Records over deference: after Loper Bright, an agency’s best defense sat in its files

The week

For decades, a simple idea shaped how Washington regulated. When a law was ambiguous, judges generally deferred to the agency’s reasonable reading of it. In the space of five days in the summer of 2024, the Supreme Court took that idea away and reshaped several other rules of the game at the same time.

On June 28, 2024, the Supreme Court decided Loper Bright Enterprises v. Raimondo and overruled the Chevron doctrine. Chevron, named after a 1984 Supreme Court case, told courts to accept an agency’s reasonable interpretation of an unclear statute. That held even if the judges might have read the law differently. Under Loper Bright, courts must now use their own judgment on what a statute means, though they may give respect to an agency’s reasoned views, and earlier cases decided under Chevron keep their effect.

The facts behind the case were small. The National Marine Fisheries Service had adopted a rule under the Magnuson-Stevens Act requiring Atlantic herring fishing companies to pay for the federal observers placed on their boats. The companies argued the statute did not authorize the agency to make them bear that cost. Lower courts, applying Chevron, had sided with the agency. A companion case, Relentless v. Department of Commerce, raised the same question about a different fishing fleet.

Chief Justice Roberts wrote for a six-justice majority. The opinion rested on the Administrative Procedure Act itself, which tells reviewing courts to decide all relevant questions of law. Chevron, the Court said, could not be squared with that instruction. Justice Kagan wrote the principal dissent. Justice Jackson did not take part in Loper Bright, which she had heard as an appeals judge, and joined the dissent in Relentless.

The majority left two doors open. Courts may still give weight to an agency’s views under Skidmore v. Swift, a 1944 decision that ties the weight to how thorough, well reasoned and consistent the agency’s position is. And where a statute expressly delegates discretion, for example by letting an agency set “appropriate” limits, the court confirms that the agency stayed within that grant and reasoned its way to a decision.

On July 1, in Corner Post v. Board of Governors of the Federal Reserve System, the Court held that the six-year window to challenge an agency action under the Administrative Procedure Act opens when the plaintiff is injured. The Administrative Procedure Act, or APA, is the law that sets out how federal agencies make rules and how courts review them. The practical effect of Corner Post is that the clock does not start when a rule is issued. A newly formed business can therefore challenge an old rule.

The facts made the point vivid. Corner Post is a truck stop and convenience store in North Dakota that opened in 2018. It joined a 2021 lawsuit by retail trade groups against Regulation II, the Federal Reserve’s 2011 rule under the Durbin Amendment to the Dodd-Frank Act. That rule caps the interchange fees large banks may charge on debit card transactions. The lower courts dismissed the claim as too late, because the rule was more than six years old.

The Supreme Court reversed, six to three. Justice Barrett wrote for the majority, reading the general federal statute of limitations to run from the date each plaintiff is injured. Justice Jackson dissented, warning that the ruling would let new entrants reopen settled rules indefinitely. Justice Kavanaugh wrote separately to argue that the APA lets courts vacate unlawful rules outright, a question that matters as much as timing to anyone who relies on a rule.

Two other decisions landed in the same week. On June 27, SEC v. Jarkesy held that the SEC must bring civil penalty claims for securities fraud before a jury. Until then, the agency could pursue many such cases in its own in-house proceedings, heard by an administrative law judge employed within the agency system rather than by a federal court.

The case began with an SEC in-house proceeding against George Jarkesy, a hedge fund adviser, and his firm, for misleading investors. Dodd-Frank had widened the SEC’s power in 2010 to seek civil penalties in its own forum against anyone, including people outside registered firms. The Fifth Circuit ruled for Jarkesy in 2022 on three separate constitutional grounds. The Supreme Court affirmed, six to three, on the Seventh Amendment right to a jury alone, and did not decide the other two.

The key move in the opinion was to limit the “public rights” exception, the doctrine that lets Congress assign some matters to agencies without a jury. Because securities fraud claims closely resemble common-law fraud, the Court held the exception did not apply. Any agency that pursues penalties for conduct resembling a traditional legal claim had to read the decision against its own enforcement program.

