We have the law. To compensate for the destruction of environmentally protected wetlands, a permit must identify acreage apart that will protect the environment and compensate for what is destroyed. The administrative record must demonstrate how that decision was made, such that we uphold the decision not for its correctness but for its rational support.
The district judge carefully studied the justification here and saw that gaps exist without more than conclusions. Now, the circuit court skips over those gaps. I dissent and explain myself in two respects.
A. Out-of-Kind Mitigation and the Clean Water Act
The pipeline project will clear 262 acres of wetlands in the Atchafalaya Basin. That process will impact two resource types: cypress-tupelo swamp and bottomland-hardwood *705forest. In turn, the Corps applied its functional assessment tool (the Louisiana Rapid Assessment Method, or LRAM) and determined that the project's impact called for the purchase of 232.8 acres of cypress-tupelo swamp and 80 acres of bottomland-hardwood forest from mitigation banks. But in what the Corps labels an "unfortunate[ ]" turn of events, one of the chosen mitigation banks did not have the number of cypress-tupelo acres necessary to match a fully in-kind mitigation. So the Corps sanctioned instead the purchase of 69 cypress-tupelo acres and 243.8 bottomland-hardwood acres. In other words, the Corps offset cypress-tupelo harm with 69 in-kind cypress-tupelo acres and 163.8 out-of-kind bottomland-hardwood acres. The Corps thereby swapped each acre of unaccounted-for cypress tupelo with an acre of surplus bottomland hardwood-it treated the two resource types interchangeably.
Under the Clean Water Act and its corresponding regulations, before the Corps could order the above out-of-kind swap, it bore a duty to (1) determine that such mitigation "will serve the aquatic resource needs of the watershed" and (2) so document that "basis for authorization ... in the administrative record." 33 C.F.R. § 332.3(e)(2).
The court believes a satisfactory explanation lies in the Corps' LRAM tool. I disagree. The LRAM lacks a critical explanatory component and thereby leaves the Corps' out-of-kind mitigation unsubstantiated.
The court explains the LRAM's function as follows: "[T]he LRAM scores wetlands impact based on [various] factors ... [and] assigns values to the quality of the wetlands and of the mitigation banks, converts the values into credits, and determines on a watershed basis how many acres in mitigation banks must be purchased by the prospective permittee." In elementary terms, the LRAM compares land to land (impact site to mitigation bank) and calculates a ratio that, when applied to impacted acres, produces a suggested quantity of mitigation acres.
However, the Corps still must accommodate another variable: resource type . The regulations prefer in-kind over out-of-kind mitigation precisely because different resource types supply different functions, or said another way, similar resource types are "most likely to compensate for the functions and services lost at the impact site." 33 C.F.R. § 332.3(e)(1). To that end, the LRAM identifies a laundry list of habitats and groups them into six resource categories: bottomland-hardwood forest, cypress-tupelo swamp, pine flatwoods-savanna, coastal prairie, fresh-intermediate marsh, and brackish-saline marsh. Each category encompasses habitats that either provide "similar wetland functions or naturally exist together as a community." The LRAM then highlights the presumption that "in-kind habitat replacement" will "assure similar functions and services that are lost at an impact site are gained at a mitigation site." Thus, when the Corps applies in-kind mitigation to the LRAM's calculated acreage, there is no need to manipulate the end product because the Corps' path is self-explanatory-the ecological functions intrinsic to both parcels and resource types are fully documented on both sides of the mitigation equation.
But when the Corps substitutes on the back end a resource that is out of kind-defined by the LRAM as "a resource of a different structural and functional type from the impacted resource"-the LRAM can no longer rely on a presumption of like functions for like resources. How, then, does the LRAM go about accounting for the variation between the resource impacted on the front end and the one purchased *706on the back end? The LRAM's ratio itself does not factor in the resource type purchased on the back end. So, lest we assume that the LRAM's calculated acreage is entirely fungible across all resource types-something no party or the court goes so far as to suggest-there must be something else in the LRAM to translate impacts from one resource to another (in this case, to justify the one-to-one substitution of bottomland hardwood for cypress tupelo).
In that crucial respect, the LRAM is conspicuously silent. It mentions "out of kind" a single time: to define the term. Nowhere does the LRAM explain how to quantify impacts to one resource in terms of another, much less how cypress tupelo and bottomland hardwood-habitats of a "different structural and functional type"-can swap seamlessly for each other in terms of the basin's resource needs. As useful as it otherwise may be, the LRAM is simply not a tool for out-of-kind mitigation.
