Full opinion text
WOOD, Chief Judge. The invention of carbonless copy paper by NCR Corporation in the mid-1950s solved a small problem and created a large one. Though it alleviated the messy side effects of carbon paper for those who wanted copies in the pre-photocopy era, over the next quarter-century it became clear that the cost of this convenience was large-scale environmental contamination. That is because, until the early 1970s, the substance coating the paper included poly-chlorinated biphenyls (PCBs), a highly toxic pollutant. In the course of producing the carbonless paper, large quantities of PCBs were dumped into the Lower Fox River in Wisconsin, the site of the paper’s production. (References to the River in this opinion mean the Lower Fox, unless the context requires otherwise.) Recyclers poured yet more PCBs into the River. In time, the problem attracted the attention of the federal government, which, invoking the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA) (popularly known as the Superfund), eventually ordered the responsible parties to clean up the mess. See 42 U.S.C. § 9601 et seq. This case requires us to decide who should foot the considerable bill. Once the Environmental Protection Agency (EPA) identifies the site of an environmental hazard that requires remediation under CERCLA, the statute’s financial responsibility rules are triggered. CERCLA imposes a “pay-first, split-the-bill-later” regime. Any individual persons or corporations meeting certain statutory criteria can be required to pay for the cleanup. Anyone who paid can then recover contribution from other responsible parties in accordance with that entity’s equitable share of the costs. NCR was the exclusive manufacturer and seller of the emulsion that gave treated paper its “carbonless-copy” character during what the parties call the Production Period (1954 to 1971). That emulsion, unfortunately, used Aroclor 1242 as a solvent, and Aroclor 1242 is a PCB. Given its role in the pollution, NCR has thus far picked up the lion’s share of the cleanup tab for the River site. In this action it seeks contribution from several other paper mills along the river. Those firms were in the recycling business; they bought NCR’s leftover scraps of carbonless copy paper, washed the harmful chemicals off into the River, and recycled the pulp to make new paper. Several ancillary questions and counterclaims were raised along with NCR’s contribution claim, and we will address each in turn. The main event, though, relates to the equitable allocation of costs. The district court, after holding a first phase of discovery on the question of when each party became aware that the primary chemical ingredient of carbonless copy paper was harmful, held that NCR was not entitled to any equitable contribution from the paper mills. Worse than that, from NCR’s vantage point, the court held that the mills had meritorious counterclaims for cost recovery from NCR. NCR appeals that decision, and the defendant recyclers cross-appeal a handful of matters decided against them. Before addressing these matters, we begin with some background about the cleanup effort. I. Background Facts The Lower Fox River begins at Lake Winnebago in northeastern Wisconsin and winds northeasterly for 39 miles until it discharges into Green Bay, which flows into Lake Michigan. For decades, the River has been to papermaking what Pittsburgh once was to steel: the heart of the industry, and home to the highest concentration of paper mills in the world. See Region 5 Cleanup Sites: Background, Environmental Protection Agency (Aug. 3, 2011), http://www.epa.gov/region05/ cleanup/foxriver/background.htm (all websites cited last accessed Sept. 24, 2014). In addition to their infamous smell, paper mills produce a good deal of solid byproduct, which they long disposed of by dumping it into the River. This had a deleterious effect on the River and its ecosystem, and by 1970 the sorry state of the River was visible to the naked eye. Among the solid matter suspended in the mills’ effluent were PCBs, which were the pollutant that attracted the attention of the EPA in the mid-1990s. PCBs are carcinogenic for humans and animals alike, and they have harmful noncarcinogenic effects on the immune, reproductive, neurological, and endocrine systems, as well as the skin. See Health Effects of PCBs, EPA (June 13, 2013), http://www.epa.gov/ waste//hazard/tsd/pcbs/pubs/effects.htm. By the time their use was banned by the EPA in 1979, some 250,000 pounds of PCBs had been released into the River. See Lower Fox River and Green Bay Site, EPA (Apr. 2, 2014), http://www.epa.gov/ region05/cleanup/foxriver/. The PCBs in the River can all be traced back to NCR’s carbonless copy paper, which it (along with other companies with which it contracted) produced between the mid-1950s and 1971. As the name suggests, carbonless copy paper permitted a writer or typist to make instant copies of documents without the use of carbon paper. This effect is achieved by coating the back of a top sheet of paper with an emulsion containing “microcapsules” of dye and solvent; the microcapsules burst when a user writes on the sheet and thereby reproduce the same image on the lower sheet. A critical ingredient of the emulsion was Aroclor, a PCB-based chemical sold by Monsanto. NCR manufactured the emulsion, which it then sold to two companies (Appleton Coated Paper Company and Combined Paper Mills), which coated the paper and sold the finished product back to NCR for commercial distribution. Those two companies were formally independent from NCR until 1969 and 1970, respectively, when they became NCR’s wholly owned subsidiaries. The PCBs used to make carbonless copy paper ended up in the Lower Fox River in two principal ways. First was the straightforward one: some of the emulsion used to coat the paper was necessarily lost in the production process and was mixed with the wastewater that the mills released into the River. Explaining the second way requires us to give a bit more background about the paper industry. Producing paper from raw materials is relatively expensive. The production process creates a fair amount of waste, scraps, and undersized rolls that are unusable by the original manufacturer; these are called “broke” in the trade. Making paper from recycled broke is cheaper than making it from scratch, a fact that spurred the growth of a sub-industry of “recycling mills.” The nongovernmental defendants in this case are such mills. These mills purchase broke from other paper mills through middlemen and use it to make paper. The companies that coated paper with NCR’s emulsion also participated in the normal industry practice of selling broke to recycling mills. Upon receipt of the broke, the recycling mills would process it to separate the usable fibers from the coating, thus removing the PCBs from the portion of the paper that went into the new product. The waste (including of course the PCBs) was then dumped into the River with the mills’ wastewater. Another problem with PCBs is that they attach readily to solids and do not degrade. As more and more PCBs were dumped into the River each year, they accumulated in the riverbed. Eventually, people became aware of the dangers of PCBs, and both private and governmental entities began to take action. Monsanto stopped selling products containing PCBs in August 1970, and NCR ceased using PCBs in its carbonless copy paper once its