The same day, in Ohio v. EPA, the Court paused an EPA air rule, finding the agency had likely failed to explain its approach in light of concerns raised in comments. When agencies propose rules, the public can submit comments, and the agency is expected to consider and respond to the significant ones. Here, the Court found the response likely fell short.

The rule in question was EPA’s “good neighbor” plan under the Clean Air Act, which set ozone controls for upwind states whose emissions drift into downwind states. EPA had designed the controls on the assumption that a large group of states would be covered. Commenters had asked what would happen if some states fell out of the plan. After courts paused EPA’s action for a number of states, that question became real. The Court, five to four in an opinion by Justice Gorsuch, found EPA had likely not given a reasoned answer. Justice Barrett wrote the dissent.

The ruling came on an emergency application, before the D.C. Circuit had decided the merits, which made its focus on the agency’s explanation more striking. The week was crowded. The presidential immunity decision on July 1 drew most of the public attention, and the administrative law cases were read mainly by lawyers in the days that followed.

Taken one at a time, each case was a legal story for lawyers. Taken together, they changed where agencies would win and lose. Less would turn on how an agency read the law, and more on what it could prove from its own files.

After West Virginia v. EPA in 2022, a rule’s survival odds already belonged in the compliance plan, and agencies had reason to review which programs rested on broad readings of old statutes. Loper Bright went further on legal interpretation.

It also left one ground largely intact. Courts still review an agency’s factual findings, technical judgments and policy choices under the APA’s reasoned-decision standard. Where a statute delegates discretion, the court checks that the agency stayed within it and explained itself. In practice, that shifts an agency’s defense from its reading of the statute to the quality of its record: what it found, what it considered, and how it answered objections.

The four decisions together test the record in three different settings. A rule’s record is tested when it is adopted. It is tested again across its whole life, because Corner Post keeps it open to challenge. And enforcement records are tested in a courtroom, because Jarkesy moves penalty cases to juries. Each setting puts a different demand on how agencies keep information. That makes a legal shift, surprisingly, into an information management problem, and one that technology can help with or hurt.

Who feels this first depends on the kind of organization. Federal agencies, both executive departments and independent commissions, carry the direct burden: their rules and enforcement cases are the ones under review. State and local agencies are affected less directly. Loper Bright governs how federal courts treat federal agencies, and state courts follow their own doctrines. Several states, among them Wisconsin, Arizona, Florida and Mississippi, had already moved away from deference to state agencies in 2018, by court decision, statute or constitutional amendment.

Regulated organizations sit on the other side of the same files. A bank, a broker-dealer, a health system or a federal contractor relies on rules that may now be challenged, and faces enforcement cases that may now be tried differently. Their technology and compliance leaders need to understand how agency records are judged, because those records decide which rules survive. Their own records are judged the same way when a dispute reaches court.

What “the record” actually is

For a non-lawyer, the phrase “administrative record” can sound abstract. It is, quite literally, the file. When an agency adopts a rule, the record is everything it relied on. That includes the proposal, the public comments, the studies and data, the cost-benefit analysis, internal technical memos that were considered, and the agency’s final explanation. When a court reviews the rule, it generally looks at that file, and judges the agency on what it shows.

Most of that file lives in a docket, the official collection of public documents for a rulemaking. The agency’s final explanation is usually in the preamble, the long section published before the rule’s text that explains what the agency decided and why, and answers the main comments.

The rules that make the file decisive

Three long-standing principles explain why the file carries so much weight. First, courts generally review an agency’s action on the record that was before the agency when it decided, a principle the Supreme Court set out in Citizens to Preserve Overton Park v. Volpe in 1971. New evidence produced in litigation usually cannot rescue a decision. Second, under SEC v. Chenery, decided in the 1940s, a court can uphold an agency only on the reasons the agency actually gave at the time. Lawyers cannot supply better reasons afterwards.

