The issue before us presents a question of statutory interpretation: Under the Comprehensive Environmental Response, Compensation and Liability Act (CERC-LA), when does the limitations period for suing to collect remedial action costs from a party responsible for hazardous substances begin? One would expect a fairly straightforward answer to that question in the statute. Potential 'plaintiffs need to know when to file suit, and potential defendants would surely prefer clear notice as to when their legal liability, if any, lapses. True, in the “maze”-like structure and “baffling language” of CERCLA, clarity is rare. Carson Harbor Vill. Ltd. v. Unocal Corp., 270 F.3d 863 , 880, 883 (9th Cir.2001) (en banc). The provision we grapple with today appears at first blush to be no exception. But as one works one’s way through the statute as a whole, a fairly definite answer emerges. As will appear, we conclude that the limitations period for bringing an initial suit for recovery of remedial action costs under CERCLA cannot accrue until after the final adoption of the remedial action plan required by the statute.

FACTUAL BACKGROUND

For 35 years, Neville Chemical Company (Neville) manufactured at its industrial facility in Santa' Fe Springs, California, various chemical compounds for use in insecticides, solvents, metal working lubricants, and flame retardants. These activities contaminated the groundwater and soil at the facility. In 1986, the California Department of Toxic Substances Control (the Department) issued a Remedial Action Order, directing Neville to (1) begin the process of cleaning the site; (2) conduct a remedial investigation and feasibility study; (3) submit a draft remedial action plan (RAP); and, once the draft RAP was finalized, (4) implement the plan.

The Department sent Neville a letter on September 29, 1989, informing Neville of its obligation to pay an “activity fee” to the Department. The letter explained that the activity fee — $46,636.38—was “to partially cover the Department’s cost of overseeing [Neville’s] actions to characterize and satisfactorily remediate this site.” At that time, the Department had a formal policy of “only collect[ing] direct program expenditures (generally laboratory or contract expenditures) beyond activity fees in cases where the responsible parties are being cooperative.” In 1992, the Department rescinded this policy in favor of pursuing the full cost recovery of overseeing a clean-up, regardless of whether the responsible party was recalcitrant or cooperative.

In August 1991, Neville presented the Department with preliminary findings from the Remedial Investigation. In October 1991, the Department directed Neville to prepare a Groundwater Removal Action Proposal (the Proposal), in which Neville was to propose an expedited response to the contamination. The Department stated that the Proposal “should be consistent with a final cleanup strategy for groundwater as it may ultimately become the final remedy presented in the Remedial Action Plan.” Neville submitted its Proposal on September 1, 1992. It included “three major components: an extraction system, a temporary on-site treatment system, and an effluent disposal system.”

The Department reviewed the Proposal and, in January 1993, directed Neville to implement the extraction and treatment system. In a letter to Neville, the Department stated: “The proposed system will potentially become part of the final remedial alternative for the site,” and “[t]he ground water extraction and treatment system is envisioned as part of the final remedial alternative.... [H]owever, the Department may order the discontinuation of its use in the event it is not effective or if it enhances the migration of contaminants from the Site.”

Neville submitted a Feasibility Study Technical Memorandum in August 1993, listing alternative possible remedies. In response to this memorandum, the Department stated that it:

Additionally, the letter stated:

Neville began to excavate three extraction wells at the site in April 1994. A month later, Neville submitted a Draft Feasibility Study, again proposing several alternative remedies. The Department responded with comments to this draft in June of 1994, including the following: “The Department has never stated that the Ground Water Removal Activity ... is the final ground water remedy, but has to be tested to determine the efficiency of the system.” In October of the same year, the Department sent Neville a letter expressing concern because Neville had not started construction of the Groundwater Removal System. The Department also noted, “Neville will need to compare several sample results to determine the effectiveness of the System. Neville will use the information to determine whether this or a modified System will be incorporated into the draft Remedial Action Plan.”

Neville submitted a final Feasibility Study, discussing seven alternative groundwater remedial options, in December of 1994. Later that month the Department approved it.

Neville then submitted a draft remedial action plan. On May 8, 1995, after having circulated the draft for public review and comment and holding a public meeting to discuss the plan, the Department approved the final remedial action plan. The groundwater containment and treatment system originally designed as an interim removal action remained part of the final RAP.

