Opinion for the Court filed by Senior Circuit Judge STEPHEN F. WILLIAMS.
STEPHEN F. WILLIAMS, Senior Circuit Judge:For the third time, we must review the lawfulness of a Federal Aviation Administration regulation establishing fees for air traffic control services for “overflights”— flights by planes that go across some part of the United States but do not take off or land there. For the third time, we find that the FAA disregarded its statutory mandate.
In 1996 Congress directed the FAA to set and collect fees for the provision of air traffic control services for “overflights.” 49 U.S.C. § 45301(a); see also § 273 of the Federal Aviation Reauthorization Act, Pub.L. 104-264. “Services for which costs may be recovered include the costs of air traffic control, navigation, weather services, training and emergency services ... and other services provided by the [FAA] or by programs financed by the [FAA].” Id. § 45301(b)(1)(B). In the form originally enacted (and applicable when the agency promulgated the rules at issue here), it said that the fees must be “directly related to the Administration’s costs of providing the service rendered.” Id.
The FAA published its first interim final rule establishing fees in 1997, but we found that its methodology did not yield fees “directly related to ... the costs.” Asiana Airlines v. FAA, 134 F.3d 393, 402 (D.C.Cir.1998). In 2000 it promulgated a new interim final rule, Fees for FAA Services for Certain Flights, 65 Fed.Reg. 36002, with a new cost methodology. It calculated its total costs for providing service to all aircraft (overflights and non-overflights), took out a class of “overhead” charges, and then divided that figure by the total number of miles flown within the United States to arrive at a per-mile rate. The rule also excluded costs incurred by aircraft while they were on the ground (the so-called “Surface” airspace) or within forty miles of an airport (the so-called “Terminal” airspace); it included costs incurred for all other U.S.-controlled airspace (known as the “Enroute” and “Oceanic” airspaces). For the included airspaces it relied on an assumption that the FAA incurred the same per-mile costs in servicing overflights and non-overflights. Since the agency provided no support for this necessary assumption, we remanded the case to the agency to explain and support its rule. Air Transport Ass’n of Canada v. FAA, 254 F.3d 271 (D.C.Cir.), modified 276 F.3d 599 (D.C.Cir.2001).
In August 2001 the FAA re-promulgated the overflight rule in largely the same form as before. Fees for FAA Services for Certain Flights, 66 Fed.Reg. 43,680 (2001) (the “Final'Rule”). Most importantly for this case, the agency retained the assumption that it incurred virtually identical per-mile costs in servicing overflights and non-overflights in the so-called Enroute and Oceanic airspaces. During the notice and comment period, petitioners reiterated their prior objections that labor costs incurred in providing traffic control services for overflights were substantially lower than for non-overflights, and submitted supporting declarations by two air traffic control experts. These experts most crucially contended that (1) traffic controller costs are not “fixed,” because the FAA varies the number of controllers on duty “depending on the volume of aircraft operating within the particular geographical area or sector”; (2) flights in the “High-Altitude” sector (18,000 feet and above) *336require far less controller attention per mile than flights in the “Low-Altitude” sector; (3) overflights occur almost exclusively in the High-Altitude sector; and (4) by virtue of these differences and FAA practices, there are means to allocate controller time as between overflights and non-overflights far more precisely than in the agency’s per-mile calculation. Supplemental Declaration of Joseph A. Beaudoin, Joint Appendix (“J.A.”) 251-53.
In response, the agency provided four arguments:
(1) The agency incurs the “vast majority of costs by making its comprehensive [air traffic control] system available to all flights (regardless of the type of aircraft ...);”
(2) “[T]he FAA’s marginal cost, including labor cost, for providing services to any flight is close to zero”;
(3) “[T]he majority of FAA’s costs are common and fixed costs”; and
(4) “[T]he controllers’ responsibilities for Overflights are not fundamentally any different than for non-Overflights.”
Final Rule, 66 Fed.Reg. at 43,685/1. In support of these positions, the agency cited a new report by the economic consulting firm of Capital Economics; but the report failed to confront the challengers’ claims with evidence more specific, or with an analysis more compelling, than those of the agency’s comments cited above.
On October 11, 2001 the petitioners filed in this court for review of the Final Rule. On November 19, 2001 Congress adopted a statute with language amending § 45301’s standards for computing the overflight fee and limiting jurisdiction for judicial review. See § 119(d) of the Aviation and Transportation Security Act of 2001 (the “2001 Act”), Pub.L. 107-71. We save discussion of those provisions for later, but note in the meantime that the 2001 Act also included a savings clause, § 141(d), saying,
This Act shall not affect suits commenced before the date of the enactment of this Act [with certain irrelevant exceptions]. In all such suits, proceedings shall be had, appeals taken, and judgments rendered in the same manner and with the same effect as if this Act had not been enacted.
