Opinion for the Court filed by Circuit Judge STEPHEN F. WILLIAMS.

In International Union, UAW v. OSHA, 938 F.2d 1310 (D.C.Cir. 1991), we remanded a regulation to the Occupational Safety & Health Administration on the ground that, although we believed the controlling statute was susceptible of a construction narrow enough to satisfy the nondelegation doctrine, see A.L.A. Schechter Poultry Corp. v. United States, 295 U.S. 495, 55 S.Ct. 837, 79 L.Ed. 1570 (1935), the agency had apparently not adopted such a construction. Finding that the interpretation OSHA has set forth on remand embodies an adequate constraint on its discretion, we now dismiss the petition for review.

In 1989 OSHA issued a regulation designed to protect workers who perform maintenance or servicing operations on powered industrial equipment from the hazard of energy unexpectedly released from that equipment. “Control of Hazardous Energy Sources (Lockout/Tagout)”. 54 Fed.Reg. 36,-644 (1989) (codified as amended at 29 C.F.R. § 1910.147 (1993)). Issued pursuant to the Occupational Safety and Health Act (the “OSH Act”), 29 U.S.C. §§ 651 et seq. (1988), the standard essentially requires an employer to affix a “lock” to an energy isolating device connected to the equipment (“lockout”), or, if the employer can prove its equal efficacy (or the equipment is unlockable), to place a “tag” on the energy isolating device, warning employees not to operate the device or the equipment until the tag is removed (“tagout”). With limited exceptions, the standard applies to all powered industrial equipment in industrial workplaces.

The National Association of Manufacturers (“NAM”) and the International Union, UAW, challenged the standard in this court. The union argued that § 6(b)(5) of the OSH Act, 29 U.S.C. § 655(b)(5), required OSHA to promulgate a stricter standard than it had, while *668NAM argued that that section was inapplicable to safety regulation and that, in the absence of an intelligible congressional guide, OSHA could not lawfully promulgate any safety regulation at all. The petitioners also alleged several defects in the standard and the procedures leading to its adoption.

We found that OSHA’s authority in regulating safety hazards (i.e., hazards that produce immediately noticeable harm) was not governed by § 6(b)(5), which applies only to health hazards, but by § 3(8), 29 U.S.C. § 652(8), which defines an “occupational safety and health standard” as a standard “reasonably necessary or appropriate” to the agency’s overall goals of ensuring “safe or healthful employment and places of employment.” International Union I, 938 F.2d at 1313-17. OSHA’s then-stated interpretation of § 3(8), however, appeared to assume in the agency an authority to choose freely among levels of stringency, from adopting no standard at all to adopting the most stringent standard feasible. Id. at 1317, 1321. Concluding that such free-wheeling authority might well violate the nondelegation doctrine, we remanded the case to the agency to give it the opportunity to adopt an interpretation of § 3(8) that would be both “reasonable and consistent with the nondelegation doctrine.” Id. at 1313. We also remanded for OSHA to reveal the reasoning behind a number of its choices about the structure of the rule. Id. at 1322, 1323-24.

OSHA issued a Supplemental Statement of Reasons on March 30, 1993. 58 Fed.Reg. 16,612. Arguing that this Statement faded to cure the defects we identified in the first round, NAM again asks that we vacate the standard. Because we find that OSHA’s current interpretation of its statutory authority to issue safety standards is consistent with the nondelegation doctrine and that its explanations of the other disputed decisions are adequate, we uphold the regulation.

The Supplemental Statement of Reasons delineates OSHA’s view of the statutory bounds of its authority to issue safety standards. The agency points primarily to several principles — most of them not derived from § 3(8) itself but from other sections, including some not directly applicable, such as § 6(b)(5) — that constrain its discretion in choosing a safety standard. The agency must find that (1) “the standard will substantially reduce a significant risk of material harm”, cf. Industrial Union Dep’t v. American Petroleum Institute (“Benzene”), 448 U.S. 607, 639-59, 100 S.Ct. 2844, 2862-73, 65 L.Ed.2d 1010 (1980); (2) & (3) compliance will be economically and technologically feasible, cf. American Textile Mfrs. Inst. v. Donovan, 452 U.S. 490, 513 n. 31, 101 S.Ct. 2478, 2493 n. 31, 69 L.Ed.2d 185 (1981); and (4) the standard “employs the most cost-effective protective measures”. In addition, it must (5) for any standard differing from an existing national consensus standard, publish its reasons why its standard would better effectuate the purposes of the Act; and (6) support its choice of standard with evidence in the rulemaking record and explain any inconsistency with prior agency practice. 58 Fed. Reg. at 16,614/2-3.

