Opinion for the court filed by Circuit Judge BROWN.
Concurring opinion filed by Chief Judge SENTELLE.
BROWN, Circuit Judge:
Petitioners NetworkIP, LLC, and Network Enhanced Telecom, LLP, (collectively “NET”) seek review of a pair of final orders of the Federal Communications Commission (“FCC”) — one finding liability, APCC Servs. Inc., 21 F.C.C.R. 10488 (2006) (Order on Review) (“Liability Order”), and the other imposing damages, APCC Servs., Inc., 22 F.C.C.R. 4286 (2007) (Memorandum Opinion and Order) (“Damages Order”). Because the FCC reasonably interpreted its own prior orders, we deny the petition as to liability. We grant in part NET’s petition as to damages, however, because the FCC’s failure to enforce its filing deadline was arbitrary and capricious.
I.
In a terabyte generation in which even three-year olds carry GPS-equipped wireless phones,1 the payphone industry may seem like a Technicolor afterthought. Nonetheless payphones still fill an important, though decreasing, role in communications, and Congress has sought to keep them around.
“Two types of calls may be placed from a payphone. The first and most common type is the ‘coin call,’ in which the caller inserts a coin directly into the payphone before making the call; the rates for coin calls are set by State commissions.” Sprint Corp. v. FCC, 315 F.3d 369, 371 (D.C.Cir.2003). Increasingly common, however, is “the second type of call — ‘coin-less calls’ — which a caller places by using a service such as directory assistance, operator service, an access code, or a subscriber 800 number.” Id. The rules governing this second category of calls are at issue here.
To ensure payphone service providers (“PSPs”) are compensated for these dial-around “calls to 800 numbers or 10XXX numbers that the caller uses to reach the long-distance carrier of his choice,” and thus to encourage the availability of payphones, “Congress enacted § 276 of the Telecommunications Act of 1996.” III. Pub. Telecomm. Ass’n v. FCC, 117 F.3d 555, 559 (D.C.Cir.1997) (citing 47 U.S.C. § 276). The FCC must “establish a per call compensation plan to ensure that all payphone service providers are fairly compensated for each and every completed intrastate and interstate call using their payphone....” 47 U.S.C. § 276(b)(1)(A).
The concept is simple: Telecommunications carriers must compensate PSPs for calls made from payphones with calling cards. Application, alas, is complicated, because long-distance calls often involve multiple carriers. For instance, a local exchange carrier (“LEC”) initially might receive a call, and then route it to a non-LEC — “typically an interexchange carrier (TXC’)[ ] ... such as Sprint, AT & T, and WorldCom” — that then transmits the call to yet another carrier. Sprint Corp., 315 [119]*119F.3d at 371. “If the recipient of the call is a customer of the IXC, the IXC will simply transmit the call to the LEC that serves the customer,” but “[i]f the call recipient is not a customer of the IXC, ... the IXC transfers the call to a ‘reseller’ of the IXC’s services.” Id.
We have noted that “[t]wo types of resellers exist. The first, known as switch-less resellers, do not possess their own switching facilities and must rely on an IXC to perform the switching and transmission functions that are required to complete a call.” Id. “By contrast, the second type, switch-based resellers (‘SBRs’), possess their own switching capacities.... ” Id. “[I]n some instances the SBR transfers the call to another SBR, which in turn routes the call to yet another SBR, and so on.” Id.