Pub.L. 107-71, § 141(d).
As a result of the 2001 Act, FAA requested and received from this court a remand in this case to determine the extent to which the statutory amendments might require a modification of the Final Rule. The FAA received public comments but then terminated the remand proceeding in June 2002, deciding that § 119 “merely clarifies and amplifies congressional intent so as to provide further validation” of the Final Rule, so that no new rulemaking was required. 67 Fed.Reg. 42462, 42464/1 (2002). Thus, the appeal resumed.
The FAA argues at the outset that § 119 of the 2001 Act deprives us of jurisdiction over the key issue of this case — the FAA’s cost-allocation judgment. Section 119(d)(3) modifies § 45301(b)(1)(B) by adding at the end the following sentence: “The Determination of such costs by the Administrator is not subject to judicial review.” Pub.L. 107-71, § 119(d)(3).
We pretermit the parties’ competing interpretations of just what “cost” determinations § 119(d)(3) screens from judicial review. The language of the savings clause quoted above clearly and explicitly bars the application of § 119(d)(3) to this suit — which indubitably “commenced before the date of enactment” of § 119. We note that the savings provision contains two explicit exceptions, but it is plain — and the FAA does not argue to the contrary— that neither has any relation to this case.
*337Thus the FAA is reduced to an exceptionally lame reliance on legislative history, pointing to language in the conference report to the effect that § 119 was intended “to clarify longstanding Congressional intent that the FAA expeditiously and continuously collect the fees authorized under section 45301(a) of title 49.” H. R. Conf. Rep. No. 107-296, at 72 (2001) (emphasis supplied by respondents). It argues that if we apply the savings clause as written this goal cannot be effected. But the FAA identifies no ambiguity in § 141(d) that this committee reference might be thought to resolve; ordinarily, we do not read legislative history to create otherwise non-existent ambiguities. See, e.g., Ratzlaf v. United States, 510 U.S. 135, 147-48, 114 S.Ct. 655, 662-63, 126 L.Ed.2d 615 (1994). To be sure, we have held that “legislative history may ‘shed new light on congressional intent, notwithstanding statutory language that appears superficially clear,’ ” U.S. Telecom. Ass’n v. F.B.I., 276 F.3d 620, 625 (D.C.Cir.2002) (quoting Natural Resources Defense Council, Inc. v. Browner, 57 F.3d 1122, 1127 (D.C.Cir.1995)), but nothing in this conference report even suggests (much less shows) that whoever drafted this snippet of legislative history had ever intended § 119(d) to contain some sort of implicit exception to § 141(d). Rather, the history seems designed only to ensure that the modifications of § 119(d) — added only at conference — did not affirmatively engender delay by seeming to create a substantively different standard for the agency. See H. R. Conf. Rep. No. 107-296, at 72 (explaining that conference amendment “is not intended to require revision of the fees recently promulgated by the FAA”).
Thus, we reach the merits. We have summarized above the assertions of petitioners’ experts, which make a substantial case refuting the agency’s unexplicated insistence that miles of overflights and non-overflights in the Enroute and Oceanic airspaces are approximately equivalent in their per-mile generation of costs. In response the agency offers what amount to little more than conclusory denials.
Although consideration of the agency’s specific arguments may seem like flogging a dead horse, we proceed with the exercise so that it may be assured that we have scrutinized its arguments. It said in the rulemaking, for example, that the controllers’ responsibilities in servicing overflights are not “fundamentally any different” from those involved in servicing non-overflights. Final Rule, 66 Fed.Reg. at 43,685/1. But there was never any contention by petitioners that the work differed in some inherent quality. The fact that the type of labor does not vary with flight altitude tells us little about the quantity of labor required to service the different altitude sectors. Petitioners’ experts say without contradiction not only that more air traffic controllers are needed on a per-mile basis to monitor and communicate with low-altitude aircraft than with high-altitude aircraft, but that it would be simple for the agency to calculate how many more, because the FAA treats high- and low-altitude control as different shifts. Beaudoin Declaration, J.A. 251-52. Each individual controller may service both altitude sectors over the course of his career, and maybe even over the course of the day, but no controller services both types of flights at the same time. Or so say petitioners’ experts, without contradiction. Thus, assuming the truth of petitioners’ other largely uncontradicted claim — that overflights are overwhelmingly at altitudes above 18,000 feet — the agency need only count the number of shifts devoted to high-altitude flights to calculate the range within which overflights are generating costs.