OSHA’s items (1), (2), and (3) together are the same criteria we rejected in the previous round as insufficiently cabining the agency’s discretion in adopting safety standards. See International Union I, 938 F.2d at 1317. Item (4) narrows that discretion somewhat more, forcing the agency at a minimum to adopt the cheapest standard that will achieve the desired level of safety, or to adopt a more protective standard over an equally costly but less effective alternative. It does very little, however, to narrow the agency’s discretion to choose among levels of safety.

Items (5) and (6) also fail to channel agency discretion in the substantive manner demanded by the nondelegation doctrine. The requirement of justifying a deviation from national consensus standards, imposed by § 6(b)(8), 29 U.S.C. § 655(b)(8), presupposes guiding principles from some other source, “the purposes of this chapter”, id., as the norm against which a standard’s deviation from a national consensus standard is to be measured. The requirements of evidence in the rulemaking record and of an explanation for any inconsistency with prior agency practice also assume the existence of an intelligible legislative mandate; indeed, both exist at least in part as devices to assure *669agency compliance with such a mandate. The required explanation of any change in policy is a component of the more general requirement of reasoned decisionmaking, Greater Boston Television Corp. v. FCC, 444 F.2d 841, 850-52 (D.C.Cir. 1970), which itself is necessary for the reviewing court “to satisfy itself that [the agency’s reasons for acting] do not deviate from or ignore the ascertainable legislative intent.” Id. at 850. The requirement of supportive evidence operates to assure a link between the agency’s substantive mandate and the real-world circumstances in which the agency operates. If an agency could claim to be applying a statutory constraint merely by asserting the existence of some fact, or without connecting the facts to the congressionally specified standards in a reasoned decision, it would be free to defeat the underlying purpose of the constitutional limits on delegation: to make sure that the regulatory principles as actually applied have their origin in a judgment of the legislature. See Panama Refining Co. v. Ryan, 293 U.S. 388, 415, 430, 55 S.Ct. 241, 252-53, 79 L.Ed. 446 (1935); see also A.L.A. Schechter Poultry Corp., 295 U.S. at 551, 553, 55 S.Ct. at 852, 853 (1935) (“The delegated power of legislation which has found expression in this code is not canalized within banks that keep it from overflowing_ This is delegation running riot. No such plenitude of power is susceptible of transfer.”) (Cardozo, J., concurring).

Were the six itemized criteria the full statement of OSHA’s interpretation of its statutory mandate, we might have to vacate the rule, because the agency might still have too much freedom to “roam between the rigor of § 6(b)(5) standards and the laxity of unidentified alternatives”. International Union I, 938 F.2d at 1317. But OSHA has gone on to infer from various sections— § 2(b), 29 U.S.C. § 651(b) (stating congressional purpose “to assure so far as possible ... safe ... working conditions”); § 6(a), 29 U.S.C. § 655(a) (in first two years of Act’s application, agency was to choose among existing standards the one “which assures the greatest protection”); § 6(b)(8), 29 U.S.C. § 655(b)(8) (setting out terms for justifying deviation from national consensus standard); and § (5)(a)(l), 29 U.S.C. § 654(a)(1) (requiring each employer to provide each employee a place “free from recognized hazards that are causing death or serious physical harm”) — that the Act’s “overriding purpose” is “to provide a high degree of employee protection.” 58 Fed.Reg. at 16,614/3-15/1. Thus the agency reads the Act to require it, once it has identified a “significant” safety risk, to enact a safety standard that provides “a high degree of worker protection”. Id. at 16,615/1. It is not permitted to “do nothing at all”, as we had earlier suggested. Id. (quoting International Union I, 938 F.2d at 1317). Rather, OSHA reads the Act to permit it to deviate only modestly from the stringency required by § 6(b)(5) for health standards. Accordingly, as construed by OSHA, the Act guides its choice of safety standards enough to satisfy the demands of the nondelegation doctrine. See International Union I, 938 F.2d at 1317-18, for a discussion of cases.

OSHA’s Statement rejected the idea that the Act requires OSHA to perform what the Statement called “formal cost-benefit analysis”. 58 Fed.Reg. at 16,622/2. Although in the original ease NAM had argued, as an alternative to its nondelegation claim, that § 3(8) required cost-benefit analysis, it does not here argue that the omission of cost-benefit analysis renders OSHA’s interpretation invalid; indeed, except for its nondelegation claim and a brief attack on OSHA’s use of § 2(b)’s statement of congressional purpose as a source of constraint on its discretion, it does not dispute the validity of any aspect of OSHA’s interpretation. Accordingly, we heed not pass on the validity of the omission.

