Opinion for the Court filed by Circuit Judge STEPHEN F. WILLIAMS.
Concurring opinion filed by Circuit Judge RUTH BADER GINSBURG.
STEPHEN F. WILLIAMS, Circuit Judge:
When heated water sits in a water heater in a cooler environment, it loses heat, i.e., energy, thus exemplifying the second law of thermodynamics. Under authority granted by the Energy Conservation Standards for
[1043]*1043New Buildings Act of 1976 (“Buildings Act”), 42 U.S.C. §§ 6831-40 (1988), the Department of Energy promulgated “standby loss” rules to limit these losses in water heaters installed in new federal construction projects. The Gas Appliance Manufacturers Association (“GAMA”) and five of its members challenge the Department’s 1990 version of the rule, claiming its requirements are not in accord with the statutory authority and are arbitrary and capricious. Applying, a concept developed from different statutory language in the context of occupational health regulation, the district court held that the rule conformed to the statutory mandate because it did not “threaten massive dislocation to, or imperil the existence of, the industry”. Gas Appliance Manufacturers Ass’n v. Department of Energy, 773 F.Supp. 461, 466 (D.D.C.1991) (“GAMA II”) (internal quotations and citations omitted). Because the district court arrived at an incorrect statutory interpretation and DOE failed to apply the correct interpretation adequately, we reverse and remand to the district court with instructions to remand to DOE.
Background
The Buildings Act requires the Secretary' of Energy to develop interim energy efficiency standards that are to be binding for new federal government construction projects but merely guidelines for others (though the-others include 'state and local regulators). See 42 U.S.C. §§ 6833(a), 6835, 6839. DOE is to assure that the standards “are adequately analyzed in terms of energy efficiency, stimulation of use of nondepletable sources of energy, institutional resources, habitability, economic cost and benefit, and impact upon affected groups.” 42 U.S.C. § 6839.
In 1987 DOE proposed revisions to its standby loss standards for commercial storage 1 water heaters, cutting the maximum allowable standby loss to 1.9 W/sq ft (watts per square foot of tank surface area) for electric water heaters and to 1.3- + 38/V percent per hour for oil and gas water heaters.2 See 52 Fed.Reg. 17,175 (1987). In addition, DOE proposed to restrict the definition of non-storage water heaters so that some heaters formerly considered non-storage would become subject to the standards. These revisions followed a proposal by the American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (“ASH-RAE”) to make its pre-existing standard, ASHRAE Standard 90.1-1980, more stringent. See ASHRAE Standard 90.1-1989 (1989). (The ASHRAE standards are developed by industry experts; the nature of their enforcement is not revealed by this record.) Although GAMA argued that the stricter 1987 proposal could not satisfy cost/benefit standards, DOE adopted it as final in 1989.
GAMA then filed the first of two suits against DOE in the United States district court. In Gas Appliance Manufacturers Ass’n, Inc. v. Secretary of Energy, 722 F.Supp. 792, 798 (D.D.C.1989) (“GAMA I”), the court set the standards aside, ruling that neither ASHRAE nor DOE had considered the Buildings Act’s requirement that the regulations be “ ‘practicable’ and justifiable in terms of economic cost and benefit and impact on affected groups.” Id. at 797 (quoting 42 U.S.C. § 6831(b)(2) (1988) and drawing on § 6839). It enjoined enforcement' of the rules and remanded the case for DOE to develop a statement of reasons, to allow for comment by interested parties, and then to issue its final rule and response to comments. Id. at 798.
On remand, DOE published a “Preliminary Statement of Reasons for Adoption of Standby Loss Criteria” (“Preliminary Statement”), 54 Fed.Reg. 49,724 (1989), as corrected, 54 Fed.Reg. 50,341 (1989), which attempted to defend the standards enjoined. GAMA criticized the Preliminary Statement as inadequate and proposed alternatives. Although DOE rejected appellants’ proposal and kept the initial standard for oil and gas heaters, it [1044]*1044relaxed the requirements for electric heaters from 1.9 W/sq ft to 3.0 W/sq ft. See 55 Fed.Reg. 23,842 (1990) (“Revised Rule”).