*338The agency observes that some overflights spend some time at low altitudes. But this does not undermine petitioners’ claim that they fly predominately in the high altitude sector, to a degree far greater than do non-overflights. If the agency has reason to believe that overflights occur non-trivially in the low-altitude range, obviously it can make adjustments to reflect that proposition.
The agency also insists that the benefits of the air traffic control system are somehow indivisible, so that throughout an aircraft’s time in U.S. airspace it benefits from “the entire [air traffic control] infrastructure and full scope of services ..., regardless of the type of flight, user, or aircraft.” Appellee’s Br. at 44-45. We may assume that planes at low altitudes benefit from services that keep high-altitude planes on course, and even the converse. But that in no way contradicts petitioners’ evidence that the costs of keeping planes on course vary markedly as between the high and low altitudes.
Nor does the agency fare better with its conclusions that “the FAA’s marginal cost, including labor cost, for providing services to any flight is close to zero,” and, in what amounts to the same thing, “the majority of FAA’s costs are common and fixed.” Final Rule, 66 Fed.Reg. at 43,685/1. These conclusions are based largely upon one section of the Capital Economics report, in which it explains that air traffic controllers are paid the same amount no matter whether they service overflights or non-overflights; all controllers receive the same training; radar and navigation equipment are common and fixed costs; and telecommunications capabilities are common and fixed costs. Final Rule, 66 Fed. Reg. at 43,685/2-3. Conspicuously absent from this list of considerations is any analysis of the quantities of labor consumed by the two types of flights and the divisibility of the resultant labor costs — i.e., the factors that petitioners addressed in their comments.
The agency attempts to remedy this deficiency by arguing that because the FAA “has a set number of controllers to provide ... services nationwide ..., [and t]hese numbers do not change daily to manage an additional Overflight, or a non-Overflight,” Final Rule, 66 Fed.Reg. at 43,709/3, fixed and common costs dominate; thus any difference in the marginal cost of servicing an additional overflight as opposed to an additional non-overflight is immaterial, id. at 43,709/3-710/1. But the lumpiness of costs, which is what the FAA points to here, is quite separate from the notion that costs do not systematically diverge as between the two services. Petitioners are not advocating that the FAA should somehow compute either the incremental or the marginal cost of servicing each overflight. On that subject, the FAA’s observation that “the metering costs of identifying ... differences in marginal costs would be substantial,” id. at 43,709/3, is both pertinent and uncontradicted by petitioners. But it has no bearing on petitioners’ actual claim: that overflights occur almost exclusively in the high-altitude range, for which the FAA makes assignments of controllers separate from assignments for the low-altitude, and that per-mile servicing costs are systematically lower in the high-altitude range.
The agency finally turns to the substantive provisions of the 2001 Act. Where the statute previously required that the fees be “directly related to the ... costs,” 49 U.S.C. § 45301(b)(1)(B), the 2001 Act changed it to read “reasonably related to the ... costs.” 2001 Act, § 119(d)(1). Of course as we have noted the Act’s savings clause prevents it from having any direct effect in this case, but the agency resourcefully suggests that even then the new language merely “clarifies” the old, pointing to the legislative history (quoted above) that in fact suggests such a pur*339pose. Thus, in the FAA’s view, new language that merely “clarifies” the old, and which Congress explicitly provided should not apply to a suit commenced when this one was, in fact should apply in the guise of shedding light on the true meaning of the old language. A paradox, a most ingenious paradox. But we can reject this theory without entering the thicket of using subsequent congressional pronouncements to infer the meaning of earlier provisions. Cf. Consumer Product Safety Comm’n v. GTE Sylvania, 447 U.S. 102, 118 n. 13, 100 S.Ct. 2051, 2061 n. 13, 64 L.Ed.2d 766 (1980). The savings clause, § 141(d)(1) of the 2001 Act, states that “judgments are to be rendered ... with the same effect as if this Act had not been enacted.” Using the Act to give the word “directly” a meaning different from our prior understanding, whether we call the process clarification or modification, would violate that command.
Finding the Final Rule “not in accordance with law,” 5 U.S.C. § 706(2)(A), we set it aside and order the case remanded, id. § 706(2).
So ordered.