supplies of Aroclor ran down the following year. The EPA issued regulations largely banning the use of PCBs in 1979. See 40 C.F.R. § 761.1 et seq. Congress passed CERCLA in 1980 to spur the environmental cleanup of sites contaminated by hazardous substances. See Pub.L. No. 96-510, 94 Stat. 2767 (1980) (codified as amended at 42 U.S.C. § 9601 et seq.). The EPA, in coordination with the Wisconsin Department of Natural Resources (WDNR), set its sights on the Lower Fox River as a CERCLA cleanup target in 1998; it issued a final cleanup plan in 2002. See United States v. NCR Corp., 688 F.3d 833, 836 (7th Cir.2012). The plan divided the River into five “operable units,” which is jargon for geographic sections, and ordered a combination of dredging the riverbed and capping contaminated areas so as to remove and contain the PCBs and prevent them from reaching Lake Michigan, where it would be all but impossible to do anything about them. When the EPA determines that environmental remediation is required, CERCLA shifts the cost of that cleanup to the parties responsible for creating the hazard and away from taxpayers, who otherwise would be left to pick up the bill. The statute identifies who is regarded as a Potentially Responsible Party (PRP) in a cleanup action: it includes the current owners and operators of the cleanup site; the owners and operators at the time that the hazardous substance was disposed; parties that “arranged for” disposal of the substance; and parties that accepted the substance for transportation to a disposal site of their choosing. CERCLA § 107(a), 42 U.S.C. § 9607(a). All PRPs are liable for costs incurred by the state and federal governments related to the remediation. At the liability stage, no consideration is given to a party’s relative fault in contributing to the hazardous conditions at the site; any party meeting one of the statutory definitions is potentially liable for the full cost of cleanup. With one important exception to which we will return, all of the nongovernmental litigants before us are admitted PRPs for the Lower Fox River cleanup. Many of these parties have challenged their liability to the government in separate litigation related to this case, but that question is not before us here. We are concerned with separate CERCLA provisions that allow a PRP that believes it has paid cleanup costs in excess of its fair share to sue to recover contribution from other PRPs jointly liable for cleanup at the same site. See CERCLA § 113(f), 42 U.S.C. § 9613(f). In such an action, the district court is directed to allocate contribution costs under section 113(f) “using such equitable factors as the court determines are appropriate.” Id. § 9613(f)(1). After the EPA issued a unilateral administrative order in 2007 governing the cleanup of operable units 2 through 5 at the Lower Fox River site, NCR took the lead in responding and bore much of the cost of remediation from that point onward. Believing that its financial burden has by now exceeded its fair share, NCR filed this suit for contribution against the other PRPs named in the administrative order. Over the next several years, the district court disposed of a number of issues in the case. It held that NCR and its indemnitor and former subsidiary Appvion were not entitled to any equitable contribution for the expenses they incurred as part of the Lower Fox River cleanup. It further ruled that the recycling mills should receive contribution from NCR for their costs, including those incurred in the form of “natural resources damages.” Between the appeal and the cross-appeal, we have a considerable array of issues before us. They include the following: (1) whether NCR (and Appvion, which stands in the same position as NCR for this purpose, and so unless otherwise required we use “NCR” to refer to both) are entitled to recover any portion of the cost of clean-up under either section 107 or 113 of CERCLA, 42 U.S.C. §§ 9607, 9613; (2) whether NCR has any arranger liability; (3) whether the district court erred by awarding costs to Glatfelter without offsetting its insurance recoveries; (4) whether NCR is hable for natural resource damages; (5) whether the court erroneously dismissed Glatfelter’s claims based upon discharges at Portage, Wisconsin; and (6) whether the district court wrongly dismissed Glatfelter’s common-law counterclaims as preempted. We address the issues in that order. II. Cost Recovery or Contribution We begin with NCR’s assertion that it should not be required to proceed by way of a contribution action under section 113(f) at all, but instead should be able to sue the defendant mills under the more plaintiff-friendly provision for cost recovery found at CERCLA § 107(a), 42 U.S.C. § 9607(a). We review this question of statutory interpretation de novo. Storie v. Randy’s Auto Sales, LLC, 589 F.3d 873, 876 (7th Cir.2009). A Defining the relation between cost-recovery suits under section 107 and contribution actions under section 113 has proven vexing for courts. Section 107(a) is meant to support a claim for parties to recover costs incurred during a self-initiated environmental cleanup, while section 113(f) creates a right to contribution for parties already subject to liability in either a section 107 action or an action by the government under CERCLA § 106, 42 U.S.C. § 9606. See United States v. Atlantic Research Corp., 551 U.S. 128, 138-39, 127 S.Ct. 2331, 168 L.Ed.2d 28 (2007). Proceeding by way of section 107(a) holds advantages for a plaintiff insofar as it can recover “any ... necessary costs of response incurred,” and defendants can assert only the statutory defenses enumerated in section 107(b), such as acts of God, acts of war, and third-party omissions. See 42 U.S.C. § 9607(a)(4)(B). See also California v. Neville Chem. Co., 358 F.3d 661, 672 (9th Cir.2004). Equity plays no role in a section 107(a) action, in contrast to a section 113(f) action, where the entire allocation of costs is equitable, and even a defendant who concedes statutory liability may argue that it should bear none of the costs of response. The defendant in a section 107(a) action can always bring a section 113(f) counterclaim if the plaintiff is a PRP, but the burden of proof would then be on the counterclaiming defendant to demonstrate an entitlement to contribution. See Atl. Research Corp., 551 U.S. at 140, 127 S.Ct. 2331 (explaining availability of 113(f) counterclaim); Goodrich Corp. v. Town of Middlebury, 311 F.3d 154, 168 (2d Cir.2002) (discussing burden of proof). Whether a party must proceed under section 107(a) or 113(f) depends on the procedural posture of the claim. Atl. Research Corp., 551 U.S. at 139-40, 127 S.Ct. 2331 (citing Consol. Edison Co. of N.Y. v. UGI Utils., Inc., 423 F.3d 90, 99 (2d Cir.2005)). If a party already has been subjected to an action under section 106 or 107, or has “resolved its liability to the United States or a State for some or all of a response action or for some or all of the costs of such action in an administrative or judicially approved settlement,” it must proceed under section 113(f). 