GAMA responded with the present lawsuit, GAMA II. Rather than reviewing the decision in light of the statutory reference to economic ’ costs and benefits, the court applied the two-step standard of United Steelworkers of America, AFL-CIO-CLC v. Marshall, 647 F.2d 1189, 1272 (D.C.Cir.1980), determining first that the standards were technologically feasible, and second that, because the federal water heater market was so small, the standard could not adversely affect a substantial 'part of the industry. GAMA II, 773 F.Supp. at 465-66. The court upheld the rule as within DOE’s statutory authority and adequately explained. GAMA appealed.
During the pendency of the appeal, two relevant events occurred. First, ASHRAE agreed, in settlement of a case brought by GAMA, Gas Appliance Manufacturers Ass’n, Inc. v. American Soc’y of Heating, Refrigerating and Air-Conditioning Eng’rs, Inc., No. 89 Civ. 3319 (D.D.C. Dec. 11, 1989), to revise ASHRAE Standard 90.1-1989, adopting a new and relaxed standby loss standard for all commercial water heaters, ASHRAE Standard 90.1b-1992 (1992). Second, Congress adopted the new ASHRAE standard for all such equipment manufactured after January 1, 1994. Energy Policy Act of 1992, Pub.L. No. 102-486, § 122(d), 106 Stat. 2776, 2812 (1992) (42 U.S.C.A. § 6313(a)(5) (West Supp.1993)). For future federal construction projects, the 1992 Act directed DOE, within two years after enactment (on October 24, 1992), to adopt standards that “meet or exceed” those of ASHRAE Standard 90.1-1989, the new standards to take effect no less than a year after issuance. Id., § 101, 106 Stat. 2784-85 (1992) (42 U.S.C.A. § 6834(a)(1)-(2) (West Supp.1993)). The statute leaves the prior rules undisturbed until they are supplanted by the new ones (presumably about October 1995). 106 Stat. 2785 (42 U.S.C.A. § 6834(d)). Thus the issue is not moot. Statutory Interpretation
The substantive criteria expressed in Steelworkers emerged as a construction of § 6(b)(5) of the Occupational Safety and Health Act, 29 U.S.C. § 655(b)(5), and are inapplicable to the standards to be issued under the Buildings Act. Whereas § 6(b)(5) concerns protection of human health and involves risks difficult to quantify, the Buildings Act concerns protection of energy resources and consumers, involves costs and benefits that are relatively easy to quantify, and explicitly asks that they be computed. Thus, while the OSH Act calls for standards stringent enough to safeguard workers from health risks in the workplace to the “extent feasible”, id., the Buildings Act instead requires that standards be set to “achieve the maximum practicable improvements in energy efficiency”, 42 U.S.C. § 6831 (emphasis added).
Both explicitly and implicitly, Congress made clear that the practicability of rules promulgated under the Buildings Act entails cost-benefit analysis. Congress declared explicitly that such standards must be “adequately analyzed in terms of ... economic cost and benefit, and impact upon affected groups.” 42 U.S.C. § 6839. The' Act’s preamble emphasizes economic efficiency and makes clear that Congress was concerned with the economic impact of energy waste on consumers’ pocketbooks — and on those of taxpayers as the ultimate suppliers of funds loaned by the government. “[T]he failure to provide adequate energy conservation measures in newly constructed buildings increases long-term operating costs that may affect adversely the repayment of, and security for, loans made, insured, or guaranteed by Federal agencies or ... regulated instrumentali-ties_” 42 U.S.C. § 6831(a)(3). Congress was thus concerned that the costs of energy waste were reducing the ability of mortgagors to meet them mortgage payments. This financial concern would hardly be well served by a standard that saved mortgagors $100 in annual energy bills at a capital cost that raised their mortgage payments by $120 a year.
Of course 42 U.S.C. § 6839 refers to factors other than “economic cost and benefit”; these include “energy efficiency, stimulation of use of nondepletable sources of energy, institutional resources, habitability, ... and impact on affected groups.” There might be cases where the statute would allow one or [1045]*1045several of the other factors to trump the cosVbenefit analysis, justifying adoption of a standard with costs exceeding benefits. But perhaps not: “impact on affected groups” sounds like a constraint on stringency, at least if, as DOE and the district court supposed, Congress aimed to prevent undue disruption of the industry. See Appellee’s Brief at 19 n. 9; GAMA II, 773 F.Supp. at 466 & n. 4. Three other factors, “stimulation of use of nondepletable sources of energy,” “institutional resources,” and “habitability,” are subject to vague, speculative analysis at best (and “habitability” would seem to be captured in the costs and benefits). Only “energy efficiency” and “economic cost and benefit” are susceptible to comparatively precise analysis. Of these two, “energy efficiency” suggests no reasonable limits in and of itself, since it may be increased to the limits of technology so long as nonenergy resources are regarded as free. As the limiting principle inherent in “economic cost and benefit” is both reasonable and relatively determinate, any override of a negative .cost/beriefit analysis would seem to require a very careful justification.
