STEPHEN F. WILLIAMS, Circuit Judge:

Representatives of labor and industry challenge a regulation of the Occupational Safety and Health Administration, "Control of Hazardous Energy Sources (Lockout/Tagout)". 54 Fed.Reg. 36,644 (1989). The regulation deals not with the effects of such subtle phenomena as electrical energy fields but with those of ordinary industrial equipment that may suddenly move and cut or crush or otherwise injure a worker. "Lockout" and "tagout" are two procedures designed to reduce these injuries. Lockout is the placement of a lock on an "energy isolating device", such as a circuit breaker, so that equipment cannot start up until the lock is removed. See 29 CFR § 1910.147(b) (1990). Tagout is the similar placement of a plastic tag to alert employees that the tagged equipment "may not be operated" until the tag is removed. See id. Although OSHA had previously issued specific standards governing especially dangerous equipment, the present rule extends lockout/tagout to virtually all equipment in almost all industries. See 29 CFR § 1910.147(a)(1)(ii) (1990). It generally requires employers to use lockout procedures during servicing and maintenance, unless the employer can show that tagout will provide the same level of safety. See id. § 1910.147(c)(2)(ii).

The first issue we address is the claim of petitioner UAW that § 6(b)(5) of the Occupational Safety and Health Act, 29 U.S.C. § 655(b)(5) (1988), provides the statutory criteria for the lockout rule. Its claim is that the hazard involved is a "harmful physical agent" as that term is used in the first sentence of § 6(b)(5), and that, even if the first sentence of § 6(b)(5) is not applicable, the remaining sentences are. OSHA resists both theories, and we find its interpretation reasonable.

The exclusion of § 6(b)(5) from the picture takes us to the claim of the National Association of Manufacturers that Congress has given so little guidance for rules issued under § 6(b) but not covered by § 6(b)(5) that as to such rules the Act invalidly delegates legislative authority. Although we reject that claim, we find that the interpretation offered by the Secretary is, in light of nondelegation principles, so broad as to be unreasonable. We note, however, the existence of at least one interpretation that is reasonable and consistent with the nondelegation doctrine.

Thus, after addressing some objections to the rule that appear likely to survive any reasonable interpretation the Secretary may adopt, we remand the case to the Secretary for further consideration.

I

Section 6(b)(5) of the Act limits the Secretary's discretion when he is promulgating standards that deal with "toxic materials or harmful physical agents". He must adopt "the standard which most adequately assures, to the extent feasible, . . . that no employee will suffer material impairment of health or functional capacity." Id. The Supreme Court has interpreted this language to require that the proposed standard be both technologically and economically "feasible", American Textile Mfrs. Inst., Inc. v. Donovan, 452 U.S. 490, 101 S.Ct. 2478, 69 L.Ed.2d 185 (1981) ( "Cotton Dust"), a criterion the Court appeared to regard as satisfied so long as the costs of a standard would not "threaten the competitive stability of an industry", id. at 530 n. 55, 101 S.Ct. at 2501 n. 55. OSHA and the courts have since embellished that concept. See, e.g., National Cottonseed Products Ass'n v. Brock, 825 F.2d 482, 487-88 (D.C. Cir. 1987). The union argues that § 6(b)(5) applies to this case. We agree with OSHA that it does not.

OSHA interprets § 6(b)(5) as applicable only to "health" standards. It views these as coextensive with standards governing latent hazards, such as carcinogens, "which are frequently undetectable to the casual observer because they are subtle or develop slowly or after latency periods", Brief of OSHA at 24, and contrasts them with "safety" standards, such as the lockout regulation, which address hazards that cause immediately visible physical harm. We accord considerable weight to an agency's construction of a statutory scheme it is entrusted to administer, rejecting it only if unreasonable. See Chevron U.S.A., Inc. v. NRDC, 467 U.S. 837, 843, 104 S.Ct. 2778, 2781, 81 L.Ed.2d 694 (1984).

The union plays a dictionary game to support its view, noting definitions of "physical" as "of or pertaining to matter or energy", and of "agent" as "an active force or substance producing an effect". The lockout rule of course relates to both "matter" and "energy" and controls their "effects". But it is hard to imagine a workplace hazard within Congress's reach that involves neither matter nor energy, and that produces no "effect". Indeed, the union's notion of physical agents would engulf § 6(b)(5)'s companion term, "toxic materials". More important, the union's reading would obliterate a distinction that Congress drew between "health" and "safety" risks. It referred in the Act's preamble to efforts aimed at

exploring ways to discover latent diseases, establishing causal connections between diseases and work in environmental conditions, and conducting other research relating to health problems, in recognition of the fact that occupational health standards present problems often different from those involved in occupational safety. . . .

29 U.S.C. § 651(b)(6) (1988) (emphasis added).

Other sections of the Act confirm the more limited reading for "harmful physical agents". Section 8(c)(3), 29 U.S.C. § 657(c)(3) (1988), for instance, uses the phrase "toxic materials or harmful physical agents" in association with words that make sense primarily (if not exclusively) for the sort of gradually accumulating hazards depicted by OSHA. It speaks of records of workers' "exposures" to "harmful physical agents", and of notice to workers when their exposure is "in concentrations or at levels" exceeding those of a standard. Similarly, § 20(a)(3), 29 U.S.C. § 669(a)(3) (1988), directs the Secretary of Health and Human Services to develop criteria for safe "exposure levels" for "toxic materials and harmful physical agents and substances". And § 20(a)(5), 29 U.S.C. § 669(a)(5) (1988), authorizes regulations for reporting workers' "exposure" to hazardous "substances or physical agents" and for medical exams and tests to develop information on the subject. OSHA's idea of harmful physical agents fits all these terms well, while many hazards covered by the union's reading do not fit them at all.

