In these consolidated cases, petitioning oil companies challenge effluent limitations contained in permits issued to them under section 402 of the Federal Water Pollution Control Act.1 In cases 75-3794 and 75-3796, the permits provide for the limited discharge of deck drainage, produced water, and sanitary wastes from the companies’ offshore oil and gas platforms. The permits in case 75-3795 allow the limited discharge of produced water from onshore facilities designed to separate gas and crude oil from the produced water. Petitioners contend that the permits require a greater level of pollution control than is achievable through the use of “the best practicable control technology currently available,” Federal Water Pollution Control Act §§ 301(b)(1) and 304(b)(1).2 More specifically, petitioners argue that the permit standards are the result of an invalid statistical analysis and, moreover, do not provide for currently inevitable upsets and malfunctions. Petitioners also raise significant statutory and constitutional objections to the procedure followed by the Environmental Protection Agency (EPA) in issuing permits under section 402 of the Act. We conclude that while the petitioners miss the mark with certain of their complaints there is sufficient merit in several of them to require that we remand this proceeding to the EPA for the limited purpose of considering certain specific matters delineated in the course of this opinion.
I.
History of the Dispute.
A. Offshore Oil Drilling and Waste Problems.
Petitioners in cases 75-3794 and 75-3796 operate offshore platforms in Cook Inlet, *1257Alaska.3 The platforms are self-contained structures with live-on crews that can number up to 50 men during drilling operations. In addition to drilling, a number of related and necessary operations are conducted regularly on board the platforms and contribute to the platforms’ waste. Particularly relevant to this dispute are cleaning activities, rig moving, and rig conditioning.
Three forms of waste must be disposed of from the platforms and are the subject of the permit provisions under challenge— deck drainage, produced water, and sanitary wastes. Deck drainage collects on the curbed platform decks from a variety of sources such as platform equipment, drilling equipment, deck washings, and rain; the drainage is composed of a shifting composite of substances that include fresh water, salt water, oil, grease, drilling muds, and soaps. Produced water rises inescapably along with the desired oil and gas from their subsurface formations. Sanitary wastes pass from the lavatories, showers, etc., used by the workers stationed on the platforms. The wastes can either be disposed of offshore or be pumped to shore and then disposed of. At present, only sanitary wastes are disposed of offshore by all the platforms in Cook Inlet. Some of the platforms pump their deck drainage to shore.4 All of the platforms pump their produced water to shore for treatment before disposal. However, onshore disposal of deck drainage and produced water is not expected to continue indefinitely. Ultimately all of the platforms expect to discharge all three of the wastes offshore.5 Thus, with the exception of Shell’s permits,6 the permits were requested to cover and do cover all three forms of waste.
Petitioners in case 75-3795 operate three onshore facilities which separate gas and crude oil from produced water. As noted earlier, offshore platforms in Cook Inlet currently pump their produced water to shore for treatment and disposal at these facilities.
B. Statutory Scheme.
Under the Federal Water Pollution Control Act, as amended in 1972 (hereinafter the Control Act),7 it is unlawful to discharge any pollutant into a navigable water without first obtaining a permit under section 402 of the Control Act.8 33 U.S.C.A. § 1311(a) (1977 Supp.). Section 402(a) of the Control Act9 authorizes the Administrator to issue a permit if he determines, inter alia, that the discharge would meet the requirements of section 301 of the Control Act.10 Section 301(b)(1)(A),11 in turn, limits discharges prior to July 1, 1983, to those achievable by use of “the best practicable control technology currently available” (hereinafter referred to as BPCTCA), as defined by section 304(b)(1) of the Control Act.12
*1258It is the practice of the EPA, before issuing a permit under section 402, to obtain from the appropriate state government a certificate stating that the permit limitations will comply with relevant state law. See 40 C.F.R. § 125.32(e)(8). A state can waive its certification right either explicitly or by failing to respond to a certification request within 30 days. Id. In summary, before an effluent limit is set by the EPA, the Agency must determine that the limit is representative of BPCTCA and obtain a certification by the state government.
As to the proper procedure to be followed in drafting and issuing the permits, section 402 of the Control Act requires the Administrator to provide an “opportunity for public hearing.” The EPA has not interpreted this mandate as requiring a full adjudicatory hearing under sections 554, 556, and 557 of the Administrative Procedure Act (hereinafter the APA).13 However, pursuant to 40 C.F.R. § 125.36, “interested persons” may request and the Administrator, in his discretion, may grant an adjudicatory hearing. This adjudicatory hearing, however, will not necessarily incorporate all of the procedural requirements of sections 554, 556, and 557 of the APA.
C. Administrative Proceedings.
1. Cases 75-3794 and 75-3796: Offshore platforms.
In late 1971 and early 1972, petitioners applied for discharge permits from the Army Corps of Engineers pursuant to the Federal Refuse Act.14 Before the Corps could take action on petitioners’ applications, however, the Refuse Act was superseded by the Federal Water Pollution Control Act Amendments of 1972.15 Therefore, in mid-1973 petitioners applied to the EPA for permits under section 402 of the Control Act. Draft permits were circulated by the Seattle Regional Administrator in September of 1973; after informal public hearings in October, the permits were tentatively issued in December of 1973.
These initial permits contained limits on the discharge of deck drainage of 25 milligrams per liter (mg/1) daily average and 75 mg/1 daily maximum (typically expressed as a combined limit of “25/75 mg/1”).16 The permits made no provision for upsets or malfunctions in the pollution control system. The permits also prohibited bypassing of the control system “except (i) where unavoidable to prevent loss of life or severe property damage, or (ii) where excessive storm drainage or runoff would damage any facilities necessary for compliance with the effluent limitations and prohibitions of their permit.” See, e.g., J.R.17 23.
Pursuant to petitioners’ request under 40 C.F.R. § 125.36, the Administrator granted further consideration of the proposed permits, including an adjudicatory hearing with the right of cross-examination. Shell Oil Company requested and was granted a hearing separate from the other petitioners. Both the Shell hearing and the “consolidated” hearing were held in August of 1974.
All petitioners argued that the discharge standards had been set at too low a level. Petitioners in the consolidated hearing contended that their permits should contain a produced water limit of 75/100 mg/1 and a deck drainage limit of 85/100 mg/1. Shell argued for an even higher deck drainage limit of 100/200 mg/1. One of Shell’s witnesses also computed recommended deck drainage limits based on pounds of pollu*1259tants discharged per day. In addition, petitioners in both hearings attacked the failure of the permits to provide for malfunctions and upsets in the pollution control system and argued for more liberal bypass provisions that would cover prolonged but necessary periods of equipment repair and maintenance.
After reviewing the record certified by the Administrative Law Judge, the Regional Administrator reaffirmed his 25/75 mg/1 limit with respect to deck drainage and refused to enact any limit with respect to produced water due to a paucity of data and the “uncertainty” of petitioners’ plans to begin discharging produced water offshore.18 The Regional Administrator also declined to add upset provisions to the permits or to liberalize the bypass provisions already contained in the permits.
The petitioners thereupon requested and were granted direct review by the Administrator pursuant to 40 C.F.R. § 125.36(n). The Administrator accepted written briefs and held oral argument in August of 1975. The Administrator, with the aid of a Chief Judicial Officer and EPA technical personnel, issued his decisions on September 25, 1975, raising the limits for deck drainage to 65/90 mg/1, setting an alternative standard for deck drainage of 5/9.25 pounds per day,19 and setting limits for produced water of 48/72 mg/1.20 The Administrator affirmed the Regional Administrator’s decision to deny upset provisions and liberalized bypass provisions.
This action followed. Since the filing of the action, the EPA has issued final permits in line with the Administrator’s decision. Prior to issuance of the final permits, the Regional Administrator requested certification from the State of Alaska. Alaska initially notified the Regional Administrator that the discharge limits for both deck drainage and produced water should be tightened to 25/50 mg/1; on the other hand, Alaska also asked that upset provisions be added to the permits.21 However, on June 30,1976, Alaska waived its right to certify the permits.
2. Case 75-3795: Onshore facilities.
Tentative permits for the discharge of produced water from petitioners’ onshore facilities were issued by the EPA in December of 1973. As in the case of the offshore platforms, petitioners had originally applied for discharge permits from the Corps of Engineers pursuant to the Federal Refuse Act. However, as explained above, the Federal Refuse Act was superseded by the *1260Federal Water Pollution Control Act Amendments of 1972.