In its First Payphone Order, the FCC said “faeilities-based carriers [‘FBCs’] should pay the per-call compensation for the calls received by their reseller customers.” Implementation of the Pay Telephone Reclassification and Compensation Provisions of the Telecommunications Act of 1996, 11 F.C.C.R. 20541, 20586, ¶86 (1996) (Report and Order). Later that year, in its First Payphone Reconsideration Order, the FCC said an FBC “maintains its own switching capability, regardless if the switching equipment is owned or leased by the carrier.” Implementation of the Pay Telephone Reclassification and Compensation Provisions of the Telecommunications Act of 1996, 11 F.C.C.R. 21233, 21277, ¶ 92 (1996) (Order on Reconsideration). After two unsuccessful attempts to set a per call dial-around rate, see III. Pub. Telecomm. Ass’n, 117 F.3d 555, 564 (D.C.Cir.1997) (remanding $.35 rate); MCI Telecomms. Corp. v. FCC, 143 F.3d 606, 608 (D.C.Cir.1998) (remanding $.284 rate), the FCC established $.24 per call as the applicable rate, Implementation of the Pay Telephone Reclassification and Compensation Provisions of the Telecommunications Act of 1996, 14 F.C.C.R. 2545, 2632, ¶ 191 (1999) (Third Report and Order), which we upheld on review, Am. Pub. Commc’ns Council v. FCC, 215 F.3d 51, 58 (D.C.Cir.2000).
NET, headquartered in Texas, is a telecommunications carrier that owns switches. Using an innovative web interface, NET empowered various other carriers to develop prepaid calling cards. Traditionally, carriers were obligated to purchase or lease their own switches in order to fully control calling-card functions. NET developed a new technology that (it says) allowed its customers to control switches as if they possessed them, thus severing the technologically out-dated link between switching and physical possession of switches. NET’S customers could “modify, in real time, the key economic parameters vital to the prepaid business,” such as “how to set up accounts, how much to charge, which domestic or foreign destinations could be reached with the cards, and by which methods.” Pet’r’s Br. 6. NET likewise instructed its customers that they alone were responsible for compensating PSPs, and often language to that effect was included in its contracts. Between October 1999 and November 2001, the relevant period for our purposes, upwards of eleven million calls were placed with calling cards distributed by NET’S customers, using NET’s switches.
In 2002, a group of PSPs including APCC Services, Inc. (“APCC”), a billing clearinghouse for PSPs, filed an informal complaint with the FCC against NET; a formal complaint followed in 2003. There were two proceedings, one for liability, and the other for damages. Ultimately, the FCC ordered NET to pay $2,789,505.84, plus interest at 11.25%. NET has peti[120]*120tioned for review of both the Liability and Damages Orders, and our review has been consolidated. APCC intervened, filing a motion to dismiss because of an alleged jurisdictional defect in NET’s petition for review of the Damages Order, this motion has since been withdrawn.2
II.
We first consider jurisdiction, though it is no longer contested. “It is axiomatic that subject matter jurisdiction may not be waived, and that courts may raise the issue sua sponte." Athens Cmty. Hosp., Inc. v. Schweiker, 686 F.2d 989, 992 (D.C.Cir.1982). Indeed, we must raise it, because while arguments in favor of subject matter jurisdiction can be waived by inattention or deliberate choice, we are forbidden — as a court of limited jurisdiction — from acting beyond our authority, and “no action of the parties can confer subject-matter jurisdiction upon a federal court.” Akinseye v. District of Columbia, 839 F.3d 970, 971 (D.C.Cir.2003); see also Wilks v. U.S. Marshals Serv., No. 92-5287, 1993 WL 118285, at *1 (D.C.Cir.1993) (per curiam).
At first blush, jurisdiction seems euclidean. By statute, federal appellate courts have “exclusive jurisdiction to enjoin, set aside, suspend (in whole or in part), or to determine the validity of’ final FCC orders. 28 U.S.C. § 2342(1). We are called upon to “determine the validity of’ a pair of FCC final orders, so we have jurisdiction. QED. But the existence of parallel provisions — one to challenge final agency action and the other to enforce compliance — complicates this otherwise straightforward equation, particularly in light of a Supreme Court precedent attempting to “harmonize” a superficially similar statutory scheme. ICC v. Atlantic Coast Line R.R., 383 U.S. 576, 586, 86 S.Ct. 1000, 16 L.Ed.2d 109 (1966).