42 U.S.C. §§ 9613(f)(1), (f)(3)(B); see Bernstein v. Bankert, 733 F.3d 190, 201-02 (7th Cir. 2012). Conversely, a party that has not been subjected to an enforcement or liability action, and that is not party to a settlement, may proceed under section 107(a). See Cooper Indus., Inc. v. Aviall Servs., Inc., 543 U.S. 157, 166-68, 125 S.Ct. 577, 160 L.Ed.2d 548 (2004). Section 113(f) is closed to a litigant without a preexisting or pending liability determination against it even if it wants to proceed by that route, because that statute creates a right to contribution, and contribution exists only among joint tortfeasors liable for the same harm. See Restatement (Second) of Torts § 886A. See also Cooper Indus., 543 U.S. at 166-68, 125 S.Ct. 577 (unavailability of section 113(f)); Atl. Research Corp., 551 U.S. at 135-36, 127 S.Ct. 2331 (availability of section 107(a)). Thus, although a strict reading of the phrase “necessary costs of response” in section 107(a) might suggest that parties who pay pursuant to an enforcement action might be able to sue under section 107(a), this court — like our sister circuits — restricts plaintiffs to section 113 contribution actions when they are available. Bernstein, 733 F.3d at 206. See also Hobart Corp. v. Waste Mgmt. of Ohio, Inc., Nos. 13-3273, 13-3276, 2014 WL 3397147 at *7 (6th Cir. July 14, 2014) (agreeing with Bernstein that sections 107(a) and 113(f) provide mutually exclusive remedies). B The question whether NCR may sue under section 107(a) is controlled by our decision in Bernstein. In that case we held that a settlement with the EPA “resolves” a party’s liability when the agreement, by its own terms, releases a party from CERCLA liability in an enforcement action. Bernstein, 733 F.3d at 204-15. We contrasted two Administrative Orders of Consent between the PRP and the EPA, both of which contained explicit language stating that the PRP would not be protected from suit by the EPA until after it had completed its cleanup obligations under the orders. The earlier order dealt with a project on which the PRP had completed work. The PRP was therefore limited to a contribution action under section 113(f) to recover its costs. The later order stipulated that “nothing in this Order constitutes a satisfaction or release from any claim or cause of action against the Respondents,” and that “[tjhese covenants [not to sue] are conditioned upon the complete and satisfactory performance by Respondents of their obligations under this Order.” Id. at 203, 207. The agreement memorialized in that order disclaimed any admission of liability by the PRP, Id. at 204, and the terms of the agreement gave the EPA the right to sue up until the point when the PRP completed its obligations. The PRP’s liability therefore could not be considered “resolved” by the order for CERCLA purposes. This meant that the order could form the basis of a section 107(a) suit because the PRP was still in the process of complying with it. NCR’s response costs at the River site arose under three orders: a consent decree following a 2001 suit by the EPA and WDNR, an Administrative Order of Consent for design work in 2004, and a Unilateral Administrative Order for remedial work in 2007. The company concedes that its costs under the 2001 consent decree must be recovered under section 113(f), if at all. It continues, however, to assert its ability to sue the recycling mills under section 107(a) for costs incurred pursuant to the 2004 and 2007 orders. Its argument with respect to the 2007 order is easily dispatched. The government filed a lawsuit to enforce that order in 2010. Under CERCLA’s express terms, a party may seek contribution from any other PRP “during or following any civil action under section 9606 [CERCLA § 106] of this title or under section 9607(a) [CERCLA § 107(a)] of this title.” CERCLA § 113(f)(1); 42 U.S.C. § 9613(f)(1) (emphasis added). Thus, a section 113(f) action is available to NCR for the costs incurred under the 2007 order. This means that section 107(a) is not available. See Bernstein, 733 F.3d at 206. We are unpersuaded by NCR’s contention that the costs it incurred under the order before the action was filed in October 2010 were “voluntary,” and thus not part of the costs recoverable under section 113(f). Such slicing and dicing of costs incurred under the same administrative order makes little sense when a party’s liability for all of those costs will ultimately be determined in the enforcement action. As for the 2004 Administrative Order of Consent, our analysis again is guided by Bernstein. The question whether NCR has resolved its liability to the government through the consent order — and thus is limited to section 113(f) — is a matter of contract interpretation. On that score, the consent order here diverges in every meaningful way from the one in Bernstein that left section 107(a) available. In Bernstein, the covenants not to sue were “conditioned upon the complete and satisfactory performance” of the PRP’s obligations. 733 F.3d at 203. In contrast, under the 2004 order here both the EPA and WDNR “eovenant[ed] not to sue or to take administrative action against Respondents [under CERCLA or state law] for performance of the work.” It explicitly provided that “[t]hese covenants not to sue shall take effect upon the Effective Date ” (emphasis added); in Bernstein, the covenants did not take effect until completion of the work. To be sure, the NCR order also has language conditioning the covenants on “satisfactory performance” of NCR’s obligations. But this means only that the federal or state government could sue NCR if it breaches the agreement — a standard arrangement that is consistent with the fact that neither the EPA nor Wisconsin could sue NCR if it complied with its obligations. The agreement resolved NCR’s liability, and so the district court correctly held that it limited NCR to proceeding under section 113(f). To hold otherwise would mean that no consent order could resolve a party’s liability until the work under it was complete. Such a rule would be contrary both to the analysis in Bernstein and to common sense. C Appvion finds itself in a materially different position from NCR when it comes to the choice between cost recovery and contribution. In fact, it appears to be in an unusual, possibly unique, position among parties incurring costs under CERCLA: it was initially identified as a PRP by the government and paid response costs in that capacity, but later it was held to fall outside of CERCLA’s statutory grounds for liability. It is now on the hook for response costs only as NCR’s indemnitor pursuant to an agreement signed when the companies split up. It is seeking the costs of response it paid directly while it was regarded as a PRP. To understand how Appvion ended up in this position, it is helpful to look at its corporate history. It started out as Appleton Coated Paper Company, one of the two mills that coated copy paper with NCR’s PCB-based emulsion and then sold the finished carbonless product back to NCR. In 1970, NCR acquired all of its stock, and it became NCR’s wholly owned subsidiary. It then merged with another NCR subsidiary and was renamed “Appleton Papers, Inc.,” which merged with NCR in 1973 and became an unincorporated division. NCR later sold the assets of that division to an outside corporation called Lentheric, which changed its name in 1978 to Appleton Papers Inc. (without a comma, unlike its previous iteration). It finally became Appvion after yet one more name change in 2013. The 2007 Unilateral Administrative Order identified Appvion as a PRP, and the government named it as a defendant in its 2010 enforcement action under section 106. In late 2011, however, the district court decided that Appvion had not assumed Appleton Coated Paper Company’s CERCLA liability when