Before Graves, Ho, and Duncan, Circuit Judges. Stuart Kyle Duncan, Circuit Judge: The PredictIt Market is an online marketplace that lets people trade on the predicted outcomes of political events. Essentially, it is a futures market for politics. In 2014, a division within the Commodity Futures Trading Commission (“CFTC”) issued PredictIt a “no-action letter,” effectively allowing it to operate without registering under federal law. But, in 2022, the division rescinded the no-action letter, accusing PredictIt of violating the letter’s terms but without explaining how. It also ordered all outstanding PredictIt contracts to be closed in fewer than six months. Various parties who participate in PredictIt (collectively, “Appellants”) challenged the no-action letter’s rescission in federal district court and moved for a preliminary injunction. The district court has not ruled on that motion, though, despite PredictIt’s looming shutdown. Appellants now seek our review, treating the district court’s inaction as effectively denying a preliminary injunction. We granted Appellants an injunction pending our consideration of their appeal. The CFTC has since raised a host of objections to our even hearing the appeal, arguing that it is moot, that there has been no final agency action, that revoking the no-action letter was within the agency’s discretion, and that Appellants lack standing. These threshold objections are all meritless. We now conclude that a preliminary injunction was warranted because the CFTC’s rescission of the no-action letter was likely arbitrary and capricious. So, we remand for the district court to enter a preliminary injunction while it considers Appellants’ challenge to the CFTC’s actions. I. Background Launched in 2014 by the Victoria University of Wellington in New Zealand, PredictIt was conceived as a data-gathering tool for academic researchers. It allows people to make small investments based on predicting political events, like future elections or the passage of federal legislation. For instance, in recent markets predicting the 2024 presidential nominees, Donald Trump “shares” were trading at $0.56, while Ron DeSantis “shares” were trading at $0.22 (on 47.5 million shares traded). Joe Biden was outpacing Gavin Newsom by $0.66 to $0.21 (16.4 million shares). And in trading on whether Alexandria Ocasio-Cortez would run for president in 2024, “No” was beating “Yes” $0.97 to $0.03 (361,000 shares). If a trader accurately predicts an event’s outcome, each of his shares will cash out at $1.00.[1] Offering these sorts of “event contracts” typically requires registering as “a designated contract market or swap execution facility” under the Commodity Exchange Act (“CEA”) and CFTC regulations. See 7 U.S.C. § 7a-2(c)(5)(C)(i); 17 C.F.R. § 40.11.[2] But the CFTC can exempt certain transactions from the CEA. See 7 U.S.C. § 6(c)(1)–(2). And a division within the agency, the Division of Market Oversight (“DMO”), can issue various “letters” concerning the CEA. See 17 C.F.R. § 140.99 (setting out DMO authority to issue “exemptive, no-action, and interpretative letters”). Relevant here, a “no-action letter” provides that, as to a proposed transaction or activity, the DMO “will not recommend enforcement action to the [CFTC] for failure to comply with a specific provision of the Act or of a Commission rule, regulation or order.” See id. § 140.99(a)(2). Only the division that issued the no-action letter is bound by it and “[o]nly the Beneficiary may rely upon the no-action letter.” Ibid. In 2014, seeking to operate PredictIt without registering under the CEA, Victoria University sought a no-action letter. The university proposed a small-scale, not-for-profit market that would serve as a valuable academic tool for researchers. This market, the university explained, would abide by certain limits, such as capping trader investment at $850 and restricting each event contract to 5,000 total traders. In October 2014, DMO issued Victoria University’s requested no-action letter. The letter stated that “based upon [Victoria University's] representations” to abide by certain terms—such as maintaining nonprofit status and allowing researchers to access generated data—the DMO would “not recommend that the Commission take any enforcement action.” The letter also explained that its position “represent[ed] the views of DMO only, and d[id] not necessarily represent the positions or views of the Commission.” And the DMO purported to “retain[] the authority to condition further, modify, suspend, terminate or otherwise restrict the terms of the no-action relief . . . in its discretion.” Nearly eight years later, in August 2022, the DMO rescinded the no-action letter. The revocation stated that “[t]he University has not operated its market in compliance with the terms of [the no-action letter]” and that, therefore, the no-action letter was “ hereby withdrawn.” The DMO provided no explanation about which terms of the letter had been violated. Instead, the revocation directed that “remaining listed contracts and positions comprising all associated open interest in such market should be closed out and/or liquidated no later than 11:59 p.m. eastern on February 15, 2023.” In September 2022, various parties affiliated with PredictIt (“Appellants”) sued the CFTC in federal court.