APCC initially complained that this court’s jurisdiction to hear NET’s challenge to the orders should not trump APCC’s right to seek enforcement under 47 U.S.C. § 407. The language of the enforcement statute at issue in Atlantic Coast, 49 U.S.C. § 16(2) (1964), was for all relevant purposes identical to § 407. Like § 407 does, § 16(2) allowed “any person for whose benefit [an agency’s] order was made” (an “adjudged-injured party”) to file a suit against a party who “d[id] not comply with an order for the repayment of money” (an “adjudged-injuring party”). In a situation somewhat similar to that here, the Atlantic Coast Court construed this language to mean that “a[n adjudged-injuring party] may obtain review of the Commission’s order only in the court where the [adjudged-injured party] commences its enforcement action — or where the [adjudged-injured party] seeks review of the Commission’s order.” 383 U.S. at 579, 86 S.Ct. 1000.
[121]*121The similarities between this case and Atlantic Coast are obvious, but we decline to extend Atlantic Coast, even assuming that case was jurisdictional and not merely venue-related. Unlike the review provision in Atlantic Coast, 49 U.S.C. § 17(9) (1964), § 2342(1) places jurisdiction “exclusively]” in the courts of appeal; in Atlantic Coast, both were district courts. There are institutional differences between trial and appellate courts, and when Congress has spoken so explicitly as to the particular type of court it wants to review agency action, as it has in § 2342(1), that explicit statement should not be set aside lightly. Likewise, the Court in Atlantic Coast put weight on the possibility of a § 17(9) cross-proceeding in a § 16(2) action. 383 U.S. at 601-02, 86 S.Ct. 1000. But in the FCC context, a § 2342(1) cross-proceeding in a district court is impossible.
Although there is some potential for vitiating congressional policy enhancing the injured party’s ability to choose its forum and encouraging prompt payment of reparation awards, we think the FCC context is distinct enough to justify the exercise of jurisdiction here. What finally tips the scale in favor of our having jurisdiction is a statute enacted in 1988, 47 U.S.C. § 208(b). As happened here, agencies can bifurcate a single grievance into separate proceedings for liability and damages. Under § 208(b), an adjudged-injuring party can in some instances seek federal appellate review of an FCC liability order even before a damages order has been issued. See Verizon Tel. Co. v. FCC, 269 F.3d 1098, 1103-06 (D.C.Cir.2001). Thus, in our “harmonizing” of competing statutes, we have a new input: § 208(b). Consequently, an adjudged-injured party already may have to forego its favorite forum; if it wants to defend a liability order, it may have to intervene in the § 2342(1) action. This scenario undercuts much of the reasoning in Atlantic Coast. We therefore conclude we have jurisdiction.3
III.
We now address the Liability Order. NET appears to concede the FCC’s interpretation of the First Payphone Reconsideration Order, including its emphasis on some kind of possessory interest, is reasonable. Pet’r’s Br. 28 (“For these reasons, NET’s interpretation certainly is as reasonable as the FCC’s.... ”). This is no act of charity. Final agency orders are upheld unless “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law,” or not supported by “substantial evidence.” 5 U.S.C. § 706(2). The FCC’s “interpretation of its own orders and rules is entitled to substantial deference,” AT & T Corp. v. FCC, 448 F.3d 426, 431 (D.C.Cir.2006), just as “an agency’s interpretation of one of its own regulations commands substantial judicial deference.” Drake v. FAA, 291 F.3d 59, 68 (D.C.Cir.2002).
[122]*122In this case, even without substantial deference, the FCC’s interpretation of its earlier payphone orders was appropriate.4 Because a challenge to the appropriateness of the Liability Order itself is unavailing, NET’s stronger argument does not go to the reasonableness of the FCC’s construction of the First Payphone Reconsideration Order, but instead to whether NET was fairly warned and thus should not have to pay damages.