In any event, although DOE’s brief stressed the impact on affected groups, the Department did not, at the time of the rule-making, consider whether or why that factor (or any other criterion of § 6839) would justify adoption of a standard whose marginal costs exceed its marginal benefits.3 Thus we may say (1) that Congress clearly determined that DOE must perform a proper cosVbenefit analysis, Chevron U.S.A. Inc., v. NRDC, 467 U.S. 837, 842-43, 104 S.Ct. 2778, 2782, 81 L.Ed.2d 694 (1984), and (2) that in the present ease DOE has so far identified no bases for disregarding the outcome of such an analysis,to impose a standard that fails that test.
Application of the Statutory Standard
As DOE purported to apply cost/benefit analysis on the remand from GAMA I, we have before us a record susceptible to review under that standard. And we need not remand to the district court to apply the correct standard, as we may review the administrative record directly. “We do not accord any particular deference to the decision of the District Court where, as here, ‘the District Court and this court are both reviewing an administrative record, and no additional testimony was taken in the District Court....’” Costello v. Agency for Int’l Den, 843 F.2d 540, 543 n. 5 (D.C.Cir.1988) (quoting Walter O. Boswell Memorial Hosp. v. Heckler, 749 F.2d 788, 790 n. 2 (D.C.Cir. 1984)). See also Novicki v. Cook, 946 F.2d 938, 941 (D.C.Cir.1991).
Before proceeding to that review, we note a number of special issues. First, the parties have disputed whether the statute requires DOE to demonstrate a “path to compliance”. It concededly does not do so in explicit terms. But the concept developed in the context of occupational health rules under 29 U.S.C. § 655(b)(5), where the agency need show little more than that the standards are technically feasible and that the industry will survive the costs of compliance. Even in that context, we expressed wonder at how OSHA “could arrive at even general estimates of compliance costs when it is so unclear as to how employers can comply”, Steelworkers, 647 F.2d at 1296, and we demanded a clear explanation, id. The need for a discernible path to compliance is surely more compelling where the agency .must perform a direct balance of costs and benefits.
Second, plaintiffs object that the agency has unduly relied on computer modeling. Our approach to this is fairly well established. We have noted that although “computer modeling is a useful and often essential tool for performing the Herculean labors Congress impose[s] on [administrative agencies]”, “[s]uch models, despite their complex design and aura of scientific validity, are at best imperfect and subject to manipulation”. Sierra Club v. Costle, 657 F.2d 298, 332 (D.C.Cir.1981). Since the accuracy of [1046]*1046any computer model “hinges on whether the underlying assumptions reflect reality,” id.,
the agency must sufficiently explain the assumptions and methodology used in preparing the model; it must provide a complete analytical defense of its model and respond to each objection with a reasoned presentation. The technical complexity of the analysis does not relieve the agency of the burden to consider all relevant factors and to identify the stepping stones to its final decision. There must be a rational connection between the factual inputs, modeling assumptions, modeling results and conclusions drawn from these results.
657 F.2d at 333 (brackets, footnotes, internal quotation marks, and citations omitted). The agency’s burden becomes heavier when a method of prediction “is being relied on to overcome ... ‘adverse’ actual test data...." International Harvester Co. v. Ruckelshaus, 478 F.2d 615, 642 (D.C.Cir.1973). As we shall see, that was the case here.