it was sold by NCR, and that the liability remained with NCR as a matter of contract. This ruling created a conceptual problem, because Appvion already had incurred costs of compliance under the 2007 order. The district court sidestepped the question — apparently of first impression— whether a party formerly identified as a PRP but later found not to have that status (and thus not to be liable under CERCLA) could recoup its costs under section 107(a), 113(f), neither, or both. Instead, the court found that Appvion had agreed to indemnify NCR for CERCLA costs as part of an earlier settlement agreement between the companies. (The existence of this agreement was the basis for Appvion’s successful argument that it was not directly liable in the government’s 2010 enforcement action.) The court held that whatever costs Appvion had incurred could not be traced to CERCLA; they were incurred instead pursuant to the indemnity agreement. The agreement provided, the court concluded, that Appvion’s “rights, limitations, and defenses” were the same as NCR’s. Appvion was subrogated to NCR’s contribution claim through CERCLA § 112(c), 42 U.S.C. § 9612(c)(2) (providing for. subrogation), and could recover its costs exclusively through NCR and this subrogation arrangement. In the run-of-the-mill case, a rule that a CERCLA indemnitor (here, Appvion) is limited to proceeding through its indemnitee would be sound. The leading case in this area is Chubb Custom Ins. Co. v. Space Systems/Loral, Inc., 710 F.3d 946 (9th Cir.2013). There, the Ninth Circuit held that an indemnitor insurance company had not incurred costs “under CERCLA” because the money it disbursed did not correspond to “costs of response,” but rather to independent contractual obligations based on the indemnified party’s costs of response. Id. at 952-53. If the rule were different, the court feared, insurers could exploit it to make an end-run around section 113(f): they could make payments directly to jointly liable parties instead of filtering the money through their indemnitees. They could then turn around and bring a section 107(a) action against those parties because the insurer (by hypothesis) was not a PRP. Chubb reflects the fact that an indemnitor normally is able to satisfy the full amount of any claim it would otherwise have brought under section 107(a) by standing in the shoes of its indemnitee in a section 113(f) action. Permitting it to operate outside section 113(f) would unjustly improve its litigating position. This case, however, does not fit the normal pattern. Appvion was not acting as an indemnitor when it paid the response costs it now seeks to recover; as of then, it was a PRP. Even if it were to stand in NCR’s place, it could not hope to recover for what it paid, because NCR would not be entitled to contribution for response costs that it did not bear. The indemnity agreement is a one-way street: it does not expand NCR’s costs of response and thus allow it to sue for contribution toward Appvion’s costs for paying at least part of NCR’s contribution share. It is conceivable that in some circumstances, an indemnitor-former PRP that incurs direct response costs and is later found not liable under CERCLA may wish to recover costs even from the party it is indemnifying. Imagine, for example, a hypothetical situation in which Appvion paid $20 million in response costs before it was found not to be a PRP, and later NCR was assigned to pay $5 million in contribution to its fellow PRPs in a section 113(f) action. Appvion would be on the hook to NCR for $5 million under the indemnity agreement (assuming it indemnifies at 100%). Even though in principle it starts out wanting to recover the $20 million it paid in response costs because it was not a liable party in the first place, in the end the sum needed to make it whole would be $25 million — its own response costs of $20 million plus the $5 million paid under the indemnity agreement. There is good reason to take the position that at least the $20 million should be allocated equitably among the PRPs, not irrevocably assigned to a party that was erroneously identified as a PRP and in that capacity complied with the EPA’s order while contesting its liability. Principles of contract should govern any rights between the parties under the indemnity agreement. It is not readily apparent which statutory mechanism is the proper one for reimbursing the erroneously imposed costs paid by the non-PRP. But it seems apparent that something should be available. This is true regardless of any independent indemnity agreement with another PRP, insofar as costs that fall outside the agreement are concerned. A section 113(f) action is a poor fit for this situation, because contribution exists only among joint tortfeasors. Restatement (Second) of Torts § 886A. A party such as Appvion that is no longer a PRP logically cannot be a joint tortfeasor for CERCLA purposes. That leaves us with section 107(a). It turns out to be a reasonably good fit, if one characterizes Appvion’s response payments as constructively voluntary. When it turned out that Appvion could not legally be bound to pay the response costs that it had already paid under the order, the nature of those payments had to be reconsidered. Had Appvion been properly characterized from the start, any payments it might have made would have been wholly voluntary. It makes sense, we think, to apply that lack of compulsion retroactively. Under Atlantic Research Corp., Appvion is therefore entitled to bring a section 107(a) action against the PRPs sharing liability for the Lower Fox River site. See 551 U.S. at 135-36, 127 S.Ct. 2331. We stress that this action is available not because Appvion is NCR’s indemnitor, but precisely because it was not indemnifying NCR when it incurred these response costs. Section 107(a) is not available for the recovery of any costs that arise through Appvion’s indemnity agreement; we agree with the Ninth Circuit’s holding in Chubb that the company qua indemnitor is limited to its indemnitee’s CERCLA remedies. Maintaining the bright-line distinction between costs incurred as a former PRP and costs incurred as an indemnitor should prevent the undesirable scenario that would permit an indemnitor to exploit the use of direct CERCLA payments to enhance its litigating position, while safeguarding the rights of non-responsible parties not to have to contribute to the costs of clean-up. Because the district court held that Appvion could not sue under section 107(a), we must reverse its decision and remand for further proceedings. III. Rights to Contribution We now turn to two related questions on the merits: whether the district court abused its discretion when it held on summary judgment that NCR was not entitled to contribution for any of its response costs at the Lower Fox River site; and whether the court correctly ruled that the recycling mills were entitled to 100% contribution from NCR for their own costs, as they asserted in counterclaims. We discuss these two points together, as they are just two different ways of asking whether NCR can be held responsible for all response costs, no matter who paid initially. We generally review a grant of summary judgment de novo. A party is entitled to summary judgment if, viewing the evidence in the light most favorable to the non-moving party, there are no genuine issues of material fact and the moving party is entitled to judgment as a matter of law. See