The union extracts an argument from § 6(c)(1), which grants the Secretary authority to promulgate emergency standards when he finds that "employees are exposed to grave danger from exposure to substances or agents determined to be toxic or physically harmful or from new hazards". See 29 U.S.C. § 655(c)(1)(A) (1988) (emphasis added). It says that this section "plainly encompasses the kind of harm at issue here", and since it uses the "physically harmful" phrase, the use of similar language in § 6(b)(5) should be construed broadly. See Brief for Union at 31. Neither element of the linkage is sound. First, we think it more likely that § 6(c)(1) applies only to new hazards or to materials whose hazardous character is newly discovered. Such revelations seem improbable about saw blades, drill presses or pipes filled with steam. Second, even if § 6(c)(1) did apply to obvious safety hazards, the language in § 6(b)(5) is not identical, and Congress could well have believed that a distinction between health and safety hazards that was useful for permanent regulations was unnecessary for temporary emergency standards.

The legislative history also supports OSHA. In an early version of the bill § 6(b)(5) did apply to all occupational safety and health standards. An amendment added the language restricting its application to "standards dealing with toxic materials or harmful physical agents". Leg. Hist. at 502. This amendment was a pointless exercise if the term "harmful physical agent" embraces all health and safety risks.

Legislative discussion reflects the same distinction. For example, Senator Williams noted specifically that, as to "toxic materials and harmful physical agents", workers "are often unaware of the nature of such exposure or its extent." Leg.Hist. at 415. Throughout the debate, the terms "toxic materials" and "harmful physical agents" were used together, the former to refer to carcinogens, poisons and the like and the latter to cover hazards such as extreme noise and vibration. See, e.g., S.Rep. No. 1282, 91st Cong., 2d Sess. 2-3 (1970), U.S. Code Cong. Admin.News 1970, pp. 5177, 5179, reprinted in Leg.Hist. at 142-43 (discussing 1967 Surgeon General study that found that 65% of employees in industrial plants "were potentially exposed to harmful physical agents, such as severe noise or vibration, or to toxic materials") (emphasis added); Leg.Hist. at 412 (remarks of Sen. Williams) (same); id. at 446 (remarks of Sen. Harris) (speaking of workers' exposure to "harmful physical effects such as severe noise and toxic materials") (emphasis added); id. at 516 (remarks of Sen. Hart) (similar); id. at 845 (H.R. Rep. No. 91-1291) (similar).

There is no warrant for the union view that obviously dangerous industrial machines fall within the statutory definition of "harmful physical agents". The Secretary's decision that the lockout regulation does not constitute regulation of a "harmful physical agent" is reasonable.

The union next argues that even if the first sentence of § 6(b)(5) does not apply to the lockout regulation, the rest — with an allusion to "feasibility" at least as a relevant factor — does. A footnote in Industrial Union Dept., AFL-CIO v. American Petroleum Institute, 448 U.S. 607, 100 S.Ct. 2844, 65 L.Ed.2d 1010 (1980) ( "Benzene"), left the matter open, noting that "the term `subsection' used in the second sentence" might refer "to the entire subsection 6(b) (which sets out procedures for the adoption of all types of health and safety standards), [or] simply to the toxic materials subsection, § 6(b)(5)." Id. at 612 n. 1, 100 S.Ct. at 2849 n. 1 (emphasis added). As this footnote suggests, there is a grammatical argument that Congress's use of the term "subsection" in the second sentence of § 6(b)(5), as opposed to "subsubsection", argues for reading the latter portion of § 6(b)(5) as universally applicable. But in Moore v. District of Columbia, 907 F.2d 165, 171-72 (D.C. Cir. 1990) ( en banc), we declined to place serious weight on Congress's use of "subsection" instead of "subsubsection", viewing it as unlikely to have been a focus of drafting attention. At most, its use creates ambiguity. OSHA's view that Congress intended all of § 6(b)(5) to apply only to latent-hazards standards must prevail if reasonable, see Chevron, 467 U.S. at 844, 104 S.Ct. at 2782, and we agree with OSHA and with National Grain Feed Ass'n v. OSHA, 866 F.2d 717, 732-33 (5th Cir. 1989), that it is reasonable to conclude that the two sentences do not reach beyond toxic materials and "harmful physical agents".

OSHA's view, that the two sentences simply amplify the first sentence's limits on the Secretary's authority over those hazards, fits with Congress's decision to introduce the paragraph with a sentence referring only to the latent-hazard subset. It is also consistent with the Supreme Court's unitary treatment of § 6(b)(5). The Benzene plurality, for example, in justifying its conclusion that § 3(8)'s definition imposed substantive constraints, reasoned that otherwise "there would be no statutory criteria at all to guide the Secretary in promulgating either national consensus standards or permanent standards other than those dealing with toxic materials and harmful physical agents." 448 U.S. at 640 n. 45, 100 S.Ct. at 2863 n. 45 (emphasis added); see also id. at 642, 100 S.Ct. at 2864 ("The standards promulgated pursuant to § 6(b)(5) are just one species of the genus of standards governed by the basic requirement [of Section 3(8)]."); Cotton Dust, 452 U.S. at 513-14 n. 32, 101 S.Ct. at 2492-93 n. 32 (quoting Benzene footnote 45); id. at 512, 101 S.Ct. at 2492 (saying that because health standards presented special problems Congress chose in § 6(b)(5) to impose separate requirements for "a subcategory of occupational safety and health standards dealing with toxic materials and harmful physical agents").

The substance of the two sentences' requirements also supports OSHA. Standards are to be based upon "research, demonstrations, experiments, and such other information as may be appropriate"; the Secretary may consider a number of factors including "the latest available scientific data in the field" and "experience gained under this and other health and safety laws." 29 U.S.C. § 655(b)(5) (1988). Concern for "scientific data" and "experiments" makes complete sense for regulation of carcinogens but sounds out of place when the hazard is a bacon-slicing machine, a spinning saw blade or a moving conveyor. For them, actual injury rates seem far more likely to be controlling.