The tentative permits set discharge limits of 15/30 mg/1. Like the offshore permits, no provision was made for upsets or malfunctions in the pollution control system. The permits also strictly limited bypassing to those cases where bypassing was necessary to prevent loss of life or severe property damage or where excessive storm drainage would otherwise damage the treatment facilities.
Petitioners were granted a consolidated hearing on their permits pursuant to their request under 40 C.F.R. § 125.36. Petitioners challenged the failure to include upset provisions and more liberal bypass provisions in their permits. The petitioners also urged that the effluent limitation be liberalized. According to petitioners Marathon and Atlantic Richfield, the permit standard should be set at 68/100 mg/1; Shell argued for an even higher limit of 75/120 mg/1. After reviewing the record certified by the Administrative Law Judge, the Regional Administrator set a new effluent limitation of 25/50 mg/1, but refused to authorize upset provisions or to liberalize the bypass provisions.
Pursuant to petitioners’ request under 40 C.F.R. § 125.36(n), direct review by the Administrator was then granted. After an informal conference, written briefs were accepted and oral arguments were held in August and September of 1975. The Administrator issued his decision on September 25, 1975 with the help of a Chief Judicial Officer and EPA technical personnel. The Administrator set an effluent limit of 48/72 mg/1; he refused, however, to include either upset or more liberal bypass provisions in the permits.
Before issuing final permits in line with the Administrator’s decision, the Regional Administrator requested certification from the State of Alaska. Alaska initially notified the Regional Administrator that the discharge limitations should be set at 15/30 mg/1, as in the tentative permits; on the other hand, Alaska asked that upset provisions be added to the permits. Alaska later waived its right to certify the permits and final permits were issued on September 3, 1976.
II.
Procedural Attacks.
A. Statutory Challenges.
Petitioners contend that the EPA, before issuing permits under section 402 of the Control Act, must afford applicants formal adjudicatory hearings that adhere to the requirements of sections 554, 556 and 557 of the APA.22 In line with this contention, petitioners argue that, while they were given adjudicatory hearings, those hearings diverged in certain important respects from the strict requirements of the APA. Specifically, petitioners urge that the Regional Administrator, rather than the Administrative Law Judge, rendered the initial decision, in contravention of sections 554(d) and 557(b) of the APA,23 and that the Administrator considered evidence outside of the record, in conflict with section 556(e) of the APA.24
Support for the petitioners’ first contention is provided by the recent opinion of the Seventh Circuit in United States Steel Corp. v. Train, 556 F.2d 822 (7th Cir., 1977), holding that the APA provisions are indeed applicable to section 402 proceedings. We agree with the Seventh Circuit that the APA adjudicatory hearing requirements apply. However, we reach this result by a slightly different route.25 As to *1261the EPA’s adherence to the APA in this instance, the Administrator apparently did consider evidence outside of the record in making his determination, requiring a limited remand. In all other respects, however, the APA requirements were followed.
1. Applicability of the APA.
In setting out procedures that an agency must follow in making “adjudicatory” determinations, Congress recognized that certain administrative decisions closely resemble judicial determinations and, in the interest of fairness, require similar procedural protections.26 These “quasi-judicial” proceedings determine the specific rights of particular individuals or entities. And, like judicial proceedings, the ultimate decision often turns, in large part, on sharply-disputed factual issues. As a result, such APA procedures as cross-examination of key witnesses are needed both for the protection of affected parties and to help achieve reasoned decisionmaking. At the opposite end of the pole are agency determinations that depend less on the resolution of factual disputes and more on the drawing of policy; 27 such “rulemaking” decisions must by necessity be guided by more informal procedures. See Bi-Metallic Investment Co. v. State Board of Equalization, 239 U.S. 441, 36 S.Ct. 141, 60 L.Ed. 372 (1915). According to the Attorney General’s Manual on the APA, issued shortly after the APA was passed,
“the entire act is based upon [this] dichotomy between rule making and adjudication . . . Rule making . is essentially legislative in nature, not only because it operates in the future but also because it is primarily concerned *1262with policy considerations Typically, the issues relate not to the evidentiary facts, as to which the veracity and demeanor of witnesses would often be important, but rather to the policy-making conclusions to be drawn from the facts . . . . Conversely, adjudication is concerned with the determination of past and present rights and liabilities . . In such proceedings, the issues of fact are often sharply controverted.” Attorney General’s Manual on the Administrative Procedure Act, at 14-15 (1947).
Working from this basic dichotomy, the setting of effluent limitations under section 402 of the Control Act is clearly “adjudicatory” in nature and requires the special protections of sections 554, 556 and 557 of the APA. Unlike, for example, section 304 proceedings which lead to the promulgation of industry-wide effluent limitation guidelines and which are in large measure policymaking, section 402 proceedings focus on whether particular effluent limitations are currently practicable for individual point sources.28 As the instant proceeding well demonstrates, the factual questions involved in the issuance of section 402 permits will frequently be sharply disputed. Adversarial hearings will be helpful, therefore, in guaranteeing both reasoned decisionmaking and meaningful judicial review. In summary, the proceedings below were conducted in order “to adjudicate disputed facts in particular cases,” not “for the purpose of promulgating policy-type rules or standards.” The protections of sections 554, 556 and 557 of the APA are therefore particularly appropriate.29 See United States v. Florida East Coast Railway Co., 410 U.S. 224, 245, 93 S.Ct. 810, 35 L.Ed.2d 223 (1973).
The EPA agrees that the challenged proceedings were adjudications. The APA specifically provides that licensing proceedings, which term includes the issuance of permits, are adjudications. 5 U.S.C. §§ 551(6)-(7) & (9) (1970). However, the EPA observes that sections 554, 556 and 557 of the APA only apply to adjudications “required by statute to be determined on the record after opportunity for an agency hearing.” 5 U.S.C. § 554(a) (1970). Since section 402(a)(1) of the Control Act requires only an “opportunity for public hearing” and fails to specify that permit limitations must be “determined on the record,” the EPA argues that the formal adjudicatory hearing provisions of the APA are inapplicable. We disagree.30
*1263First, the provision for judicial review of permit proceedings contained in section 509(b)(1) of the Control Act indicates that section 402 determinations must be made “on the record,” despite the failure of section 402 so to provide. According to the Attorney General’s Manual on the APA, a requirement that administrative decisions be “on the record” can be “clearly implied in the provision for judicial review . in the circuit courts of appeal.” Attorney General’s Manual, supra, at 41. Thus, even if the requirements for invoking the adjudicatory hearing provisions of the APA are read rigidly, they are met in this case.
Moreover, whether the formal adjudicatory hearing provisions of the APA apply to specific administrative processes does not rest on the presence or absence of the magical phrase “on the record.” Absent congressional intent to the contrary, it rests on the substantive character of the proceedings involved. The 79th Congress’ purpose in limiting the APA provisions to determinations made “on the record” after opportunity for a hearing was not to provide future Congresses with a talisman that they would use to signify whether or not sections 554, 556 and 557 of the APA should apply. It was to limit the sections’ applications to those types of adjudications, discussed above, needing special procedural safeguards. The APA defines “adjudication” broadly as an agency process leading to a final disposition “other than rulemaking.” 5 U.S.C. §§ 551(6H7) (1970). But not all “non-rulemakings” are “adjudications” of the nature outlined above and calling for special protective proceedings. Thus, Congress inserted section 554’s prefatory language cited by the EPA to exclude from the residual definition of adjudication “governmental functions, such as the administration of loan programs, which traditionally have never been regarded as adjudicative in nature and as a rule have never been exercised through other than business procedures.” Attorney General’s Manual, supra, at 40. The setting of effluent limitations under section 402 of the Control Act, however, falls squarely within the mainstream of traditional adjudications.