ANALYSIS

I. Accrual of Cause of Action

Neville first argues that the district court erred in denying Neville’s summary judgment motion because the statute of limitations for bringing a cost recovery action under CERCLA barred California’s suit. A party may appeal a denial of summary judgment once a final judgment has been entered in the suit. Comsource Indep. Foodservice Cos. v. Union Pac. R.R. Co., 102 F.3d 438, 442 (9th Cir.1996). We review a denial of summary judgment de novo. Id. (citing Warren v. City of Carlsbad, 58 F.3d 439, 441 (9th Cir.1995), and Pomerantz v. County of Los Angeles, 674 F.2d 1288, 1290 (9th Cir.1982)).

California brought suit on behalf of the Department under § 107 of CERCLA, 42 U.S.C. § 9607 . This statute provides that the owner and operator of a facility “shall be held liable for — (A) all costs of removal or remedial action incurred by ... a State ... not inconsistent with the national contingency plan....” An “initial action” for recovery of costs “must be commenced ... for a remedial action, within 6 years after initiation of physical on-site construction of the remedial action.” 42 U.S.C. § 9613 (g)(2). The present suit was brought on September 21, 2000. Therefore, the suit is time-barred if and only if the “initiation of physical on-site construc tion of the remedial action” occurred on or before September 21,1994.

The facts of the case are not in dispute. The only question is which of the enumerated events constitutes the “initiation of physical on-site construction of the remedial action,” thereby triggering the limitations period. Neville maintains that the statute of limitations began to run in April 1994, when it started excavating the extraction wells. California argues that no remedial action could have occurred until the final remedial action plan was approved by the Department on May 8, 1995. This Court has yet to decide when an action is remedial for the purpose of triggering the statute of limitations in cost recovery suits under 42 U.S.C. § 9613 (g)(2).

A. Statutory Interpretation

The purpose of a limitations period is to “clearly define the time period in which suit must be commenced.” United States v. Colvin, 204 F.3d 1221, 1226 (9th Cir.2000). Here, the statute of limitations is invoked to bar the government from collecting the costs it expended in cleaning up a hazardous waste site, a situation in which we have been specially instructed by the Supreme Court to construe limitations periods in favor of the government. See Badaracco v. Comm’r, 464 U.S. 386, 391-92, 104 S.Ct. 756 , 78 L.Ed.2d 549 (1984) (“Statutes of limitation sought to be applied to bar rights of the Government, must receive a strict construction in favor of the Government.”) (citing E.I. Dupont De Nemours & Co. v. Davis, 264 U.S. 456, 462 , 44 S.Ct. 364 , 68 L.Ed. 788 (1924)). Additionally, if at all possible, the statute should be interpreted to provide a clear accrual date, so that each party — but especially the State as plaintiff — knows when the time to bring suit runs out. The text of the statute, read as a whole rather than in pieces, specifies that ascertainable date.

Title 42 U.S.C. § 9613 (g)(2) provides that the “initiation of physical on-site construction of the remedial action” triggers the statute of limitations. CERCLA defines “remedial action” in section 9601(24):

42 U.S.C. § 9601 (24) (emphasis added). “Removal,” in turn, is defined thus:

42 U.S.C. § 9601 (23). The plain meaning of the definition of “remedial,” read together with the statute of limitations in § 9613(g)(2) and the use of that same term in the rest of the statute, supports the conclusion that “the initiation of physical on-site construction of the remedial action” can only occur after the final remedial action plan is adopted.

The first half of the definition of “remedial action” provides a general description of how such an action fits into the entire scheme of the clean-up required by the statute. Remedial actions, the statute provides, must be “consistent with permanent remedy taken instead of or in addition to removal actions.” The second half of the definition lists some activities that could constitute remedial actions. These examples, however, must be read in light of the more general description of the first half.

For example, the “provision of alternative water supplies” is listed as both a type of “remedial action” and as a type of “removal.” The provision of alternative water supplies will only be “remedial,” therefore, if it is done “consistently] with permanent remedy....” On the other hand, if alternative water supplies are provided on a more temporary basis, the very same activity would be a “removal” action. That is, “removal actions generally are immediate or interim responses, and remedial actions generally are permanent responses.” Geraghty & Miller, Inc. v. Conoco Inc., 234 F.3d 917, 926 (5th Cir.2000). In this case, therefore, even if the completed extraction wells do fall under one of the types of activities listed in the second half of the definition of “remedial action” (e.g., “onsite treatment”), the excavation of those wells must still be “consistent with permanent remedy” to have triggered the statute of limitations.