[3] They claimed the no-action letter’s rescission was arbitrary and capricious because it failed to explain the agency’s decision. See 5 U.S.C. § 706. They also claimed the revocation constituted a withdrawal of a license without the necessary procedural steps. See 5 U.S.C. § 558. Appellants moved for a preliminary injunction. In response, the CFTC moved to dismiss on the grounds that none of Appellants’ claims was justiciable. In December 2022, a magistrate judge recommended the case be transferred to Washington, D.C. During this time, spanning three months, the district court did not rule on the preliminary injunction motion, even after Appellants moved to expedite its consideration in light of the looming deadline for closing PredictIt contracts. Given this inaction, Appellants appealed what they deemed the effective denial of a preliminary injunction. The CFTC moved to dismiss the appeal for lack of jurisdiction. A motions panel of our court denied that motion, citing Carson v. Am. Brands, Inc., 450 U.S. 79 (1981). Under Carson, a court of appeals may review a district court’s order that, while not explicitly denying a preliminary injunction, “nonetheless ha[s] the practical effect of doing so” and might cause irreparable harm absent immediate appeal. Id. at 83; see also, e.g., Thomas ex rel. D.M.T. v. Sch. Bd. of St. Martin Par., 756 F.3d 380, 384 & n.7 (5th Cir. 2014); 28 U.S.C. § 1292(a)(1). The motions panel carried with the case Appellants’ motion for an injunction pending appeal. Our panel granted that injunction on January 26, 2023, and heard argument in February 2023. Less than a month later, in March 2023, the CFTC withdrew its August 2022 rescission of the no-action letter. Notwithstanding the injunction pending appeal, the agency substituted a new letter that “determined as a preliminary matter that [the no-action letter] is void and should be withdrawn.” This new letter gave some explanation for rescinding the no-action letter and gave Victoria University a chance to respond. Given these developments, the CFTC moved to dismiss this appeal as moot. Appellants opposed and cross-moved for sanctions, arguing the CFTC had violated our earlier injunction. On May 1, 2023, we denied both motions. At the same time, we clarified that CFTC “is ENJOINED from closing the PredictIt Market or otherwise prohibiting or deterring the trading of Market contracts until 60 days after a final judgment in this matter.” II. Threshold Issues Before addressing whether a preliminary injunction is warranted, we consider several threshold issues raised by the CFTC. Those are: (1) whether the appeal is moot; (2) whether withdrawal of the no-action letter is “final agency action”; (3) whether that withdrawal is unreviewable prosecutorial discretion; and (4) whether Appellants have standing. A. Mootness The CFTC contends this appeal is moot because the August 2022 rescission of PredictIt’s no-action letter is no longer in effect, having been replaced by the March 2023 letter. And that new letter, the CFTC argues, gives Appellants “the full extent of post-remand relief available to [them],” by providing an explanation for the rescission and a chance for Victoria University to be heard. Moreover, because the March 2023 letter expresses only a “preliminary” determination, the CFTC argues there is “nothing before this Court to review.” In opposition, Appellants invoke the doctrine of voluntary cessation and also argue that the March 2023 letter remains procedurally deficient. The appeal is not moot. Post-filing events do not moot a case “[a]s long as the parties have a concrete interest, however small, in the outcome of the litigation.” Knox v. Serv. Emps. Int’l Union, Loc. 1000, 567 U.S. 298, 307 (2012) (citation omitted) (alteration in original). That is true here. The parties continue to spar over whether PredictIt can operate outside the CEA’s strictures. Although the DMO has now taken down its August 2022 rescission of the no-action letter, its March 2023 replacement continues to say the letter “is void and should be withdrawn.” It makes no difference that the DMO calls this new action “preliminary” and allows Victoria University to lodge objections. The fact that Victoria University can try to change the DMO’s mind does not change the fact that the DMO has declared the no-action letter “void.” A case is not moot when the government rescinds one law only to enact a different version that “disadvantages [the plaintiffs] in the same fundamental way.” Ne. Fla. Chapter of Associated Gen. Contractors of Am. v. City of Jacksonville, 508 U.S. 656, 662 (1993).[4] Nor is it the true that the March 2023 letter gives Appellants all they ask for. That letter actually gives nothing to Appellants—it lets Victoria University object to the no-action letter’s withdrawal but says nothing about Appellants. And, in any event, Appellants continue to assert that the March 2023 letter, despite giving some explanation for the rescission, falls short of what the APA requires when an agency changes course. See, e.g., Wages & White Lion Invs., LLC v. FDA, 16 F.4th 1130, 1139 (5th Cir. 2021) (“When an agency changes course, . . . it must be cognizant that longstanding policies may have engendered serious reliance interests that must be taken into account.” (citation omitted)). B. Final Agency Action The CFTC also argues that withdrawal of the no-action letter is unreviewable because it is neither “agency action” nor “final.” See 5 U.S.C. § 704 (providing judicial review of “final agency action”). We disagree on both points. First, agency action. “Under the APA, ‘agency action’ is a defined term, limited to an ‘agency rule, order, license, sanction, relief, or