Opinion for the Court filed by Senior Circuit Judge WILLIAMS.
Opinion concurring in part and dissenting in part filed by Circuit Judge ROGERS.
WILLIAMS, Senior Circuit Judge.
This case arises from our remand in Shieldalloy Metallurgical Corp. v. NRC, 624 F.3d 489 (D.C.Cir.2010). It concerns, above all, an important question of public safety that demands great clarity and precision, neither of which the agency has supplied: What rules govern the means by which the owner of a licensed nuclear facility may decommission that facility and dispose of its radioactive materials? The Nuclear Regulatory Commission claims that it has taken a clear and consistent approach to answering this question. The clarity and consistency are not apparent to us.
The order now under review, Shieldalloy Metallurgical Corp., CLI-11-12, 74 NRC-(Oct. 12, 2011) (“Order”), presents that issue and several others. All stem from the interaction between the NRC’s regular decommissioning process and a statutory provision (§ 2021 of the Atomic Energy Act, 42 U.S.C. § 2021) authorizing the NRC to transfer regulatory authority over classes of nuclear materials located within a state to the government of that state. Except as to the primary issue just highlighted — the Commission’s basic standards for decommissioning — we find no legal error in the Commission’s Order; as to that issue, however, our inability to understand the key regulatory materials purportedly guiding the agency’s exercise of control over decommissioning requires another remand.
* * *
Petitioner Shieldalloy Metallurgical Corporation is an industrial company that manufactured metal alloys in Newfield, New Jersey between 1955 and 1998. Its manufacturing process generated radioactive byproducts, which the company held at the Newfield facility under license from the NRC. In the early 1990s, Shieldalloy began consulting with the NRC on the development of a proposal to decommission the facility and dispose of its radioactive materials on-site. It proposed that the disposal would be “restricted,” an NRC term of art requiring special conditions designed to minimize the risks to public health and safety. Restricted disposal- is in contrast to “unrestricted,” under which remediation and reduction of radiation levels have proceeded to the point where there is no need for limits on public access to the disposal site. See 10 C.F.R. § 20.1003 (definitions of “restricted area” and “unrestricted area”). These consultations stretched on for many years and culminated in a formal decommissioning plan, which Shieldalloy submitted to the agency in 2002.
The NRC rejected the 2002 plan as technically deficient. After further consultations with NRC staff, Shieldalloy submitted a second, revised plan in 2005. The NRC rejected that plan, too, with comments. In 2006, Shieldalloy submitted a third plan, which responded to the NRC’s comments, as well as to NRC guidance on decommissioning of licensed facilities issued earlier in 2006. The NRC accepted this third plan for the purposes of technical review, but still sought further clarification from Shieldalloy on various aspects. In the summer of 2009, Shieldalloy submit[374]*374ted a fourth plan that was responsive to these last inquiries. The agency never reviewed that document.
In 2008, while the NRC was reviewing the 2006 plan, New Jersey applied for a transfer of regulatory authority over instate nuclear materials, pursuant to § 2021 of the Act. The NRC made an initial determination that the transfer to New Jersey would satisfy § 2021(d)’s requirement that the state’s regulatory program be “compatible” with that of the NRC, and sought comments from the public. Shieldalloy filed a letter asserting incompatibility between the programs. To no avail: the agreement transferring authority to New Jersey went into effect in September 2009, barely a month after Shieldalloy submitted the fourth version of its decommissioning plan. Less than two weeks later, New Jersey informed Shieldalloy that its 2009 plan — which the NRC had forwarded to the state — did not meet the state’s remediation requirements.
Fearing that New Jersey would require it to abandon its plans for on-site disposal and pursue a more expensive off-site alternative (again, in NRC parlance, “unrestricted”), Shieldalloy appealed to this court. We agreed with the company in part and remanded the case to the agency. See Shieldalloy Metallurgical Corp., 624 F.3d at 489. Specifically, we found that the NRC had not responded meaningfully to Shieldalloy’s invocation of criterion 25 of the Commission’s “Criteria Document:”1
Existing NRC Licenses and Pending Applications. In effecting the discontinuance of jurisdiction, appropriate arrangements will be made ... to ensure that there will be no interference with or interruption of ... the processing of license applications, by reason of the transfer.