For an action to be “consistent with permanent remedy,” a permanent remedy must already have been adopted. Neither party can know for sure whether a given action is consistent with permanent remedy until that permanent remedy is determined. The first point at which both parties can be certain that any construction is consistent with a permanent remedy is when the permanent remedy is actually selected. In this case, as in most cases, the permanent remedy was selected when the final RAP was approved.

Until after the adoption of the RAP, then, California could not have brought suit to recover remedial costs. “The standard rule [is] that the limitations period commences when the plaintiff has a complete and present cause of action.” Bay Area Laundry & Dry Cleaning Pension Trust Fund v. Ferbar Corp. of Cal., Inc., 522 U.S. 192, 201 , 118 S.Ct. 542 , 139 L.Ed.2d 553 (1997) (quoting Rawlings v. Ray, 312 U.S. 96, 98 , 61 S.Ct. 473 , 85 L.Ed. 605 (1941)) (internal quotation marks omitted). “[U]ntil the plaintiff can file suit and obtain relief,” a limitations period ordinarily does not commence. Bay Area Laundry, 522 U.S. at 201 , 118 S.Ct. 542 ; see also Reiter v. Cooper, 507 U.S. 258, 267 , 113 S.Ct. 1213 , 122 L.Ed.2d 604 (1993) (“While it is theoretically possible for a statute to create a cause of action that accrues at one time for the purpose of calculating when the statute of limitations begins to run, but at another time for the purpose of bringing suit, we will not infer such an odd result in the absence of any such indication in the statute.”).

The Department could not have brought suit for costs of remedial action at the time Neville began excavating the extraction wells. At that time, neither party could have known if the wells would eventually be “consistent with permanent remedy,” because no final remedial action plan had been decided upon at that point. As we must assume that the limitations period did not begin to run until at least the time California could have brought suit to recover remedial costs, that period could not have been triggered by the excavation of the wells. To rule otherwise would be to hold, as a practical matter, that California was required to bring suit before April 19, 2000, less than five years after its action for recovery of remedial costs first accrued, even though the statute specifies a six-year limitations period.

In the ease at hand, neither party could have known before the final RAP was approved whether any particular construction projects would be “consistent with the final remedy”: Before the final remedial action plan was approved, the letters from the Department emphasized and re-emphasized that it did not know whether any of the measures already taken by Neville would or would not be consistent with the final remedial plan. See Letter from the Department to Neville, January 1993 (stating that the Department may discontinue the use of the groundwater extraction and treatment system “in the event it is not effective or if it enhances the migration of contaminants from the Site”); Letter from the Department to Neville, August 1993 (stating that the Department “has not gathered sufficient information and public comment to require any of the alternatives to be implemented as of yet”); id. (stating that the Department “reviewed and approved of the[Groundwater] Removal Action as an interim measure” but that the Groundwater Removal Action “is not a Department-approved final Remedial Action, and cannot be construed as such.... Therefore, whether the [Groundwater] Removal Action constitutes the groundwater portion of the final Remedial Action cannot be determined at this point.”).

Section 9617 of CERCLA, which provides for public participation in selection by the President or a State of a remedial action plan, reinforces our interpretation of “remedial action” as action taken after the final remedial action plan has been approved. After .mandating a public notice and comment period and a public meeting regarding the proposed remedial action plan, 42 U.S.C. § 9617 (a)(1) & (2), the statute provides: “Notice of the final remedial action plan adopted shall be published and the plan shall be made available to the public before commencement of any remedial action.” 42 U.S.C. § 9617 (b) (emphasis added). So, under the statute, remedial action may not commence — and therefore, the “initiation of physical on-site construction of the remedial action” cannot begin — until after a final remedial action plan is adopted.

The statutory provision limiting the time in which a party may commence a suit for natural resource damages also supports this interpretation of “remedial action.” Title 42 U.S.C. § 9613 (g)(1) states, “In no event may an action for damages under this chapter with respect to [a. facility at which a remedial action is scheduled] be commenced ... before selection of the remedial action if the President is diligently proceeding with a remedial investigation and feasibility study....” The reason for this limitation was illuminated in the House Report issued by the Committee on the Judiciary: Congress sought to integrate cost recovery and damages actions. H.R. Rep. 99-253(III), reprinted in 1986 U.S.C.C.A.N. 3038, 3044. The premise behind the Report’s reasoning is clear: natural resource damages actions and cost recovery actions could not be integrated if damages actions were brought before a remedy was selected, because cost recovery suits could not be brought before that point.