the equivalent or denial thereof, or failure to act.’” Indep. Equip. Dealers Ass’n v. EPA, 372 F.3d 420, 428 (D.C. Cir. 2004) (quoting 5 U.S.C. § 551(13)). The parties joust over whether the no-action letter is a “license” under this definition. Appellants say yes, contending the letter is a “form of permission” to operate a proposed market. See 5 U.S.C. § 551(8) (defining “license” as “an agency permit, certificate, approval, registration, charter, membership, statutory exemption or other form of permission” (emphasis added)). The CFTC says no, because “ [n]othing in the CEA or any regulation permits staff to license trading facilities” and that the no-action letter, on its face, granted “no affirmative entitlement to do anything.” We agree with Appellants. The no-action letter qualifies as agency action under the APA. “Agency action” has a broad sweep: the term “is meant to cover comprehensively every manner in which an agency may exercise its power.” Whitman v. Am. Trucking Ass’ns, 531 U.S. 457, 478 (2001).[5] Here, the whole point of Victoria University’s requesting the no-action letter was to obtain permission to operate an unregistered event futures market, and to get that green light before plunging significant resources into it. The no-action letter itself characterizes the university as seeking “no-action relief that would allow Victoria University . . . to operate” the proposed market. Furthermore, the letter details the proposed market, states that operating it outside the CEA’s strictures would not be “contrary to the public interest,” and affirmatively “allow[s]” proposed variations from a different event market. Thus, by the letter’s own terms, the no-action relief granted is a “ form of permission.” See 5 U.S.C. § 551(8). Courts have previously found that such grants of permission to avoid compliance with administrative requirements constitute agency action. See, e.g., Atl. Richfield Co. v. United States, 774 F.2d 1193, 1200 (D.C. Cir. 1985) (discussing one such “temporary license”); Gallagher & Ascher Co. v. Simon, 687 F.2d 1067, 1072–76 (7th Cir. 1982) (reviewing withdrawal of a special permit exempting customs brokers from ordinary requirements). Therefore, because the no-action letter here is a “license” within the meaning of the APA, its withdrawal constitutes agency action. Cf. 5 U.S.C. § 558(c) (providing procedural protections for license revocations). Next, finality. Agency action is final when it meets two requirements: “(A) the action must mark the consummation of the agency’s decisionmaking process—it must not be of a merely tentative or interlocutory nature;” and “(B) the action must be one by which rights or obligations have been determined, or from which legal consequences will flow.” Data Mktg. P’ship v. U.S. Dep’t of Lab., 45 F.4th 846, 853 (5th Cir. 2022) (citations and internal quotation marks omitted); see generally Bennett v. Spear, 520 U.S. 154, 177–78 (1997). “This is generally a ‘pragmatic’ inquiry.” Data Mktg., 45 F.4th at 853 (quoting U.S. Army Corps of Eng’rs v. Hawkes Co., 578 U.S. 590, 599 (2016)). And it is a pragmatic inquiry colored by the APA’s embodiment of the “basic presumption of judicial review.” Abbott Lab’ys, 387 U.S. at 140. The CFTC argues that neither finality prong is met. Granting or revoking no-action relief, it claims, does not “consummate” the agency’s decisional process because it is interlocutory—meaning, it pertains only to whether DMO staff will recommend enforcement action to the CFTC. Nor does the letter’s withdrawal trigger any legal consequences. The agency assures us that PredictIt “is free to continue unabated with or without any staff no-action relief,” and that the CFTC can commence enforcement “with or without a staff no-action letter.” Countering this, Appellants argue that the no-action letter’s withdrawal is final because it is unappealable and subjects impacted parties to enforcement proceedings. We again agree with Appellants and find that both finality prongs are met. As to the “consummation” prong, the key question is whether withdrawal of the no-action letter is “subject to further agency review.” Data Mktg., 45 F.4th at 854 (quoting Sackett v. EPA, 566 U.S. 120, 127 (2012)); see also Louisiana v. U.S. Army Corps of Eng’rs, 834 F.3d 574, 581 (5th Cir. 2016) (same). It is not: the DMO’s decision to issue or withdraw the letter is unappealable. So, it does not matter that the letter pertains only to the staff’s recommendation to the agency. Once the staff decide to issue or withdraw the letter, there is no further appeal within the agency. Illustrating that reality, CFTC regulations state that a beneficiary “may rely” on the DMO’s issuing a no-action letter. 17 C.F.R. § 140.99(a)(2). As to the “legal consequences” prong, once more our recent decision in Data Marketing is instructive. As we explained, it is “well-established that ‘where agency action withdraws an entity’s previously held discretion, that action alters the legal regime, binds the entity, and thus qualifies as final agency action.’” Data Mktg., 45 F.4th at 854 (quoting Texas v. EEOC, 933 F.3d 433, 442 (5th Cir. 2019)). That condition was satisfied in Data Marketing because the relevant regulation stated that requestors may “rely” on an advisory opinion. Ibid. This reliance “bound the Department to some degree and withdrew its previously held discretion.” Ibid. The same can be said about PredictIt’s no-action letter: it withdrew some of the CFTC’s discretion because regulations state a beneficiary “may rely” on it. 17 C.F.R § 140.99(a)(2). Thus, for the same reasons as in Data Marketing, legal consequences flowed from the 2014 no-action letter issued by the DMO.