Preliminarily, we confront the FCC’s contention that the fair notice issue was not presented to the agency. We cannot review “questions of fact or law upon which the Commission, or designated authority within the Commission, has been afforded no opportunity to pass.” 47 U.S.C. § 405(a). If a petitioner could have called a question of law or fact to the agency’s attention, but did not, the issue is waived. Freeman Eng’g Assocs. v. FCC, 103 F.3d 169, 182-83 (D.C.Cir.1997). However, an issue need not be raised explicitly; it is sufficient if the issue was “necessarily implicated” in agency proceedings. Time Warner Entm’t Co. v. FCC, 144 F.3d 75, 79-80 (D.C.Cir.1998).
Here, NET adequately raised the fair notice issue. Before the FCC, NET argued “the Enforcement Bureau disregarded the plain language of the Commission’s payphone compensation rules, and ignored NET’s business, which NET structured in reliance on the rules.” Application for Review at 2, APCC Services Inc., et al. v. NetworkIP, LLC, et al. (FCC, 2006) (No. EB-03-MD-011). NET also averred “the Enforcement Bureau’s determination is in conflict with the Commission’s regulations, decisions, and established policy,” and that it “brushed aside” the FCC’s prior statements. Id. at 3, 11. NET even quoted one of our cases for the proposition that “ ‘there is a need for a clear and definitive interpretation of all agency rules so that the parties upon whom the rules will have an impact will have adequate and proper notice concerning the agency intentions.’ ” Id. at 13 (quoting FTC v. Atlantic Richfield Co., 567 F.2d 96, 103 (D.C.Cir.1977)). The Enforcement Bureau addressed NET’s contention, APCC Services Inc., 20 F.C.C.R.2073, 2081, ¶ 19 n. 43 (2005) (Memorandum Opinion and Order) (“Bureau Liability Order ”), and the Commission “affirm[ed] the Bureau Liability Order .... ” Liability Order, 21 F.C.C.R. at 10488-49, ¶ 1. This is sufficient to preserve the issue.
Though agencies are entitled to deference, they may not retroactively change the rules at will. Indeed, that “[ejlementary considerations of fairness dictate that individuals should have an opportunity to know what the law is and to conform their conduct accordingly” has been well-established for “centuries.” Landgraf v. USI Film Products, Inc., 511 U.S. 244, 265, 114 S.Ct. 1483, 128 L.Ed.2d 229 (1994). Anything less ought not to be dignified with the title of law.5 These “[tjraditional concepts of due process in-[123]*123corporated into administrative law preclude an agency from penalizing a private party for violating a rule without first providing adequate notice of the substance of the rule.” Satellite Broad. Co. v. FCC, 824 F.2d 1, 3 (D.C.Cir.1987).
At the same time, however, agencies are authorized to make policy choices through adjudication, and giving a decision retroactive effect is “not necessarily fatal to its validity.” SEC v. Chenery Corp., 332 U.S. 194, 203, 67 S.Ct. 1575, 91 L.Ed. 1995 (1947). After all, “[e]very case of first impression has a retroactive effect, whether the new principle is announced by a court or by an administrative agency.” Id. And, as is common with comprehensive regulatory schemes, often “every loss that retroactive application ... would inflict on [one party] is matched by an equal and opposite loss that non-retroactivity would inflict on [another].” Qwest Servs. Corp. v. FCC, 509 F.3d 531, 540 (D.C.Cir.2007). This case potentially stands at the pivot point between these competing principles.
There are “two conflicting modes of judicial review to agency interpretations,” with “[o]ne longstanding line of [our] cases allow[ing] agencies to apply new interpretations of regulations retroactively,” while another requires “revers[ing] agency action where regulated parties do not have fair warning of the agency’s interpretation of its regulations.” Kieran Ringgenberg, Comment, United States v. Chrysler: The Conflict Between Fair Warning and Adjudicative Retroactivity in D.C. Circuit Administrative Law, 74 N.Y.U. L.Rev. 914, 916 (1999). NET attacks with fair notice cases like United States v. Chrysler Corp., 158 F.3d 1350 (D.C.Cir.1998); the FCC parries with retroactivity cases like Qwest.
