STEPHEN F. WILLIAMS, Circuit Judge:

In its 1977 amendments to the Clean Air Act, Congress ratified a previously established program for the "prevention of significant deterioration of air quality." See § 127, Pub.L. No. 95-95, 91 Stat. 685, 731-42, adding to the Act §§ 160-69, codified at 42 U.S.C. §§ 7470-79 (1982). The stated purpose of these "PSD" provisions was (roughly) to protect the air quality in national parks and similar areas of special scenic or recreational value, and in areas where pollution was within the national ambient standards, while assuring economic growth consistent with such protection. Clean Air Act § 160, 42 U.S.C. § 7470. For two pollutants, sulfur dioxide and particulate matter ("Set I pollutants"), Congress followed the Environmental Protection Agency's earlier approach, fixing a maximum allowable increase (an "ambient air quality increment") over baseline concentrations. See § 163, 42 U.S.C. § 7473; compare prior rules, Prevention of Significant Air Quality Deterioration, 39 Fed.Reg. 42,510, 42,514-17 (1974). For several other pollutants ("Set II pollutants"), including nitrogen oxides, Congress took no immediate action. Instead it provided the EPA with general guidance as to regulations that it was to promulgate within two years. § 166, 42 U.S.C. § 7476. We deal here with the regulations governing nitrogen oxides, the only Set II pollutant that EPA has yet regulated. The case turns on the meaning, and above all the interrelationship, of the two main guides, § 166(c) and (d):

(c) Such regulations shall provide specific numerical measures against which permit applications may be evaluated, a framework for stimulating improved control technology, protection of air quality values, and fulfill the goals and purposes set forth in section 7401 of this title [statement of purposes of Clean Air Act] and section 7470 of this title [statement of purposes of the PSD provisions].

(d) The regulations . . . shall provide specific measures at least as effective as the increments established in section 7473 of this title [the Set I rules] to fulfill such goals and purposes, and may contain air quality increments, emission density requirements, or other measures.

§ 166(c), (d), 42 U.S.C. § 7476(c), (d). Because the EPA did not adequately consider the requirements of subsection (c), we reverse.

I. An Overview of the Regulatory Scheme

Although in 1980 the EPA noted ten possible strategies for Set II, see Prevention of Significant Deterioration for Hydrocarbons, Carbon Monoxide, Nitrogen Oxides, Ozone, and Lead (PSD Set II) (Advanced Notice of Proposed Rulemaking), 45 Fed.Reg. 30,088 (1980) ("Advanced Notice"), it ultimately decided to mimic Congress's approach for Set I. The first step, not here controverted, was the use of the same three-tiered scheme for classifying protected areas: Class I — comprising mainly large national parks and national wilderness areas; Class II — regions where the ambient air quality levels more than meet the national standards; and Class III — regions meeting the definition of Class I or Class II areas but redesignated at the behest of a state for higher levels of industrial development. §§ 162, 164, 42 U.S.C. §§ 7472, 7474; compare Prevention of Significant Deterioration for Nitrogen Oxides ("Proposed Rules"), 53 Fed.Reg. 3698, 3699/2-3 (1988) (noting statutory classifications and adopting them for Set II). (According to the EPA, no Class III areas have been established to date. Id.; Brief for Respondent at 11.)

EPA's second step, again not controverted here, was its decision to base the nitrogen oxide PSD program on "ambient air quality increments" similar to those for the Set I pollutants. Though Congress contemplated that EPA might use increments for the Set II program, it did not require their use. See § 166(d), 42 U.S.C. § 7476(d). EPA tentatively rejected the alternatives to increments that it had considered in its 1980 Advanced Notice, but left the door open for a state to adopt an alternative strategy if it could show that its choice was as effective as EPA's. See Proposed Rules, 53 Fed.Reg. at 3709/2.

EPA's third step was to formulate the permissible increments "by reference to" the National Ambient Air Quality Standards ("NAAQS" or "ambient standards") established under § 109 of the Clean Air Act, 42 U.S.C. § 7409. In so doing, it rested on its view that Congress had used the ambient standards "as the benchmark for determining what constitutes `significant deterioration'" for Set I pollutants and that the ambient standards were "the basic measure of air quality under the Act." See Proposed Rules, 53 Fed.Reg. at 3700/3. The three challenged aspects of the nitrogen oxide PSD program stem directly from this decision.