Third, the Court reinforced the ban on after-the-fact explanations in Department of Homeland Security v. Regents of the University of California in 2020. In Department of Commerce v. New York in 2019, it also showed that it would look behind a stated rationale it found to be contrived. Taken together, these rules mean an agency fights with the file it built at the time. If the reasoning is not written down when the decision is made, it effectively does not exist.

What survives of deference, and why it depends on records

Loper Bright did not tell courts to ignore agencies. It pointed them to Skidmore, which gives an agency’s interpretation weight according to how thoroughly it was considered, how valid its reasoning is, and how consistent it has been over time. Each of those factors is a property of a record. Thoroughness shows in the analysis. Validity shows in the explanation. Consistency shows only if the agency can produce what it said before, in guidance, interpretive letters, FAQs and earlier preambles.

Many agencies cannot do that easily. Interpretive letters may sit in a legacy correspondence system, and old FAQs may have been overwritten on a website without an archive. Staff may have answered questions by email in ways that contradict official guidance. An agency that wants Skidmore weight needs a searchable, dated catalog of its own interpretations, with superseded versions retained. Without one, a challenger can find an inconsistency the agency did not know it had.

The court’s review now resembles a teacher marking a student’s working on a math exam. Under Chevron, a court reviewing a disputed reading of the law would often accept the agency’s answer if it was reasonable. After Loper Bright, the court works out the legal answer itself. But on facts, technical judgments and policy choices, the court still asks to see the working, and the working is the record. An agency with excellent working but a contestable reading of the statute may still lose. An agency with a clear legal mandate but sloppy working can lose too.

Suppose a federal safety agency issued a technical standard fifteen years ago. The analyst who built the supporting model has retired. The model ran on software the agency no longer licenses. The comments were summarized in a spreadsheet that sits on a shared drive nobody can find. The rule itself may be perfectly sound. The agency’s ability to prove it is not.

The same exposure applies to regulated organizations in a different form. A bank that adopted a control because of a regulator’s interpretive letter needs to keep that letter, the date it relied on it, and the analysis behind its approach. If the interpretation is later disowned or struck down, the bank’s file shows it acted in good faith on the guidance available at the time.

How a rule gets made, step by step

Each step in the life of a typical federal rule leaves something in the file.

1. Analysis. Agency staff study a problem, gather data and often build a model estimating costs and benefits. That work becomes the factual foundation.
2. Proposal. The agency publishes a proposed rule with an explanation of what it intends and why, and invites public comment.
3. Comment. Companies, trade groups, advocates, states and individuals respond. For a contested rule, that can mean a very large volume of submissions, some short, some running to hundreds of pages with their own data.
4. Response. Staff sort the comments, decide which raise significant issues, and change the rule or explain why not.
5. Final rule. The agency publishes the final text and preamble.
6. Life of the rule. The rule takes effect, is enforced, and, ideally, is monitored to see whether it works.

Several of those steps involve parties outside the agency, and each leaves its own trail. For a significant rule, the Office of Information and Regulatory Affairs at OMB reviews drafts under Executive Order 12866, and other agencies weigh in during that review. Small business analyses under the Regulatory Flexibility Act add their own documents. Many agencies collect comments through regulations.gov, the shared federal docket site. Others, such as the FCC with its Electronic Comment Filing System, run their own.

Comment volume has grown harder to manage. Advocacy campaigns can generate large numbers of near-identical comments, and some comments are submitted under false names or produced by software. The Administrative Conference of the United States, the federal body that studies agency procedure, issued recommendations in 2021 on handling mass, computer-generated and falsely attributed comments. The volume problem makes the risk of missing the one comment that matters greater, not smaller.

Every one of those steps is, at heart, an information task: collecting, organizing, analyzing, retaining and retrieving. Each of the four decisions puts pressure on a different step. Ohio v. EPA pressed on step four. Corner Post stretched step six across decades. Jarkesy moved enforcement into a forum with much higher evidence demands. Loper Bright made the factual and analytical work of steps one through four more important, because it is now where agencies can most clearly earn a court’s respect.