[6] None of this is changed by the fact that the DMO has now issued its March 2023 letter. Like the August 2022 letter it supersedes, the March 2023 letter cancels PredictIt’s no-action relief. It states: “As a result of the University’s non-compliance with the terms of [no-action letter], DMOhas determined as a preliminary matter that [no-action letter] is void and should be withdrawn.” True, the letter purports to make that decision “as a preliminary matter,” and it “invite[s] the University to submit any objections it may have” by March 20, 2023. But the letter does not promise to reconsider its decision that the no-action letter “is void and should be withdrawn.” But, again, the possibility that the DMO may reconsider is irrelevant to our inquiry. “[T]he mere fact that the agency could—or actually does— reverse course in the future does not change” an action’s finality. Data Mktg., 45 F.4th at 854 (citing Biden v. Texas, 142 S. Ct. 2528, 2545 (2022)). The March 2023 letter does not say the DMO is merely considering withdrawing no-action relief; it accuses the university of violating the no-action letter’s term in numerous ways and declares the letter “void.” This forces Appellants “either to alter [their] conduct, or to expose [themselves] to potential liability.” Texas v. EEOC, 933 F.3d at 446 (quoting Texas v. EEOC, 827 F.3d 372, 383 (5th Cir. 2016)). For these reasons, the DMO’s withdrawal of no-action relief constitutes final agency action. C. Committed to Agency Discretion The CFTC briefly argues that withdrawing no-action relief is unreviewable as “committed to agency discretion by law.” See 5 U.S.C. § 701(a)(2). It contends that no-action letters are like agency decisions not to prosecute or enforce and, as such, are the “classic illustration of a decision committed to agency discretion.” Bd. of Trade of Chi. v. SEC, 883 F.2d 525, 530 (7th Cir. 1989); see generally Heckler v. Chaney, 470 U.S. 821, 831–32 (1985) (discussing why agency decisions to refuse enforcement are generally unsuitable for judicial review). We disagree. This case does not challenge an agency’s discretionary decision to enforce(or not enforce) the law. What is challenged, rather, is the withdrawal of a regulatory instrument (the no-action letter) that ensured the DMO would not recommend that the agency enforce the CEA against PredictIt. And, as we have pointed out, the agency’s own regulations allow beneficiaries to rely on such letters. See 17 C.F.R. § 140.99(a)(2). The cases the CFTC cites involve the distinct scenario where third parties try to compel an agency to enforce penalties against recipients of no-action letters. Cf. Chicago Bd. of Trade, 883 F.2d at 530 (a challenge to the issuance of a no-action letter is unreviewable as committed to agency discretion). Those cases might apply if we had some third party challenging PredictIt’s no-action letter, arguing the DMO should never have issued it. This case is different: the no-action letter has been rescinded, and the affected parties claim the agency failed to do so properly. We thus conclude that the decision to rescind a no-action letter is not “committed to agency discretion by law.” D. Standing The CFTC also argues that Victoria University’s absence spoils Appellants’ standing. “An individual has standing to sue if his injury is traceable to the defendant and a ruling would likely redress it.” Tex. State LULAC v. Elfant, 52 F.4th 248, 253 (5th Cir. 2022) (citations omitted). In other words: (1) injury, (2) traceability, and (3) redressability. See Lujan v. Defs. of Wildlife, 504 U.S. 555, 560–61 (1992); U.S. Const. art. III, § 2. Appellants say they satisfy each prong. We agree. Appellants—market operators, traders, and academics claiming to be impacted by the no-action letter’s rescission—easily satisfy the standing requirements. At this stage, they have shown numerous injuries stemming from the letter’s withdrawal and the resulting impact on the PredictIt Market. Academics will lose a research tool that was PredictIt’s raison d’être. Traders will lose value in compromised contracts. And PredictIt’s service providers will incur costs from having to prematurely shut down operations. Indeed, Appellants have shown that financial harm was already ongoing before this court issued a stay pending appeal, with “the CFTC’s prohibition on new markets and the impending shutdown order” causing market distortions and “a significant withdrawal of funds.” These injuries, moreover, are directly traceable to the no-action letter’s withdrawal. Operation of the PredictIt market depended on the 2014 no-action relief; withdrawing it would obviously imperil the market, resulting in harms to Appellants. Finally, a favorable ruling would redress these injuries by allowing trading to continue on the same terms as before while the district court adjudicates Appellants’ challenge to the CFTC’s action. The CFTC resists these conclusions. It argues that because “[o]nly the Beneficiary may rely on the no-action letter,” 17 C.F.R. § 140.99(a)(2), only a no-action letter’s beneficiary (here, Victoria University) would have standing to sue. It also observes that all of Appellants’ alleged injuries “reflect[] a downstream harm flowing directly from Victoria University’s hypothetical decision to continue or cease operating PredictIt.” It