Finally, we note DOE’s argument that particular deference is due when an agency “is making predictions, within its area of special expertise, at the frontiers of science.” Baltimore Gas and Electric Co. v. Natural Resources Defense Council, 462 U.S. 87, 103, 103 S.Ct. 2246, 2255, 76 L.Ed.2d 437 (1983). Plaintiffs respond that in reality “this case is about hypothesized incremental improvement in the exceedingly conventional and well-plowed field of heat transfer technology, where any new ‘solution’ for increased energy efficiency can be readily and easily tested by modifying existing state-of-the-art equipment in accordance with the ‘solution’ and simply measuring the unit’s heat transfer performance.” Appellants’ Brief at 16-17. In fact, DOE’s expectations about compliance with its standards derive not from any daring new insight into particle physics, but from purported applications of accepted propositions. The issue is whether rather simple inferences are supported. Of course we defer to any relevant scientific or technical expertise, but that does not authorize us to gloss over the critical steps of DOE’s reasoning process.
DOE’s Cost/Benefit Analysis
A. Electric Water Heaters
Appellants argue that DOE’s electric storage water heater standby loss standard of 3.0 W/sq ft is arbitrary and capricious. Although appellants claim that no such water heater currently on the market can meet the standard,4 that fact, if true, would not invalidate the standard if DOE could demonstrate that firms could attain the standard at costs satisfying the demands of § 6839. That is where DOE has fallen down.
1. The 40% Fudge Factor. GAMA objects that DOE’s computer modeling is faulty and does not correspond to the results of the limited testing that DOE directed. ' On the face of the tests, the objection is valid. See Revised Rule, 55 Fed.Reg. at 23,853-54. Whereas the DOE model had predicted heat loss for a modified electric water heater to be 193.4 Btu/h, the actual loss was 294.2 Btu/h. And whereas the computer model predicted heat loss for the same unit covered in an R-11 fiberglass insulation blanket and batt to be 162.3 Btu/h (or 2.19 W/sq ft),5 the actual loss was 237.9 Btu/h (or 3.21 W/sq ft),6 in excess of DOE’s standard. DOE responded with a focus primarily on the marginal impact of adding the R-ll batt, which yielded savings greater than predicted by the model (56.3 Btu/h father than 31.1 Btu/h, or a 25.2 Btu/h difference). Since the model was only intended to predict jacket savings, DOE inferred that the additional 25.2 Btu/h in savings was caused by the batt’s reducing heat loss from the controls. Maybe so. This theory, however, did nothing to solve the problem of high absolute heat losses. To [1047]*1047explain these away, DOE assumed them to be due to losses from fittings, which had not been modified during the tests. DOE made the assumption that losses from fittings would equal 69.9 Btu/h. 55 Fed.Reg. at 23,-854. This nearly made up the otherwise unexplained difference between the model prediction and the test results (to within 5.7 Btu/h).7
In its.revised computer model, DOE then postulated a 40% reduction in .losses at the fittings and controls, but nowhere did it specify how these reductions are to be achieved, This percentage was incorporated into the model and has affected all subsequent predictions. .See 55 Fed.Reg. at 23,874 (Table 8). Following this model, DOE specifies an addition of 3.5 inches of 75% foam/25% fiberglass insulation as a path to compliance for a 50-gallon water heater for the purposes of cost/benefit analysis. DOE Statement of Reasons, Tab 7.
Defending its handiwork, DOE asserts that its estimates of losses due to fittings and controls are accurate and were derived from various studies that it properly deemed reputable. 55 Fed.Reg. at 23,853. Let us assume so, and also that DOE is correct in assuming that there was no problem in applying the studies-which were based oh residential heaters-to commercial ones. Even with these assumptions, the figures do not support DOE’s assumption that 40% of these losses can be eliminated at a reasonable cost. DOE cannot accurately estimate the costs and benefits of compliance without some notion of how industry will achieve the loss reduction. The task may be complex; the fittings vary according to the model and manufacturer. 55 Fed.Reg. at 23,853. Yet some method of compliance must be proposed in order to provide a legitimate foundation for the cost-benefit analysis.
2. DOE’s Testing. Plaintiffs object to DOE’s failure to pursue its testing to the point of having on hand at least one real-world electric water heater that satisfied the standard. We have applied a common sense test to such matters, saying that an agency can rely on its own expertise to resolve doubtful factual issues, but that it “may not tolerate needless uncertainties in its central assumptions when the evidence fairly allows investigation and solution of those uncertainties.” Natural Resources Defense Council, Inc. v. Herrington, 768 F.2d 1355, 1391 (D.C.Cir.1985); see also National Lime Ass’n v. EPA, 627 F.2d 416, 454 (D.C.Cir. 1980). An important, easily testable hypothesis should not remain untested. Where creation of a fully complying prototype is not very costly, as appears to be the case for water heaters here, it seems reasonable to do so in order to establish the validity of the energy-saving benefits that the agency believes can be achieved at a reasonable cost. Thus, DOE should consider the costs and benefits of testing options, taking into account the importance of the hypothesis, its uncertainty, the likelihood that testing would resolve the uncertainty, and the cost of testing. While of course we would defer to any reasoned decision on incremental testing, here we cannot discern even the faintest glimmer of an effort to make such a decision.