Envtl. Transp. Sys., Inc. v. ENSCO, Inc., 969 F.2d 503, 507 (7th Cir. 1992); Fed.R.Civ.P. 56(a). The equitable allocation of response costs in a CERCLA action, however, is for the district court to decide in its discretion, and we do not start from scratch in determining whether we would allocate every penny in the same way. See Browning-Ferris Indus, of Ill., Inc. v. Ter Maat, 195 F.3d 953, 957, 959 (7th Cir.1999) (holding that district court “did not abuse [its] discretion”). Instead, this court has implicitly accepted, and some of our sister circuits have more explicitly stated, that we review the district court’s allocation for abuse of discretion, even when the issue comes to us on appeal from summary judgment. See ENSCO, 969 F.2d at 506, 509 (indicating district court decided case on summary judgment and discussing court’s discretion under CERCLA); United States v. R.W. Meyer, Inc., 932 F.2d 568, 573 (6th Cir.1991) (applying abuse of discretion standard when evaluating summary judgment); see also Goodrich Corp. v. Town of Middlebury, 311 F.3d 154, 170 (2d Cir.2002); Tosco Corp. v. Koch Indus., Inc., 216 F.3d 886, 894 (10th Cir.2000); Boeing Co. v. Cascade Corp., 207 F.3d 1177, 1187 (9th Cir.2000). We begin with the text of section 113(f): Any person may seek contribution from any other person who is liable or potentially liable under section 9607(a) [CERCLA § 107(a) ] of this title, during or following any civil, action under section 9606 [CERCLA § 106] of this title or under section 9607(a) of this title. Such claims shall be brought in accordance with this section and the Federal Rules of Civil Procedure, and shall be governed by Federal law. In resolving contribution claims, the court may allocate response costs among liable parties using such equitable factors as the court determines are appropriate. Nothing in this subsection shall diminish the right of any person to bring an action for contribution in the absence of a civil action under section 9606 of this title or section 9607 of this title. 42 U.S.C. § 9613(f) (CERCLA § 113(f)). In a contribution action the court is not concerned with deciding whether a party is hable under CERCLA. That is a question for the enforcement action; the contribution action merely imports its conclusions. Even so, meeting a statutory liability trigger renders a party only potentially liable for contribution under CERCLA. ENSCO, 969 F.2d at 507. It is within the district court’s power to require all, none, or some intermediate share of contribution from a PRP, depending on the court’s weighing of the equities. We have described the district court’s authority in this area as “broad and loose.” Browning-Ferris, 195 F.3d at 957. CERCLA not only entrusts the district court to make the ultimate equitable allocation of costs, but it also grants the court the authority to decide which equitable factors will inform its decision in a given case. Kerr-McGee Chem. Corp. v. Lefton Iron & Metal Co., 14 F.3d 321, 326 (7th Cir.1994). In that connection, “a court may consider several factors, a few factors, or only one determining factor ... depending on the totality of circumstances presented to the court.” ENSCO, 969 F.2d at 509. The fact-intensive inquiry is “particularly suited” to case-by-case analysis, and we have resisted attempts to impose upon district courts a requirement to either include or to ignore any particular factors in its ultimate decision, even ones that might strike an unbiased observer as salient facts. Id. Even so, we have a responsibility to do more than rubber-stamp the district court’s decision, and so we begin by exploring how the court arrived at its decision to place 100% of the financial burden for cleaning up the Lower Fox River on NCR. A After the court sorted out the questions of cost-recovery and contribution, it held a scheduling conference under Rule 26(f) and asked the parties to lay out a discovery plan. NCR and Appvion proposed that the court set a firm trial date and permit full discovery into all relevant matters at once. The defendants countered with the suggestion that the court limit initial discovery to “(1) when each party knew, or should have known, that recycling NCR-brand carbonless paper would result in the discharge of PCBs to a water-body, thereby risking environmental damage; and (2) what, if any, action each party took upon acquiring such knowledge to avoid the risk of further PCB contamination.” Defendants argued that actual or constructive knowledge of the risk and any failure to take immediate action to prevent the risk would be the key factors in determining the equitable allocation of costs. The court adopted the defendants’ proposal for targeted discovery. It explained that, in the relatively wide-open world of equitable-contribution actions, “fairness would seem to dictate that those who were in the best position to know about possible contamination ... should bear more responsibility than parties further down the stream (so to speak)” and that “under some circumstances, joint tortfeasors having substantially greater culpability can be denied contribution altogether.” A ruling reached on this basis, the court hoped, would obviate the need for much of the discovery that NCR and Appvion were seeking. The parties undertook discovery in accordance with this plan, and both sides filed motions for summary judgment. Their efforts yielded a hefty record, with some 900 exhibits, covering expert and government reports; deposition testimony; laboratory notes; corporate records; and correspondence. Based on these materials, the court distilled the undisputed material facts. The court began by finding that NCR used Aroclor 1242, a solvent containing PCBs and manufactured by Monsanto, from 1954 until it found an adequate alternative in 1971. During this time, which it called the “production period,” roughly 30 million pounds of the PCB-laden emulsion were used. Though it had not yet called for discovery into the quantities that were discarded into the River, the court noted that the WDNR calculated that 98% of the PCBs in the Lower Fox River entered it during the production period. WDNR estimated that 39% of the PCBs entered the river as a result of NCR’s own manufacturing process, and 56% entered as a result of the recycling operations of the defendants during the 1954 to 1971 period. The remainder, it said, was dumped by recycling mills continuing to recycle carbonless copy paper after the production period ended in 1971. These estimates, however, were outside the scope of the discovery plan, and NCR did not concede their accuracy. , We therefore disregard this data for present purposes. The court next considered evidence about knowledge: what did each party know about the environmental dangers posed by PCBs, and when did it acquire this knowledge. For this purpose, the court relied heavily on correspondence by the British company Wiggins Teape, which was NCR’s exclusive licensee for carbon-less copy paper production in Europe. The Wiggins Teape documents showed that the company was aware as early as 1964 that Aroclor was toxic, and that it knew that recycled NCR paper could not be used for food packaging unless it was sufficiently cleansed. There were also records of scientific tests beginning in 1965 demonstrating the toxicity of Aroclor 1242. NCR’s own studies showed that PCBs had a “defatting effect” when they came into contact with skin, and indicated that the company was