When to apply which line of cases has not been resolved definitively by our precedents. We too leave for another day the question of how these two lines interplay, because under either one, NET loses. That NET has an unwinnable case under the retroactivity line is obvious; the correctness of NET’s interpretation was anything but “settled”, and many PSPs will be harmed if NET escapes liability. Qwest, 509 F.3d at 540-41. NET, however, also loses under the fair notice line, because its interpretation of the First Payphone Reconsideration Order is less plausible than the FCC’s. The FCC’s, in fact, is the most reasonable interpretation. We have never applied the fair notice doctrine in a case where the agency’s interpretation is the most natural one.6
Consider the language of the First Payphone Reconsideration Order. The operative phrase reads, “We clarify that a carrier is required to pay compensation and provide per-call tracking for the calls originated by payphones if the carrier maintains its own switching capability, regard[124]*124less if the switching equipment is owned or leased by the carrier.” 11 F.C.C.R. at 21277, ¶ 92. NET focuses on the words “maintain” and “capability.” Quoting Webster’s Dictionary, NET defines “maintain” as “ ‘to provide for,’ ‘to continue,’ ‘to keep in existence: to sustain,’ and ‘to preserve or keep in a given existing condition, as of efficiency or good repair,’ ” and “capability” is defined as “ ‘the quality or state of being capable,’ or the ‘capacity to be used, treated or developed for a particular purpose....’” Pet’r’s Br. 17 (quoting Webster’s II New College Dictionary 661, 164 (1999)). NET argues these words are satisfied provided a carrier has the ability to control switches, which its customers did. It then interprets the words “regardless if the equipment is owned or leased” as “it does not matter whether a switching capability is maintained via an ownership or lease or some other means.” Id.
NET’S is not an impossible interpretation, but it is not the most natural one. When the words “maintains its own switching capability” are read in light of the phrase “regardless if the switching equipment is owned or leased by the carrier,” then “maintains its own switching capability” is best understood as shorthand for either owning or leasing, but nothing else. Indeed, even NET’S proffered definitions suggest the need for a physical connection or possession of some sort; one usually “keep[s] in existence” or “preserve[s]” something in “good repair” by means of physical access. Thus, though the language can be read the way NET does, we agree with the FCC’s Enforcement Bureau that “rather than rejecting a possessory interest requirement, the sentence simply clarifies the kinds of posses-sory interests that will suffice.” Bureau Liability Order, 20 F.C.C.R. at 2081, ¶ 18.
NET also turns to other sources of potential ambiguity. For instance, it points to additional language from the First Order on Reconsideration that arguably permits a carrier to “maintain its own switching capability” by contract. From this, NET asserts the FCC’s insistence on a possessory interest in switches is oceans apart from what NET reasonably perceived as the earlier, more flexible rule. Again, we are unpersuaded.
The critical language reads: “If a carrier with a switching capability has technical difficulty in tracking calls from origination to termination, it may fulfill its tracking and payment obligations by contracting out this duty to another entity....” First Order on Reconsideration, 11 F.C.C.R. at 21277, ¶ 92. If tracking is synonymous with switching capability, the sentence borders on incoherence — a carrier with the capability to track calls is technically unable to track calls? Thus, defining “switching capability” as the mere technical ability to track calls is not the most reasonable approach. Instead switching capability and tracking capability are separate, and a “tracking and payment obligation” does not lodge until after a carrier is already an FBC. A carrier can have the ability to track without being an FBC, and a carrier can be an FBC without having the ability to track. But if a carrier is an FBC, it has a legal duty to track, either directly or by means of contract.