46 Fed.Reg. 7540, 7543 (Jan. 23, 1981). Shieldalloy had argued that as its license application had already been years in processing, transfer to New Jersey would clearly interrupt the process. See Shiel-dalloy, 624 F.3d at 493. Further, Shieldal-loy had suggested that if the Commission wished to generally transfer authority over in-state nuclear materials to New Jersey, there was nothing to prevent it from reserving the Newfield facility from any such transfer.
To this twofold claim, the NRC offered a twofold response. As to the risk of an interruption in seeming violation of criterion 25, it expressed confidence that there would be a “smooth transition” in the processing of pending licensing actions (including decommissioning applications) because New Jersey’s regulatory scheme recognized existing NRC licenses and would continue “any licensing actions that are in progress.” Id. at 494 (quotations removed). As to Shieldalloy’s preferred solution — excepting Newfield from the transfer — the NRC claimed that the legislative history of § 2021 showed that Congress did not contemplate its allowing “concurrent regulatory authority over licensees.” 624 F.3d at 493. We rejected both elements of the defense for reasons elaborated in our earlier opinion. Id. at 493-95.
Shieldalloy’s initial suit raised a further claim, which we noted but did not address because Shieldalloy had not raised it sufficiently with the Commission. See 624 F.3d at 495-97. This other claim pertained to a cost-benefit analysis principle embedded in the NRC’s regulatory program called the ALARA principle, which, [375]*375as we explain more thoroughly below, refers to the requirement that residual radiation at a site be “as low as reasonably achievable.”
The core of Shieldalloy’s ALARA claim was that the New Jersey decommissioning regime, unlike the NRC’s, requires exhumation and off-site disposal of radioactive materials even where that course of action would expose the public to higher doses of radiation than on-site disposal. Thus, Shieldalloy contended, “while New Jersey’s standards may be more stringent, they are actually less safe.” 624 F.3d at 496. And while a Commission policy statement, its Compatibility Guidance Document,2 allowed transfer where a state’s standards were “more stringent,” that allowance was subject to the proviso that the “more stringent requirements do not preclude or effectively preclude a practice in the national interest without an adequate public health and safety or environmental basis related to radiation protection.” 62 Fed.Reg. 46,517, 46,520/2. The same policy statement also affirmed the NRC’s mission as being to assure that civilian use of nuclear materials “is carried out with adequate protection of public health and safety.” Id. 46520/1; see also Shieldalloy, 624 F.3d at 496. Section 2021(d)(2) itself requires the state program to be “adequate to protect the public health and safety.” 42 U.S.C. § 2021(d)(2). In light of these principles, we found that Shieldalloy’s depiction of the effects of transfer held out a “troubling prospect.” 624 F.3d at 496.
On remand, the NRC afforded Shieldal-loy a “fresh opportunity” to comment on the transfer agreement in order to “assure a full airing of the matter.” Order at 6-7. In its submission, Shieldalloy of course pursued the issues occasioning the remand, but it also pursued the contention that the New Jersey rules were more stringent but less safe.
The NRC’s Order rejected Shieldalloy’s objections and reinstated the transfer agreement. Shieldalloy again petitions us for review. We find that, on its second attempt, the agency adequately addressed Shieldalloy’s claims arising out of criterion 25 and the parties’ conflicting interpretations of § 2021. The NRC’s response to Shieldalloy’s ALARA claim, however, appears divorced from the authorities on which it purports to draw, and accordingly we vacate the Order and remand the case again for a more responsive analysis of that issue.
We begin with the question of concurrent federal and state jurisdiction under § 2021. On remand the NRC determined that § 2021 requires it to accept a state’s request for a partial transfer of regulatory authority if the request satisfies the conditions laid out in § 2021(d), but forbids the agency from proposing partial transfers either at the request of a regulated entity or on its own initiative. In other words, the NRC reads § 2021 as a grant of authority to accept or reject, but never to modify, a proposed transfer agreement. See Order at 8-20.