First, though by its terms § 166 demands that the regulations cover "nitrogen oxides," the EPA regulated only one nitrogen oxide compound; nitrogen dioxide or NO[2], as this is the only compound for which it had established an ambient standard. Second, as its ambient standard for NO[2] imposes direct limits only for the annual average concentration, EPA defined the permissible increments only in terms of an annual average. Final Rules, 53 Fed.Reg. at 40,660/3. Short-term concentrations, which are only indirectly and incompletely limited by an annual average, may have adverse health and welfare effects (see, e.g., EDF Comments at 18 (noting high concentration "acid pulse" of nitric acid in springtime can cause ecological damage)), but EPA at present appears to believe the evidence too uncertain to justify including a short-term limit among the ambient standards for NO[2]. See Retention of National Ambient Air Quality Standards for Nitrogen Dioxide, 50 Fed.Reg. 25,532, 25,536/3 (1985). Third, and most sharply contested, the EPA set the permissible increments of nitrogen dioxide at the same percentage of the nitrogen dioxide ambient standard as the percentage that the Set I increments were of their annual ambient standards (at the lower of the two percentages when they varied). The table below sets forth the lower of the primary and secondary national ambient air quality standards for Set I and II pollutants at the time the increments for each were created (date in parentheses), and the increments for the pollutants both as absolute concentrations and as percentages of the ambient standards (in parentheses). Both standards and increments are expressed in micrograms per cubic meter. Ambient Standards: Class I Increment Class II Increment Class III Increment

II. The Interrelation of Subsections (c) and (d)

First, there are at least three possible meanings to subsection (d)'s "at least as effective" test. The EPA, with some hesitation, appears to have read it as requiring that the Set II rules be at least as stringent as those for Set I, i.e., that increments be set no lower, as percentages of a pollutant's ambient standards, than the Set I increments were as percentages of their respective ambient standards. See Proposed Rules, 53 Fed.Reg. at 3701/1. (EDF appears at times to accept this basic formulation. EDF Comments at 19.) Proposed Rules, 53 Fed.Reg. at 3700/3. A second possible meaning is that the Set II rules must be at least as protective of the various environmental values threatened as the Set I increments were of the environmental values threatened by those pollutants. See id. at 3701/3 (asking whether statute should be read as calling for assessment of severity of effects of the pollutants). (EDF supports this reading, but realizes it is subsumed under its argument based on § 166(c). Brief for Petitioner EDF at 45.) Finally, there is a literal interpretation, for the most part overlooked by both parties, under which the Set II rules would have to achieve as optimal a balancing of the environmental and economic growth values made relevant by § 160 as Congress achieved for the Set I pollutants. The agency recognized that the PSD program necessarily involved a balancing of goals, see Proposed Rules, 53 Fed.Reg. at 3699/1, and to a degree assessed its NO2 increments in terms of effectiveness in "avoiding unreasonable economic impacts" on adversely affected industries, see id. at 3704/1-2, but it did not attempt to use optimality as an ultimate measuring stick.

The more faithful any of these views is to the literal meaning of the "goals and purposes" with respect to which effectiveness is to be measured, the less workable it is. Thus, while the "optimality" standard appears best to reflect the wording, its practical difficulties are breathtaking. First, it would entail gathering data about not only the Set II pollutants but also those of Set I. On even so narrow an issue as acid rain effects, the EPA made a comparison, see Final Rules, 53 Fed.Reg. at 40,661/1-2, but concluded that a more comprehensive comparison was impossible given the current data, id. at 40,661/2. Second, there are two Set I pollutants, and the optimality of the congressionally adopted balance for the two may differ. Third, it is hard even to imagine a conceptual apparatus for evaluating optimality. Would the EPA identify the ratio of marginal tradeoff between environmental and "growth" values for the Set I pollutants at which those increments maximized "utility" (e.g., one util of environment equals two utils of growth, thus giving the environment a two-to-one predominance), and then replicate it for the Set II pollutants? And how would the two different types of utility be measured? The "protectiveness" reading of subsection (d) escapes the extreme intractability of the optimality reading, but it accomplishes this only by slighting the "economic growth" goal of § 160. Moreover, it is also far from readily workable: how can degrees of protectiveness be established for two pollutants that inflict their primary harms on different environmental values (e.g., one most injures visibility, the other vegetation)? The "stringency" reading is the most workable, but at the price of using ambient concentration percentages as a proxy for all the relevant goals. Proposed Rules, 53 Fed.Reg. at 3701/2.