3. Use of an Arbitrary Multiplier for Projecting Costs. Key to the Department’s cost projection's were (1) an unexplained assumption that the costs of including insulation in a residential electric water heater will typically be 2.5 times the costs of the insulation itself and (2) an assumption that the first assumption could be extended to the commercial market. Plaintiffs vigorously resist the second step, adducing figures indicating that the residential market is 170 times the size of the commercial market. Thus, they say, the unit costs of product redesign and retooling are far higher in the diminutive commercial market. They also claim unit production costs are higher, due to relatively low-speed, labor-intensive production' lines. See GAMA Comments (February 28,1990) at 15-16.
The difference in total • size of the two markets may not necessarily undermine DOE’s extension of the 2.5 multiplier to the commercial market. It is conceivable that, [1048]*1048despite the aggregate difference, firms and plants in both industries might be the same size and have similar production characteristics. But if so, DOE has conspicuously failed to make the ease-or, more narrowly, to answer the focused and plausible contentions of plaintiffs. See, e.g., International Fabricare Institute v. EPA, 972 F.2d 384, 389 (D.C.Cir.1992). DOE points to a report by Arthur D. Little Inc. as evidence that the market size differences have no impact on cost, Appellee’s Brief at 22, but the report is expressly directed to the residential heater market, see Consumer Products Efficiency Standards Engineering Analysis Document, by Arthur D. Little Inc. at D3-1, and appears nowhere to address the differences between the two markets. Because of its failure to answer these critiques, DOE’s action did not manifest reasoned decisionmak-ing and was to that extent arbitrary and capricious.
In a related claim, GAMA argues that DOE has not adequately addressed its comments about increased shipping costs. See GAMA Comments (February 28, 1990) at 15. Standing alone GAMA’s objection seems rather weak; its contentions about shipping of commercial heaters make no comparison with residential ones, and thus do not appear inconsistent with the 2.5 multiplier. Depending upon the course of the remand, however, the issue may become pertinent.
4. Use of an Incorrect Temperature Differential in Cost/Benefit Analysis. Rather obscurely, see Appellants’ Brief at 23 n. 22, GAMA objects that DOE used a lower “temperature differential” to estimate the energy savings in its cost/benefit analysis (67.7°F) than it used in the standard itself (80°F).8 DOE itself ultimately acknowledged the discrepancy, saying that its “cost benefit analysis calculations utilized a smaller temperature differential ... which in effect moderated the resulting cost benefit analysis.” See Opposition to Motion for Pretrial Conference, etc., at 4 (emphasis added). “Moderated”? It simply exaggerated the supposed energy saving from DOE’s standard. In its cost/benefit analysis, DOE calculated energy loss to be 122.21 Btu/h (2.47 W/sq ft) with 3.5-inch 75% foam/25% fiberglass insulation. Although DOE later recognized that with the differential used in the standard itself the loss would have been 2.77 W/sq ft, id., it made no effort to recalculate benefits using this corrected figure, thus leaving the benefit component of the eost/benefit analysis skewed. In addition, as both figures profit from the 40% fudge factor, the mere fact that the corrected figure (2.77 W/sq ft) would comply with the standard is no evidence of a reasonable path to compliance. DOE cannot use one set of conditions for the standard itself, and another, more favorable set, to estimate the proposed compliance method’s likely achievements for cost/benefit purposes.
B. Oil and Gas Water Heaters
Here again,9 DOE failed to justify a path to compliance or to demonstrate that its standard’s benefits exceeded their costs. The proposed paths are (1) for “high-input” heaters,10 the addition of two inches of foam insulation or three inches of fiberglass insulation and (2) for “low-input” heaters,11 the addition of insulation and the installation of flue dampers and IIDs. DOE points to neither a current model nor a prototype that meets the standard.