The failure of Congress to provide for any hearing whatsoever within an administrative process may well be a valid indication that Congress either did not feel that it was providing for an “adjudication” in the traditional sense of the word or did not intend the APA procedures to apply. See, e. g., Yong v. Regional Manpower Administrator, 509 F.2d 243, 245 (9th Cir. 1975); Reed v. Morton, 480 F.2d 634, 643 (9th Cir.), cert. denied, 414 U.S. 1064, 94 S.Ct. 571, 38 L.Ed.2d 469 (1973). But see Wong Yang Sung v. McGrath, 339 U.S. 33, 70 S.Ct. 445, 94 L.Ed. 616, modified, 339 U.S. 908, 70 S.Ct. 564, 94 L.Ed. 1336 (1950). However, if a statute provides for a hearing, similar weight should not typically be accorded to Congress’ failure to specify that determinations must be made “on the record.”31 The legislative history of the APA does not suggest that the specific words “on the record,” or words of similar effect, must be contained in a statute in order to invoke the adjudicatory hearing requirements of the APA. According to the report of the Senate Judiciary Committee, section 554’s prefatory language, including the phrase “on the record,” was intended to limit applicability to “cases in which other statutes require the agency to act upon or after a hearing.” Legislative History of the Administrative Procedure Act, 79th Cong., 2d Sess., at 202 (1946) (emphasis added). See also id. at 226, 260, & 359-60. Neither is there any indication in the legislative history of the Federal Water Pollution Control Act Amendments of 1972 that Congress, by omitting the phrase “on the record,” intend*1264ed to deny applicants for effluent discharge permits the protection of the APA’s adjudicatory hearing provisions.32
In summary, the crucial question is not whether particular talismanic language was used but whether the proceedings under review fall within that category of quasi-judicial proceedings deserving of special procedural protections. See United States v. Florida East Coast Railway Co., 410 U.S. 224, 251, 93 S.Ct. 810, 35 L.Ed.2d 223 (1973) (Douglas, J., dissenting); United States v. Allegheny-Ludlum Steel Corp., 406 U.S. 742, 757, 92 S.Ct. 1941, 32 L.Ed.2d 453 (1972). The focus of our inquiry should be on the nature of the administrative determination before us. International Telephone & Telegraph v. Local 134, International Brotherhood of Electrical Workers, 419 U.S. 428, 439, 95 S.Ct. 600, 42 L.Ed.2d 558 (1975). As discussed in detail above, permit proceedings under section 402 of the Control Act would appear to fall precisely within the category of proceedings Congress sought to address in the APA. Sections 554, 556 and 557 of the APA must, therefore, be applied. “[I]t would be a disservice to our form of government and to the administrative process itself if the courts should fail, so far as the terms of the Act warrant, to give effect to its remedial purposes where the evils it aimed at appear.” Wong Yang Sung v. McGrath, supra.33
2. EPA’s Adherence to the APA.
a. Did the Regional Administrator err in making the initial decision?
Petitioners concede that they were granted hearings which met in large measure the adjudicatory hearing requirements of the APA. However, as outlined earlier, following the hearings, the hearing records were certified to the Regional Administrator who proceeded to make initial permit decisions on the basis of the records. Petitioners contend that, under sections 554(d) and 557(b) of the APA, the Administrative Law Judge was required to make the initial decisions or, at least, recommend decisions to the Regional Administrator. We disagree, as did the Seventh Circuit in the similar case of United States Steel Corp. v. Train, supra, at p. 833.
Sections 554(d) and 557(b) of the APA do not apply to adjudications involving “applications for initial licenses.” 5 U.S.C. §§ 554(d)(A) & 557(b) (1970). Section 551(8) of the APA defines “license” as, inter alia, “the whole or a part of an agency permit.” An initial license is an “original license as contradistinguished from renewals or amendments of existing licenses.” Legislative History of the APA, supra, at 226. See Chotin Towing Corp. v. FPC, 102 U.S.App.D.C. 69, 250 F.2d 394 (1957). There is nothing in the record to suggest that petitioners currently hold a discharge permit; while permits were requested from the Corps of Engineers, the permits were never granted. Thus, the EPA was determining an application for an “initial license” and sections 554(d) and 557(b) of the APA were inapplicable.
b. Consideration of extra-record materials.
The APA requires that an agency confine itself to the record in making adjudicatory determinations. 5 U.S.C. § 556(e) (1970). See NLRB v. Johnson, 310 F.2d 550, 552 (6th Cir. 1962). Nor do we read the Administrator’s regulation, 40 C.F.R. § 125.-*126536(n)(12), to authorize going beyond the record in such determinations. The evidence indicates that the Administrator of the EPA made excursions outside of the record in issuing the decisions under challenge here.34 However, “to constitute fatal error it must appear that an administrative agency’s journey outside the record worked substantial prejudice.” Id.; see United States v. Pierce Auto Freight Lines, 327 U.S. 515, 528, 66 S.Ct. 687, 90 L.Ed. 821 (1946). To determine whether substantial error resulted from the Administrator’s actions here, it is necessary to examine his excursions from the record in the context of the substantive issues posed by the permit applications. Therefore, after examining petitioners’ constitutional challenge to the EPA’s procedures, we return in part III to consider the Administrator’s study of extra-record materials in the course of an examination of those substantive issues.
B. Constitutional Challenge.
Whether or not the APA requires the delegation of decisional authority to the Administrative Law Judge, petitioners contend that review by the Regional Administrator of the draft permits which he originally issued violates constitutional due process. We disagree.
At this early stage in the development of environmental law and procedure, an area whose technical complexity poses novel difficulties for the adjudicatory process, we should be circumspect in importing rigid procedural requirements resting on constitutional grounds and developed in quite dissimilar contexts. Circumspection, however, need not be our only basis for rejecting the petitioner’s constitutional arguments because we conclude that existing authorities applicable to procedural due process do not invalidate the procedure followed in this case.
In a number of cases, the probable bias of an adjudicator or reviewing agent was so self-evident that he could not be permitted, compatibly with due process, to make any adjudication at all. Morrissey v. Brewer, 408 U.S. 471, 92 S.Ct. 2593, 33 L.Ed.2d 484 (1972); In re Murchison, 349 U.S. 133, 75 S.Ct. 623, 99 L.Ed. 942 (1955); Turney v. Ohio, 273 U.S. 510, 47 S.Ct. 437, 71 L.Ed. 749 (1927). It is neither self-evident, nor have petitioners presented evidence to suggest that the review procedure here poses the “unacceptable risk of bias” present in Morrissey, Murchison or Turney. Rather, vesting the initial decision in the Regional Administrator, who after further hearings before an Administrative Law Judge reviews the matter prior to the issuance of an initial license, more nearly resembles the permissible combination of investigation and adjudication sanctioned by Withrow v. Larkin, 421 U.S. 35, 95 S.Ct. 1456, 43 L.Ed.2d 712 (1975), than, for example, the situation in which a judge receives as compensation part of a fine which he levies, Turney v. Ohio, supra. We see no more reason to assume that a Regional Administrator cannot objectively review his earlier decision in light of additional hearings than to assume that a federal judge cannot retry or review a case after its remand following the initial trial or review. The presumption of objectivity extended to administrative law judges (then trial examiners) in N.L.R.B. v. Donnelly Garment Co., 330 U.S. 219, 67 S.Ct. 756, 91 L.Ed. 854 (1947) is equally appropriate in this case. This presumption does not disappear merely because the procedures employed by the Regional Administrator in this case bear some functional resemblance to initial and appellate adjudication.
III.
The Effluent Limitations.
Petitioners contend that the effluent limitations set by their permits are not at*1266tainable by use of the “best practicable control technology currently available,” 33 U.S.C.A. § 1311(b)(1) (Supp.1977). In order to sustain the limitations, we must find that they are supported by “substantial evidence.” 5 U.S.C. § 706(2)(E) (1970). This is a more demanding standard than the “arbitrary or capricious” standard under which actions of the EPA are typically reviewed. See Citizens to Preserve Overton Park v. Volpe, 401 U.S. 402, 415, 91 S.Ct. 814, 28 L.Ed.2d 136 (1971). The record must include “such relevant evidence as a reasonable mind might accept as adequate to support” the EPA’s determinations. Consolidated Edison Co. v. NLRB, 305 U.S. 197, 59 S.Ct. 206, 83 L.Ed. 126 (1938); see Jones v. Priebe, 489 F.2d 709, 710 (6th Cir. 1973); John W. McGrath Corp. v. Hughes, 264 F.2d 314, 315 (2d Cir.), cert. denied, 360 U.S. 931, 79 S.Ct. 1451, 3 L.Ed.2d 1545 (1959).35
A. The EPA’s Mode of Analysis.
Before turning to petitioners’ specific objections, it is useful to outline briefly the' statistical approach taken by the EPA in setting the instant discharge standards. The approach closely paralleled the EPA’s typical method of setting effluent limitations. Broadly stated, the EPA sets effluent standards by looking to the average performance of the best existing pollution technology in the industry. More particularly, the EPA first determines the “best practicable control technology currently available.” (BPCTCA) If possible, the EPA then collects effluent discharge data from those plants in the industry that utilize BPCTCA in an exemplary fashion. The EPA’s final task is to utilize the data to compute an effluent discharge standard average of this data, which typically will be a number that exemplary BPCTCA plants can meet. This is done by using so-called “confidence intervals.” Since the influents being treated typically vary in quality and makeup from the time of one data collection to the time of another, the quality of the effluent discharged will also vary considerably. For example, in the instant case, one of the offshore platforms viewed by the EPA to be using BPCTCA in the treatment of deck drainage showed effluent qualities ranging from 1.1 mg/1 on June 3 and June 24 to 86.5 mg/1 on June 27, with considerable variance between these two figures. J.R. 1247-48, 1250. In computing a standard based on a “confidence interval” of, say, 99 percent, the EPA determines a mg/1 standard that the effluent discharge data from an exemplary plant can be expected to fall inside of 99 percent of the time.36 The EPA used “confidence intervals” of 99 percent and 97.5 percent, respectively, in setting the produced water and deck drainage limitations under challenge here. J.R. 1220 & 122237
Petitioners do not attack the EPA’s general approach towards computing effluent limitations representative of BPCTCA.38 *1267Instead, petitioners attack the manner in which the EPA applied this approach in determining the particular effluent standards under challenge here. With respect to deck drainage, petitioners complain that the Administrator incorrectly determined BPCTCA, excluded relevant data from the BPCTCA data base, and failed to take into account critical discharge variances between platforms in setting alternative fixed poundage limitations. As to produced water, petitioners argue that the Administrator again excluded relevant data within the record and furthermore relied on new data that was not included in the record.