Whether § 2021 straightjackets the NRC into treating state transfer applications on an “all or nothing” basis is reviewable under Chevron USA, Inc. v. NRDC, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984), which means that, as Entergy Corp. v. Riverkeeper, Inc., 556 U.S. 208, 218, 129 S.Ct. 1498, 173 L.Ed.2d 369 (2009), observed, a “reasonable agency interpretation prevails.”
[376]*376As we mentioned in our earlier opinion, there appears to be a contradiction between § 2021(b), which provides that the NRC “is authorized to enter into agreements ... with respect to any one or more ” classes of materials, and § 2021(d), which states that the agency “shall enter into an agreement” under § 2021(b) provided the conditions of compatibility, adequacy and certification are met. 42 U.S.C. § 2021(b) & (d) (emphases added); see Shieldalloy, 624 F.3d at 495. The latter provision appears to state an imperative, whereas the former suggests some flexibility to negotiate which classes of materials will be governed by the transfer.
In its Order, the Commission reconciles these two provisions by reading § 2021(b) as a “general grant of legal authority” to hand a state authority over designated nuclear materials, and § 2021(d)’s “shall” as mandating that transfer when the other statutory criteria are met. Order at 11-14. The Commission bolsters this reading with references to the legislative history suggesting a strong congressional determination to recognize the interests of states. Id. at 14-15. It also plausibly construes the “any one or more” language of § 2012(b) as reflecting no more than an anticipation that agreements might cover fewer than all the three relevant categories of nuclear materials at a time. Id. at 16. The Commission also attributed its earlier approval of a transfer to Oklahoma of less than all the sites with materials of a specific class, see Shieldalloy, 624 F.3d at 493-94, to Oklahoma’s express unwillingness to assume authority over the specific materials, which the legislative history again suggested might be a sound basis for accepting a proposal of partial transfer. Order at 16-20.
In other words, the NRC saw § 2021 as conferring authority to accept but not request partial transfers. We can imagine contexts where this bright-line distinction would break down. Suppose, for example, the NRC rejected or made clear its intent to reject a proposed transfer agreement based solely on concerns about a single facility. Suppose the state then intuited that a partial proposal which excluded that facility would be a shoo-in and decided it was better not to throw the baby out with the bathwater? Would this not be an NRC-negotiated partial transfer in substance if not in name? But even taking into account this conceptual weakness, the Commission’s interpretive exercise appears reasonable and consistent with the statutory scheme.
Shieldalloy argues that the permissive language found at § 2021(c)(4) & (j) demonstrates Congress’s intent “for the NRC to retain flexibility in determining how to fashion and manage its agreement with a state.’,’ Pet’r Br. 28. But both seem to cut in favor of the Commission’s view. Section 2021(j) gives the NRC discretion to suspend “all or part” of a transfer agreement in the event of an “emergency situation” or where the state has failed to live up to its responsibilities under an agreement. And § 2021(c)(4) concerns only situations where the agency has determined “by regulation or order” that the materials over which the state seeks regulatory authority pose such “hazards or potential hazards” that they must be disposed under NRC license. 42 U.S.C. § 2021(c)(4). These very narrow provisions for special treatments tend to confirm the NRC’s view that it has no general authority to reject part of a state’s qualifying request for transfer.
There remains Shieldalloy’s argument that the Commission violated criterion 25 by failing to ensure that there would be no interruption in the processing of license applications. Of course, once the Commission has precluded its exercise of [377]*377any discretion to cut part of a transfer out of a state’s request, the only remedy would be the seemingly drastic one of denying the request in full.
Indeed, if one reads criterion 25 as Shieldalloy contends it should be read, then the NRC in fact should have rejected New Jersey’s application once it determined that a partial transfer was unavailable. The company argues, not unreasonably, that where a regulated entity has engaged in an extensive process of dialogue and collaboration with an agency in pursuit of specific licensing outcome, then a sudden relinquishment of regulatory authority to a different sovereign with a radically different regulatory framework necessarily constitutes “interference with or interruption of’ the processing of its licensing application.
