, but the letter reads, to a lay reader, like a final rejection.

Marc: Because it is a final rejection at the agency level.

Gwen: Which is exactly why the wording creates a consequential mismatch. 'Denied' in ordinary usage means 'you lost.' 'Denied' in this administrative-law usage means 'you've now exhausted administrative remedies, the sixty-day clock to federal court is running.' Those are practically opposites, and the letter doesn't make it clear it's the second one.

Marc: So unrepresented claimants see the letter and apply the ordinary-language meaning.

Gwen: Yes — they think it's over, don't file, run out the clock. And by the time they figure out they had a right they didn't exercise, the clock is gone and so is the case. Represented claimants, meanwhile, know the letter is the checkpoint before federal court. They've been expecting it. They file the appeal in federal court, and the case proceeds.

Marc: Which connects back to everything we said in the last episode about privatized knowledge.

Gwen: Right. The meaning of the denial letter — which is the most consequential piece of paper the claimant will receive in this process — is information available to experienced counsel and not meaningfully available to the unrepresented. The letter doesn't explain itself, at least not in a way most people it's impacting can understand. Again — nobody designed the letter to mislead. It's the artifact of administrative-law vocabulary meeting ordinary-language expectation. The effect on unrepresented claimants who take the letter at face value is still that they lose cases they didn't have to lose.

Marc: So in this model, review mostly doesn't happen, and the review denial itself produces outcomes through a mechanism unrelated to the merits.

Gwen: Right. The summary denial functions as a case-ending event for a significant percentage of unrepresented claimants. The ALJ's decision stands. Whatever inconsistency existed across the thousands of ALJs is preserved, not corrected. Model one is an exhaustion step, not necessarily a correction mechanism — and the exhaustion step itself is where a meaningful percentage of unrepresented claimants lose by default.

Marc: Model two?

Gwen: Immigration. Immigration judge decisions get forwarded to the BIA if either party appeals. The BIA then either conducts full review or issues a summary affirmance — a brief order, sometimes just a sentence, saying 'affirmed for the reasons stated by the immigration judge.'

Marc: But when the BIA does do full review —

Gwen: That's where it gets different from Social Security. In model one, the cases the Appeals Council doesn't review just disappear — the ALJ's decision stands, and nothing about it travels anywhere. The BIA can designate decisions as precedential, and when it does, those decisions bind every immigration judge in the country. So the same review body that issues one-sentence affirmances in most cases is also, selectively, making law that governs every case that comes after it. The stakes of which cases get full review are much higher than they look.

Marc: Before we get to which cases get reviewed and what becomes precedent — back to the spectrum we set up earlier. You said immigration walks back the APA default the furthest. Read the regulation.

Marc [reading]: 8 C.F.R. § 1003.1(d)(3)(i): The Board will not engage in de novo review of findings of fact determined by an immigration judge. Facts determined by the immigration judge, including findings as to the credibility of testimony, shall be reviewed only to determine whether the findings of the immigration judge are clearly erroneous.

Marc: Clear error. That's the standard a federal appeals court uses on a district judge's findings.

Gwen: Same standard. So the BIA can't do what the SEC Commission can do. It can't sit down with a transcript, decide the asylum applicant's flat affect reads as evasive rather than traumatized, and substitute its own credibility call. If the IJ found the testimony credible, the Board is stuck with that unless it's left with a definite and firm conviction the IJ got it wrong. Immigration's constraint on fact-finding is the tightest of the three.

Marc: Which makes immigration look like the agency that has its house in order. The fact-finder is protected.

Gwen: On facts. But the review problem at immigration doesn't live at the fact level. The BIA reviews application of law to fact de novo — whether a credible applicant's experience adds up to “persecution,” whether a group counts as a “particular social group,” whether there's “nexus” to a protected ground. Those are where outcomes live. And the bigger lever sits above the Board entirely.

Marc: Which sounds more like what a listener might hope internal review looks like.

Gwen: It's closer. But here's the piece that makes immigration different from the model you might be imagining. The Attorney General can reach into any case, at any stage, and decide it himself.

Marc: Just to be clear — the Attorney General personally?

Marc [reading]: 8 C.F.R. § 1003.1(g)(2): Selected decisions designated by the Board, decisions of the Attorney General, and decisions of the Secretary of Homeland Security as provided in paragraph (h)(2)(i) of this section will be published and serve as precedents in all proceedings involving the same issue or issues.