cites National Wrestling Coaches Ass’n v. Department of Education, 366 F.3d 930 (D.C. Cir. 2004), where several interested parties in the collegiate men’s wrestling world (though not the universities and colleges themselves) challenged a policy interpretation of Title IX. Id. at 934–36. The D.C. Circuit found that the plaintiffs lacked standing because they failed to establish causation and redressability: it was unclear that the third-party colleges would eliminate their men’s wrestling programs in response to the Title IX guidance’s being enjoined. Id. at 938–45. According to the CFTC, that same deficiency is present here because Victoria University may choose to operate or close PredictIt independent of any no-action letter. These counterarguments miss the mark. Whatever CFTC regulations might say, the APA permits suit by anyone “adversely affected or aggrieved by agency action.” 5 U.S.C. § 702. Appellants fall into that category. And National Wrestling is distinguishable. In that case, the court reasoned that even if the challenged policy was enjoined, “Title IX and the 1975 Regulations would still be in place,” serving as an independent obligation for federally funded schools to equally accommodate both genders in athletics. Nat’l Wrestling, 366 F.3d at 939–40. Thus, the third-party schools would not necessarily behave any differently than they otherwise would. Id. at 940. Here, by contrast, enjoining the withdrawal of no-action relief would reinstate the 2014 no-action letter, permitting PredictIt to continue operating as before. And there would be no independent obligations to register with the CFTC because of the promise that Victoria University “may rely” on its no-action relief. 17 C.F.R. § 140.99(a)(2). In sum, Appellants have standing. III. Preliminary Injunction We now turn to whether the district court abused its discretion by denying a preliminary injunction. See Moore v. Brown, 868 F.3d 398, 402 (5th Cir. 2017) (per curiam). To obtain a preliminary injunction, Appellants must show: (1) a substantial likelihood of success on the merits, (2) a substantial threat of irreparable harm if the injunction does not issue, (3) that the threatened injury outweighs any harm that will result if the injunction is granted, and (4) that granting the injunction is in the public interest. Id. at 402–03. A. Substantial Likelihood of Success We first ask whether Appellants are substantially likely to show that the no-action letter’s revocation was arbitrary and capricious. “The APA’s arbitrary-and-capricious standard requires that agency action be reasonable and reasonably explained.” Data Mktg., 45 F.4th at 855 (quoting FCC v. Prometheus Radio Project, 141 S. Ct. 1150, 1158 (2021)). The court can only consider the reasoning “articulated by the agency itself,” and cannot consider “post hoc rationalizations for agency action.” Motor Vehicle Mfrs. Ass’n of U.S., Inc. v. State Farm Mut. Auto Ins. Co., 463 U.S. 29, 50 (1983). “[W]e must set aside any action premised on reasoning that fails to account for ‘relevant factors’ or evinces ‘a clear error of judgment.’” Data Mktg., 45 F.4th at 855 (quoting Univ. of Tex. M.D. Anderson Cancer Ctr. v. HHS, 985 F.3d 472, 475 (5th Cir. 2021)). The August 2022 revocation fails these standards for the obvious reason that it gives no explanation whatsoever. Instead of “reasonably explain[ing]” the withdrawal, ibid., the DMO delivered this terse missive: “The University has not operated its market in compliance with the terms of [the no-action letter].” Not a word discloses which terms were violated or what evidence supports the charge. Nor is any reason given why PredictIt must swiftly close all contracts by a certain date or why the agency rejected less draconian measures, given the significant reliance interests in play. See, e.g., Nat’l Shooting Sports Found., Inc. v. Jones, 716 F.3d 200, 215 (D.C. Cir. 2013) (agency action is not upheld if it fails to consider “significant and viable and obvious alternatives” (cleaned up)). This is the epitome of arbitrary and capricious action. See Sprint Nextel Corp. v. FCC, 508 F.3d 1129, 1132 (D.C. Cir. 2007) (“We . . . require more than a result; we need the agency’s reasoning for that result.” (emphasis added)). Less than a month after oral argument, the agency tried to fix these glaring defects by issuing the March 2023 letter. As noted, this letter purports to “supersede” the August 2022 rescission while reaffirming the agency’s decision that the no-action letter “is void and should be withdrawn.” It also provides some explanation for withdrawing the no-action letter, such as the charge that Victoria University violated the letter’s terms by allowing a for-profit company (Aristotle, Inc.) to operate PredictIt. We have already explained why the March 2023 letter does not moot this appeal. See supra II(A). The March 2023 letter should also have no bearing on whether the withdrawal of the no-action letter is arbitrary and capricious. That is because theletter violates the injunction pending appeal our panel previously entered. Appellants had asked us to “enjoin the enforcement of the Commission’s February 15, 2023, liquidation mandate and allow the PredictIt Market event contracts that were offered as of the date of the agency’s decision . . . to continue trading pending resolution of this appeal.” We granted that requested injunction on January 26, 2023. The March 2023 letter violates that injunction by purporting to withdraw no-action relief, thereby