As NET suggests, the FCC has not always insisted a possessory interest is a necessary attribute of the phrase “facilities based.” In the narrow context of “unbundled network elements,” the FCC has taken a loosey-goosey approach to ownership. See Federal-State Joint Bd. on Universal Serv., 12 F.C.C.R. 8776, 8862-70 (1997). The FCC reasonably responds, however, that “facilities-based” — as the plain words suggest — typically connotes some sort of possessory interest, and “the commonly understood meaning of the term ‘facilities-[125]*125based’ ” among those regulated requires a possessory interest of some sort. Liability Order, 21 F.C.C.R. at 10490, ¶ 6.7 The interpretation of the phrase “facilities based” in the “unbundled network elements” context is noteworthy, but it was unreasonable for NET to assume that an idiosyncratic exception should define the rule. Because the FCC’s interpretation of its First Order on Reconsideration is the most natural, we hold the fair notice doctrine has been satisfied.8
IV.
Congress has set a two-year statute of limitations for “[a]ll complaints against carriers for the recovery of damages not based on overcharges_” 47 U.S.C. § 415(b). The FCC recognizes both formal and informal complaints. 47 C.F.R. § 1.711. Informal complaints are forwarded “to the appropriate carrier for investigation,” and the carrier must, “within such time as may be prescribed, advise the Commission in writing, with a copy to the complainant, of its satisfaction of the complaint or of its refusal or inability to do so.” Id. § 1.717. If the informal-complaint process proves ineffective, “the complainant may file a formal complaint,” and “[s]uch filing will be deemed to relate back to the filing date of the informal complaint” if, inter alia, it “[i]s filed within 6 months from the date of the carrier’s report”; but “[i]f no formal complaint is filed within the 6-month period, the complainant will be deemed to have abandoned the unsatisfied informal complaint.” Id. § 1.718. However, “[a]ny provision of the [FCC’s] rules may be waived by the Commission on its own motion or on petition if good cause therefor is shown.” Id. § 1.3.
In the fall of 2002, APCC filed an informal complaint with the FCC against NET. On the absolutely last day it could be timely, May 19, 2003, APCC unsuccessfully attempted to file a formal complaint. The filing was deficient in two respects: APCC submitted a single check (rather than a check for each of the two defendants in the formal complaint), and the filing fee prof[126]*126fered for each defendant was $5.00 short. APCC explained to the Enforcement Bureau “it submitted the wrong filing fee (and missed the six-month deadline under rule 1.718) because its counsel consulted only the hard-copy version of the Code of Federal Regulations (‘CFR’), dated October 1, 2002, which contained a filing fee amount — $165 per defendant — that had been superseded by the time APCC filed its formal complaint in May 2003.” APCC Services Inc., 20 F.C.C.R. 16727,16729, ¶ 6 (2005) (“Bureau Waiver Order ”).
About two weeks later, on June 3, 2003, APCC finally filed its formal complaint. The Enforcement Bureau accepted it, pursuant to the “good cause” exception to its rules, notwithstanding “the errors by APCC’s counsel [were] difficult to excuse, given that they were easily avoidable, and APCC’s law firm is highly experienced, resourceful, and knowledgeable in communications law....” Id. at 16732, ¶ 12. If the FCC had enforced the deadline, much of the Damages Order would be invalid.9
In affirming the Enforcement Bureau, the Commission considered it inappropriate to permit “a $5.00 fee error by APCC’s counsel — as negligent as it may have been — ” to stand in the way of fair compensation for PSPs, especially when the “formal complaint was otherwise submitted and served on time and in good faith, with advance notice to [NET].” Damages Order, 22 F.C.C.R. at 4297, ¶ 23. Thus, “under these specific circumstances, strict enforcement of [the] six-month relation-back deadline would unduly conflict with the public interest in ensuring the payment of compensation necessary to ‘promote the widespread deployment of payphone services to the benefit of the general public....’” Id. (quoting 47 U.S.C. § 276(b)(1)).