Two problems familiar from the electric heater context are repeated and require no further discussion. First, DOE again extended the 2.5 multiplier from the residential to the commercial market, see Preliminary Statement, 54 Fed.Reg. at 49,733, 49,740-41, [1049]*1049without justifying the extension, and second, it failed to justify the absence of additional real-world testing. The other problems are distinctive to the gas and oil heater market.
1. Disregarding the Cost of Adding Flue Dampers and IIDs to Low-Input Heaters. Low-input water heaters constitute a significant portion-25%—of the market for commercial oil and gas water heaters. Second Declaration of Joseph M. Mattingly (May 28, 1991) at 7. All agree that insulation alone will not enable these to meet the standard and that installing flue dampers and IIDs is necessary. DOE has not, however, included any cost for such installation, and its cosi/benefit analysis in the Preliminary Statement did not even mention those costs. GAMA protested that these modifications would lead to a price' increase of $500 or more, which in some cases represents 50% of current unit price. Appellants’ Brief at 32 & n. 30; GAMA’s Comments (Feb. 28, 1990) at 17. DOE responded by claiming that since equally stringent ASHRAE standards would soon be in effect for the entire commercial oil and gas water heater market, DOE’s requirement would impose no additional' burden: “Thus the added cost of these devices is virtually nil since they would have been required anyway in the absence of the DOE standard,” Revised Rule, 55 Fed.Reg. at 23,863.
In fact, DOE’s predictions regarding industry standards proved to be false, as ASH-RAE adopted less stringent standards for low-input models12— standards with which they can evidently comply without flue dampers or IIDs (or so appellants tell us without contradiction, see Appellants’ Brief at 33 n. 32). In any event, DOE had no basis to assess the costs of new devices at zero simply because another regulatory body contemplated requiring those devices. Congress did not delegate the standard-writing to ASHRAE, nor authorize DOE to do so. Thus DOE’s regulation must.stand on its own.
As manufacturers have not found it worthwhile to add these devices to low-input heaters independently of the standby-loss requirement, it is the standby-loss concern that will cause the cost of installing them to be incurred (if indeed DOE ultimately adopts a standard that requires them). Accordingly, the cost must be- included if the cost-benefit analysis is to be a coherent marginal analysis. (To the extent that the devices contribute to the value of the water heaters apart from standby-loss reduction, that windfall can be taken-into account as an offset to the cost.)
2. The Computer Model. As with electric water heaters, DOE encountered inconvenient facts in its limited testing of gas water heaters and attempted to brush off the flaws thus revealed in its computer model.' When the National Institute of Standards and Technology (“NIST”) tested a 78-gallon single-flue gas water heater wrapped in an R-ll fiberglass insulation blanket, measured standby losses were 2.37% per hour, as opposed to the 1.76% per hour required by the standard. Letter from A. Fanney to DOE (May 31, 1990) at 2-3.13 In response, DOE recommended for compliance “the addition of one inch of fiberglass to the base two inches [the R-ll blanket] or ... the use of approximately two inches of foam insulation.”- 55 Fed.Reg. at 23,877. GAMA attacks DOE’s computer model on several grounds and contends that the proffered path to compliance is not proven to be effective.
a. The percentage of loss through jacket. The percentage of loss through the heater’s jacket is significant because the main path to [1050]*1050compliance DOE outlines-increased jacket insulation-would affect only these losses. DOE asserts that jacket loss accounts for 70% of the total standby loss in gas water heaters with flue dampers and IIDs, 55 Fed. Reg. at 23,857, while GAMA asserts that the corresponding figure is approximately 50%, GAMA’s Comments (Feb. 28, 1990) at 8-9. DOE has not pointed us to anything in the record that supports its assumption of 70% jacket loss.14 In response to GAMA’s objections, DOE declared that “it is ... conceivable that better design and production and the more predominant use of electronic flue dampers can rectify much of the residual heat loss.” Revised Rule, 55 Fed.Reg. at 23,858. DOE’s resort to speculation as to what is “conceivable” simply illustrates the bankruptcy of its effort to delineate a path to compliance.