subjecting PredictIt—and all of its existing contracts—to regulation. Although we exercised our discretion to deny Appellants’ sanctions motion, we will not allow the enjoined agency to game the system by retrofitting its previous rescission with “reasons” after oral argument. See Texas v. Biden, 10 F.4th 538, 558–59 (5th Cir. 2021) (“It is a fundamental precept of administrative law that an administrative agency cannot make its decision first and explain it later.”). But even if we were to consider the March 2023 letter, we would still find serious problems with its reasons for voiding the no-action letter. To begin with, we have concluded that the no-action letter qualifies as a “license” under the APA. See supra II(B). The March 2023 letter, however, does not purport to follow the procedural requirements for withdrawing a license. See 5 U.S.C. § 558(c). The agency only provided Victoria University with an opportunity to respond to objections. It offered no opportunity for Victoria University “to demonstrate or achieve compliance” with the requirements that were purportedly violated. Id. § 558(c)(2). The withdrawal of no-action relief is therefore procedurally deficient on that basis alone. Aside from that defect, there are other evident flaws in the March 2023 letter’s substance. For instance, the letter does not meaningfully explain why the DMO rejected alternatives like allowing currently existing markets to expire on their own terms. It says only that such alternatives would not “be appropriate,” given the likelihood of recurrence due to past violations. But the letter does not explain why past violations suggest a likelihood of recurrence in the future. This is hardly the “reasoned decision making” required of administrative agencies. Michigan v. EPA, 576 U.S. 743, 750 (2015) (quoting Allentown Mack Sales & Serv., Inc. v. NLRB, 522 U.S. 359, 374 (1988)). Nor does the DMO justify its conclusion that monitoring future compliance would require an “unreasonable use of taxpayer resources.” It says nothing about the magnitude of the resources required and does not explain why they would not be justified given longstanding reliance interests. See Encino Motorcars, LLC v. Navarro, 579 U.S. 211, 221–22 (2016) (“[A]n agency must also be cognizant that longstanding policies may have ‘engendered serious reliance interests that must be taken into account.’” (quoting FCC v. Fox Television Stations, Inc., 556 U.S. 502, 515 (2009))); Michigan, 576 U.S. at 750–51 (requiring consideration of both sides of the cost-benefit ledger). Finally, the letter engages in obvious post hoc rationalization. It tries to partially justify the agency’s charge that Victoria University “ceded operational control” of PredictIt to a for-profit company by referring to remarks made by the company’s counsel at oral argument. That is verboten. What counsel said at argument cannot justify actions the agency took months if not years before. See DHS v. Regents of the Univ. of Cal., 140 S. Ct. 1891, 1907 (2020) (“It is a ‘foundational principle of administrative law’ that judicial review of agency action is limited to ‘the grounds that the agency invoked when it took the action.’” (quoting Michigan, 576 U.S. at 758)); see also SEC v. Chenery Corp., 318 U.S. 80, 95 (1943) (“[A]n administrative order cannot be upheld unless the grounds upon which the agency acted in exercising its powers were those upon which its action can be sustained.”). In sum, we conclude that the revocation of the no-action letter was likely arbitrary and capricious because the agency gave no reasons for it. And the agency’s attempts to retroactively justify the revocation after oral argument—and in the face of our injunction—only underscore why Appellants are likely to prevail. B. Irreparable Injury We now turn to irreparable injury. Appellants have alleged a number of harms they will suffer absent a preliminary injunction. First, investors and traders will not be able to see their contracts through and realize any gains from having predicted events correctly. Even if they wanted to cash out now, the prices for those contracts would be distorted due to the market disruptions that the no-action letter’s rescission engendered. Second, as traders have attempted to salvage their investments due to a looming and impending shutdown order, academics have had their research compromised by the trading irregularities that corrupted the integrity of their data. Finally, PredictIt’s operators have been saddled with heavy compliance costs given the market’s closure. As it did in the standing context, the CFTC claims that all of these harms are inherently speculative. It asserts that any possible injuries could be undone through monetary remedies. And, although the United States would enjoy sovereign immunity, Appellants could sue the market operators. See Dennis Melancon, Inc. v. City ofNew Orleans, 703 F.3d 262, 279 (5th Cir. 2012) (“The possibility that adequate compensatory or other corrective relief will be available at a later date . . . [weighs] heavily against a claim of irreparable harm” (quoting Morgan v. Fletcher, 518 F.3d 236, 240 (5th Cir. 1975))). We disagree and conclude that Appellants are likely to suffer irreparable harm. As noted, Appellants have shown they were already undergoing harm before we issued the stay pending appeal. Some of these harms, such as the academic value of accurate data, would be difficult to restore with monetary damages. And to the extent some of these harms are economic, the