searching for a replacement. (In other words, PCBs are an irritant that chemically dissolves the fats in the skin and leaves it cracked and vulnerable to infection.) An independent study published by the Swedish scientist Soren Jensen in 1966 called PCBs “as poisonous as DDT,” and said they were harmful to the liver and the skin. NCR scientists were aware of this report by February 1967 at the latest. Monsanto apparently began to worry about the environmental effects of Aroclor in the late 1960s: on March 12, 1969, it circulated an internal memo with advice about how to deal with customers’ concerns about Aroclor’s safety. It essentially advocated silence, except with NCR— which by itself comprised 40% of the market for Aroclor 1242 in 1968. Monsanto decided to follow NCR’s approach toward the increasing worries about PCBs in crafting its own response. When Monsanto employees visited NCR headquarters in late March 1969, NCR officials told them that it wanted to stay informed about new developments, but it was not going to make any changes unless another article appeared specifically naming NCR as a source of the pollution. Samples taken by Monsanto in late 1969 and early 1970 showed that the effluent from NCR’s emulsion plants had “quite high” levels of Aroclor 1242. Wiggins Teape documents from 1970 show that it was becoming acutely aware of the problem and that British regulators were zeroing in on Aroclor as an environmentally toxic substance. Nevertheless, documents from that year reveal that NCR pushed Wiggins Teape not to disclose that NCR was the source of the Aroclor-coated paper. This request made Wiggins Teape executives uncomfortable. An internal memo from the latter written on February 13, 1970, indicated that although no study had definitely established a causal link between PCBs and health problems at that point, Wiggins Teape had known about potential toxicity since 1955. The memo also said that the Jensen report in 1966 awoke a “sleeping tiger.” It worried that a comparison could be drawn between the PCB threat and the evolving public knowledge of the health dangers of cigarettes. Monsanto stopped selling Aroclor 1242 in 1970, explaining to its customers that PCBs’ “use in synthetic resin compositions may be a source of ... alleged environmental contamination.” NCR decided to draw down its remaining stock of Aroclor; it ceased using Aroclor to make carbonless copy paper in April 1971. The following year, NCR circulated an internal memorandum stating, “In the late 1960’s accumulative evidence began to show that PCBs may have adverse effects on certain forms of animal life.... The resistance to breakdown — such as thermal and biodegradation — -was shown to lead to accumulations in the environment.” The authenticity of these documents was not contested, and in light of them the district court concluded that it could draw certain inferences that could not be disputed. First, it concluded that NCR knew that PCBs posed a risk of environmental harm by the late 1960s. This is not to say that NCR knew definitively that PCBs were harmful, but it was aware that there was a danger that they were toxic to human and animal life when released in waterways, and that they did not break down in the environment. At the latest, said the court, NCR knew this much when Monsanto sent its letter about environmental concerns in early 1970. The court next concluded that none of the defendants knew about the risks of PCBs until after NCR had ceased to produce carbonless copy paper in 1971. It noted that NCR had produced no evidence that any defendant knew that NCR’s paper contained PCBs at all, or that the paper could lead to environmental damage, in the period before April 1971. In drawing this inference, the court discounted three pieces of evidence that NCR now highlights: 1) an industry-wide report stating that the presence of PCBs in paper products and mill effluents had been recognized since the late 1960s; 2) testimony from a Menasha purchasing agent that it was instructed not to buy carbonless copy paper broke as early as 1950 because it might contain PCBs; and 3) testimony by a Fort Howard employee that Monsanto’s decision to stop selling Aroclor 1242 in 1970 was “well known.” The court discounted the industry report because it was prepared in 1976 and did not show that the mills were specifically alerted to PCBs in NCR’s products; it found the Menasha agent’s testimony unhelpful because he later repudiated it and the plaintiffs themselves rejected it in a different document before the court; and it found the Fort Howard employee’s testimony to be beside the point, because the deponent dated the knowledge to 1974, not 1970. In the district court’s view, the record left no doubt that NCR and Appvion knew long before others that PCBs posed a long-term risk to the environment. The importance of this early knowledge, it thought, drowned out all other equitable factors. It added that NCR actually increased its production of PCB-laden carbonless copy paper even as its knowledge of the risks increased in the late 1960s, and NCR’s response to the evidence was sluggish at best. It also reasoned that NCR was in a far better position to learn about the risks, even if it was not fully aware of them. Finally, it found that the equities strongly favored the defendants even in the period from 1954 until the mid-1960s, when all parties were ignorant of the risks posed by PCBs, because “between parties who produced the product and those who merely processed it and recycled it along with all other paper products or water sources, these latter parties are significantly less blameworthy.” The court recognized that this was not a claim for indemnification by the defendants, but it felt that such actions provided a useful analogy. The principles behind indemnification — particularly when one party provides another with a defective product — argued for denying contribution to NCR. The district court decided that no equitable adjustment was needed for the period when the defendant recycling plants continued to use broke from NCR’s carbonless copy paper after the product was discontinued in April 1971, and indeed past the point where the court concluded that all parties had knowledge of the harmfulness of PCBs. There was little that would have alerted the recyclers to the danger of this use. In 1976, Wisconsin approved the continued use of wastepaper containing PCB in recycling operations. See Wis. Stat. § 144.50(3)(c) (1976). The EPA promulgated a regulation in 1979 that allowed the continued use of PCB-containing carbonless copy paper, 44 Fed.Reg. 31,514 (May 31, 1979), and in 1984 it granted an express exemption for the use of “recycled PCBs,” 49 Fed.Reg. 28,172, 28,175 (July 10, 1984). A further amendment to allow the “safe level” of recycled PCBs to be measured in a more flexible way (to benefit the paper recyclers) followed in 1988, 53 Fed.Reg. 24,206 (June 27, 1988). For the interim periods when the defendants continued to recycle PCB-contaminated paper before these express authorizations, the district court used a WDNR study to conclude that the amount of PCBs dumped into the River during this period was less than two percent of the total (although it did not conduct discovery on this point). It would be a “Herculean task,” the court feared, to try to isolate and apportion responsibility for such a small portion of the overall PCBs in the River. At any rate, said the court, there