Gwen: So when the BIA does decide to write, it can bind every immigration judge in the country. Except themselves — they're still free to change their mind. Here.

Marc [reading]: 8 C.F.R. § 1003.1(g)(3): Except as Board decisions may be modified or overruled by the Board or the Attorney General, decisions of the Board and decisions of the Attorney General are binding on all officers and employees of DHS or immigration judges in the administration of the immigration laws of the United States.

Marc: And the regulation that lets the AG pull a case in the first place —

Gwen: 8 C.F.R. § 1003.1(h)(1).

Marc [reading]: 8 C.F.R. § 1003.1(h)(1): The Board shall refer to the Attorney General for review of its decision all cases that the Attorney General directs the Board to refer to him.

Marc: So the Attorney General can pick a case — any case, pending or decided — and issue a decision that binds every immigration judge in the country.

Gwen: Any case. And this has been used in ways that dramatically shift doctrine. Take asylum for survivors of domestic violence. In 2014, the BIA published Matter of A-R-C-G- — held that a Guatemalan woman in a relationship she couldn't escape could be a member of a particular social group for asylum purposes. Survivors of domestic violence had a precedential anchor. Then in 2018, AG Sessions certifies a case to himself — Matter of A-B- — overrules A-R-C-G-, and writes that claims involving domestic violence by private actors generally won't qualify. One certification. Every immigration judge bound. No statute changed. No notice-and-comment rulemaking. In 2021, AG Garland certifies A-B- to himself, vacates Sessions's decision, reinstates A-R-C-G-. Domestic violence is back as a viable basis. Then in 2025, AG Bondi certifies a different case, Matter of S-S-F-M-, and uses it to overrule Garland and reinstate A-B-. Three swings on the same doctrine in seven years, all through the same mechanism. Same story for family-based asylum — Matter of L-E-A- has been certified and re-certified across administrations. These aren't individual case decisions. They're substantive reinterpretations of who qualifies for protection, imposed through a single referral.

Marc: And because it's binding precedent, it flows down to every immigration judge going forward.

Gwen: To every immigration judge going forward. This is why review in immigration is so different. Review isn't just error-correction, and it isn't just case-by-case consistency-building. At immigration it's also an active policy instrument wielded by a politically appointed head of department who can reach into pending adjudications and rewrite the rules. The BIA exists formally as the appellate body. The AG sits above it with the power to pull any case and change the doctrine. That's not error correction. That's policy control executed through the review mechanism.

Marc: And the reason to emphasize that structurally is that other agencies have versions of the same arrangement.

Gwen: The commissions at enforcement agencies have less sweeping versions — their review authority is real but it doesn't carry the AG certification's sheer doctrinal reach. The Appeals Council at Social Security has weaker versions — it reviews little, doesn't designate precedent, doesn't make doctrine. What the immigration example shows is what the review mechanism can look like when the agency head has both the legal authority and the political will to use it.

Marc: That's a different kind of thing from what the Appeals Council does.

Gwen: Completely different in degree, same kind of thing in kind. The Appeals Council at Social Security mostly doesn't review. The BIA plus AG certification reviews a lot, and the review includes reshaping doctrine in real time through the cases the AG selects. The variable across agencies isn't whether the review mechanism exists. It's how aggressively the mechanism gets used.

Marc: Model three?

Gwen: Enforcement agencies. SEC, FTC, NLRB. At these agencies, review isn't discretionary — it's automatic. When an ALJ issues an initial decision, it goes to the Commission or Board by default. The ALJ's decision is essentially a recommendation. It doesn't become final until the agency heads either affirm it, modify it, reverse it, or let the review window expire.

Marc: So at the SEC, every case gets full review by the commissioners.

Gwen: Every enforcement case. The Commission is expected to issue its own final decision. The ALJ hearing is always preliminary. The Commission has the last word.

Marc: Why the difference? Why does Social Security barely review while the SEC reviews everything?

Gwen: Volume and design. Social Security processes millions of disability claims and files hundreds of thousands of appeals a year. It's mechanically impossible for the Appeals Council to meaningfully review every case. The SEC brings maybe a couple hundred enforcement actions a year. Five commissioners can genuinely look at a couple hundred cases.