If the second and third sentences of § 6(b)(5) provided a clear criterion for rules not covered by the first sentence, one might argue that the advantage of allowing them to perform that service should overcome the arguments from language and context. But in fact they merely list conflicting factors, including "the highest degree of health and safety protection", "feasibility", "experience" and "such other information as may be appropriate". Thus they do not provide much of a benchmark. Again, the Secretary's interpretation is a reasonable one.

The union's final theory is that § 6(b)(5) must state the substantive test for all permanent standards, since § 3(8) (our other candidate for limiting the Secretary's discretion) merely sets a definition. In fact, however, the Supreme Court has read § 3(8) as a source of substantive criteria. See Benzene. We address its meaning in detail below.

We uphold the Secretary's conclusion that § 6(b)(5) does not govern occupational safety standards that regulate hazards causing immediately noticeable physical harm.

II

The removal of § 6(b)(5) as a direct constraint on OSHA regulations outside the area of toxics (the term we use hereafter as shorthand for "toxic materials or harmful physical agents") gives point to the NAM's claim of an excessive delegation of legislative power. The only evident source of constraints remaining is § 3(8). It defines an "occupational safety and health standard" as

a standard which requires conditions, or the adoption or use of one or more practices, means, methods, operations, or processes, reasonably necessary or appropriate to provide safe or healthful employment and places of employment.

29 U.S.C. § 652(8) (1988) (emphasis added). Though the language is exceedingly vague, the Benzene plurality found it the source of a threshold requirement of "significant risk", without which OSHA was not to act under § 6(b) at all. It justified this narrowing construction with the argument (among others) that otherwise "the statute would make such a `sweeping delegation of legislative power' that it might be unconstitutional under the Court's reasoning in A.L.A. Schechter Poultry Corp. v. United States, 295 U.S. 495, 539 [ 55 S.Ct. 837, 847, 79 L.Ed. 1570 (1935)] and Panama Refining Co. v. Ryan, 293 U.S. 388 [ 55 S.Ct. 241, 79 L.Ed. 446]." 448 U.S. at 646, 100 S.Ct. at 2866.

The Benzene construction was, of course, a manifestation of the Court's current general practice of applying the nondelegation doctrine mainly in the form of "giving narrow constructions to statutory delegations that might otherwise be thought to be unconstitutional." Mistretta v. United States, 488 U.S. 361, 373 n. 7, 109 S.Ct. 647, 655 n. 7, 102 L.Ed.2d 714 (1989); see also National Cable Television Ass'n v. United States, 415 U.S. 336, 342, 94 S.Ct. 1146, 1149, 39 L.Ed.2d 370 (1974); FPC v. New England Power Co., 415 U.S. 345, 94 S.Ct. 1151, 39 L.Ed.2d 383 (1974); Kent v. Dulles, 357 U.S. 116, 129, 78 S.Ct. 1113, 1119, 2 L.Ed.2d 1204 (1958) (in face of First Amendment objection to Secretary of State's exercise of discretion, Court adopts narrow construction, observing that where citizen seeks to engage in constitutionally protected activity, the Court will "not readily infer that Congress gave . . . unbridled discretion to grant or withhold" permission); Zemel v. Rusk, 381 U.S. 1, 17-18, 85 S.Ct. 1271, 1281, 14 L.Ed.2d 179 (1965) (rejecting delegation claim on basis of Kent's conclusion that the statute should be read as incorporating prior administrative practice); Synar v. United States, 626 F.Supp. 1374, 1384 (D.D.C. 1986) (three-judge panel) (noting evolution of nondelegation rule largely into one of construction), aff'd sub nom. Bowsher v. Synar, 478 U.S. 714, 106 S.Ct. 3181, 92 L.Ed.2d 583 (1986). See generally Cass R. Sunstein, "Law and Administration After Chevron", 90 Colum.L.Rev. 2072, 2111-14 (1990) (arguing that maxims of statutory interpretation based on constitutional norms should trump agency authority under Chevron, in part to assure legislative consideration of troublesome constitutional issues). In effect we require a clear statement by Congress that it intended to test the constitutional waters.

We thus turn to possible constructions.

A

One can imagine broader constructions than the one proposed by OSHA, but not easily. It essentially identifies two boundaries. First, in its preamble section entitled "Basis for Agency Action", OSHA found the Benzene decision's "significant risk" requirement satisfied, see 54 Fed.Reg. at 36,652/3, evidently regarding it as binding even outside the realm of toxics. Second, it made findings of economic and technological feasibility, see id. at 36,667/2 ("OSHA views this type of lockout to be both technologically and economically feasible"); id. at 36,662/2; id. at 36,667/3; id. at 36,668/2, implicitly acknowledging feasibility as a limit on the stringency of its rules. Cf. Cotton Dust, 452 U.S. at 513 n. 31, 101 S.Ct. at 2492 n. 31 ("any standard that was not economically or technologically feasible would a fortiori not be `reasonably necessary or appropriate' under the Act") (emphasis omitted). But on the agency's reading, feasibility works only as a ceiling, and not, as for toxics, as a floor; this was evidently the basis for the agency's rejection of union claims that more stringent rules than those adopted were feasible. The upshot is an asserted power, once significant risk is found, to require precautions that take the industry to the verge of economic ruin (so long as the increment reduces a significant risk, compare Building Construction Trades Dep't v. Brock, 838 F.2d 1258, 1270-71 (D.C. Cir. 1988)), or to do nothing at all. All positions in between are evidently equally valid.