Petitioners also urge that where standards have been set by the use of “confidence levels,” so that even exemplary plants will meet the limits less than 100 percent of the time, upset provisions must be provided for the percentage of time that the standards will be exceeded. Otherwise, petitioners argue, the permits would be requiring more than the BPCTCA standard demands. In a similar vein, petitioners argue that the permits must allow them to bypass the pollution control equipment during necessary periods of equipment maintenance.
B. Deck Drainage Limitations.
1. BPCTCA.
As noted earlier, the first step in determining effluent limitations is to determine BPCTCA. The Administrator decided that “[ejxemplary flotation systems are representative of the Best Practicable Control Technology Currently Available . for treating deck drainage” and used data from the only three flotation systems currently operating in Cook Inlet to compute the numerical standards contained in the challenged permits. J.R. 1084, 1119-20. Petitioners argue that the Administrator is not supported in this determination by the record.39
According to the Administrator, a study prepared by Brown and Root, Inc., and introduced into the record by petitioners (hereinafter referred to as the Brown and Root Study) “concludes that exemplary flotation systems are representative of [BPCTCA] and provide a reasonable basis for establishing numerical effluent guidelines and limitations.” J.R. 1119. While the Brown and Root Study did conclude that flotation systems are representative of BPCTCA for produced water, the study did not address deck drainage. Furthermore, the Brown and Root Study did not limit BPCTCA to flotation systems, but also included in its definition a large number of other systems, some of which are presently used in Cook Inlet.40 We agree that the Brown and Root Study does not by itself support the Administrator’s determination of BPCTCA for the disposal of deck drainage. It is invalid to conclude from the fact that flotation systems are BPCTCA for pro*1268duced water that they are also representative of BPCTCA for deck drainage, particularly in light of the testimony of several witnesses that BPCTCA depends heavily on the particular influent and situation.41
We believe, however, that there is other substantial evidence in the record which supports the Administrator’s determination of BPCTCA for the treatment of deck drainage.42 In particular, the testimony of at least one of petitioners’ witnesses directly supports the Administrator’s determination. Mr. W. T. Strickland, Senior Research Engineer for Shell Development Co., testified before the EPA that he agreed with the statement of an EPA expert in an earlier hearing that BPCTCA for Cook Inlet Platforms is “systems such as dissolved air or gas flotation incorporating a combination of heat addition and/or chemical addition with surge control and recycle capability.” J.R. 163. Furthermore, the effluent discharge levels for Cook Inlet Platforms that do not use flotation units are considerably higher than those for platforms treating their drainage with flotation systems. See generally J.R. 1057-63, 1243-50. The EPA could conclude from this difference in treatment results that flotation systems represent BPCTCA for deck drainage.43
2. Exclusion of Relevant Data.
Once the Administrator determined that flotation units represent BPCTCA for deck drainage treatment in Cook Inlet, he validly limited the data base used in computing the permit standards to the three platforms in Cook Inlet then using flotation systems.44 Petitioners urge, however, that the Administrator further narrowed the data base by excluding data collected on these platforms during the periods of drilling and follow-up operations such as drill rig moving and cleanup. Since these are normal and necessary operations on petitioners’ platforms, petitioners urge that this exclusion of data was invalid. We agree. Substantial evidence does not support a conclusion that the present permit standards can be met during periods of drilling and cleanup. We therefore remand to the Administrator for reconsideration of the proper effluent discharge limits during these periods of operation.
At oral argument, the EPA asked for a voluntary remand, apparently recognizing their error with respect to at least part of these exclusions. However, the EPA only sought a remand with respect to
“the applicability of deck drainage effluent limitations for oil and grease discharges during periods of drilling. In all other respects, including oil and grease *1269effluent limitations on deck drainage, to the extent they apply during periods of when drilling is not taking place, EPA continues to contend that the limitations are supported by the record.” Letter from Alan W. Eckert, Deputy Associate General Counsel, Water Quality Division, EPA, February 16, 1977, at p. 1.
Our remand is not so limited. The EPA deck drainage standards were computed from a data base that assumed “neither drilling activities or post-drilling cleanup activities.” J.R. 1222 (emphasis added). The record indicates that the EPA not only excluded from the flotation system data base data collected during periods of drilling but also excluded data collected during oil rig moving and various cleanup activities. See J.R. 1222, 1250. These are necessary and normal operations aboard the petitioners’ platforms. See J.R. 133-34. They are also operations that lead to greater oil and grease concentrations in discharged deck drainage. See J.R. 1083. Their exclusion from the data base therefore led to an understating of the level of pollution control achievable by the use of BPCTCA during those periods. On remand, the Administrator must either compute new, more liberal limitations to be applied during periods of drilling, oil rig moving, and cleanup activities,45 or, alternatively, reopen hearings for the purpose of introducing substantial evidence demonstrating that the present limits can be met during these periods.46
3. Poundage Limitations.
The original permits stated the discharge limitations for deck drainage in terms of milligrams per liter. In the past, the EPA has avoided stating discharge standards in terms of milligrams per liter because of their fear that “the manufacturers could [evade the standards] by simply diluting the wastewater.” FMC Corp. v. Train, 539 F.2d 973, 980 n. 10 (4th Cir. 1976). Instead, the EPA has justifiably turned to standards expressed in pounds of pollutant per a specified quantity of product produced.47 See, e. g., id. Recognizing that the same danger exists in the instant dispute, the Administrator wrote a 5/9.25 Ibs/day limitation on deck drainage into the final permits, applicable to all Cook Inlet platforms. J.R. 1121-22. Surprisingly, however, the pounds per day limit is stated in the alternative: the petitioners have their choice of meeting either the milligrams per liter standard or the pounds per day limitation. This form of statement would seem to defeat the entire purpose of using a poundage standard.48
*1270Petitioners, nonetheless, object to the uniform pounds per day standard on the ground that it fails to take into account the variation in effluent flow between the platforms. Typically, as noted above, discharge standards are set in terms of pounds of pollutant per quantity of product produced. Such a standard will to some extent automatically adjust to the variation in manufacturer size and the resulting variation in amount of influents needed to be treated. Here, in comparison, each platform is limited to the same number of pounds of pollutant discharged per day even though some of the platforms will have significantly greater quantities of influent to treat because of significantly larger operations. The EPA itself has recognized this inevitable variance in quantities of influent by authorizing flows of from 18,000 to 200,000 gallons per day. Compare J.R. 1192 with J.R. 1155. A pounds per day limit found to represent BPCTCA for one platform in Cook Inlet will not automatically represent BPCTCA for any other platform. Depending on the influent flow of the other platforms, the standard may prove either too high or too low.49
The courts have been hesitant to approve even standards set in terms of pounds of pollutant per quantity of product produced when the evidence suggests that the influent per quantity of product produced is variable. In FMC Corp. v. Train, supra, the court reviewed a standard written in terms of pounds of pollutant per 1,000 pounds of product produced. The Administrator computed the standard by multiplying a milligrams per liter standard by “the amount of water used by a ‘typical’ plant to manufacture 1,000 pounds of product.” Id. at 980 n.10. The Fourth Circuit remanded on the ground that there was “no uniform water usage per unit of product manufactured.” Id. at 980. A fortiori, we believe that a remand is dictated here.50
The EPA notes, however, that the poundage limitation is stated in the alternative. Since petitioners are free to ignore the pounds per day standard in favor of the milligrams per liter limitation, the EPA reasons, petitioners cannot complain of imperfections in the poundage standard. We disagree. The EPA was free, and is still free upon remand, to insert only a milligrams per liter limitation into the permit. But having chosen to include a poundage limitation, the standard set must be supported by substantial evidence.51
C. Produced Water Limitations.
1. Cases 75-3794 and 75-3796: Offshore platforms.
As described earlier, none of the Cook Inlet platforms currently dispose of produced water waste offshore. Thus, the EPA could not use Cook Inlet data to determine *1271produced water standards for the offshore platforms, as it was able to do in computing deck drainage limits. Instead, the EPA turned to a study of produced water treatment in the Gulf of Mexico prepared by Brown and Root, Inc., the Brown and Root Study mentioned earlier. The Administrator began by adopting the study’s conclusion that flotation systems are representative of BPCTCA for produced water. J.R. 1123. The Administrator then utilized the data contained in the study, plus one year of extra data collected on the same Gulf platforms to compute the permit standards. J.R. 1218-20.