NET insists the FCC’s decision to allow the formal complaint to relate back was erroneous for two reasons. First, it claims the FCC unlawfully extended the two-year statute of limitations under Section 415(b) of the Act. In response, the FCC avers that it reasonably found that an informal complaint “constitutes a ‘complaint’ within the meaning of section 415.... ” Id. at 4294, ¶ 16. We need not resolve this specific issue because, as discussed below, we find NET’S alternate argument persuasive.
NET also argues that even if the FCC did not violate § 415(b), it nonetheless acted arbitrarily and capriciously in excusing APCC’s sloppiness, because under the adamantine standard set forth in the FCC’s Meredith/New Heritage Strategic Partners, L.P., 9 F.C.C.R. 6841, 6842-43, ¶¶ 6-9 (1994), deadlines can only be waived under “unusual or compelling circumstances” involving “a calamity of a widespread nature that even the best of planning could not have avoided, such as an earthquake or a city-wide power outage which brings transportation to a halt,” id. at 6842, ¶ 6. APCC cannot even begin to meet that standard. The FCC rejoins that Meredith only applies to “filing deadlines for pleadings that ‘initiate adjudicatory proceedings,’ ” which does not include formal complaints when an informal complaint has already been filed, and Meredith likewise only applies to late filings, not to pleadings that are timely offered but technically defective. Damages Order, 22 F.C.C.R. at 4298-99, ¶¶ 26-27 (quoting Meredith, 9 F.C.C.R. at 6843, ¶ 10).
Whether Meredith applies is not essential to our analysis; in any event, “the [127]*127Commission has implied that the Meredith standard might not materially differ from the [FCC’s] general waiver standard.” Id. at 4299, ¶ 26 n. 84. We have repeatedly “discourage[d] the Commission from entertaining late-filed pleadings ‘in the absence of extremely unusual circumstances.’ ” BDPCS, Inc. v. FCC, 351 F.3d 1177, 1184 (D.C.Cir.2003) (quoting 21st Century Telesis Joint Venture v. FCC, 318 F.3d 192, 200 (D.C.Cir.2003)). Consistent with this warning — which applies to any FCC decision to accept late pleadings, even in non-Meredith contexts — we hold the FCC’s failure to apply its six-month filing deadline was arbitrary and capricious. We do so reluctantly; given the deference we afford to an agency’s decision whether to waive one of its own procedural rules. See AT & T Corp. v. FCC, 448 F.3d 426, 431 (D.C.Cir.2006). But even deference has limits.
As we explained in Northeast Cellular Telephone Co. v. FCC, 897 F.2d 1164 (D.C.Cir.1990), before the FCC can invoke its good cause exception, it both “must explain why deviation better serves the public interest, and articulate the nature of the special circumstances to prevent discriminatory application and to put future parties on notice as to its operation,” id. at 1166. The reason for this two-part test flows from the principle “that an agency must adhere to its own rules and regulations,” and “[a]d hoc departures from those rules, even to achieve laudable aims, cannot be sanctioned, for therein lie the seeds of destruction of the orderliness and predictability which are the hallmarks of lawful administrative action.” Reuters Ltd. v. FCC, 781 F.2d 946, 950-51 (D.C.Cir.1986). This basic tenet is especially appropriate in the context of filings. When an agency imposes a strict deadline for filings, as the FCC has done, many meritorious claims are not considered; that is the nature of a strict deadline. The power to waive that strict deadline is substantial, because it allows an agency to decide which meritorious claims get considered. The inverse is true too — the power to waive allows an agency to decide which otherwise liable parties are off the hook.
The criteria used to make waiver determinations are essential. If they are opaque, the danger of arbitrariness (or worse) is increased. Complainants the agency “likes” can be excused, while “difficult” defendants can find themselves drawing the short straw. If discretion is not restrained by a test more stringent than “whatever is consistent with the public interest (by the way, as best determined by the agency),” then how to effectively ensure power is not abused? The “special circumstances” requirement is that additional restraint. Otherwise, we are left with “nothing more than a ‘we-know-it-when-we-see-it’ standard,” and “future [parties] — and this court — have no ability to evaluate the applicability and reasonableness of the Commission’s waiver policy.” Northeast Cellular, 897 F.2d at 1167.