United States cannot be sued due to its sovereign immunity. See Wages & White Lion Invs., 16 F.4th at 1142 (“[C]omplying with an agency order later held invalid almost always produces the irreparable harm of nonrecoverable compliance costs . . . because federal agencies generally enjoy sovereign immunity for any monetary damages.” (cleaned up, quotation omitted)). To the extent the CFTC argues that the market operators could always be sued, that argument neglects the simple fact that at least some of those operators are themselves parties to this lawsuit. We therefore conclude that Appellants have established a substantial likelihood of suffering irreparable harm absent a preliminary injunction. C. Balance of the Equities and the Public Interest Finally, we consider the remaining preliminary injunction factors: the balance of the equities and the public interest. These factors “merge when the Government is the opposing party.” Nken v. Holder, 556 U.S. 418, 435 (2009). When addressing these factors, “courts ‘must balance the competing claims of injury and must consider the effect on each party of the granting or withholding of the requested relief.’” Winter v. NRDC, 555 U.S. 7, 24(2008) (quoting Amoco Prod. Co. v. Village of Gambell, 480 U.S. 531, 542 (1987)). These factors weigh in favor of granting an injunction. On Appellants’ side, the harms include all those just discussed: investor losses, corrupted academic data due to market distortions, and heavy compliance costs on market operators. Moreover, “[t]he public interest is served when administrative agencies comply with their obligations under the APA.” Northern Mariana Islands v. United States, 686 F. Supp. 2d. 7, 21 (D.D.C. 2009). As for the other side of the ledger, the CFTC points to the systemic harms that would arise by permitting litigation on informal no-action letters. It argues that requiring full-dress APA litigation on these sorts of informal letters would discourage the practice of giving them in the first place, and result in “a net loss of far greater proportions to the average citizen than any possible gain which would accrue.” Taylor-Callahan-Coleman Cntys. Dist. Adult Prob. Dep’t v. Dole, 948 F.2d 953, 959 (5th Cir. 1991) (citation omitted). While mindful of that possibility, that sort of a high-level, systemic consideration cuts both ways: agency decisionmaking is legitimated in part by the agency’s providing adequate reasons. Especially where, as here, longstanding policies have engendered serious reliance interests, agencies must take those considerations into account before abruptly changing course. See Encino Motorcars, 579 U.S. at 221–22. Accordingly, we conclude that the balance of the equities and the public interest weigh in favor of granting a preliminary injunction IV. Conclusion We REVERSE the district court’s effective denial of a preliminary injunction and REMAND with instructions that the district court enter a preliminary injunction pending its consideration of Appellants’ claims. James C. Ho, Circuit Judge, concurring: Plaintiffs’ theory of final agency action admittedly conflicts with the precedents of our sister circuits. To my knowledge, no circuit has held that a no-action letter or its withdrawal is sufficient to constitute “final agency action” under the Administrative Procedure Act. And some have held the opposite. See, e.g., New York City Employees’ Retirement System v. SEC, 45 F.3d 7, 12 (2nd Cir. 1995) (“No-action letters . . . do not impose or fix a legal relationship upon any of the parties.”); Trinity Wall Street v. Wal-Mart Stores, Inc., 792 F.3d 323, 331 (3rd Cir. 2015) (“[N]o-action letters are not binding—they reflect only informal views of the staff and are not decisions on the merits.”); Bd. of Trade of City of Chicago v. SEC, 883 F.2d 525, 531 (7th Cir. 1989) (“The petition for review of the no-action letter . . . is dismissed for want of a reviewable order.”). Cf. Paul v. Petroleum Equipment Tools Co., 708 F.2d 168, 174 n.5 (5th Cir. 1983) (“[T]his ‘no action’ position is not equivalent to an exemption.”). That said, we need not reach a definitive conclusion on this issue at this time. As detailed in the majority opinion, the issues presented in this case are sufficiently close that Plaintiffs have demonstrated a substantial likelihood of success, and satisfied the remaining elements required for a preliminary injunction as well. “The purpose of a preliminary injunction is merely to preserve the relative positions of the parties until a trial on the merits can be held.” Univ. of Tex. v. Camenisch, 451 U.S. 390, 395 (1981). “[F]indings of fact and conclusions of law made by a court granting a preliminary injunction are not binding at trial on the merits.” Id. See also Feds for Medical Freedom v. Biden, 63 F.4th 366, 389 (5th Cir. 2023) (“We hasten to emphasize that this case only involves a preliminary injunction.”). Accordingly, I concur. James E. Graves, Jr., Circuit Judge, dissenting: Although I agree that this case is not moot, I would not issue a preliminary injunction in this case. As reiterated by this court on numerous occasions, the issuance of a preliminary injunction is an exceptional remedy that should be granted only when the moving party has clearly shown that they can meet all four requirements. See, e.g., Guy Carpenter & Co. v. Provenzale, 334 F.3d 459, 464 (5th Cir. 2003) (“A preliminary injunction is an extraordinary remedy which courts grant only if the movant has clearly carried the burden as to all