We certainly sympathize with Shieldal-loy’s frustration at having been jilted by the NRC after a decade of dogged courting. But under the extremely narrow standard by which we review an agency’s interpretation of its own regulations, see, e.g., Thomas Jefferson Univ. v. Shalala, 512 U.S. 504, 512, 114 S.Ct. 2381, 129 L.Ed.2d 405 (1994), we are constrained to accept the NRC’s argument that it has never read criterion 25 as anything other than a “purely administrative” provision requiring the orderly transfer of records from the agency to the transferee state. Order at 30; see generally id. at 28-34. Nothing in the text plainly forbids the NRC from proceeding with a transfer under § 2021 even though a pending license application will meet a different fate under a state regulatory regime than it would under the federal equivalent.
Shieldalloy’s ALARA argument concerns the operation of that principle in the context of a 1997 NRC final rule on the decommissioning of licensed facilities, termed the License Termination Rule (“LTR”). 10 C.F.R. §§ 20.1401-06. The LTR applies to a broad range of NRC licenses, including those required for the operation of nuclear reactors, the possession of spent fuel and byproduct radioactive materials, and the disposal and storage of high-level radioactive wastes. See id. § 20.1401 (stating that the LTR applies to the decommissioning of licensed facilities under 10 C.F.R. parts 30 (byproduct material), 40 (source material), 50 (production and utilization facilities), 52 (nuclear power plants), 60 (high-level radioactive waste), 61 (land disposal of radioactive waste), 63 (waste disposal in Yucca Mountain), 70 (special nuclear material) and 72 (spent fuel and other wastes)). Shieldalloy holds a part 40 license to possess source material containing uranium and thorium at the Newfield site. Shieldalloy Metallurgical Corporation, NRC.GOV, http:// www.nrc.gov/info-finder/decommissioning/ complex/shieldalloy-metallurgical-corporation-smc-.html.
Under the LTR, a site can be decommissioned for either unrestricted or restricted use. 10 C.F.R. § 20.1402-03; see also id. § 20.1003 (definition of “decommission”). Again, unrestricted use describes a site that has been remediated and at which radiation levels have been reduced to the point where there is no need to limit public access. Id. §§ 20.1003 (definition of “unrestricted area”), 20.1402. Restricted sites are those that will be remediated to some point short of the conditions required for unrestricted use, but where public health and safety can be ensured through the establishment of adequate controls. Id. §§ 20.1003 (definition of “restricted area”), 20.1403. The regulations also put a ceiling on dose exposure in the event that the institutional controls are “no longer in effect,” id. § 20-1403(e), a ceiling higher than Shieldalloy’s estimate of the dose ex[378]*378posure in that event under its current decommissioning plan. See Order at 38-39.
Section 20.1402 of the LTR sets out the conditions under which a site will be released for unrestricted use on termination of a license. Section 20.1403 does the same for restricted release. Under § 20.1402, a licensee can pursue unrestricted release after reducing residual radiation at the site to “levels that are as low as reasonably achievable (ALARA),” and in any event no greater than 25 millirem (“mrem”). Id. § 20.1402.3 The prerequisites for restricted release under § 20.1403 are more elaborate, and require the licensee to arrange and set aside funds for “legally enforceable institutional controls” that will limit human exposure to no greater than 25 mrem. Id. §§ 20.1403(b)-(c).
In addition to these technical instructions, the LTR requires a licensee seeking restricted release to make what the NRC calls (Order at 41) an initial showing of “eligibility” set out at § 20.1403(a):
A site will be considered acceptable for license termination under restricted conditions if:
(a) The licensee can demonstrate that further reductions in residual radioactivity necessary to comply with the provisions of § 20.1402 would result in net public or environmental harm or were not being made because the residual levels associated with restricted conditions are ALARA Determination of the levels which are ALARA must take into account consideration of any detriments, such as traffic accidents, expected to potentially result from decontamination and waste disposal;
Id. § 20.1403(a) (emphasis added). The NRC characterizes the “net public or environmental” harm test alluded to in § 20.1403(a) as an “abbreviated” form of ALARA that takes into account only a subset of the costs involved in a full ALARA assessment. Order at 25-26; Resp. Br. 13-14 n. 3, 59-60 n. 13. The parties have focused exclusively on the ALARA test, and so shall we.