Marc: So at one end you have benefits claimants who mostly get no meaningful review. At the other end you have enforcement respondents who get guaranteed review — by the same institution that brought the case against them.

Gwen: And both of those sentences are doing work in that trade-off. We'll come back to the second one in a minute. But first I want to stay with the three models and say something that ties them together.

Segment 4 — The selection power

Marc: Go ahead.

Gwen: In each of the three models, the review body exercises selection power over its own docket. This is a big deal — it's a structural feature of agency review that courts don't have.

Marc: Because courts don't get to do that. They take what comes to them.

Gwen: Exactly. Think about how a circuit court gets cases. Parties appeal. The court gets whatever the parties bring. It can deny cert at the Supreme Court level, it can issue summary affirmances, but historically it can't reach down into a trial court's docket and pull up a case it wants to decide. The court is working from the pool of cases other people brought.

Marc: And agencies.

Gwen: Agencies can reach. The Appeals Council can review a Social Security case on its own motion, even if nobody appealed. The BIA can do the same. The SEC Commission can initiate review. And the Attorney General — as we just saw — can certify any case, and decide it himself.

Marc: So the reviewing body picks.

Gwen: And that's not a small thing. In the court system, law develops because litigants push questions up. The Fifth Circuit rules on the cases that come to it. Parties can force the court to confront a question by appealing. That gives litigants a meaningful role in how the law develops — they pick which issues get resolved when.

Marc: And in the agency system —

Gwen: Agency leadership picks. Which questions reach the precedent-making level. Which facts get adjudicated at that level. Which doctrine gets announced. The precedent is curated from the top down, not pushed up from the bottom.

Marc: And the selection is its own form of control.

Gwen: A significant form of control. Because if the agency wants the law to develop in a particular direction, it can pick the cases that go that direction. The Attorney General's certification power is the most visible version of this, but every agency's review body exercises some version of it. Deciding what to review is deciding what becomes law.

Marc: Which means the review layer isn't neutral error-correction. It's a mechanism that combines error-correction with agenda-setting.

Gwen: That's the honest description. The textbook version of internal review treats it as quality control — the agency heads sample and correct. The real version is agenda-setting plus quality control plus, at enforcement agencies, a structural conflict we haven't talked about yet.

Segment 5 — The dual-role problem

Marc: Which is what you flagged a minute ago. Back to the enforcement side.

Gwen: Back to the enforcement side. At the SEC, the commissioners oversee the enforcement division. They approve investigation targets. They set enforcement priorities. They allocate resources to the enforcement staff. And then they serve as the appellate tribunal reviewing whether that same staff proved its case.

Marc: The investigator-adjudicator overlap we've been talking about — at the institutional level.

Gwen: In an ordinary appellate system, the reviewing body has no institutional stake in the dispute. The Ninth Circuit doesn't supervise the plaintiff's lawyers. The D.C. Circuit doesn't set policy for the agencies whose rules it reviews. Enforcement-agency review looks completely different.

Marc: How has that pattern shown up in the data?

Gwen: When the SEC Commission reverses ALJ decisions, the large majority of those reversals — around three-quarters in studies that have been done — favor the enforcement division. Either finding violations the ALJ didn't find, or increasing penalties the ALJ imposed.

Marc: So when the Commission steps in, it usually goes against the respondent.

Gwen: Most of the time. And you don't need bad faith to see why. The enforcement division works for the Commission. If the Commission regularly rejects enforcement cases, they're effectively telling their own staff they're getting it wrong. The incentives line up in one direction, and they do so without anyone necessarily acting improperly.

Marc: And they're exercising that while applying the 'all the powers' authority to redo credibility from a transcript.

Gwen: Right. So the combination is: the reviewing body has an institutional preference, and the review includes the authority to overturn credibility findings made by someone who was actually in the room. That's not a neutral error-correction mechanism. That's something else.

Marc: What about benefits agencies? The Appeals Council doesn't have the same prosecutorial stake.

Gwen: It doesn't. Nobody at SSA 'wins' when a claimant loses. But the Appeals Council operates within an agency that has budget constraints, processing targets, and institutional views about how disability standards should be applied. If agency leadership believes ALJs are being too generous, the review function can become a tool for tightening outcomes. The structural pressure is different from enforcement, but it's not absent.

Marc: Is there a counterargument to all of this? Because I can hear someone saying — of course agency heads review their own agency's ALJs. They're the experts. They're accountable. Somebody has to be the final word.