b. Adjustment for an atypical test prototype. Unlike most high-input ,gas water heaters, • the 78-gallon water heater DOE tested, which was used to correct the model, had a pilot light and not an IID. Instead of ordering a different model, DOE attempted to adjust for this peculiarity by adding a factor, “Q ADD.” As DOE explained, “ ‘Q ADD’ is the amount of losses that must be added to the jacket losses ... to match total losses with [testing agency] measured losses.” Revised Rule, 55 Fed.Reg. at 23,877/3. Because DOE apparently underestimated the efficiency of a pilot light in such circumstances, it overestimated the benefit of replacing the prototype’s pilot light with the standard IID. This in turn led DOE to underestimate non-jacket heat loss in a normal high-input gas water heater (one with an IID) — and thus to overstate jacket loss, i.e., the only kind of loss addressed by DOE’s proposed path to compliance for high-input heaters.
Once again, DOE’s difficulties stem from, a dubious analogy between the residential and commercial markets. DOE put pilot light efficiency at 22%, a figure based on a 1977 study for residential water heaters, R. Hos-kins & E. Hirst, Oak Ridge National Laboratory, Energy and Cost Analysis of Residential Water Heaters (June 1977) at 33, which in turn was based on a 1978 study (at that time forthcoming), G.G. Slaughter and D.E. Spann, An Efficiency Evaluation and Consumer Economic Analysis of Domestic Water Heaters, ORNL/ Con-5 (1978) at 19. GAMA points out that residential water heaters typically do not have flue dampers, and that that lack results in lower pilot light efficiency. Indeed, the 1978 study points out that if a flue is severely restricted, as it would be by a flue damper, pilot light efficiency rises to 71% at 160°F. Id. Thus DOE’s effort to relate its rather idiosyncratic prototype to the real world appears based on a misconception.
C. Redefining Storage Units
DOE also failed to perform an adequate cost/benefit analysis, or indeed any analysis, on its redefinition of storage units. After initially defining storage water heaters as those with an input rating of less than 4,000 Btu/h per gallon and a storage capacity of more than ten gallons, Preliminary Statement, 54 Fed.Reg. at 49,727, DOE ultimately defined them as those with an input rating of less than 4,000 Btu/h per gallon or a storage capacity of more than ten gallons. 55 Fed. Reg. at 23,865. In doing so, DOE adopted an ASHRAE standard that broadened the definition of storage water heaters to include some that were formerly considered instantaneous. See ASHRAE 90.1-1989, printed in 55 Fed.Reg. at 23,865/1. DOE claimed that the new standard “was consciously established to eliminate any confusion over the definition of storage water heaters and instantaneous water heaters.” 55 Fed.Reg. at 23,865/1. So far as appears, however, no confusion had surfaced; DOE was simply following ASHRAE standards. More importantly, specific analysis of the standard’s economic effects is entirely lacking. Although DOE here asserts that “the cost/benefit analysis that DOE performed is valid for the entire range of storage water heaters covered by the standby loss criteria”, Appellee’s Brief at 42, it points to nothing supporting that contention, and indeed its own figures— which don’t address very small heaters— show benefit/cost ratios gradually declining [1051]*1051with size. See, e.g., Preliminary Statement, 54 Fed.Reg. at 49,740. To be meaningful, DOE’s economic analysis must take into account the standards’ differing effects on the major segments of the market regulated. ■
The defects identified substantially undermine the Department’s justification of the Revised Rule. To the extent that GAMA has properly15 raised other claims not explicitly addressed here, it has not persuaded us that they had substantial merit. This is not to deny, however, that more polished versions of these critiques raised in the course of the remand to DOE may require consideration by DOE.
Whether a preliminary injunction should issue against enforcement of all or part of the rule, compare GAMA 722 F.Supp. at 797-98, rests within the sound discretion of the district court, under the principles governing interim relief when an agency rule is found wanting. See, e.g., International Union, UAW v. FMSHA, 920 F.2d 960, 967 (D.C.Cir. 1990) (pointing to “the seriousness of the order’s deficiencies (and thus the extent of doubt whether the agency chose correctly) and the disruptive consequences of an interim change that may itself be changed”); Allied Signal, Inc. v. U.S. Nuclear Regulatory Comm’n, 988 F.2d 146, 151 (D.C.Cir.1993). Accordingly, we reverse and remand the case to the district court for any preliminary relief that may be appropriate and to remand to DOE for further proceedings.
So ordered.