Shieldalloy argues that the NRC provides for an eligibility test for restricted release predicated on a comparative application of the ALARA principle. More specifically, its response to the Commission’s solicitation of views on remand argued that the Commission’s rules and interpretations allow a licensee to employ restricted release if it would result in a lower radiation exposure than unrestricted release (which necessarily involves a comparison of the two). Joint Appendix (“J.A.”) 712, 714-15; Pet’r Br. at 47. In other words, if we understand Shieldalloy correctly, the proper application of the emphasized language would entail a comparison between restricted and unrestricted release, and the former would win when it yielded lower risks than unrestricted. By contrast, Shieldalloy asserts, New Jersey does not contemplate any form of radiation dose comparison between restricted and unrestricted release, and may require unrestricted release even where restricted release would have been safer. On this analysis, and given the preferences for safety expressed in the statute, the Criteria Document and the Compatibility Guidance Document, Shieldalloy argues that the New Jersey rules are not “compatible” with NRC’s, as required by § 2021(d)(3).
In its Order and before us, the NRC rejects outright this reading of § 20.1403(a). It maintains that its regulations “neither explicitly nor implicitly require a comparison between the levels of protection afforded by the unrestricted [379]*379and restricted decommissioning options.” Order at 37; see also Resp. Br. at 57. It further declares that “unrestricted release and restricted release are simply not susceptible to being compared meaningfully,” Order at 37, because of the “uncertainties” associated with restricted release, id.
As for ALARA, the agency states that the principle’s operation is confined to comparisons of different remediation options of the same class — that is, for comparing two unrestricted-release proposals, or two restricted-release proposals, but never for comparing a restricted release option with an unrestricted one. Order at 40-41. The purpose of what it views as the correct understanding of ALARA, the agency avows, is twofold: First, it serves as a tool for measuring the reductions in radiation levels that it will require below the maxima set out in §§ 20.1402-03 in cases where the benefits of such additional reductions do not exceed the costs. Id. 41. Second, through its incorporation into § 20.1403(a) it provides “a criterion to limit the use of restricted release.” Id.
With respect to the precise meaning of § 20.1403(a), the agency tells us that the provision requires unrestricted release, except where “remediation to the level of 25 mrem per year for unrestricted release [set forth in § 20.1402] would not be beneficial from a cost standpoint.” Resp. Br. at 55. The agency further asserts that the ALARA principle, as used in § 20.1403(a), is the analytic prism for such a weighing of costs and benefits. Order at 25, 41. As explained in its brief, the NRC sees the role of ALARA in § 20.1403(a) as helping to ascertain whether even the highest radiation levels permissible for unrestricted release under § 20.1402 are attainable in a cost-beneficial manner. Resp. Br. at 59.
We are quite baffled by the NRC’s current interpretation of § 20.1403(a). The agency’s order and its brief do not quote from the provision or make any effort to engage with its text. Instead, they state in bald and conclusory fashion that the regulation does not mean what Shieldalloy says it means. Unlike the NRC, we do not find the meaning of § 20.1403(a) self-evident. Rather, we think that its text neither precludes the reading given to the provision by Shieldalloy nor, at least without exegesis that is completely missing here, supports that proposed by the NRC.
The first sentence of § 20.1403(a) provides that a licensee is eligible for restricted release if it can prove that the remedial measures required of unrestricted release “were not being made because the residual levels associated with restricted conditions are ALARA.” 10 C.F.R. § 20.1403(a). Although the passive construction of this clause leaves us to guess who exactly is not making the residual reductions required for unrestricted release, we think that it can only be the licensee or its agents, since the licensee is ultimately responsible for satisfying the decommissioning requirements of the LTR. Thus § 20.1403(a) appears to stand for the proposition that a licensee is eligible for restricted release upon showing that it has performed an ALARA analysis of restricted release decommissioning options, and the results of that analysis have caused it not to pursue unrestricted release.