The claimed power to roam between the rigor of § 6(b)(5) standards and the laxity of unidentified alternatives would, we believe, raise a serious nondelegation issue. As was true of the standard upset in Schechter, the scope of the regulatory program is immense, encompassing all American enterprise. "When the scope increases to immense proportions (as in Schechter) the standards must be correspondingly more precise." Synar, 626 F.Supp. at 1386; cf. Schechter, 295 U.S. at 553, 55 S.Ct. at 853 (characterizing challenged regulation as "as wide as the field of industrial regulation" and amounting to "delegation run riot") (Cardozo, J., concurring). Cases upholding delegations governing a single industry are thus inapposite. See, e.g., Fahey v. Mallonee, 332 U.S. 245, 249-50, 67 S.Ct. 1552, 1553-54, 91 L.Ed. 2030 (1947); Federal Power Commission v. Hope Natural Gas Co., 320 U.S. 591, 64 S.Ct. 281, 88 L.Ed. 333 (1944); American Power and Light Co. v. SEC, 329 U.S. 90, 67 S.Ct. 133, 91 L.Ed. 103 (1946); United States v. Rock Royal Co-Op, 307 U.S. 533, 59 S.Ct. 993, 83 L.Ed. 1446 (1939); Sunshine Coal Co. v. Adkins, 310 U.S. 381, 60 S.Ct. 907, 84 L.Ed. 1263 (1940); cf. Arizona v. California, 373 U.S. 546, 83 S.Ct. 1468, 10 L.Ed.2d 542 (1963) (delegation to Secretary of Interior to apportion Colorado river water, within a stated range, among seven states).

Nor is the delegation claimed by OSHA defensible under any of the special theories that have been invoked for many others that have survived, such as delegations to the President of "war" powers, see, e.g., Yakus v. United States, 321 U.S. 414, 419-423, 64 S.Ct. 660, 665-67, 88 L.Ed. 834 (1944); Lichter v. United States, 334 U.S. 742, 778-83, 68 S.Ct. 1294, 1313-16, 92 L.Ed. 1694 (1948); Woods v. Miller Co., 333 U.S. 138, 68 S.Ct. 421, 92 L.Ed. 596 (1948), or over foreign policy issues, see, e.g., Knauff v. Shaughnessy, 338 U.S. 537, 542-43, 70 S.Ct. 309, 312-13, 94 L.Ed. 317 (1950); United States v. Curtiss-Wright Corp., 299 U.S. 304, 319-20, 57 S.Ct. 216, 220-21, 81 L.Ed. 255 (1936), or delegations to entities with attributes of sovereignty themselves, such as Indian tribes, United States v. Mazurie, 419 U.S. 544, 556-57, 95 S.Ct. 710, 717-18, 42 L.Ed.2d 706 (1975), or territorial governments, District of Columbia v. Thompson Co., 346 U.S. 100, 106-10, 73 S.Ct. 1007, 1010-13, 97 L.Ed. 1480 (1953).

It is true that price and wage controls blanketing the entire economy have been sustained under quite vague legislative directions. See Amalgamated Meat Cutters v. Connally, 337 F.Supp. 737, 745-63 (D.D.C. 1971) (three-judge court). But in view of the inevitable tensions in such controls between such purposes as price stabilization on the one hand and the need for adjustments on ground of changes in cost and other market conditions on the other, compare Stephen G. Breyer Richard B. Stewart, Administrative Law and Regulatory Policy 86-90 (1985), an insistence on greater clarity from Congress would deny it any power to impose price controls at all. Not so here. Congress can readily articulate some principle by which the beneficent health and safety effects of workplace regulation are to be traded off against the adverse welfare effects. "Policy direction is all that was ever required, and policy direction is what is lacking in much contemporary legislation." John Hart Ely, Democracy and Distrust 133 (1980). OSHA's reading of the Act finds no such direction.

We note that OSHA's claimed discretion is procedurally confined. The agency sets "standards", which would normally apply across an industry, or to a category of machines, or to some other reasonably broad category. Thus, even under its view OSHA would normally not be free to single out the Jones Company for standards embodying strict feasibility while letting the Smith Company off on ones reflecting some different principle. But even the use of general standards leaves opportunities for dangerous favoritism. The cost of compliance with a standard will vary among firms in an industry, so the power to vary the stringency of the standard is the power to decide which firms will live and which will die. At the simplest level, for example, compliance may involve economies of scale, so that a tough standard will erase small, marginal firms and leave the field to a small group of larger ones. Compare Ann P. Bartel Lacy Glenn Thomas, "Direct and Indirect Effects of Regulation: A New Look at OSHA's Impact", 28 J.L. Econ. 1, 23-25 (1985).

OSHA's proposed analysis would give the executive branch untrammelled power to dictate the vitality and even survival of whatever segments of American business it might choose. Although in Benzene the Court focused perhaps more on the severity of the power claimed by OSHA than on its variability, the plurality's point is apt here: "In the absence of a clear mandate in the Act, it is unreasonable to assume that Congress intended to give the Secretary the unprecedented power over American industry that would result from the Government's view. . . ." 448 U.S. at 645, 100 S.Ct. at 2865. At least if reasonable alternative readings can be found, OSHA's must be rejected as unreasonable.

B

The NAM argues (as a fallback to its nondelegation claim) that Congress's use of "reasonably necessary or appropriate" in § 3(8) contemplates "cost-benefit" analysis. See NAM Brief at 24-25. Under this interpretation, in imposing standards under § 6(b) but outside the realm of toxics, OSHA may adopt a safety standard if its benefits outweigh its costs, and not otherwise.