The record also has owners scattered across the agency. Program staff hold the analysis. The docket office holds the comments. The general counsel’s office holds the drafts of the explanation. The records officer sets the retention rules, and the CIO runs the systems all of it lives in. Nobody owns the file as a whole, and that gap explains many of the failures described below.

The record when a rule is adopted

Ohio v. EPA was decided the day before Loper Bright and drew less attention. Its holding was narrow and procedural: the agency had not explained why its plan still made sense if some states dropped out, a concern raised in comments. That is a records failure as much as a legal one.

Comment handling in many agencies runs on spreadsheets and email. Thousands of comments are grouped by staff, summarized and answered in a preamble. When a significant comment is lost in that process, the gap can decide the case. It is the same risk any large organization runs when it handles high volumes of customer feedback by hand: the important item that falls between two people’s inboxes.

The fix is a system. A comment management tool that tags each comment by issue, links it to a response and a responsible analyst, and tracks closure lets counsel see unaddressed issues before publication. No comment can be closed until someone has written an answer.

How the comment workflow should run

A defensible workflow starts at intake. Each submission is logged with its date, source and any attachments, and duplicates are grouped without being discarded. Staff then code each comment against an issue list drawn from the proposal, adding new issues as they appear. Every issue gets an owner, a draft response and a pointer to where the preamble answers it. Before publication, counsel reviews the open items and confirms that each significant issue has a written answer.

Software can speed this up. Deduplication and clustering tools can group thousands of form letters in hours, and language models can suggest which issues a long comment raises. The trade-off is accountability. A court will ask whether the agency considered the significant comments, and an agency should be able to show that people reviewed the substantive ones. Tool outputs, the settings used and the human decisions that followed should all be logged.

Simple metrics help counsel and program heads see the state of the file. Useful ones include the number of significant issues still without a response, the age of the oldest open item, and the share of long technical comments that a subject expert has read in full. A comment that attaches its own data deserves particular care. If it contradicts the agency’s analysis, the response needs to engage with the data as well as the argument.

Technical analyses need the same discipline. A cost-benefit model, a security baseline or a data quality study should be stored with its inputs, code, version and assumptions, so the agency can show exactly what supported the rule. “Reproducible” is the key word: someone else, years later, should be able to rerun the analysis and get the same answer. Since courts now read the statute fresh, the agency’s expert findings are where it can still earn respect. They need to be findable and reproducible.

Reproducibility has concrete parts. The model’s code belongs in version control, with the version used for the final rule tagged. Input datasets should be archived with checksums, a digital fingerprint that proves the file has not changed. The computing environment, including software versions, should be captured in a form someone can rebuild, such as a container image. A short methods note should explain the assumptions in language a generalist judge can follow.

Assumptions are where challengers attack. A cost-benefit model may assume a compliance cost per firm, a rate of adoption or a discount rate. If those choices are documented with their sources and the alternatives considered, the agency can show reasoned judgment. If they appear only as numbers in a spreadsheet cell, a challenger can argue they were arbitrary, and the agency will have trouble proving otherwise.

There is a trade-off. A thorough record slows rulemaking, and staff time spent on comment tracking and model documentation is time not spent on new rules. For agencies with thin policy staff, the practical answer is tiering: full discipline for economically significant or contested rules, lighter handling for routine technical amendments.

A familiar objection deserves an answer. Scholars have warned for decades that rising procedural demands “ossify” rulemaking, making agencies slower and more cautious, sometimes to the point of avoiding rules altogether. Adding records discipline could make that worse. The objection is fair, and the response is proportion. Most of the effort described here is work agencies already do badly by hand. Doing it in a system costs less per rule once established, and the alternative, losing a major rule on a gap in the file, is the most expensive outcome of all.

State agencies face a similar choice under their own procedure acts. Many state statutes require agencies to issue a concise statement of reasons or a response to comments when they finalize a rule. As state courts move away from deference, those statements will be read more closely, and the same tiered discipline applies.

The record across a rule’s life

Corner Post removed the comfort of age. A rule adopted in 2005 can be challenged in 2030 by a company formed in 2028. The agency will then need the 2005 docket, the analyses behind it, and evidence of how the rule has worked since.