Nonetheless, the simple "stringency" reading of § 166(d) is well supported by the legislative history of the 1977 Amendments. The House bill made no distinction between Set I and Set II pollutants in defining increments; it set them all as simple percentages of the ambient standards for each pollutant covered by ambient standards (2% for Class I areas, 25% for Class II, 50% for Class III). See H.R. 6161, 95th Cong., 1st Sess. § 108(a), at 294-95 (April 6, 1977), reprinted in 4 Legislative History at 2281-82. But for the Set II pollutants, the House bill would have allowed the states to supplant the proposed statutory increments with a program of their own design so long as the Administrator determined that it "carr[ied] out the purposes [of the PSD program] at least as effectively as" the statutory program. See id. § 108(a), at 308, reprinted in 4 Legislative History at 2293. (This language survives in § 166(e), 42 U.S.C. § 7476(e).)

In that context, it seems plain that the stringency concept would be both workable and completely faithful to a broad vision of the relevant goals and purposes. State increments for a Set II pollutant set at exactly the same percentages of ambient standards as were the House-proposed statutory ones would surely be at least as effective in fulfilling the goals and purposes of the program.

Compromise with the Senate, however, altered the context of the phrase. See H.R.Conf.Rep. No. 95-564, 95th Cong., 1st Sess. (1977), U.S. Code Cong. Admin. News 1977, p. 1077, reprinted in 3 Legislative History at 531 (tersely describing compromise). The Senate bill had merely commanded the EPA to study the Set II pollutants and report back to Congress with proposed increments. S. 252, 95th Cong., 1st Sess. § 6, at 20-21 (May 10, 1977), reprinted in 3 Legislative History at 594-95. In the ensuing compromise, the House gave up the statutory fixing of Set II increments, and accepted a statute assigning to the EPA a discretion somewhat paralleling that which the House bill had extended to the states. The Senate gave up its requirement of an EPA report back to Congress, and instead allowed the EPA itself to promulgate the program.

The wrenching of the "at least as effective" formula from its moorings in the House bill created the range of choices that we have reviewed. The practical problems of the "protective" and "optimal" readings arise from the need to compare the "effectiveness" of regulations across pollutants. When we consider both the practical problems and the legislative history, we believe EPA's choice of the percentage stringency test permissible for purposes of subsection (d). Not surprisingly, the nitrogen dioxide increments now under review are almost identical to the increments that would have become law under the House bill.

As a reading of § 166 as a whole, however, EPA's approach overlooks both the language of subsection (c) and the vector of forces represented by the Senate bill. The Senate originally wanted more study conducted on the Set II pollutants, with Congress to make the final choice. Subsection (c) appears to manifest much of this intention, merely substituting the EPA for Congress as decisionmaker. Thus, except to the extent that EPA may properly read subsection (d) as limiting subsection (c), a failure to assess a pollutant in terms of the PSD goals breaches the agency's duty to consider all the relevant statutory factors. See Motor Vehicle Manufacturers Ass'n of U.S. v. State Farm Mutual Auto. Ins. Co., 463 U.S. 29, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983); Specialty Equipment Market Ass'n v. Ruckelshaus, 720 F.2d 124, 132 (D.C. Cir. 1983). The EPA candidly admits it did not make that inquiry. See, e.g., Proposed Rules, 53 Fed.Reg. at 3701/2.

Now we must turn to some possible relations between the two subsections, for the moment identifying subsection (d) with EPA's stringency interpretation. First, one might, as EPA evidently did, read the two as providing the agency an absolute safe harbor: if the regulations mirror the Set I restrictions on a percentage basis, the EPA is free to disregard whatever the inquiry under subsection (c) might yield. The difficulty with this reading, seemingly incurable, is that it turns subsection (c) into no more than an EPA option, despite mandatory wording.

Second, while subsection (c) commands a broad weighing of factors, subsection (d) might operate as a limited safe harbor (a presumptive baseline) — authorizing regulations that follow the House approach if but only if the Administrator determines (without being arbitrary and capricious) that the criteria under subsection (c) do not call for a more, or a less, stringent standard. EDF appeared at oral argument to endorse this view, and, while we by no means regard the position as any kind of binding concession, we do find it at least superficially conformable to the statutory language.

If we relax the assumption that subsection (d) imposes a simple stringency test, further alternatives open up. For example, if EPA should read subsection (d)'s "at least as effective" test in terms of optimality, it could function as a refinement of subsection (c)'s mandate to fulfill the specified "goals and purposes," telling EPA how to balance the competing environmental and economic goals. At one time EPA appears to have contemplated such a melding of the two sections, though recognizing some of its formidable difficulties.