four elements.”); Allied Marketing Group., Inc. v. CDL Marketing, Inc., 878 F.2d 806, 809 (5th Cir. 1989) (stating that preliminary injunctive relief “is an extraordinary remedy and should be granted only if the movant has clearly carried the burden of persuasion with respect to all four factors”). We do not grant such relief unless we find: (1) a substantial likelihood of success on the merits; (2) a substantial threat of irreparable injury; (3) the threatened injury to the movant outweighs the threatened harm to the party sought to be enjoined; and (4) granting the injunctive relief will not disserve the public interest. City of Dallas v. Delta Air Lines, Inc., 847 F.3d 279, 285 (5th Cir. 2017). I am not convinced that Appellants have satisfied this high burden. In my view, Appellants have failed to demonstrate a substantial likelihood that they will prevail on the merits, as there is no final agency action in this case. For agency action to be “final,” two conditions must be met: (1) “the action must mark the ‘consummation’ of the agency’s decisionmaking process”; and (2) “the action must be one by which ‘rights or obligations have been determined,’ or from which ‘legal consequences will flow.’” Bennett v. Spear, 520 U.S. 154, 177–78 (1997) (citations omitted). CFTC’s no-action letters fail to satisfy either condition: they neither mark the consummation of the agency’s decision making process nor determine Appellants’ legal rights or obligations. CFTC rule 140.99 outlines the procedure for requesting Commission staff letters. See 17 C.F.R. § 140.99. The rule, among other things, requires that the request must be made by or on behalf of the person subject to the request, must relate to a proposed transaction or activity, and must set forth as completely as possible all material facts and circumstances. See id. § 140.99(b). When the CFTC staff reviews a request, the rule makes clear that the “[i]ssuance of a [l]etter is entirely within the discretion of Commission staff.” Id. § 140.99(b)(1). Rule 140.99 further explains that no-action letters are “a written statement” that the issuing staff, here DMO, “will not recommend enforcement action to the Commission,” and that such a statement “binds only the issuing Division . . . and not the Commission.” Id. § 140.99(a)(2). Thus, no-action letters are informal and advisory, inherently staff-level statements about whether the issuing staff might (or might not) recommend to the CFTC that the Commission, at the Commission’s sole discretion, vote to authorize civil proceedings against a non-compliant entity. Accordingly, these letters plainly do not mark the consummation of the agency’s decision making. Nor do the letters represent a decision determining rights or obligations, or one from which legal consequences flow as it does not commit the CFTC to taking enforcement action. Despite this, the majority concludes that the 2014 no-action letter effectively constituted a “license.” See ante at 9. Under the APA, a “license” is defined as “an agency permit, certificate, approval, registration, charter, membership, statutory exemption or other form of permission.” 5 U.S.C. § 551(8). With such a sweeping definition at hand, the majority concludes that “by the letter’s own terms, the no-action relief granted is a form of permission.” See ante at 9 (internal quotation marks omitted). I remain unconvinced by this argument, as the word “permission” is commonly understood as a formal authorization.1 What happened here is in stark contrast to the concept of explicit consent. On its face, the no-action letter does not grant Appellants the right to do anything. Instead, the letter simply expresses DMO’s intention to “not recommend that the Commission take any enforcement action in connection with the operation of [the] proposed market.” The DMO’s decision was contingent upon information furnished by Appellants and was subject to certain conditions. The letter explicitly states that any alterations, omissions, or discrepancies in the facts or circumstances may render the granted no-action relief null and void. Thus, to maintain that the absence of a recommendation to prosecute equates to formal consent stretches the bounds of credulity. See Paul v. Petroleum Equip. Tools Co., 708 F.2d 168, 174 n.5 (5th Cir. 1983) (observing that a “no-action” letter “is not equivalent to an exemption”) (Higginbotham, J.). I have not come across any instance where a court has ruled that a “no-action letter” constitutes a final action taken by the agency. Tellingly, the majority cites no such case. Contrarily, no-action letters have been regularly found to be non-binding and devoid of legal authority, precluding their review. See, e.g., Trinity Wall St. v. Wal-Mart Stores, Inc., 792 F.3d 323, 331 (3d Cir. 2015) (recognizing that “no-action letters are not binding—they reflect only informal views of the staff and are not decisions on the merits”); Board of Trade of City of Chicago v. SEC, 883 F.2d 525, 530 (7th Cir. 1989) (concluding that SEC no-action letters are not reviewable because they do not constitute a “final” decision concerning the status of the parties); New York City Emps.’ Ret. Sys. v. SEC, 45 F.3d 7, 12 (2d Cir. 1995) (“No-action letters are deemed interpretive because they do not impose or fix a legal relationship upon any of the parties.”). Because I am not persuaded that we should be the first court to draw the conclusion that a “no-action letter” constitutes “final agency action,” I respectfully dissent.