That collides with how government records are usually kept. Records schedules, the official rules that set how long each type of document must be kept before it can be destroyed, often treat rulemaking support files as temporary after a set period. Legacy dockets may sit in retired systems or paper archives. Models may run only on software that no longer exists. The record may technically exist and still be practically unusable.

Why the timing collided with records policy

The decisions arrived at an awkward moment for federal records programs. OMB and the National Archives had set June 30, 2024 as the point by which agencies were expected to manage permanent records electronically. After it, the Archives would generally stop accepting new analog transfers. The deadline had already slipped once. It fell in the same week as Corner Post, and it left many agencies still converting paper and legacy files while the legal value of old dockets was rising.

Email is a particular weak point. Many agencies manage email under the Archives’ Capstone approach, which keeps the accounts of senior officials permanently and applies shorter retention to most other staff. That approach works for history. It may not keep the technical exchanges between working-level analysts that explain how a rule was built. Agencies that expect to defend a rule for decades should check whether those exchanges are kept somewhere other than a mailbox scheduled for deletion.

The case that started it

Corner Post itself shows the stakes. The Federal Reserve was asked to defend choices made in 2011 about which bank costs could count toward the debit fee cap. The people who made them, the data they used and the systems that held both had all changed in the intervening years. The Fed had proposed in October 2023 to lower the cap using newer data. That proposal shows a second route to a defensible rule: refreshing the record through a new rulemaking instead of relying on the original file.

For banks and card networks, the case had a practical edge. A rule that shapes their revenue was reopened by a merchant that did not exist when it was adopted. Their planning had to treat the cap as a live legal question, not a settled feature of the market. Other long-standing rules that move money between industries, such as fee caps and pricing rules, carry the same exposure.

Agency records officers and general counsel should agree which rules are high value and exposed. For those, they should extend retention, migrate dockets into current systems, and keep models reproducible. Migration here means moving old files into systems people use today, in formats they can open.

Existing review requirements give this work a home. Section 610 of the Regulatory Flexibility Act already requires agencies to review, within ten years, rules that have a significant economic impact on a substantial number of small entities. Executive Order 13563, issued in 2011, asked agencies to plan retrospective reviews of existing rules. Those reviews have often been thin. After Corner Post, they become a way to identify exposed rules and to refresh their records before a challenger arrives.

Choosing which rules count as high value is a judgment, and a simple screen helps. Counsel can rank rules on four factors. They are economic significance, how often the rule is litigated or criticized, how far it depends on a contestable reading of the statute, and the age of its supporting analysis. A rule that scores high on all four deserves a docket migration and a refreshed analysis first. A routine rule with a clear statutory basis can wait.

A short program performance file for each major rule, updated periodically with outcomes data, gives the agency evidence that the rule works as intended. That evidence also helps if Congress is asked to clarify authority. It costs far less to maintain steadily than to reconstruct under a litigation deadline.

For a technology rule, that file might hold adoption rates, incident or complaint trends, and the agency’s own assessment of whether the requirement still fits current systems. Where the answer is no, the file also tells the agency to amend the rule before a challenger forces the question. The file then serves as a management tool as well as a defense: an early warning that a rule has aged.

The costs and owners need to be explicit. Records officers rarely control the budget for migrating legacy dockets, and CIOs rarely see old rulemaking files as a priority against current systems. The general counsel has the clearest view of legal exposure but no systems budget. A short memorandum agreed among the three, naming the high-value rules and assigning each task, prevents the work from falling between offices.

The record in enforcement

Jarkesy concerned the SEC, but its reasoning reaches other agencies that seek civil penalties in in-house proceedings for claims resembling common-law actions. Those are the kinds of claims, such as fraud, that historically went before juries. Cases moved to federal court face the Federal Rules of Evidence, full discovery and a lay jury. Discovery is the process in which each side can demand documents and testimony from the other. Agencies used to building cases for an administrative law judge must now build them for twelve citizens and opposing counsel with subpoena power.