While rejecting “formal cost-benefit analysis”, OSHA asserted in the Statement its belief that the six criteria discussed above “assure that the costs of safety standards are reasonably related to their benefits.” Id. at 16,621/2; see also id. at 16,221/3. We note that whether this reasonable relationship in fact exists, it does not appear to be a logically necessary implication of OSHA’s six guide*670lines.1 For example, OSHA has often taken the Benzene opinion’s reference to risks of “one in a thousand”, Benzene, 448 U.S. at 655, 100 S.Ct. at 2870-71, as establishing that such a risk is significant. See International Union, UAW v. Pendergrass, 878 F.2d 389, 392 (D.C.Cir. 1989). Applied to an industry that can easily pass a very large cost increase on to consumers (because demand for its product is highly inelastic), but which has very few workers exposed to the risk, OSHA’s six criteria could lead to adoption of a regulation costing billions, yet likely to save few if any lives. It may well be that the nature of safety issues is such that this hypothetical is really a purple cow, but that would be true from the happenstance of safety issues, not the logic of the agency’s criteria. In any event, the current case does not require us to decide whether the statute requires a reasonable relationship between a rule’s costs and its benefits. Indeed, here OSHA found “that the relationship between the benefits secured by the lockout/tagout standard and the costs it imposes is reasonable”, 58 Fed.Reg. 16,621/3, and NAM does not contest the finding.

* * *

In International Union I we directed the agency on remand to explain (1) why, in light of differential injury rates across industries, with some showing an observed injury rate of zero, 938 F.2d at 1322, it failed to disaggre-gate industries for purposes of finding the statutorily required “significant risk”, and (2) how it arrived at its preference for mandatory lockout over a lockout-tagout option.

Failure to Disaggregate Industries

In its Supplemental Statement of Reasons, the agency explained that it adopted a single standard applicable to all general industry employers because, regardless of the industry in which they work, “workers face a significant risk of material harm every time they perform service or maintenance work on powered industrial equipment.” 58 Fed.Reg. 16,620/2 (citing 54 Fed.Reg. 36,647-48, 36,652-53). In industries where there is little such equipment, or where such service and maintenance operations are rarely performed, or where such work is restricted to a small group of the total employees, the total injury rate will likely be low. See 58 Fed.Reg. at 16,620/3. But even in industries with low or negligible overall accident rates, the workers who engage in the operations covered by the standard face a “significant risk of material harm.” Id. at 16,620/2.

Since the standard “impose[s] a compliance duty on employers only to the extent that hazardous servicing and maintenance activities in fact take place in their workplaces,” id. at 16,620/1, and since even those “sectors with ‘negligible’ concentrations of powered equipment ... occasionally have workers involved in servicing or maintenance,” id. at 16,620/3, OSHA believed that disaggregating was neither necessary nor productive. Because the burdens the standard imposes on an industry correlate directly with the incidence of the safety hazard in that industry, OSHA reasoned, id. at 16,621/1, the standard will not be overinelusive. And, since OSHA was concerned “[t]o avoid having the standard be underinclusive,” id. at 16,620/1, it preferred not to exclude any industry, regardless of that industry’s reported accident rate.

NAM’s objections to this reasoning assert that OSHA has disregarded “wide variations in the risk to workers”. NAM Initial Brief at 20. But the variation, as OSHA has explained, exists only by reference to industry categories established with other purposes in view. If, as OSHA asserts and NAM appears not to dispute, the regulation applies simply to machines that pose a significant risk and to workers subjected to that risk, we see no reason why OSHA should be concerned with industry classifications that appear essentially irrelevant to its task.

Preference for Lockout

OSHA’s Supplemental Statement of Reasons also justifies its decision to abandon its original proposal, which would have allowed employers an option between lockout *671and tagout regardless of the relative efficacy of the two methods, and to adopt instead a standard that prefers lockout over tagout safety programs. 58 Fed.Reg. 16,618-20. In support of its decision, OSHA adduces data from the Regulatory Impact Analysis of the rule, demonstrating, it says, that the costs of the preference for universal lockout are minimal and, in any event, reasonable in relation to its benefits. NAM has not disputed the accuracy of OSHA’s data, the validity of OSHA’s analysis, nor the propriety of that analysis under the statutory construction OSHA adopted. Instead, NAM hangs this issue on its nondelegation claim, arguing that the propriety of the lockout preference cannot be evaluated in the absence of an interpretation of the statute consistent with the nondelegation doctrine. As NAM’s nondele-gation claim fails, this argument fails in tandem.

Accordingly we deny NAM’s motion to enforce the judgment in International Union I and dismiss its petition to review the Supplemental Statement of Reasons.

So ordered.