Petitioners do not object to the EPA’s basic approach. They concede that flotation units are representative of BPCTCA for produced water. And they do not object to the use of Gulf Coast data in computing effluent limitations for Cook Inlet. Petitioners do urge, however, that the EPA misanalyzed the Gulf Coast data. First, petitioners contend that relevant data was excluded from the Administrator’s computation. Second, petitioners argue that the Administrator erred in assuming that the data followed a lognormal distribution. More critically, petitioners complain that they were never shown or given an opportunity to comment upon either the extra year of data used by the EPA or the internal EPA memorandum concluding that the data was lognormally distributed.
On the face of the record before us, we find nothing substantively wrong with EPA’s analysis of the Gulf Coast data. EPA’s various adjustments in the Brown and Root data were all valid.52 Similarly, substantial evidence supports the Administrator’s decision to treat the Brown and Root data as lognormally distributed.53 According to a memorandum written by Mr. Gary Liberson, an EPA statistician, to the EPA Administrator, petitioners’ assumption
“that the distribution of the data was unknown . . . was wrong.
“Each of the platforms in the [Brown and Root] data were tested statistically by a method developed by Lilliefors, to determine if it obeyed a lognormal distribution. All but 4 platforms obeyed the lognormal distribution for composite data. . . . The 4 platforms which failed the test obeyed a three parameter lognormal distribution.” J.R. 1220.
Unfortunately, the evidence which supports the Administrator’s decision is seriously marred by various procedural irregularities that have prevented petitioners from effectively commenting on portions of the evidence. First, the EPA failed to reveal to petitioners and still has not revealed, even to this court, the extra year of data that was used in setting the produced water standards. The amount of data, the source of the data, and most importantly the integrity of the data remain unknown. Second, petitioners had no opportunity to comment upon the Liberson memorandum before the EPA. Since the Liberson memorandum concluded for the first time that the Brown and Root data was lognormally distributed, the petitioners should have been extended an opportunity to comment and challenge.54 These procedural defects *1272do not call for a full remand of the produced water standards. However, they do call for a limited remand to the EPA with instructions (a) to reveal the extra year of data that the EPA relied upon in computing the standards and (b) to reopen hearings in order to allow petitioners an opportunity to comment upon and challenge both the data so revealed and the Liberson memorandum. The EPA is then free either to reconfirm or revise the present limits, subject, of course, to further review by this court at the discretion of petitioners.
2. Case 75-3795: Onshore facilities.
The produced water standard for the onshore facilities was computed in exactly the same manner as the standard for the offshore platforms. Petitioners begin, therefore, by raising the same arguments. discussed above in cases 75-3794 and 75-3795. Petitioners, in ease 75-3795, also note that the Administrator’s decision ignores the recommendation of the EPA’s Draft Development Document that the standard be set at 57/85 mg/1. An agency, however, is clearly free to reject an early conclusion after further study and reflection. All that is important is, as developed above, that the final permit standards in this case are supported by substantial evidence. Cf. Consolo v. FMC, 383 U.S. 607, 620, 86 S.Ct. 1018, 16 L.Ed.2d 131 (1966).
As in cases 75-3794 and 75-3796, petitioners were not given an opportunity to comment upon and challenge the Liberson memorandum. A limited remand, therefore, is called for. We instruct the EPA to reveal to petitioners the extra year of data that Liberson relied upon and to reopen hearings in order to allow petitioners an opportunity to comment upon both the data and the Liberson memorandum.
D. Upset Provisions.
As discussed earlier, a facility utilizing BPCTCA in an exemplary fashion can be expected to meet the deck drainage standards 97.5 percent of the time and the produced water standards 99 percent of the time. Conversely, even a facility that employs BPCTCA can be expected to be in violation of the standards at least a percentage of the time. It would be impossible and impracticable to set a standard that could be met 100 percent of the time assuming BPCTCA is employed in an exemplary fashion. This would require an accurate prediction of the innumerable types of upsets that could plague BPCTCA systems. Furthermore, such a standard would probably be so liberal as to be worthless as a control standard. It is better to set a standard by using a 97.5 or 99 percent “confidence interval” and then to address the expected one percent or more of “excursions” when they actually occur, by determining whether exemplary use of BPCTCA could have avoided the excursion. This has been the method adopted by the EPA. If it appears to the EPA that a violation of the standard was not the fault of the permit holder, the EPA will informally exercise its discretion not to prosecute. The issue raised by petitioners is whether a formal “upset provision” must be written into the permits. The EPA has refused to insert an “upset provision” into petitioners’ permits, arguing that excursions can be adequately dealt with informally. Petitioners argue that this is not enough: that the permits must formally provide that upsets beyond the control of the permit holder are not violations of the permit standards.
We agree. The Federal Water Pollution Control Act requires point sources of pollution to utilize the “best practicable control technology currently available” pri- or to 1983. The EPA cannot impose a higher standard without violating the Control Act. And yet the permits as currently *1273written do exactly that. The permits on their face require petitioners to meet the standards 100 percent of the time. But platforms using BPCTCA can only be expected to achieve the effluent standards 97.5 percent of the time in the case of deck drainage and 99 percent of the time in the case of produced water. We, therefore, remand to the EPA with instructions to insert upset provisions into the permits.55
It is not an adequate response that the EPA will informally take BPCTCA into account in deciding whether or not to prosecute “violators.” First, there is no guarantee that the EPA will choose to exercise this discretion. And once a prosecution is brought, the courts have no authority to dismiss the complaint on the grounds that the permit holder could not have avoided the violation. Cf. Getty Oil Co. v. Ruckelshaus, 467 F.2d 349 (3d Cir. 1972), cert. denied, 409 U.S. 1125, 93 S.Ct. 937, 35 L.Ed.2d 256 (1973). Even if we were to assume that the EPA would decline to prosecute in every case of unavoidable excursion, any concerned citizen would still be free to commence an enforcement action against the “violator” under section 505(a)(1) of the Act. “[T]here is no authorization to block a citizen’s suit under section 505 even though the agency believes that the suit should not go forward.” Bethlehem Steel Corp. v. Train, 544 F.2d 657, 660 (3d Cir. 1976).
The EPA complains, however, that formal upset provisions will place an extreme administrative burden on its already pressed resources. We disagree. There is no reason why greater resources need be used under a formal upset provision than under the EPA’s planned informal treatment of upsets, assuming that the informal program is designed to be fair and accurate. The EPA is free in writing the formal upset provision to place the burden on the permit holder of producing relevant data and proving that the upset could not have been prevented.56 Even if there were a large cost to formalizing the upset procedure, EPA’s redress would be in Congress and not here.
A similar conclusion was reached by the Fourth Circuit in FMC Corp. v. Train, 539 F.2d 973 (4th Cir. 1976). In FMC Corp., effluent limitation guidelines and standards were set with the use of confidence intervals in much the same manner as they were here. A plant using BPCTCA would be expected to meet the standards under challenge 97 to 98 percent of the time in the case of the monthly limits and 99.5 to 100 percent of the time in the case of the daily limits. The Fourth Circuit held that the EPA’s failure to include formal upset provisions in the permits was not in accordance with the law and remanded.57
*1274“[T]his court is of the opinion that EPA should provide an excursion provision that will offset the expected 2 to 3 percent and .5 percent violations. Plant owners should not be subject to sanctions when they are operating a proper treatment facility. Such excursions are provided for by the ambient air standards established under the Clean Air Act, 40 C.F.R. §§ 50.4-50.10, and this Court sees no reason why appropriate excursion provisions should not be incorporated in these water pollution regulations.”