Cost-benefit analysis is certainly consistent with the language of § 3(8). "Reasonableness" has long been associated with the balancing of costs and benefits. The "reasonable" person of tort fame is one who takes a precaution if the gravity of the injuries averted, adjusted for their probability, exceeds the precaution's burden. United States v. Carroll Towing Co., 159 F.2d 169, 173 (2d Cir. 1947).

And while the legislative history is almost blank on the subject, it suggests concern with market failures, see, e.g., S.Rep. No. 1282, supra, at 4, reprinted in Leg. Hist. at 144, and properly conducted cost-benefit analysis should yield a solution approximating that of a market undistorted by market failures. Application of cost-benefit analysis to safety standards also gives effect to Congress's distinction between slow-acting hazards and others, with its "particular concern for health hazards of `unprecedented complexity' that had resulted from chemicals whose toxic effects `are only now being discovered.'" Benzene, 448 U.S. at 692, 100 S.Ct. at 2889 (Marshall, J., dissenting).

Moreover, courts have often taken the word "reasonable" in a statute to require that burdens be justified by the resulting benefits. For example, in Consolidated Rail Corp. v. ICC, 646 F.2d 642 (D.C. Cir. 1981), this court reviewed an Interstate Commerce Commission adjudication of some shippers' claims that a rate based on the expense of certain safety precautions was not "reasonable", as required by the controlling statute, because the precautions themselves were excessive and therefore unreasonable. We read the statutory criterion as requiring cost-benefit analysis for such costs, which we defined with considerable care:

The safety measures for which expenditures are made must be reasonable ones, which means first, that they produce an expected safety benefit commensurate to their cost; and second, that when compared with other possible safety measures, they represent an economical means of achieving the expected safety benefit.

Id. at 648.

Similarly, where Congress authorized the Consumer Product Safety Commission to regulate hazards that create "an unreasonable risk" of consumer injury, we understood it to invoke the balancing test of negligence law and to authorize regulation only where the severity of the injury (adjusted for likelihood) offset the harm that the regulation would impose on manufacturers and consumers. Forester v. Consumer Product Safety Commission, 559 F.2d 774, 789 (D.C. Cir. 1977); see also Aqua Slide `N' Dive Corp. v. Consumer Product Safety Commission, 569 F.2d 831, 839 (5th Cir. 1978) (same).

In fact, the Ninth Circuit has found cost-benefit analysis a reasonable reading of § 3(8) itself, as applied to standards under § 6(a), where (as here) § 3(8) supplies Congress's sole guide. Donovan v. Castle Cooke Foods, 692 F.2d 641, 647-49 (9th Cir. 1982).

The union argues that prior cases preclude a cost-benefit interpretation here. They do not. The Supreme Court has expressly reserved the question. In Cotton Dust, upholding the Secretary's understanding that the "feasib[ility]" criterion of § 6(b)(5) was not a cost-benefit standard, the Court observed that "[w]hen Congress has intended that an agency engage in cost-benefit analysis, it has clearly indicated such intent on the face of the statute." 452 U.S. at 510, 101 S.Ct. at 2491. But it cited approvingly Forester and Aqua Slide `N' Dive (which find "unreasonable risk" to incorporate cost-benefit balancing), id. at 510-11 n. 30, 101 S.Ct. at 2491 n. 30, and went on explicitly to leave open the question of § 3(8)'s meaning apart from toxics regulation: "This is not to say that § 3(8) might not require the balancing of costs and benefits for standards promulgated under provisions other than § 6(b)(5) of the Act." Id. at 513 n. 32, 101 S.Ct. at 2493 n. 32.

Nor is the union's citation of our decision in Building Construction Trades apt. The cited passage merely restates the holding of Cotton Dust, saying that the Court "made clear that, in light of the feasibility language in § 6(b)(5) [i.e., for standards governed by § 6(b)(5)], `reasonably necessary' in § 3(8) did not require cost-benefit analysis." 838 F.2d at 1269. Building Construction Trades did not address the meaning of § 3(8) in contexts unaffected by § 6(b)(5). The same is true of our decision in the cotton dust litigation. American Federation of Labor v. Marshall, 617 F.2d 636, 662-63 (D.C. Cir. 1979) (when OSHA is regulating toxics, "no additional constraint is imposed by [ § 3(8)]"), vacated in part on other grounds, Cotton Dust, 452 U.S. 490, 101 S.Ct. 2478, 69 L.Ed.2d 185 (1981).

We briefly note the Fifth Circuit's reading of § 3(8), parts of which appear to agree with our own. It has stated that § 3(8) requires "a specie [sic] of cost-benefit justification." National Grain Feed Ass'n v. OSHA, 866 F.2d 717, 733 (5th Cir. 1989). At first glance, National Grain appears to contemplate an insistence that costs not outweigh benefits, see, e.g., Asbestos Information Ass'n v. OSHA, 727 F.2d 415, 423 (5th Cir. 1984), which accords with our understanding of cost-benefit analysis.

But the Fifth Circuit approach seemingly advocates consideration of costs exclusively in terms of impact on the regulated firms (as opposed to workers and consumers). Thus the court said in Asbestos Information Ass'n, "The protection afforded to workers should outweigh the economic consequences to the regulated industry." Id. (emphasis added). Evidently explicating that thought, it seemed to approve the idea that where "the cost of compliance is less than one cent per dollar [of industry sales]", and the danger is "grave", the agency has met its burden. Id. at 424. The court never explains the exclusive focus on firms. In an industry providing a good for which there are no close substitutes (i.e., facing an inelastic demand curve), firms would be scarcely affected at all: their reactions would be some combination of avoiding the costs by substituting equipment for labor, passing the costs back to labor in the form of reduced cash wages, and passing them forward to their consumers in the form of higher prices. We do not understand why one should call an analysis "cost-benefit" if it disregards costs borne by workers and consumers. The exclusion seems especially odd for application of a worker-protection statute; consumers are, after all, mostly workers.