Which agencies feel it, and how much

The reach of Jarkesy varies. The SEC had already been sending most contested fraud cases to federal court for several years, after a run of constitutional challenges to its in-house forum, including Lucia v. SEC in 2018. For it, the decision mostly closed an option it had stopped relying on. Other agencies use in-house proceedings far more. The banking agencies seek civil money penalties administratively, the FCC issues forfeitures, and labor and safety agencies run their own adjudication systems.

Each of those programs now needs a legal assessment of which claims resemble common-law actions and which fall within the public rights exception. The answer will differ by statute and by claim, and agencies should expect it to be litigated rather than settled by a memo. Their technology planning should assume that some share of cases will move to court, and budget for discovery capacity on that basis.

That changes evidence handling:

• Chain of custody. Examination workpapers, digital evidence and investigative files need documented chain of custody and preservation, a record showing who handled each item and that it was not altered.
• Litigation holds. A litigation hold is an instruction to preserve everything relevant to a case. Holds must reach cloud mailboxes, collaboration tools and messaging apps where staff actually discuss cases, as well as formal case files.
• E-discovery. Electronic discovery, the collection and review of digital records for litigation, is often contracted, and needs to scale to produce agency records on court timelines.
• Explainable experts. Expert analyses must be explainable to non-specialists sitting in a jury box.

Each item has a technical owner and a failure mode. Chain of custody depends on evidence systems that log every access and transfer, which many examination teams lack because their workpapers sit in general document stores. Litigation holds depend on the collaboration platform supporting preservation for chat and channel messages, and on someone applying the hold to the right accounts quickly. E-discovery depends on contracts that can scale on short notice and on data that can be exported in standard formats with its metadata intact.

Expert evidence faces its own test. The Federal Rules of Evidence were amended in December 2023 to make clear that the party offering an expert must show the testimony meets the reliability standard. An agency expert whose model cannot be reproduced, or whose assumptions are undocumented, is open to exclusion before a jury ever hears from them. The reproducibility discipline described for rulemaking applies here too.

Regulated firms will recognize the problem from the other side. Beginning in December 2021, the SEC brought a series of cases against broker-dealers and investment advisers for failing to keep employees’ business messages on personal devices and apps. Those cases made off-channel communication a recordkeeping failure with large penalties. Agencies preparing for jury trials face a version of the same question about their own staff’s messages.

Enforcement leaders should expect to bring fewer contested penalty cases and to rely more on supervision, settlements and corrective action plans. Those also depend on a clean record, since a respondent who believes the agency cannot prove its case at trial has less reason to settle.

Take the same safety agency in an enforcement case. Its examiners find what they believe is a serious violation and propose a civil penalty. Before Jarkesy, the case might have gone to an in-house hearing, where the judge was familiar with the subject and the procedures were the agency’s own. Now, in a case of this kind, it heads to federal court. Defense counsel requests every email, chat message and draft memo about the examination. Some of the key discussion happened in a messaging app the hold notice did not cover. The expert’s analysis is sound but written for specialists. The agency may still be right. Whether it can prove it in front of a jury is a different question, and the answer depends on records decisions made long before the case began.

Regulated organizations face the mirror image. Their own records of compliance decisions, control testing and correspondence with regulators will matter more when disputes reach court. Their lawyers should be tracking which of the rules that affect them are being challenged.

The mirror differs by sector. A broker-dealer already keeps business communications under SEC Rule 17a-4 and FINRA rules, so its challenge is retrieval and context more than retention. A credit union or community bank answers to the NCUA, FDIC or OCC, and its exam correspondence and remediation evidence tend to live in shared drives and email. A health system’s compliance files under HIPAA and Medicare rules are often split across departments. A federal contractor’s compliance records may be needed in False Claims Act litigation. In each case, the question is whether a decision made years ago can be shown, with its reasons, on demand.

From back-office chore to legal strategy

For an agency head, the shift reframes records management from a back-office chore into part of the agency’s legal strategy. Budgets for comment systems, model documentation and e-discovery are no longer administrative overhead. They are what allows a rule to survive and an enforcement case to be won.