Nor does the Corps' Section 404 Environmental Assessment bridge the explanatory gap. There the Corps grounded its out-of-kind swap on the bare fact that "there [were] not enough [in-kind] credits available for purchase in the basin." But lack of in-kind credits, standing alone, says nothing of the "resource needs" of the basin-the principal consideration that must accompany any order of out-of-kind mitigation. 33 C.F.R. § 332.3(e)(2).
The Corps did not meet its regulatory burden to explain out-of-kind mitigation in this case. From the administrative record, then, the Corps' "path may [not] reasonably be discerned." Nat'l Ass'n of Home Builders v. Defs. of Wildlife , 551 U.S. 644, 658, 127 S.Ct. 2518, 168 L.Ed.2d 467 (2007) (internal quotation marks omitted). The district court was therefore correct to enjoin construction.
B. Mitigated Versus Ambiguous Findings of No Significant Impact
Whatever the ultimate merits of the plaintiffs' claim under the National Environmental Policy Act, we ought to at least apply the right standard. I disagree with the court's decision to adopt various tiers of scrutiny between those so-called "mitigated Findings of No Significant Impact" (FONSIs) and those other FONSIs in which mitigation plays a prominent but facially ambiguous role.
In O'Reilly v. United States Army Corps of Engineers , 477 F.3d 225 (5th Cir. 2007), we held insufficient an Environmental Assessment "that fail[ed] to articulate how the mitigation measures will render the adverse effects insignificant." Id. at 227. The Corps argues, however, that O'Reilly 's scrutiny applies only to mitigated FONSIs, those in which an agency engages in a two-part finding: (1) project impacts alone would be significant but (2) with mitigation, the impacts are reduced to insignificance. This case, the Corps says, does not involve a mitigated FONSI because the agency considered the project impacts and mitigation all at once before issuing a single finding of no significant impact. The Corps draws its labels for this distinction from a 2011 guidance document. See 76 Fed. Reg. 3843, 3847-48. And the court appears to accept the Corps' distinction wholesale.
But that distinction is all form with no substance. O'Reilly stands for a fundamental proposition: When mitigation is a necessary part of a FONSI, the agency bears a duty to explain why the mitigation will be effective. 477 F.3d at 231-32. Thus framed, there are but two types of FONSIs under O'Reilly : (1) those in which mitigation is an integral part of the insignificant outcome and (2) those in which the mitigation is ultimately gratuitous-that is, *707when the impacts would be insignificant even without mitigation. There is no third option.
Of course, the manner in which an agency arrives at its FONSI can make the role of mitigation apparent on the face of the administrative record. When the agency issues a formal mitigated FONSI, we know for sure that mitigation was an integral piece. But, as here, when the Environmental Assessment lumps project impacts and mitigation into a single consideration with no further explication, the record obscures whether the impacts would have been significant absent the mitigation. All the same, these facially ambiguous assessments can involve necessary mitigation. And that is more than common sense talking; the Corps' own guidance document tells us that ambiguous assessments might well involve mitigation that "reduce[s] the projected impacts of agency actions to below a threshold of significance." 76 Fed. Reg. 3843, 3847. In such a case, there is zero substantive difference between a mitigated FONSI and a facially ambiguous one and, as a consequence, zero reason to treat the two any different.
So, the question becomes, was mitigation necessary to this project's insignificant impact? On the one hand, the Corps is unwilling to concede that mitigation was necessary to reduce the project's impact to insignificance. This despite the pages and pages of the Environmental Assessment detailing the hundreds of acres of shredded wetlands and corresponding compensatory mitigation. See 33 C.F.R. § 320.4(r)(1) (explaining that compensatory mitigation is meant to rectify "significant resource losses") (emphasis added). Nonetheless, that must necessarily mean the project's impacts would be insignificant even without mitigation. But as it so happens, the Corps is unwilling to say that either. And therein lies the paradox-the ambiguous record here enables the Corps to tiptoe on a nonexistent fence between the only two realities: mitigation that matters and mitigation that does not.
When an agency cloaks the importance of mitigation behind an ambiguous administrative record, I would hold the agency to the standard articulated in O'Reilly .
I respectfully dissent.