While we read the ambiguities and perplexities of the statute as delegating to the agency a broad interpretive authority, as we must under Chevron U.S.A. Inc. v. NRDC, 467 U.S. 837, 843-44, 104 S.Ct. 2778, 2781-82, 81 L.Ed.2d 694 (1984), we cannot affirm the agency's action on the current record. The reading that we have hypothesized of § 166(d) as a contingent safe harbor requires the agency first to adopt that view, then to determine that the inquiry under subsection (c) does not require a more stringent standard. It has done neither. We cannot sustain an action merely on the basis of interpretive theories that the agency might have adopted and findings that (perhaps) it might have made. See SEC v. Chenery Corp., 318 U.S. 80, 92-95, 63 S.Ct. 454, 461-63, 87 L.Ed. 626 (1943); Wisconsin's Environmental Decade, Inc. v. SEC, 882 F.2d 523, 528 (D.C. Cir. 1989); cf. Phillips Petroleum Co. v. FERC, 792 F.2d 1165, 1172 (D.C. Cir. 1986).

III. EPA's Use of the National Ambient Standards

EDF attacks not only EPA's conclusory decision to allow § 166(d) to supplant subsection (c), but also its choice of the ambient standards under sections 108 and 109 of the Clean Air Act, 42 U.S.C. §§ 7408-09, as the sole basis for deriving permissible increments. That choice resulted in EPA's defining increments for only one compound of nitrogen oxides (NO[2]), and defining them only in terms of annual averages. See page 6 above. Here the problem is that the "goals and purposes" of the PSD program, set forth in § 160, are not identical to the criteria on which the ambient standards are based, §§ 108(a) and 109(b), 42 U.S.C. §§ 7408(a), 7409(b). See also § 101, 42 U.S.C. § 7401 (overall goals of the Act). Although the statutory guidelines for the ambient standards seem to encompass everything imaginable, e.g., "the public welfare," see § 109(b)(2), § 160 emphasizes special considerations, such as national wilderness areas and their "natural, recreational, scenic, or historic value[s]." Thus a pollutant that has only mild public health effects but severe effects on wilderness areas might demand a lower increment (measured as a percentage of its ambient standards) than one with severe health effects but only mild effects on wilderness areas. Moreover, as EDF points out, the 1977 amendments had directed EPA to issue its Set II PSD limits within two years and to review and revise its § 108 criteria by January 1, 1980. Had EPA kept to the deadlines (which it didn't) and piggybacked the PSD increments on the ambient standards (which it did), the increments would have been at risk of being rendered obsolete almost immediately after promulgation. Nevertheless the ambient standards are the "basic measure of air quality under the [Clean Air Act]," Proposed Rules, 53 Fed.Reg. at 3700/3, and the controlling standards by no means exclude any value that is the subject of focus under the PSD provisions.

Our analysis of this challenge differs for the two disputed subsections of § 166. For subsection (d), we have already held that EPA's use of the Set I percentages of ambient standards was permissible. Thus, EDF cannot prevail on the argument that the NO[2] increments are not "as effective as" (or "as stringent as") the sulfur dioxide increments just because the NO[2] increments do not include short-term increments while the sulfur dioxide increments do. Subsection (c), however, commands the Administrator to inquire into a pollutant's relation to the goals and purposes of the statute, and we find nothing in the language or legislative history suggesting that this duty could be satisfied simply by referencing the ambient standards. EDF may still, therefore, make the argument on remand that under subsection (c) short-term increments or increments for other nitrogen oxide compounds are needed to "protect air quality values, and fulfill the goals and purposes" of the statute. Obviously EPA's resolution of the issue will depend on the interpretation it selects for harmonizing subsections (c) and (d). See supra pp. 188-190.

IV. Remedy

No party to this litigation asks that the court vacate the EPA's regulations, and to do so would at least temporarily defeat petitioner's purpose, the enhanced protection of the environmental values covered by the PSD provisions. Moreover, there is no way by which the court can determine what effect the EPA's reasoned exercise of its authority may have. Accordingly we remand the case for the EPA to develop an interpretation of § 166 that considers both subsections (c) and (d), and if necessary to take new evidence and modify the regulations. See Ford Motor Co. v. NLRB, 305 U.S. 364, 373, 59 S.Ct. 301, 306, 83 L.Ed. 221 (1939). While EDF asks that the remand include imposition of an immediate deadline of two years, based on the original deadline created by statute in 1977, cf. Sierra Club v. Thomas, 658 F. Supp. at 175, we refrain from any current ruling on this request.

So ordered.