For a company general counsel or chief compliance officer, the same decisions cut two ways. Rules the company relies on, for example a standard that levels the playing field with less careful competitors, may be more vulnerable than before. Rules the company dislikes may be more open to challenge, including by newly formed businesses. Either way, the company’s exposure depends partly on the quality of an agency’s files, which it cannot see, and partly on the quality of its own, which it can control.

Companies can also take part in the record directly. A well-supported comment, with data, a clear statement of the issue and a proposed alternative, becomes part of the docket. If the agency fails to answer it, that failure becomes a ground of challenge. If the agency answers it well, the rule becomes stronger. Either way, a company that comments seriously shapes the record that a court will later read, and should keep its own copy of what it submitted and why.

Boards should assume that regulatory certainty now has a shorter half-life. A compliance plan built on the assumption that an old rule is settled law may need a contingency. And when disputes with regulators reach a courtroom, the organization with the clearer, better-preserved record will usually have the stronger hand.

The strongest objection to this argument

A sceptic would say records are a sideshow. On this view, the decisions of 2024 moved power from agencies to judges, and outcomes will turn on judges’ readings of statutes and their views of regulation, whatever the files contain. A well-documented rule on a contested statutory reading will still fall. A thin record on a clear mandate may still survive.

Part of that is right. Where the case turns on the meaning of the statute, the record helps less, and Loper Bright made that question harder for agencies to win. But many challenges turn on more than one ground, and the record decides the others. Ohio v. EPA was about explanation, not statutory meaning. Corner Post put old records in front of courts. Jarkesy changed how evidence is tested, not what the law means. The share of disputes that a good file can win is large, and it is the share an agency controls.

Technology sits underneath all of it. Email retention, messaging policies, document management, version control for analytical models: these are ordinary IT decisions with very large legal consequences. They are usually made by people far from the general counsel’s office, on grounds of cost and convenience. After this week in 2024, they deserve a legal voice in the room, and a clear owner who understands that the file is now the defense.

Ownership works best when it is named. Some agencies give the role to a rulemaking records lead in the general counsel’s office, with authority to set standards for dockets, models and holds. Others give it to the chief data officer, who already owns data quality and catalogs. The choice matters less than having one person accountable for the file across its life, from proposal to the last lawsuit.

An agency that can show its working

Now suppose the same agency has spent a year on the problem. Its comment management system tags every comment to an issue and an analyst, and counsel reviews a list of unresolved issues before any rule is published. Major models live in a managed repository with their inputs, code, version and assumptions, and a different analyst reruns each one before publication. Records officers and counsel have agreed a list of high-value rules, extended their retention and moved their dockets into current systems. Each major rule has a short performance file updated with outcomes data. The enforcement division’s litigation holds reach chat and collaboration tools, and its e-discovery contract can scale to court deadlines.

The sequence matters. Agencies that made progress tended to start with the next major rule, building the comment system and model repository on live work, where staff could see the benefit. They then turned to the screen of high-value existing rules and their dockets. Enforcement holds and discovery capacity came alongside, because an incoming court case would test them without notice.

An inspector general or the Government Accountability Office reviewing that work would ask for evidence. Typical requests would include the comment tracking records for recent rules and the list of rules designated high value. Reviewers would also want the retention changes made for them, the dates of model reproductions and who performed them, and litigation hold notices with proof of delivery to messaging accounts. An agency that can hand those over quickly is also an agency that can defend itself in court.

The failures that undo that work are familiar:

• Treating comments as a mailbag. Grouping and summarizing by hand, so a significant objection goes unanswered.
• Keeping models on one person’s laptop. When that person leaves, the analysis behind the rule leaves too.
• Letting records schedules run on autopilot. Support files for a rule that can still be challenged are destroyed as routine.
• Assuming old rules are safe. After Corner Post, age is no protection.
• Holding only the formal files. Litigation holds that skip messaging apps miss where real discussion happens.
• Writing for specialists only. Expert analysis a jury cannot follow is analysis a jury may not credit.