As there appear to be many confusions about cost-benefit analysis, it may be important to make clear what we are not saying when we identify it as a reasonable interpretation of § 3(8) as applied outside the § 6(b)(5) realm. Cost-benefit analysis requires identifying values for lost years of human life and for suffering and other losses from non-fatal injuries. Nothing we say here should be taken as confining the discretion of OSHA to choose among reasonable evaluation methods. While critics of cost-benefit analysis argue that any such valuation is impossible, that is so only in the sense that pin-point figures are necessarily arbitrary, so that the decisionmaker is effectively limited to considering some range of values. In fact, we make implicit life and safety valuations each day when we decide, for example, whether to travel by train or car, the former being more costly (at least if several family members are traveling together) but safer per passenger-mile. Where government makes decisions for others, it may reasonably be expected to make the trade-offs somewhat more explicitly than individuals choosing for themselves. The difficulty of securing agreement even on a range of values hardly justifies making decisions on the basis of a pretense that resources are not scarce. In any event, OSHA has an existing obligation under Executive Order No. 12,291, 46 Fed.Reg. 13,193 (1981), to complete a cost-benefit analysis for each major rulemaking, so use of such a standard not only is doable in the qualified sense of which we have spoken but can be done without additional regulatory resources.

Thus, cost-benefit analysis entails only a systematic weighing of pros and cons, or what Benjamin Franklin referred to as a "moral or prudential algebra". Writing to a friend who was perplexed by a difficult decision, he explained his own approach:

When those difficult cases occur, they are difficult, chiefly because while we have them under consideration, all the reasons pro and con are not present to the mind at the same time. . . . To get over this, my way is to divide half a sheet of paper by a line into two columns; writing over the one Pro, and over the other Con. Then, during three or four days consideration, I put down under the different heads short hints of the different motives, that at different times occur to me, for or against the measure. When I have thus got them all together in one view, I endeavor to estimate their respective weights. . . . And, though the weight of reasons cannot be taken with the precision of algebraic quantities, yet when each is thus considered, separately and comparatively, and the whole lies before me, I think I can judge better, and am less liable to make a rash step, and in fact I have found great advantage from this kind of equation, in what may be called moral or prudential algebra.

Reprinted in Edward M. Gramlich, Benefit-Cost Analysis of Government Programs 1-2 (1981).

As we accept the NAM's contention that § 3(8)'s "reasonably necessary or appropriate" criterion can reasonably be read as requiring cost-benefit analysis, we must reject its nondelegation claim.

We hold only that cost-benefit is a permissible interpretation of § 3(8). Given the ambiguity inherent in that section, there may be other interpretations that conform to nondelegation principles. Accordingly we remand to OSHA, noting that its treatment of some of the parties' other claims, discussed below, will likely turn on its decision. We note, however, that Executive Order No. 12,291 may bear on OSHA's authority to promulgate a safety standard whose benefits fail to outweigh its costs. Section 2 of that order provides:

In promulgating new regulations . . . all agencies, to the extent permitted by law, shall adhere to the following requirements . . .

(b) Regulatory action shall not be undertaken unless the potential benefits to society for the regulation outweigh the potential costs to society;

(c) Regulatory objectives shall be chosen to maximize the net benefits to society. . . .

46 Fed.Reg. 13,193 (1981) (emphasis added).

* * *

Accordingly, in light of the NAM's nondelegation claim we reject OSHA's view that under § 3(8) it may impose any restriction it chooses so long as it is "feasible", but we also reject the NAM's nondelegation claim in light of our view that § 3(8) may reasonably be read as providing for cost-benefit analysis.

III

We now turn to the parties' narrower substantive attacks, which assert defects in OSHA's finding of "significant risk" and its decisions on the scope of the lock-out requirement.

A

The Court in Benzene construed § 3(8) of the Act to require that OSHA, before issuing any permanent standard, determine the existence of a "significant risk" to health or safety. 448 U.S. at 639, 100 S.Ct. at 2863. The NAM argues that there is not enough evidence in the record to support OSHA's finding of significant risk. Although it does not question OSHA's finding that the lockout regulation could prevent 122 fatalities and 28,400 lost workday injuries annually, see 54 Fed.Reg. at 36,652/3, it objects to the agency's failure to disaggregate the data industry by industry, even in the face of uncontested data showing that the variations are great, with injuries declining to zero in many industries.

How much risk is "significant" may well depend on the relevant substantive standard. Benzene's insistence on significant risk arose explicitly from the Court's concern that insignificant risks should not be allowed to "justify pervasive regulation limited only by the constraint of feasibility". 448 U.S. at 645, 100 S.Ct. at 2865. For regulations under § 6(b)(5), the logic of Benzene thus calls for a fairly high standard of significance. For regulations outside of § 6(b)(5) and constrained only by § 3(8), however, if the standard for such regulations is indeed cost-benefit, even a slight risk might be considered significant if it could be reduced or eliminated at a cost (including costs of enforcement and compliance) less than the resulting benefits.

Uncontrolled energy unquestionably poses greater risks in some industries than in others. Even among the manufacturing industries that OSHA classifies as "high impact" for purposes of the lockout regulation, a report by OSHA's consultants shows a nearly 20-fold difference between the high and low injury rates. III J.A. at 227. And the observed injury rate was zero in many of the "low impact" and "negligible impact" industries covered by the lockout regulation. Id.; see also I J.A. at II-14.

OSHA nowhere explains its logic. Just because paper mill equipment (which was already subject to a lockout requirement) poses a significant hazard does not mean that sewing machines do. While we have recognized OSHA's need to avoid "miniscule industry subcategories" for administrative convenience, see Building Construction Trades, 838 F.2d at 1272-73, there are no obvious barriers to disaggregation here. In fact, OSHA has in past years promulgated a wide variety of industry and equipment-specific lockout standards. As we have insisted that OSHA explain its refusal to disaggregate at the behest of unions claiming that reliance on overbroad categories denied them adequate protection, id., we similarly remand for it to explain how its aggregated approach here conforms to its interpretation of the Act.