We accept that the public interest is well-served by NET’s compensating PSPs, but that is not enough. There must also be a sufficiently “unique ... situation.” Id. at 1166. In Keller Communications, Inc. v. FCC, 130 F.3d 1073 (D.C.Cir.1997), waiver was permissible because there was a threat to public safety and the regulated party “expended] thousands of dollars of public funds in reliance on the agency’s mistaken grant of its license,” id. at 1076-77. We appreciate why that is a special circumstance. But procrastination plus the universal tendency for things to go wrong (Murphy’s Law) — at the worst possible moment (Finagle’s Corollary) — is not a “special circumstance,” as any junior high teacher can attest.
[128]*128We likewise are not convinced waiver was appropriate because NET received notice of the formal complaint prior to the deadline. After the informal complaint process has broken down, many defendants — probably most — are aware of the specific substance of a complainant’s grievance and whether a formal complaint will follow. Very few are that prejudiced when a filing occurs a day after a deadline (or a week, or a month, or maybe even a year), as opposed to the day of. Nonetheless, there is no indication the FCC’s practice is to accept those complaints; indeed, if it were, the six-month requirement would devolve into mere suggestion. The analytic difference between that common situation and this case is insufficient to satisfy the special circumstance requirement.
In so ruling, we of course do not cast doubt on the FCC’s ability to craft and apply exceptions to its procedural rules and filing deadlines; we merely hold that, under the applicable precedents and facts and circumstances of this case, the FCC’s decision to waive its filing deadline was arbitrary and capricious.
V.
We last address whether the FCC improperly ordered NET to pay interest at an annual rate of 11.25%. The FCC has previously determined that “11.25% is the appropriate cost of capital for payphone providers” because most payphones “are owned by large [LECs]” and the “authorized interstate rate of return” for LECs— 11.25% — appropriately reflects “a weighted average of debt and equity costs,” Implementation of the Pay Telephone Reclassification and Compensation Provisions of the Telecommunications Act of 1996, 13 F.C.C.R. 1778, 1806 ¶60 (1997) (Second Report and Order), even if a particular PSP is not an LEC.
NET contends, however, it only should have to pay the lower “IRS rate,” as the FCC recognized in a pair of 2002 payphone reconsideration orders. See Implementation of the Pay Telephone Reclassification and Compensation Provisions of the Telecommunications Act of 1996, 17 F.C.C.R.2020, 2032, ¶33 (2002) (Fourth Order on Reconsideration), (“Fourth Payphone Reconsideration Order”)-, Implementation of the Pay Telephone Reclassification and Compensation Provisions of the Telecommunications Act of 1996, 17 F.C.C.R. 21274, 21307-08, ¶¶ 99-101 (2002) (Fifth Order on Reconsideration). The FCC counters that the use of the IRS rate in those orders was justified by unusual circumstances. Because the FCC’s early attempts at setting a dial-around rate had been vacated by this court, “the Commission determined that PSPs had been under-compensated during one time period and over-compensated during another.” Resp’t’s Br. 41. Thus, the higher rate of 11.25% was deemed inappropriate in that narrow context, because it would not have accounted for the periods when PSPs were overcompensated.
Under the deferential arbitrary-and-capricious standard, the FCC adequately explained why it imposed the 11.25% interest rate instead of the IRS rate. There is a marked difference between “one-time ... ‘true up[ ]’ ” payments, Fourth Payphone Reconsideration Order, 17 F.C.C.R. at 2033, ¶ 33, where obligations were owed both ways, and the situation here with a financial duty owed only to the PSPs.10
[129]*129VI.
The FCC permissibly found liability and ordered interest at the rate of 11.25%. But its decision to waive for good cause APCC’s late filing was arbitrary and capricious. We therefore deny the petition as to the Liability Order, but grant in part the petition as to the Damages Order.
So ordered.