was no evidence that these additional releases contributed significantly to the cleanup costs, and they did not put a dent in NCR’s overall greater equitable responsibility for the cleanup. The court also pointed to policy reasons in support of its conclusions. It worried that encouraging “manufacturers of a toxin” to continue using it until an economically feasible replacement can be found, and then allowing the manufacturer to recover contribution from “innocent processors” of the toxin, created a moral hazard. Finally, it thought that an approach that encourages producers of toxic materials to act “swiftly and proactively” as soon as the data begin to suggest the potential for environmental damage was more compatible with CERCLA. B NCR attacks the district court’s grant of summary judgment on two fronts. Its main thrust, to which we will return in a moment, is that it was an abuse of discretion to determine at this stage that the equities so clearly favored the defendants as to make NCR responsible for 100% of the costs of response — both its own costs and the costs initially borne by the defendant recycling mills. It also makes the procedural argument that the district court impermissibly resolved disputed questions of fact on the way to reaching summary judgment. We reject the latter contention. The district court faithfully applied the appropriate standard for summary judgment, and the conclusions it reached about the parties’ relative knowledge and awareness of the risks of PCBs were consistent with the undisputed evidence in front of it. NCR’s contention that the district court resolved disputed questions about when the company knew about the dangerousness of PCBs is based on a misunderstanding or a mischaracterization of the conclusions that the district court drew. If the district court had determined, without qualification, that NCR knew how dangerous PCBs were in the mid-1960s, NCR might have a point. NCR presented evidence showing that it did not believe that Aroclor 1242 was toxic because the initial evidence of toxicity related to Aroclor products with higher amounts of chlorine. (Aroclors were numbered based on their chlorine content; the “42” in Aroclor 1242 indicates that the product was 42% chlorine.) An expert retained by NCR produced a report stating that earlier studies focused on higher-chlorinated Aroclors, and that the toxicity of PCBs was “poorly known” even through the early 1970s. NCR also presented some evidence showing that the company internally felt that PCBs were not an environmental risk. For example, notes taken from a meeting between NCR and Monsanto employees showed that the parties discussed an incident of mass bird deaths in the Irish Sea in November 1969; those notes indicate that the investigation “cleared PCBs.” On the same document, the employee taking the notes wrote, “We think it degrades.” The record also contains a letter sent from Monsanto to its customers in 1970 saying that Aroclors with a chlorine content below 54% “appear to present no potential problem to the environment.” This evidence of NCR’s subjective beliefs would call into dispute any conclusion that NCR knew for a certainty about the impact its PCB use would have on the environment in the mid-1960s. But the district court did not say that NCR had such perfect knowledge. Instead, it concluded that NCR was beginning to see the warning signs about PCBs in the mid-1960s. This is indisputably true; Wiggins Teape documents showed awareness of PCBs’ toxicity in 1964, and NCR had the Jensen Report by early 1967 at the latest. The court’s point was that by the mid-1960s NCR was aware that Aroelor posed a risk of being harmful to the environment. Evidence had begun to accumulate that PCBs might cause severe damage. The recycling mills, in contrast, had not seen that evidence, and even if they had seen it, they had no apparent way of knowing that NCR’s broke contained PCBs. Whether NCR was given different information about higher-chlorinated Aroclors, or whether some of its fears about PCBs were allayed during a meeting with Monsanto employees, does not change the fact that it was on alert by then. Given the district court’s caution in drawing its factual conclusion about NCR’s knowledge in the period between 1964 and 1971, we find nothing in the record to bring its findings into dispute. NCR’s argument could be taken to suggest that the district court should not have settled for such a modest conclusion; once it chose to decide the case based on knowledge, NCR might say, it should have proceeded (perhaps after more discovery) to resolve in full what each party knew and when. This is just another way of putting the question whether it was an abuse of discretion to decide contribution shares based on the facts as they stood — the issue to which we now turn. C The ultimate question on this part of the appeal is whether the district court abused its discretion by allocating 100% of the Lower Fox River response costs to NCR at this juncture. In the past, we have stressed that the district court’s discretion is broad, both when it determines how much weight to place on any given equitable factor before the court, and also when it chooses which factors are pertinent at all for the case before it. Firms that have failed to secure contribution have come to us looking for greater certainty. Some have argued that fidelity to volumetric shares is required, see Browning-Ferris, 195 F.3d at 958-59, and others have urged that fault must play the leading role, ENSCO, 969 F.2d at 507 n. 4. We have rejected these pleas for a one-size-fits-all bright-line rule. Courts may permissibly make the equitable allocation based on “several factors, a few factors, or only one determining factor ... depending on the totality of circumstances presented to the court,” as long as the factors chosen are rational. ENSCO, 969 F.2d at 509. See also Litgo N.J. Inc. v. Comm’r N.J. Dep’t of Envtl. Prot., 725 F.3d 369, 387 (3d Cir.2013) (courts have “tremendous discretion”). Yet discretion has limits. Even if some factors are not outcome-determinative on their own, the district court’s decision must reflect its consideration of the particulars of the ease, lest the outcome become too divorced from the purposes underlying CERCLA. For example, in Kerr-McGee Chemical Corp. v. Lefton Iron & Metal Co., we found an abuse of discretion when the district court allocated cleanup costs based on the erroneous belief that an indemnification agreement between the parties did not apply to the costs at issue. 