B

Both labor and industry are dissatisfied with the scope of the lockout/tagout regulation. The union urges elimination of four exemptions; although agency interpretation of its mandate under the Act may moot the claims, their survival appears likely enough to justify our addressing them. Industry's objection, while weak on the basis of OSHA's apparent standard, may become more significant, and we merely note the problem.

The first contested exemption allows the use of tagout instead of lockout "if an energy isolating device is not capable of being locked out", 29 CFR § 1910.147(c)(2)(i) (1990). OSHA based this on inadequacies of information in the record. 54 Fed.Reg. at 36,668/1-2. Noting that the standard was a generic one that applied across almost all industries, it said that it lacked information as to overall costs and feasibility and stressed the risk that the equipment modification itself would create greater hazards to employees. Id. at 36,668/2-3. The union brief points to findings as to the relatively low pecuniary cost of modifications; this of course does not address at all the issue of adverse safety effects. As the latter are not disputed on the current record, OSHA's decision not to push lockout further at the moment does not appear invalid under any standard likely to fit within § 3(8). Cf. Hazardous Waste Treatment Council v. EPA, 861 F.2d 277, 287 (D.C. Cir. 1988) (agency decision to take one regulatory step does not afford court jurisdiction to review its not taking a further step, apart from unreasonable delay claims).

Second, OSHA allowed the use of tagout on lockable equipment where the employer can demonstrate "that the tagout program will provide a level of safety equivalent to that obtained by using a lockout program", id. § 1910.147(c)(3)(i). This too seems unobjectionable. By definition there are no safety benefits to be gained from requiring lockout in such situations. Somewhat surprisingly, OSHA has found, 54 Fed.Reg. at 36,668/1, and the union does not dispute, Brief for Union at 25, that lockout is less expensive than tagout because it requires less employee training. If OSHA's cost estimates are correct, no rational employer will use tagout with lockable equipment. If they are wrong (as seems likely given the variety of covered operations), workers will receive equivalent safety gains at lower cost, a desirable result from almost any point of view.

Third, OSHA exempted "minor servicing activities" that are routine, repetitive and integral to the production operation. Here it said that it is "obvious that [in those circumstances] lockout or tagout cannot be performed because these procedures would prevent the machine from economically being used in production." 54 Fed.Reg. at 36,662/2. The union brief does not contest this finding, which on its face is surely not implausible. Moreover, the union's criticism omits the important fact that the exemption is conditioned on the implementation of "alternative measures which provide effective protection". 29 CFR § 1910.147(a)(2)(ii)(Note) (1990) (citing to existing OSHA machine guarding regulations).

Finally, OSHA exempted "complex group operations" from the regulation's usual principle of "one person, one lock", which in some cases might require 1000 or more locks, see II J.A. at 29-30; III J.A. at 426. But in doing so it also established elaborate alternative requirements which must in the aggregate "afford the employees a level of protection equivalent to that provided by the implementation of a personal lockout or tagout device." 29 CFR § 1910.147(f)(3)(i). OSHA noted, for example, that employees of one company with a tagout program worked 488 million hours (over an eight year period) with only one lost-time accident "which was marginally related to tagout." 54 Fed.Reg. at 36,654/3. Though contesting some of the data on which OSHA relied for its view that "one person, one lock" was inappropriate for complex operations, the union does not identify evidence suggesting that the safety-equivalence principle will not be attained. Accordingly, the union establishes no reason to believe that a stricter provision would yield any safety increment.

The NAM, on the other hand, objects that in switching from its initial proposal to allow employers to choose between lockout and tagout, to its ultimate decision to require lockout (subject to the exceptions mentioned), the final rule is unjustifiable. It adduces data suggesting that the incremental safety gains from universal lockout were modest (averting 42 injuries, of which fully 26 would not even involve a lost day of work) and the incremental cost not immaterial ($2.3 million in the first year and $400,000 annually thereafter). See NAM Brief at 33 (comparing 54 Fed.Reg. at 36,684-85 with 53 Fed.Reg. 15,496, 15,515-16 (1988)). Cf. Cotton Dust, 452 U.S. at 513-14 n. 32, 101 S.Ct. at 2493 n. 32 (suggesting that a standard under § 6(b)(5) might violate § 3(8) if it imposed a high-cost requirement whose safety gains were no greater than those of an alternative low-cost requirement). Under the pure feasibility criterion of § 6(b)(5), or under OSHA's proposed interpretation (i.e., whatever OSHA chooses, so long as it is not infeasible), these figures would scarcely raise a question. Under a cost-benefit standard, and perhaps under other standards that OSHA might lawfully adopt, they may. To the extent that the NAM presses the point on remand, OSHA should address it.

IV

The parties raise a number of procedural objections to the final lockout regulation. The first is the NAM's complaint that OSHA failed in its duty to explain the basis for its determinations, see 29 U.S.C. § 655(e) (1988), particularly its decisions that there was no need to engage in industry-by-industry analysis and that mandatory lockout was preferable to a lockout/tagout option. As remand will presumably lead to further explanation on these issues (possibly curing any procedural defects), we decline to rule on the objection.