The language of § 20.1403(a) is silent as to why an ALARA analysis of restricted release would cause a licensee not to pursue unrestricted release. If we accept the NRC’s assertion that ALARA cannot be used to compare restricted with unrestricted release, Order at 40-42, however, then the text yields a key meaning: under § 20.1403(a) a licensee can qualify for restricted release without having to make any showing about unrestricted release. In other words, assuming the NRC definition of ALARA, the availability of restrict[380]*380ed release under § 20.1403 would appear to have nothing to do with whether unrestricted release can be attained in a cost-beneficial manner, and everything to do with some property of restricted release. What that property of restricted release is we cannot say, since the NRC’s explanation of the role of ALARA in § 20.1403(a) discusses only its application to unrestricted release, and makes no reference to restricted release. Id.; see also Resp. Br. at 13-15, 54-55, 59.
Further, the understanding of § 20.1403(a) we have just sketched out (in which the critical language of § 20.1403(a) invites attention to some aspect of restricted release) jars with the NRC’s insistence that it “explicitly expressed a preference for unrestricted release in adopting” the LTR: the provision appears to permit restricted release irrespective of the merits of unrestricted release. Order at 39. The NRC is correct that it has repeatedly stated it holds that preference. See, e.g., Radiological Criteria for License Termination, 59 Fed.Reg. 43,200, 43,216 (proposed August 22, 1994) (“The Commission expects the licensee to make every reasonable effort to reduce residual radioactivity to levels that will allow unrestricted release of the site.”); Radiological Criteria for License Termination, 62 Fed.Reg. 39,058, 39.069 (July 21, 1997) (“The Commission continues to believe that unrestricted use is generally preferable for the reasons noted.”) But such words mean little if they are not reflected in the text of the rule and the Commission’s other regulatory pronouncements.
Of course, a reading of the emphasized language in the first sentence of § 20.1403(a) as conditioning eligibility for restricted release solely on some characteristic of restricted release (i.e., without the comparison with unrestricted release that Shieldalloy finds implied) seems in tension with second sentence of the provision, which states that the ALARA analysis referenced in the first sentence must account for “consideration of any detriments, such as traffic accidents, expected to potentially result from decontamination and waste disposal.” 10 C.F.R. § 20.1403(a). Traffic accidents related to waste disposal would seem to have little to do with restricted release, which involves on-site disposal of radioactive materials. By contrast, traffic accidents are an important concern for licensees pursuing unrestricted release, which ordinarily requires transfer of radioactive materials to another location. Order at 37-38 (describing restricted and unrestricted release). Yet the first sentence of § 20.1403(a) speaks only of “restricted conditions” being ALARA.
Other NRC regulations and statements pertaining to ALARA only deepen the confusion. 10 C.F.R. § 20.1003, which defines key terms used in NRC regulations pertaining to protection against radiation (which regulations include the LTR), defines ALARA as “making every reasonable effort” to cut radiation exposure “as far below the dose limits ... as is practical,” with practicality further defined as an open-ended set of “societal and socioeconomic considerations.” 10 C.F.R. § 20.1003. This definition seems to match the NRC’s claim that ALARA is a device for insisting on cost-beneficial radiation reductions below maximum dose limits (e.g., 25 mrem for unrestricted release under § 20.1402), but appears completely alien to the NRC’s reading of § 20.1403(a), under which ALARA is used to assess the cost-efficiency of attaining radiation levels at those limits. See Resp. Br. at 59.
Even harder to square with the NRC’s position are passages of an NRC policy statement called NUREG-1757, which the agency describes as a “comprehensive NRC guidance document” on license ter-[381]*381ruination. Order 24; see “Consolidated Decommissioning Guidance,” NUREG-1757, V6L 2 (Rev.l, Sept. 2006) (“NUREG-1757”), available at http://www.nrc.gov/ reading-rm/doc-colleetions/nuregs/staff/sr 1757/. The document evinces a clear expectation that a licensee must compare unrestricted and restricted release in order to establish eligibility under § 20.1403(a). For example, Chapter 6 of NUREG-1757, entitled “ALARA Analy-ses,” contains the following paragraph:
Appendix N of this volume discusses five different possible benefits.... In most comparisons between alternatives in the same class (e.g., both alternatives result in unrestricted release), the only important benefit should be the collective dose averted. In comparisons between restricted and unrestricted release, the other benefits can become important.