Id. at 986.
E. Bypass Provisions.
Petitioners join their request for upset provisions with a plea for more liberal bypass provisions allowing pollution control equipment to be bypassed if “rendered inoperative while repairs or maintenance [required to maintain the equipment in good working order] are being performed.” Petitioners’ Brief in 75-3794 and 75-3796, at 62. In contrast, the present bypass provision only allows bypassing “(i) where unavoidable to prevent loss of life or severe property damage, or (ii) where excessive storm drainage or runoff would damage any facilities necessary for compliance with the effluent limitations and prohibitions of this permit.” See, e. g., J.R. 1185. We do not believe that a more liberal bypass provision is dictated by the record; however, we remand for a clarifying amendment.
It is undisputed that preventative or corrective maintenance will at times be necessary. But such maintenance will typically not require a formal bypass exception to the permit standards. According to the Administrator, “[m]ost preventive maintenance operations can be undertaken while the equipment is operating, and therefore do not require bypassing.” J.R. 1136; see J.R. [75-3795] 612. Even where the equipment must be bypassed, the waste water usually can either be placed in reserve storage (during short-term maintenance) or, in the case of offshore platforms, be diverted to shore for treatment and disposal (during long-term maintenance). See, e. g., J.R. 115-16, 370, 1136-37; J.R. [75-3795] 611, 613. The Administrator does recognize that there may be periods when it will not be feasible to place the waste water in reserve storage or pump it to shore. Under such circumstances, prohibiting the permit holder from bypassing the control facilities may force him to “shut in” his well. See J.R. 1136; see J.R. [75-3795] 614. And as the Administrator concedes, the shutting in of a producing well can “cause significant reservoir damage and result in a permanent loss of recoverable oil and gas.” J.R. 1137. It is this situation that petitioners believe mandates a more liberal bypass provision. The Administrator, on the other hand, believes that this situation is reasonably met already under the current bypass provision. “If and when such conditions do exist, it is not inconceivable that the bypass provision relating to the prevention of severe property damage would apply, depending upon the specific circumstances presented.” J.R. 1138; see J.R. [75-3795] 614. Petitioners object that it is not clear that the present provision covers the situation where the permit holder’s only alternative to bypassing is shutting in the well; they emphasize that the Administrator merely said that it is “not inconceivable” that the present provision would apply.
We believe that the present provision, if interpreted in line with the Administrator’s comments, is adequate to meet the fears of petitioners. The bypassing of pollution control equipment and effluent discharge standards should only be allowed where the failure to bypass would impose severe costs on the permit holder. This' is what the present provision provides. However, we recognize that the present provision is vague and may be seen as ambiguous as to whether it applies when the only alternative to bypassing is “shutting in” the well. Upon remand, therefore, the present provision should be amended to clarify that the shutting in of a well, under given circumstances, can constitute “severe property damage” and, if the only alternative, that *1275such shutting in can permit the bypassing of the pollution control system.
IV.
Summary.
In conclusion, a limited remand is needed. In cases 75-3794 and 75-3796, the deck drainage limitations are remanded
(1) for reconsideration of whether the limitations are appropriate during periods of drilling, oil rig moving, and cleanup activities; and
(2) for elimination of the poundage limitations. The EPA is free to insert new poundage standards but only after the introduction of supporting evidence in the context of an adjudicatory hearing.
In all three cases, the produced water limitations are remanded for exposition of the new year of data apparently relied upon in the Liberson memorandum and for new hearings which give petitioners an opportunity to comment upon and challenge both the data and the Liberson memorandum itself. Finally, all permits are remanded
(1) for inclusion of upset provisions, and
(2) for clarification of the bypass provisions.
REMANDED.
. 33 U.S.C.A. § 1342 (1977 Supp.).
. Id. §§ 1311(b)(1) and 1314(b)(1).
. Nine Cook Inlet platforms are involved in the instant dispute — Shell Platforms A and C; Marathon’s Dolly Varden Platform; Union’s Grayling and Monopod Platforms; Atlantic Richfield’s King Salmon, Spark and A Platforms; and Mobil’s Granite Point Platform.
. The Atlantic Richfield and Union platforms pump their waste to shore, while the Marathon, Mobil, and Shell platforms treat and dispose of deck drainage offshore.
. As explained by Mr. C. P. Falls, Environmental Conservation Manager for Atlantic Rich-field, “as the volume of water increases throughout the life of the field and the water to oil ratio increases, the potentiality of freezing of the water in the pipeline and plugging the pipeline increases.
“Also there’s a possibility that we can exceed the capacity of our pipeline going to shore, and if the oil is excluded to make room for the water or production is cut back, this would represent energy that may never be recovered.” Joint Record at 593.
. According to respondent, Shell Oil Company has never applied for nor received a permit for the discharge of produced water from its Cook Inlet platforms.
. 33 U.S.C.A. §§ 1251 et seq. (1977 Supp.).
. Id. § 1342.
. Id. § 1342(a).
. Id. § 1311.
. Id. § 1311(b)(1)(A).
. Id. § 1314(b)(1).
. 5 U.S.C. §§ 554, 556 & 557 (1970).
. 33 U.S.C. § 407 (1970). See Executive Order No. 11574, 35 F.R. 19627 (1970).
. See section 402(a)(5) of the Federal Water Pollution Control Act, 33 U.S.C.A. § 1342(a)(5) (1977 Supp.).
. The “daily average” standard sets a limit on the arithmetic average of all the daily determinations of concentrations made during a calendar month. The “daily maximum” standard sets an upper limit on the arithmetic average of all grab samples taken in any one day.
. Pursuant to the request of this court, the parties submitted joint appendices which they stipulate contain all relevant portions of the record. This joint record in cases 75-3794 and 75-3796 is referenced as “J.R.” In case 75-3795, the joint record is referenced as “J.R. [75-3795].”
. Petitioners have set no specific date to begin discharging produced water offshore, but merely recognize the ultimate necessity of turning to offshore disposal. See note 5 supra.
. According to respondent, the “submission of poundage data [by a Shell expert], coupled with the statement of Shell’s counsel at oral argument before the Chief Judicial Officer that he would have no problem with a pounds limitation, encouraged the Agency to compute poundage values for the Mobil data and to express the deck drainage limitations in terms of allowable quantities of pollutant discharged as well as in terms of allowable concentrations.” Respondent’s Brief in 75-3794 and 75-3796, at 28 (citations to the record omitted). As discussed infra at note 48, the policy reason for inserting the alternative poundage standard is not clear.
. As for the Regional Administrator’s conclusion that petitioners’ plans to discharge produced water offshore were too “uncertain,” the Administrator responded that there is “no reasonable basis for failing now to specify produced water limitations for offshore discharges if and when they do occur. To conclude otherwise, when a basis now exists for reaching these determinations, would add needlessly to the already protracted nature of this proceeding and further delay a comprehensive control program for the entire production system.” J.R. 1129.
. Alaska requested that the following specific upset provision be added to the permits: “It is recognized that influent quality changes, equipment malfunctions, or uncontrollable circumstances may sometimes result in effluent concentrations exceeding the permit limitations. The permittee may demonstrate to EPA that such circumstances exist in any case where effluent concentrations exceed those set forth in this permit and they are not of a recurring nature. The Regional Administrator shall consider such showing in determining permit violations and for enforcement purposes. EPA does not waive any of its legal rights during such consideration.” J.R. 1174.
. 5 U.S.C. §§ 554, 556 & 557 (1970).
. Id. §§ 554(d) & 557.
. Id. § 556(e).
. The Seventh Circuit ultimately rested its holding on section 558(c) of the APA, 5 U.S.C. § 558(c) (1970), which reads in relevant part:
“When application is made for a license required by law, the agency, with due regard for the rights and privileges of all the interested parties or adversely affected persons and within a reasonable time, shall set and complete proceedings required to be conducted in accordance with sections 556 and 557 of this title *1261or other proceedings required by law and shall make its decision.” According to the Seventh Circuit, this section requires full adjudicatory hearings to be held whenever “application is made for a license required by law.” At p. 833. We disagree. Section 558(c) merely requires that any proceeding required to be conducted in accordance with sections 556 and 557, or any other proceeding required to be held by law, must be set and completed and a decision made, all within a reasonable time and with due regard for the interests of the parties. Section 558(c) does not independently provide that full adjudicatory hearings must be held. It merely requires any adjudicatory hearings mandated under other provision of law to be set and completed in an expeditious and judicious manner.