Gwen: Three arguments. Expertise — agency heads know the statute and regulatory landscape better than anyone, and without review, ALJ interpretations would drift in different directions. Accountability — agency heads are politically appointed by the President, more democratically accountable than career ALJs. And uniformity — without central review, different ALJs would interpret the same statute different ways with no mechanism to reconcile them.

Marc: Those are practical arguments. Do they answer the due-process concern?

Gwen: They're arguments for why we'd want some form of central review. They don't answer the question of what to do when that review is institutionally tilted. The system the APA built leaves that question open.

Segment 6 — What the three models have in common

Marc: Let me try to pull the three models together, because I want to make sure I'm seeing what unifies them.

Gwen: Go.

Marc: At Social Security, review mostly doesn't happen. The ALJ's decision stands. Hearing-level variation is preserved, not corrected. At immigration, review does happen, sometimes in depth, and sometimes it produces binding precedent through designation or AG certification — but the selection of what to designate or certify is itself politically directed, and doctrine can shift with administrations. At the SEC, review happens in every case, but with a structural institutional lean toward the enforcement division.

Gwen: That's the picture. And importantly, none of the three models primarily produces consistency. Not by design, not by accident, not as a side effect of doing something else. Each one fails to produce consistency in a different way. SSA fails by not reviewing most cases at all. Immigration fails by reviewing through a mechanism directly controlled by political leadership, whose priorities shift with administrations. Enforcement fails by reviewing through a body that has an institutional stake in the outcome. Different failure modes, same bottom line — internal review is not the mechanism that pulls similar cases toward similar outcomes.

Marc: So review can correct egregious errors. It produces some precedent where precedent gets designated. It's not zero. But as the mechanism that's supposed to fix what the hearing level couldn't — which is what the textbook description of internal review suggests — none of these three models is doing that work.

Gwen: Right. Internal review does correct some errors. It does resolve some conflicts. It produces some good precedent. This isn't a system with zero value. But as a mechanism for ensuring similar cases come out the same way — which is the thing a listener might hope it does after last week's variation episode — it's not that.

Marc: It's not the fix for the hearing-level picture.

Gwen: Right. So we've now covered two layers. ALJs don't produce consistency on their own — that was last week. The review layer doesn't produce it either — that's this week. Neither of the mechanisms the formal adjudicative system was built around is producing the thing we'd most want from it. And yet agency preferences clearly translate into outcomes. Aggregate approval rates shift across administrations. Enforcement priorities become enforcement outcomes. Doctrinal positions announced by agency leadership become practice.

Marc: Which means whatever is actually producing those patterns has to be operating somewhere other than where we've been looking.

Gwen: Right. Agencies don't have to wait for appeals to shape outcomes. They shape behavior before the ALJ even decides the case. Through metrics, through performance review, through guidance that functions like law even when technically it isn't, through the quiet institutional pressure that comes with working inside a hierarchical organization where your bosses have views. That's the mechanism we haven't looked at yet. That's next episode. How agencies actually control outcomes — not by correcting decisions, but by shaping the conditions under which decisions are made.

Closing

Marc: That's it for this episode of Administrative Remedies. If you have questions or topics you'd like us to cover, reach out to us at the University of Tulsa College of Law. And remember — you can't fix what you don't understand. I'm Marc Roark. See you next time.

Related Guides

• Why the Same Disability Claim Can Win or Lose Depending on Which Judge You Draw — The first episode in this arc. That one mapped the variation problem at the hearing level; this one explains why the review layer doesn't fix it.
• Why Two Identical Disability Cases Can Have Opposite Outcomes — The credibility findings that reviewing bodies can overturn from a cold transcript are the same ones discussed here as nearly unreviewable in the hearing room. The contrast matters.
• Who Decides: ALJs vs. Non-ALJs — Internal review is a relationship between ALJs and the agency heads above them; this guide explains who those parties are and why their authority is structured the way it is.
• Trump v. Slaughter Explained: The Case That Ended Independent Agencies — The 2026 decision that reshapes the accountability structure around this episode's SEC and FTC discussion. The reviewing commissioners who reverse ALJ decisions are now removable at will by the president, sharpening the institutional-lean problem to the point where the framing of the earlier episode's "structural institutional lean" becomes doctrinally supported rather than just observationally accurate.