NUREG-1757 at 6-3 (emphasis added).
Pursuing this cross-reference to Appendix N, we find the following passage pertaining to the “benefits” side of ALARA’s cost-benefit analysis:
Regulatory Costs Avoided
This benefit usually manifests in ALARA analyses of restricted release versus unrestricted release decommissioning goals.... When evaluating the ability of a licensee’s proposal for restricted release according to 10 CFR [§ ] 20.1403(a), avoiding these costs should be included in the benefits of the unrestricted release decommissioning alternative. These should not be included as costs related to the restricted release.
Id. at N-6 (emphasis added).
These passages do not appear to be the product of inattentive drafting. NRC officials invoked the same concept in a letter they sent Shieldalloy after accepting of the company’s 2006 decommissioning plan for technical review. That letter, entitled “Request for Additional Information” (“RAJ”), warned Shieldalloy that overestimating the cost of unrestricted release “would bias the net harm or ALARA comparison away from the unrestricted use option.” J.A. at 393.
These statements from NUREG-1757 and the RAI can reasonably be read to call for precisely the kind of comparative dose analysis that Shieldalloy claims is contemplated by § 20.1403(a). Yet the NRC waves off the statements as “additional technical information and guidance on the mechanics of properly performing the ALARA eligibility analysis” that should be “construed in context.” Resp. Br. at 61. Then, without any attempt at explaining what it is in the context that saps these words of their apparent meaning, the agency abruptly concludes that “[n]one of these NRC statements ... supports Shiel-dalloy’s unprecedented comparative dose approach.” Id. at 63. As with its discussion of § 20.1403(a), the agency appears to believe that a mere declaration of the meaning of disputed text, unsupported by any analysis, satisfies our standards of review.
Furthermore, if NUREG-1757 and the RAI mean what they appear to mean, the Order’s insistence that the choice between restricted and unrestricted dispositions can never turn on a direct comparison between the two would appear to be the sort of “swerve” from prior policy that requires explanation. Greater Boston Television v. FCC, 444 F.2d 841, 852 (D.C.Cir.1970); see also Alaska Professional Hunters Ass’n v. FAA, 177 F.3d 1030 (D.C.Cir.1999). Here, the NRC does not seem troubled by its prior inconsistent language, nor does it even “display awareness that it is changing positions.” FCC v. Fox Television Stations, Inc., 556 U.S. 502, 515, 129 S.Ct. 1800, 173 L.Ed.2d 738 (2009); cf. General Elec. Co. v. EPA, 53 F.3d 1324, 1333-34 (D.C.Cir.1995) (finding [382]*382notice inadequate where “the regulations and policy statements are unclear, where the petitioner’s interpretation is reasonable, and where the agency itself struggles to provide a definitive reading of the regulatory requirements”); Yakima Valley Cablevision, Inc. v. FCC, 794 F.2d 737, 745-46 (D.C.Cir.1986). This failure to grapple with the past is especially troubling given the Commission’s total silence on why the uncertainties involved in restricted/unrestricted comparisons (alluded to in the Order at 37) are categorically more impenetrable than those required for comparisons between different plans of restricted release, which the Commission views as entirely permissible.
The NRC trots out the familiar proposition that deference to an agency’s interpretation of its own rule “is all the more warranted when, as here, the regulation concerns a complex and highly technical regulatory program.” Resp. Br. at 53 (quoting St. Luke’s Hosp. v. Sebelius, 611 F.3d 900, 905 (D.C.Cir.2010)). But where the agency writes an opaque and ambiguous rule and then by fiat proclaims its meaning without any effort to consider its text or dispel its mysteries, the agency’s insistence on deference is misplaced. “We cannot defer to what we cannot perceive.” Int’l Longshoremen’s Ass’n v. Nat’l Mediation Bd., 870 F.2d 733, 736 (D.C.Cir.1989). Hand-waving about complexity seems especially unsuitable where the text’s opacity is all of the agency’s choosing and it concerns a complex regulatory program with immense public safety implications.