Our interpretation of section 558(c) is supported by both the legislative history of the APA and judicial precedent. According to a Senate Judiciary Committee print, the quoted language merely “requires that applications be determined promptly.” Legislative History of the Administrative Procedure Act, 79th Cong., 2d Sess., at 35 (1946). The House Judiciary Committee commented on an early version of section 558(c) that “it does not provide for a hearing where other statutes do not do so.” Id. at 275. To similar effect is the Attorney General’s Manual on the APA, published shortly after passage of the APA. “The import of this sentence is that an agency shall hear and decide licensing proceedings as quickly as possible. Should the licensing proceedings be required by statute to be determined upon the record after opportunity for an agency hearing [the language of section 554(a)], an agency will be required to follow the provisions as to hearing and decision contained in [sections 556 and 557], As to other types of licensing proceedings, the Act does not formulate any fixed procedure (just as no fixed procedure has been formulated for adjudications other than those that are required by statute to be determined on the record after opportunity for an agency hearing).” Attorney General’s Manual on the Administrative Procedure Act 89-90 (1947) (emphasis added). Many courts have held that licensing proceedings not satisfying the requirements of section 554(a) of the APA need not observe the adjudicatory hearing requirements of the APA. The only court previously to rule directly on the interpretation of § 558(c) adopted by the Seventh Circuit rejected it without discussion. See Lincoln Transit Co. v. United States, 256 F.Supp. 990, 994 (S.D.N.Y.1966) (3-judge court).
. The Supreme Court so recognized in Morgan v. United States, 298 U.S. 468, 56 S.Ct. 906, 80 L.Ed. 1288 (1936). The Court there described such “quasi-judicial” proceedings as characterized by “the taking and weighing of evidence, determinations of fact based upon the consideration of the evidence, and the making up of an order supported by such findings.” Id. at 480, 56 S.Ct. at 911.
. While factual inferences must often be made in rulemaking proceedings, they are “used in the formulation of a basically legislative-type judgment, for prospective application only, rather than in adjudicating a particular set of disputed facts.” United States v. Florida East Coast Ry. Co., 410 U.S. 224, 246, 93 S.Ct. 810, 821, 35 L.Ed.2d 223 (1973).
. The Supreme Court has stated that a section 402 permit “serves to transform generally applicable effluent limitations . . . into the obligations (including a timetable for compliance) of the individual discharger.” E.P.A. v. California ex rel. State Water Resources Control Brd., 426 U.S. 200, 205, 96 S.Ct. 2022, 2025, 48 L.Ed.2d 578 (1976).
. Cf. United States v. Florida East Coast Ry. Co., 410 U.S. 224, 247, 93 S.Ct. 810, 822, 35 L.Ed.2d 223 (1973) (Douglas, J., dissenting): “[I]t is [not] within our traditional concepts of due process to allow an administrative agency to saddle anyone with a new rate, charge, or fee without a full hearing that includes the right to present oral testimony, cross-examine witnesses, and present oral argument.”
. The Supreme Court has held that the presence or absence of the phrase “on the record” is extremely relevant in determining under section 553 of the APA whether particular rule-making proceedings must be conducted in accordance with sections 556 and 557. See United States v. Florida East Coast Ry. Co., 410 U.S. 224, 93 S.Ct. 810, 35 L.Ed.2d 223 (1973); United States v. Allegheny-Ludlum Steel Corp., 406 U.S. 742, 92 S.Ct. 1941, 32 L.Ed.2d 453 (1972). However, the Court made it clear that its ruling was limited to “exercises of legislative rulemaking power rather than adjudicatory hearings.” 406 U.S. at 757, 92 S.Ct. at 1950. This dichotomy is supported by the Attorney General’s Manual on the APA. “With respect to rule making, it was concluded, supra, that a statutory provision that rules be issued after a hearing, without more, should not be construed as requiring an opportunity for the expression of views. That conclusion was based on the legislative nature of rule making, from which it was inferred, unless a statute requires otherwise, that an agency hearing on proposed rules would be similar to a hearing before a legislative committee, with neither the legislature nor the agency being limited to the material adduced at the hearing. No such rationale applies to administrative adjudication.” Attorney General’s Manual, supra note 26, at 42—43.
. According to the Attorney General’s Manual on the APA, it should be “assumed that where a statute specifically provides for administrative adjudication . . . after opportunity for an agency hearing, such specific requirement for a hearing ordinarily implies the further requirement of decision in accordance with evidence adduced at the hearing. Of course, the foregoing discussion is inapplicable to any situation in which the legislative history or the context of the pertinent statute indicates a contrary congressional intent.” Id. at 43 (citations to legislative history omitted).
. It is conceivable that Congress chose to leave the question of whether hearings under section 402 should be conducted in accordance with the APA’s adjudicatory provisions for the courts to decide once experience indicated whether section 402 proceedings developed to be more adjudicatory or rulemaking in character.
. We might also observe that a major purpose of the APA “was to introduce greater uniformity of procedure and standardization of administrative practice among the diverse agencies whose customs had departed widely from each other. We pursue this no further than to note that any exception we may find to its applicability would tend to defeat this purpose.” Wong Yang Sung v. McGrath, 339 U.S. 33, 41, 70 S.Ct. 445, 450, 94 L.Ed. 616, modified, 339 U.S. 908, 70 S.Ct. 564, 94 L.Ed. 1336 (1950).
. It appears that the Administrator departed twice from the record in making his decisions. The Administrator heavily relied on an internal EPA memorandum that used new data not contained in the record and concluded that both the new and old data were lognormally distributed. Also, the Administrator used the record of the Shell hearing in making his decisions in the consolidated hearing, and vice versa. Petitioners, however, do not object to the latter excursion. Petitioners’ Reply Brief in 75-3794 and 75-3796, at 13.
. “[Substantial evidence was] defined by the Supreme Court even before adoption of the Administrative Procedure Act. It is evidence that affords ‘a substantial basis of fact from which the fact in issue can be reasonably inferred .... [I]t must be enough to justify, if the trial were to a jury, a refusal to direct a verdict when the conclusion sought to be drawn from it is one of fact for the jury.’ ” Watson v. Gulf Stevedore Corp., 400 F.2d 649, 652 (5th Cir. 1968), cert. denied, 394 U.S. 976, 89 S.Ct. 1471, 22 L.Ed.2d 755 (1969), quoting from N.L.R.B. v. Columbian Enameling and Stamping Co., 306 U.S. 292, 299-300, 59 S.Ct. 501, 83 L.Ed. 660 (1939).
. “Confidence intervals” are necessary because of the variance in emission quality from data sample to data sample. If the emission quality of each plant were constant (e. g., every data sample showed effluent quality of x mg/1), a standard could be computed by merely averaging each plant’s effluent quality.
. A basic discussion of confidence intervals and their use is found in A. Spurr & C. Bonini, Statistical Analyses for Business Decisions 235-39 (rev. ed. 1973).
. Such an attack would not be successful. Courts have frequently approved the EPA’s analytical approach in this area. See, e. g., American Petroleum Inst. v. EPA, 540 F.2d 1023, 1034 (10th Cir. 1976); Hooker Chemicals & Plastics Corp. v. Train, 537 F.2d 620, 632 (2d Cir. 1976); American Iron & Steel Inst. v. EPA, 526 F.2d 1027, 1057 (3d Cir. 1975); American Meat Inst. v. EPA, 526 F.2d 442, 453 (7th Cir. 1975).
. The issue is critical' since almost 90 percent of the effluent discharge data available on Cook Inlet operations comes from platforms not employing flotation systems. If the exclusion is unwarranted, therefore, it could lead to seriously biased results.
. The Brown and Root Study specifically found that “BPCTCA for produced brine treatment is either gravity separation (tank or pit), other physical separation (flotation system, plate coalescer, fibrous media coalescer or loose media coalescer) or a combination of gravity separation and other physical separation. The performance of exemplary flotation systems was selected as the basis for establishing numerical effluent concentration guidelines.” J.R. 710 (emphasis added).
It is questionable, however, whether much emphasis should be placed on the inclusion within BPCTCA of control processes other than flotation systems since the Brown and Root Study goes on in the next paragraph to note that “[l]oose media coalescers and flotation units were found to produce the lowest mean effluent oil concentrations of the processes studied. The flotation process was selected to represent exemplary produced brine treatment performance. Flotation is a widely used process [sic] and is almost universally applicable. Loose media coalescers were not selected to represent exemplary performance because of operation problems at many locations.” Id. Substantial evidence therefore supports the EPA’s conclusion that only flotation units are BPCTCA. Again, however, on the basis of the Brown and Root Study alone, this conclusion can only be applied to produced water treatment.