An intervenor, Motor Vehicle Manufacturers Association of the United States, argues that the final rule is not a logical outgrowth of the proposed rule. While it is true the proposed rule appeared to indicate that OSHA was resolved not to mandate the use of lockout over tagout, see, e.g., 53 Fed.Reg. at 15,508, interested parties clearly detected some play on the issue, and addressed remarks to OSHA both favoring and opposing employer choice. See 54 Fed.Reg. at 36,654-56. And in fact the final rule did leave an element of choice to the employer — the power to choose tagout when it could show that to be as effective as lockout. We agree with OSHA that the final rule is properly viewed as a "logical outgrowth" of the proposed rules, so that OSHA could enact it without another round of notice and comment. See, e.g., Small Refiner Lead Phase-Down Task Force v. EPA, 705 F.2d 506, 547 (D.C. Cir. 1983) (reviewing circuit authority).

The strongest of the procedural arguments relates to the exemption for routine servicing performed during normal production operations, the substance of which is addressed above. The union attacks the agency's alleged reliance on certain comments filed by employers after the close of the rulemaking record, see III J.A. at 493, a filing inspired, it appears, by OSHA's release of a then-current agency draft of these rules at a June 1989 meeting of OSHA's Shipyard Employment Standards Advisory Committee (to be used as a starting point for discussions about lockout in the maritime industry). See OSH Rep. (BNA) (June 7, 1989) at 12-13 (discussing the meeting). Not long after the filings, OSHA sent OMB — before which the draft rules were pending — an amendment creating the exception. See UAW Addendum at B14-27. It then added the late-filed documents to the public record on August 24, 1989, see UAW Brief at 44, and a few days later issued the final standard. The parties sharply dispute the role of the comments in OSHA's decisionmaking. The union urges that they "influenced the outcome", id., and says that since it had no chance to respond, the influence renders the resulting exemption unlawful; OSHA claims there is no evidence of influence, OSHA Brief at 61 n. 52.

Our principal authority on the subject, however, does not frame the issue of late-filed comments in terms of "influence". Judicial application of any such standard would surely be difficult, and likely to turn on where the burden of persuasion lay. Rather, in Sierra Club v. Costle, 657 F.2d 298, 398 (D.C. Cir. 1981), construing the special rulemaking provisions of § 307 of the Clean Air Act, 42 U.S.C. § 7607 (1988), we said that "the decisive point is that [the rule's challenger] has failed to show us any particular . . . documents to which it lacked an opportunity to respond, and which also were vital to EPA's support for the rule." Id. at 398-99 (emphasis added). The same is true here. The issue had long been a significant one in the rulemaking process, and some of the timely filed comments did address it. See III J.A. at 171-76 (comments of Motor Vehicle Manufacturers Association advocating removal of the exemption requirement that it be necessary to perform servicing with equipment energized); id. at 153-55 (comments of the American Feed Industry Association questioning the necessity of shutting down an entire operation simply to grease a bearing by slipping the nozzle of a grease gun over a fitting and pumping in the grease). Moreover, the agency set forth the support for its decision not in terms of data from the late-filed comments but in commonsensical terms — that application of lockout/tagout for the sort of servicing exempted here (routine, repetitive, and integral to operation of the machine) would uneconomically prevent actual use of the machine. The comments were plainly not "vital to [OSHA's] support for the rule".

Finally, the union claims that the exemption for complex group operations is invalid on the theory that OSHA impermissibly "amended" it with a "technical corrections notice" filed 13 months after the final rule issued. 55 Fed.Reg. 38,677 (1990). We agree with OSHA that the final rule, including the corrections, is consistent with the record evidence and would have constituted a logical outgrowth of the proposed rules if originally promulgated as corrected. The union claims a violation of the Act's requirement of notice and comment when OSHA "modif[ies]" an occupational health or safety standard. See 29 U.S.C. § 655(b)(2) (1988). Its brief, however, fails to identify a genuinely substantive change, and indeed acknowledges that the corrections notice might be viewed as a mere policy statement elucidating the original version. UAW Brief at 46. We do not read § 655(b)(2) to invalidate a clarification of this sort, issued relatively soon after the issuance of a standard.

V

We have noted above various bases for remand, notably the need for OSHA to address "significant risk" in terms of whatever substantive meaning it lawfully assigns to § 3(8), and particularly to explain its decision to impose lockout/tagout even where the risk appears to be diminutive or zero. That leaves the issue of whether we should vacate the rule pending a remand, a point the parties have not briefed. Although OSHA's reasoning on many issues is extremely obscure, partly because of its failure to identify any intelligible principle that could control its discretion under § 3(8), the likelihood that correction of these defects will alter the ultimate rule is most unclear. See International Union, UMW v. FMSHA, 920 F.2d 960, 967 (D.C. Cir. 1990); International Union, UMW v. FMSHA, 928 F.2d 1200, 1203-04 (D.C. Cir. 1991). There appear to be segments of the standard (notably its application to some of the high-impact industries) that may well have genuine life-saving effects and are highly likely to survive re-examination. Vacating the rule would interrupt the continued flow of those advantages if employers responded to vacation of the rule by eliminating the safety practices. Of course, many high-impact employers would have no choice but to continue lockout procedures as their industries are already subject to other OSHA lockout standards. Moreover, as many of those standards were promulgated as "national consensus standards", it seems likely that some employers would continue lockout even in the absence of a government command.

On the other hand, industries that ultimately may prove exempt will in the meantime have to incur substantial costs for little or no safety gain. Further, turning the rule off and then on, as it were, will not in itself impose substantial costs, compare International Union, UMW, 920 F.2d at 967, as firms can simply halt their installation of locks and their training programs if they think that the costs of lockout are not justified by any resulting increases in safety.

In view of the great uncertainty, we are asking the parties for briefing on the point, see United Mine Workers v. Dole, 870 F.2d 662, 673-74 (D.C. Cir. 1989), including the possibility of a partial vacation of the rule, and will refrain from vacating it in the meantime.

* * *

Accordingly, we remand the case to OSHA for further consideration in light of this opinion.

So ordered.