Our dissenting colleague echoes the Commission’s assertions in its Order and in its brief. See Dissent at 387-94. Thus she fully accepts the idea that § 20.1403(a), as a threshold criterion for use of a restricted release, is exclusively related to difficulties with accomplishing a satisfactory unrestricted release — despite the language in § 20.1403(a) itself that directs our attention to restricted conditions. Id. at 392-94. Like the Commission, our colleague fails to directly engage with the language of § 20.1403(a) in her defense of this reading. Further, the “exegesis” she points to in NUREG-1757, see id. at 392-OS, is merely a mathematical formula used to determine the lowest cost-effective radiation levels afforded by a particular remedial approach, which hardly settles the meaning of § 20.1403(a) or precludes a reading calling for a comparative-dose analysis — if anything, the formula seems to facilitate such a comparison.
Our colleague also suggests that the court has offered “its own interpretation of § 20.1403(a).” Dissent at 392. But far from substituting our “judgment for that of the agency,” Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443, (1983), we have merely done what courts are supposed to do with regulatory language — try to explore the validity of the agency’s interpretation. Deference does not preclude analysis. In the present case, our study of the text led to the conclusion that the Commission’s response to Shieldalloy lacked an apparent textual basis; but that finding of course does not obligate the NRC to accept Shieldalloy’s interpretation of § 20.1403(a). Rather, it requires only that the Commission explain itself in a way that rationally addre'sses the concerns we set out above.
We note finally our dissenting colleague’s contention that Shieldalloy did not properly raise its ALARA claim before the Commission. Her conviction appears to turn on Shieldalloy’s failure to (1) engage in an extensive textual analysis of § 20.1403(a); or (2) provide a “technical rationale” in support of its objections. Dissent at 4r-7. As to the first, we think it [383]*383unreasonable to expect Shieldalloy to have contested the Commission’s baffling interpretation of § 20.1403(a) before that interpretation made its debut in the Commission’s Order. Given the confusing wording of the provision and the more straightforward guidance found in NUREG-1757 and the RAI, we can hardly fault Shieldalloy for eschewing § 20.1403(a) in favor of an argument framed in terms of ALARA and the LTR. Of equal importance, our colleague ignores the extensive back-and-forth between Shieldalloy and the Commission over the past decade, including the Commission’s review of Shieldalloy’s several decommissioning plans (cited by Shiel-dalloy in its comments to the agency, J.A. at 712 n. 20), which spell out Shieldalloy’s understanding of the ALARA principle in considerable detail. See, e.g., Decommissioning Plan for the Newfield Facility 75-92, 154, Report No. 94005/G-28247, Rev. 1 (2005), available at http://pbadupws.nrc. gov/docs/ML0531/ML053190220.pdf; see also Supplemental Appendix (“S.A.”) at 27-44, 120-22. In light of this regulatory history, we are surprised that our colleague so readily accepts the Commission’s claim that it had “largely to guess” at the nature of Shieldalloy’s objections. See Order at 35; Dissent at 4.
As to Shieldalloy’s failure to provide a “technical rationale” (a term which neither the Commission nor our colleague has bothered to define), we do not see how that omission hamstrung the Commission’s ability to grasp objections based on the contention that the NRC rules and directives permitted a licensee to choose the least dangerous solution. As the Commission itself observed, its rejection of Shieldalloy’s ALARA claim turned on the company’s “fundamentally inaccurate understanding of our license termination requirements,” Order at 36, and not a flawed engineering analysis. Even assuming otherwise, the reams of technical data Shieldalloy submitted to the Commission in its decommissioning plans, see, e.g., S.A. at 3-127, certainly supplied whatever quantitative detail the Commission might have needed to appreciate Shieldalloy’s position.
For the reasons stated, we again find the NRC’s transfer of authority to New Jersey arbitrary and capricious. We therefore grant Shieldalloy’s petition, vacate the NRC’s transfer of authority, and remand for proceedings consistent with this opinion.
So ordered.