. See, e. g., J.R. 541 (“I would not recommend any specific methods for Cook Inlet platforms until it had been tried in situ [sic]”); J.R. 546 (it is “possible that a gravity separator may give lower numbers ... in treating influents which have changing characteristics than a flotation cell”); J.R. 595 (flotation units are not “a panacea for all situations”). Testimony by an author of the Brown and Root Study further weakens the Administrator’s reliance on the Study’s limited conclusion. In estimating deck drainage limits that he believed to be achievable in Cook Inlet by the use of BPCTCA, Mr. Edward Sebesta, Process Engineer for the Brown and Root Study, apparently utilized data obtained not only from Cook Inlet platforms using flotation systems but also from platforms utilizing gravity separators. J.R. 687-95.
. The “possibility of drawing two inconsistent conclusions from the evidence does not prevent an administrative agency’s finding from being supported by substantial evidence.” Consolo v. FMC, 383 U.S. 607, 620, 86 S.Ct. 1018, 1026, 16 L.Ed.2d 131 (1966).
. “The Administrator should establish the range of best practicable levels based upon the average of the best existing performances by plants of various sizes, ages, and unit processes within each industrial category.” S.Rep.No. 92—414, 92d Cong. 1st Sess., at 50 (1971), reprinted in Legislative History of the Federal Water Pollution Control Act, at 1468 (emphasis added). In essence, petitioners’ objection boils down to an attack on EPA’s practice of only considering data from plants operating in an apparently exemplary fashion, a practice that has been frequently upheld. See, e. g., FMC Corp. v. Train, 539 F.2d 973, 978 (4th Cir. 1976); Hooker Chemicals and Plastics Corp. v. Train, 537 F.2d 620, 632 (2d Cir. 1976).
. The three platforms whose data were used were the Mobil Granite Point Platform and Shell Platforms A and C.
. The Administrator is free to introduce new data or other evidence in computing these new limits. However, petitioners must be given an opportunity to comment upon and challenge such new material in the setting of an adjudicatory hearing.
. Petitioners also argue that the Administrator only considered data collected during late spring and early summer, when effluent quality might be expected to be higher than during other times of the year. What little relevant evidence appears in the record suggests that lower effluent quality can be expected in the winter months when cold temperatures are experienced and the waste stream includes rain and snow melt which do not contain electrolytes. See, e. g., J.R. 163-64, 613. While the only Shell data considered by the EPA was collected during the late-spring and summer months (winter data from the Shell platforms was validly excluded because of crankcase-oil contamination, which is a correctable problem), a majority of the Mobil data points used in the permit computations was collected in late winter and early spring. J.R. 1060, 1222. Since the Mobil and Shell data were analyzed separately and then averaged, J.R. 1222, we believe that sufficient consideration was given to winter data in computing the effluent limitations. Substantial evidence supports the Administrator’s conclusion that the permit standards can be met during all periods of the Alaskan year.
. An alternative solution is to impose limits not only on the pollution concentration but also on the effluent flow. This solution was apparently adopted in writing the permits under challenge here. See, e. g., J.R. 1155.
. The only rationale suggested in the EPA’s brief for including the alternative poundage standard was, as suggested in the text, to avoid possible evasion of the milligrams per liter standard by diluting the waste water. Respondent’s Brief in 75-3794 and 75-3796, at 27-28. In his decision, the Administrator cryptically suggested that the alternative pounds-per-day standard was included because it “reflects the variability in deck flow and oil mix, and there*1270fore is a preferred limitation for deck drainage.” J.R. 1121.
. The invalidity of applying a pounds-per-day standard uniformly to all Cook Inlet platforms is demonstrated in the internal EPA memorandum computing the permit standards. For the milligrams per liter standard, the 97.5 percent confidence limit was 83/111 mg/1 for the Shell data but only 46/64 mg/1 for the Mobil data; these figures would indicate that the Shell platforms showed lower effluent quality. However, the computations of the poundage standard would indicate that the Mobil platform shows a lower effluent quality. The 97.5 percent confidence level was 7.3/11.8 lbs/day for Mobil but only 2.7/6.7 lbs/day for Shell. This inconsistency can be at least partly explained by the lower quantity of effluent discharged by the Shell platforms.
. Even if a remand were not required by the EPA’s failure to take into account the variability in deck drainage flow, a remand might be dictated by EPA’s failure to reveal clearly how it computed the 5/9.25 lbs/day limit contained in the permits. In particular, no indication is given of how the Mobil mg/1 data was converted to lbs/day data for purposes of computing the poundage standard. See generally 1222, 1224, 1232-35. The administrative record must clearly reveal both the data used and the analysis followed in setting standards. See FMC Corp. v. Train, 539 F.2d 973, 981 (4th Cir. 1976) (failure to disclose flow figures used to compute poundage standard requires remand).
. If the EPA decides, on remand, to insert new poundage standards into the permits, new hearings will of course be required to introduce any supporting evidence and to allow petitioners the opportunity of rebuttal.
. Excluded from the Brown and Root data were effluent samples of less than 10 data points, “outliers” (data that appear to be statistical aberrations), data points collected during periods of flotation unit startup, and data points collected during an episode of heatertreater malfunction. The EPA’s exclusion of small samples and outliers was statistically valid. Cf. FMC Corp. v. Train, 539 F.2d 973, 986 (4th Cir. 1976). Since the petitioners’ permits include a 60-day grace period after startup to allow petitioners to come into compliance with the permit standards gradually, exclusion of the startup data was not invalid. Any startup difficulties not covered by the grace period should be met by the formal upset provisions mandated infra. Finally, the Administrator validly found that heater-treater malfunctions were a preventable problem. Thus, data points stemming from such malfunctions were properly excluded.
. Furthermore, in most cases, “choice of statistical methods is a matter best left to the sound discretion of the Administrator.” FMC Corp. v. Train, 539 F.2d 973, 986 (4th Cir. 1976).
. Such comment is often an invaluable source of information to a reviewing court attempting to evaluate complex statistical and technological decisions. Furthermore, the Administrator *1272himself might have been dissuaded by the petitioners’ comments. As Judge Bazelon has noted in a similar context, “If we were to [require procedures] that open the Administrator’s decision to challenge and force him to respond, we could rely on an informed ‘market’ rather than on our own groping in the dark to test the validity of that decision.” International Harvester Co. v. Ruckelshaus, 155 U.S.App.D.C. 411, 448, 478 F.2d 615, 652 (1973) (Bazelon, C. J., concurring).
. If, as would appear unlikely, the drafting of the upset provisions requires the resolution of factual issues involving new evidence, adjudicatory hearings will, of course, have to be reopened for the introduction of the evidence.
. Of course, such burden cannot be unreasonable.
. See also Essex Chem. Corp. v. Ruckelshaus, 158 U.S.App.D.C. 360, 365-366, 486 F.2d 427, 432-33 (1973), cert. denied, 416 U.S. 969, 94 S.Ct. 1991, 40 L.Ed.2d 558 (1974) (“variant provisions appear necessary to preserve the reasonableness of the standards as a whole . [T]he record does not support the ‘never to be exceeded’ standard currently in force”) Portland Cement Ass’n v. Ruckelshaus, 158 U.S.App.D.C. 308, 332, 486 F.2d 375, 398-99 n. 91 (1973), cert. denied, 417 U.S. 921, 94 S.Ct. 2628, 41 L.Ed.2d 226 (1974) (informal treatment of upsets is inadequate; “[cjompanies must be on notice as to what will constitute a violation”).
The cases cited by EPA are inapposite. In CPC Int., Inc. v. Train, 540 F.2d 1329 (8th Cir. 1976), the petitioner was not arguing for a formal upset provision; instead, the petitioner wanted a higher standard to be set that took into account the possibility of upsets. As noted earlier, the EPA’s decision to set the instant standards at 97.5 and 99 percent confidence level is not unreasonable. Its failure to include a formal upset provision, however, violates the BPCTCA standard. In American Petroleum Institute v. EPA, 540 F.2d 1023, 1036 (10th Cir. 1976), the petitioners did not request a formal upset provision excusing technical violations of the standards when the violation is shown to have resulted from no fault of the permit holder. Instead, petitioners requested that they be permitted to exceed the standards four days a year no matter what the reason for the excursions. The Tenth Circuit validly rejected this request since the “temptation to store pollutants for future discharge would be enticing.” Id. No such temptation is posed by the upset provisions requested in the instant case.