14 F.3d 321, 326 (7th Cir.1994). While acknowledging that the indemnification agreement was “not necessarily determinative” of the contribution question, we held that it should have been considered when allocating costs and that one party’s responsibility for producing most of the waste at the cleanup site did not justify “ignor[ing] other relevant considerations.” Id. The Eighth Circuit confronted a similar problem in K.C. 1986 Ltd. P’ship v. Reade Mfg., 472 F.3d 1009 (8th Cir.2007). The district court there had refused to consider pretrial settlement credits obtained by one of the parties in its contribution decision, despite CERCLA’s explicit policy against double recovery. See 42 U.S.C. § 9614(b). Reasoning that the district court “failed to consider a relevant factor that should have been given significant weight,” the appellate court found an abuse of discretion. Id. at 1017. It explained that “[i]n determining which equitable factors are appropriate, the policies articulated in CERCLA cannot be ignored.” Id. In one significant way, this case is unlike the earlier ones, or nearly any other CERCLA contribution action. Usually, a litigant dissatisfied with its contribution share will assert that the district court assigned impermissible weight to certain factors based on evidence in the record, or discounted facts that it should not have. In those situations, it is relatively easy to evaluate whether the district court neglected to consider an important factor or placed unacceptable weight on some factors while failing to accord due respect to others. Here, however, we are asked to decide whether the district court is authorized to preselect the equitable factor that it believes most likely to determine the outcome, and then conduct limited discovery into that factor to see if it can reach an equitable determination. We do not dismiss phased discovery out-of-hand as an unacceptable way to manage CERCLA contribution actions. Questions of cause and culpability surrounding an environmental cleanup can be tremendously complex, and discovery into every matter that anyone deems relevant — no matter how unlikely it is to sway a court’s equitable allocation of costs — could be prohibitively expensive and wasteful of the time of the court and parties alike. If one factor, such as knowledge, will so clearly overwhelm all others such that inquiry into other matters is unnecessary, there is no reason to force parties to bear additional costs just for the sake of appearances. Nonetheless, the court’s ultimate decision must reflect CERCLA’s equitable principles. We have always drawn a distinction between determining an action based on a single factor, and considering only certain factors on the way to the decision. The former practice is what EN-SCO approved. See 969 F.2d at 509. The latter showed up in Kerr-McGee, in which we reversed because the district court failed to show its awareness of certain relevant factors before arriving at its equitable conclusion. Similarly, in Beazer East, Inc. v. Mead Corp., the Third Circuit found an abuse of discretion when the district court “prioritize[d] a priori the parties’ relative contributions of waste over their contractual intent to allocate environmental liability among themselves.” 412 F.3d 429, 447 (3d Cir.2005). These cases demonstrate that the district court must decide what is relevant based on the record as a whole; an allocation based on otherwise permissible factors will not be rescued if it does not explain why the court has chosen to disregard other apparently relevant information. Kerr-McGee, 14 F.3d at 326; cf. R.W. Meyer, 932 F.2d at 573 (“Congress intended the court to deal with these situations by creative means, considering all the equities and balancing them in the interests of justice.”) (emphasis added). NCR argues that the district court here made the same error that we criticized in Kerr-McGee: that it preselected knowledge as the determinative factor and permitted discovery into only that aspect of the case, thus depriving NCR and Appvion of the opportunity to develop evidence of other potentially relevant factors. NCR is not suggesting (or at least it should not be suggesting) that it necessarily is entitled to a different result based on the additional evidence it would have presented. Instead, it contends that the allocation decision could not have been made based on the totality of the circumstances because the district court did not have the totality of the circumstances before it. At least three equitable factors that the court put to one side would have been relevant to the question of cost allocation, it urges: the parties’ relative volumes of PCB discharges; sources of PCBs in the river other than carbonless copy paper; and the parties’ levels of voluntary cooperation with the government’s cleanup effort. We agree with NCR that these points are potentially relevant to an understanding of who should contribute to the costs of the Fox River cleanup. For us to evaluate the district court’s allocation decision, therefore, we would need to see either that it evaluated and rejected these considerations as bases for allocation, or that it explained why one or more of them were so immaterial when compared with the ones on which it chose to rely as to make it fruitless for a party to develop evidence on those points. Unfortunately, the court did not record its thinking in this respect. We are thus unable confidently to say that the court chose knowledge as the deciding factor because undisputed facts showed that the relative volumes of discharge, alternate sources, and voluntary cooperation paled in comparison to knowledge. When explaining why it chose to proceed solely on the basis of the parties’ knowledge of the dangers posed by PCBs in the first instance, the court said that its “premise was that parties who knew, or should have known, about the dangers of PCBs should bear the brunt, or even the entirety, of any cleanup costs resulting from PCB contamination.” Its analysis of the parties’ relative knowledge was thorough and thoughtful, and reasonable as far as it went. The problem is that the court’s reasons for rejecting consideration of other factors leave us unable to say whether the record adequately supported the court’s decision to select knowledge as the decisive factor. All the court said was that apportionment on factors apart from knowledge “would require additional phases of the trial” in which the parties could prove other disputed facts like volume. It then reiterated its conclusion about knowledge, both that NCR was “not completely ignorant” of the dangers of PCBs during at least some of the period when they were producing carbonless copy paper, and that it was in the “best position” to learn of the dangers before it was fully aware. It added that in the period before any party could have been aware of the dangers, the equities favored the recycling mills because NCR actually produced the product, whereas the recycling mills “merely processed it”; it analogized the recycling mills to “innocent end-users.” It added that NCR bore more responsibility because the “defect” in its product — the PCBs — was the key to its profitability and not “incidental,” and that NCR was “sluggish” in its response to the mounting data about PCBs. As compelling a picture as this is on the knowledge issue, it does not explain why other factors, such as cooperation in the cleanup or relative volumes dumped, are so wholly irrelevant to the equitable determination that information about them need not even be gathered or presented by the parties. Without a more complete record, we are unable to endorse the court’s decision to resolve the case on the basis it chose. As a practical matter, the district court was unable to avoid mentioning basic facts such as the relative volumes of PCBs discharged by the parties, even though it qualified those references with the acknowledgment that the figures had not been subject to discovery. Other factors that NCR proposes, such as cooperation, might have influenced the court’s conclusions about NCR’s slow responses to data. We hasten to say, however, that further information may just as easily point in the other direction and have the effect of reinforcing the district court’s present conclusions. It is impossible to know at this juncture what the likely effect will be. We