Testing the file before a court does

• For our most important rules, or the rules our business depends on most, where is the record, and could we produce it within weeks?
• Could someone rerun the key analyses behind those rules today and get the same answer? Who would do it?
• How do we track public comments, and how would we know if a significant one went unanswered?
• Do our litigation holds reach the chat, collaboration and messaging tools staff actually use?
• Which rules we rely on, or object to, are now open to challenge, and what is our plan if one falls?

Asked plainly and early, these questions turn a set of Supreme Court decisions into a manageable work plan. Left until a lawsuit arrives, they become a scramble against a court deadline.

With hindsight

The decisions were applied quickly. In January 2025 the Sixth Circuit struck down the FCC’s net neutrality rules, relying on Loper Bright to read the statute itself. Net neutrality rules require internet providers to treat traffic evenly. The court did not defer to the agency’s view of its own authority, and decided the question of statutory meaning independently, exactly as Loper Bright directed.

Other early rulings followed the same path. In August 2024 a federal court in Texas set aside the Federal Trade Commission’s rule banning most noncompete agreements, finding the agency lacked authority to issue it. In November 2024 another Texas federal court vacated the Department of Labor’s 2024 rule raising the salary threshold for overtime exemptions, reading the statute without deference. Neither case turned on records alone, but both showed courts willing to read old statutes fresh against new rules.

In June 2025, in McLaughlin Chiropractic Associates v. McKesson, the Supreme Court held that district courts in enforcement cases are not bound by an agency’s interpretation of the statute and must decide its meaning themselves, giving the agency appropriate respect. That extended the logic of Loper Bright beyond challenges to rules and into cases where a rule is being enforced.

The Court also confirmed the other half of the argument. In Seven County Infrastructure Coalition v. Eagle County, decided in May 2025, it told lower courts to give substantial deference to an agency’s factual and technical judgments in environmental reviews under the National Environmental Policy Act. Deference on the meaning of statutes had gone. Deference to a reasoned, well-documented factual record had not, and in that setting the Court strengthened it.

Jarkesy’s reach was tested at the FCC. In April 2025 the Fifth Circuit relied on it to set aside an FCC forfeiture against AT&T over the sharing of customer location data. It held that the carrier was entitled to a jury. Other courts of appeals upheld similar forfeitures against other carriers later that year, leaving the scope of the decision for agencies outside the SEC unsettled.

The executive branch also used the decisions. An April 2025 presidential memorandum directed agencies to identify and repeal regulations it considered unlawful under a list of Supreme Court cases led by Loper Bright, including Jarkesy and Ohio v. EPA. The same rulings that constrain agencies in court became a basis for agencies to review their own rule books.

That memorandum built on Executive Order 14219, issued in February 2025, which had told agencies to review their regulations within 60 days for ones that might exceed their authority. The April memorandum told agencies to repeal rules that clearly exceeded their authority, giving priority to conflicts with the listed decisions. It encouraged them to rely on the APA’s good cause exception to skip notice and comment, with a brief statement explaining why it applied. It also required a one-page explanation for each flagged rule an agency chose to keep.

Many rules continued in force, and courts upheld agencies whose authority was clear and whose reasoning was well documented. That is the most practical point. The decisions did not sweep away regulation wholesale. They raised the premium on clear authority and good records.

Rescission itself needs a reasoned record under the APA. An agency cannot simply delete a rule; it has to explain why, consider relevant objections and show its working, just as it did when adopting it. The record has therefore mattered in both directions: to defend rules, and to change them.

Repeals made without notice and comment carry their own risk. The good cause exception has historically been read narrowly by courts, and a repeal that skips comment can be challenged on that ground alone. An agency that relies on it needs a written explanation strong enough to survive review, which is one more demand on the file.

For agencies, that means records discipline pays off whichever way policy moves. For regulated organizations, it means watching how well each side can prove its case, as well as which rules are challenged. With less deference, the quality of each side’s file carries more weight.