Some courts have raised the concern that if one reads the statute,' as we do, to provide that “initiation of physical on-site construction of the remedial action” can only take place after the final remedial action plan is approved, much of the definition of “remedial action” would become superfluous. See, e.g., United States v. Navistar Int’l Transp. Corp., 152 F.3d 702, 712 (7th Cir.1998); California v. Hyampom Lumber Co., 903 F.Supp. 1389, 1392-93 (E.D.Cal.1995); Advanced Micro Devices, Inc. v. Nat’l Semiconductor Corp., 38 F.Supp.2d 802, 811 (N.D.Cal.1999) (citing Hyampom, 903 F.Supp. at 1393 ). This concern is unwarranted. First, the definition of “remedial action” has roles in the statute other than defining the onset of one limitations period. The functional aspects of the definition are critical, for example, in apportioning the percentage of the costs of any clean-up between States and the Fund. See 42 U.S.C. § 9604 ; see also 42 U.S.C. § 9621 (requiring that cer tain standards be met when implementing remedial actions).

Second, even though an action can only be remedial if it is taken after the final remedial action plan is approved, that does not mean that all actions taken after the final remedial action plan is approved are remedial. In Colorado v. Sunoco, Inc., 337 F.3d 1233 (10th Cir.2003), for example, the Tenth Circuit was faced with the task of classifying certain actions, all of which took place after the EPA chose its permanent remedy, as either “remedial” or “removal” actions. After discussing the character of the various actions in light of the definitions of “remedial” and “removal,” the court determined that two of the actions were, in fact, removal actions, both because they were interim rather than permanent measures, taken in response to the threat of release of contaminated water, and because similar activities had been denominated “removal action” by the EPA in previous clean-ups. Id. at 1244-45 . Thus, as Sunoco illustrates, our interpretation of the statute of limitations does not render any part of the definition of “remedial” superfluous.

B. Decisions of Other Circuits

Our conclusion that no action can be “remedial” until a final remedial action plan is in place is consistent with the results reached by every court of appeals that has considered the onset of the limitations period for recovery of remedial action costs under CERCLA, if not with all the reasoning of those cases.

In Geraghty, the Fifth Circuit held that the installation of monitoring wells could not have triggered the statute of limitations because it occurred before the government agency overseeing the clean-up had issued its final approval of the remedial plan. See Geraghty, 234 F.3d at 927 . The Seventh Circuit also reached the same result as we do, although some of its reasoning differed from ours. See Navistar, 152 F.3d at 711-12 . In Navistar, although the final remedial design was not approved until 1990, the final remedial action plan, selecting a permanent clay cap as part of the permanent remedy, was apparently approved before that. Id. at 704 (“At the conclusion of this process[, which ended before February 1989], the EPA determined that, among other things, the landfill needed to be covered with a permanent clay cap to isolate the hazardous materials from the rest of the environment.”). Thus, although the Seventh Circuit rejected a bright-line rule in which the final remedial design had to be formally approved before an action could be considered remedial, the action that it found to be remedial—installing the clay cap—occurred after the final remedial action plan was chosen. Because our holding finds the pivotal event for defining the initiation of remedial action is the adoption of a remedial action plan—not a final remedial design—the facts in Navistar would have led us to find the suit barred by the limitations period as well.

Finally, the Tenth Circuit, while not expressly rejecting a bright-line rule, has distinguished “remedial actions” from “removal actions” based solely on the more “descriptive” parts of their definitions. See Sunoco, Inc., 337 F.3d at 1244-45 ; see also Pub. Serv. Co. of Colo. v. Gates Rubber Co., 175 F.3d 1177, 1182 (10th Cir.1999) (distinguishing “remedial actions” from “removal actions” in a context other than the triggering of the statute of limitations). In both the Tenth Circuit cases, however, it appears that the actions in question took place after a remedial action plan was in place. See Sunoco, 337 F.3d at 1237, 1244-45 ; Pub. Serv. Co., 175 F.3d at 1179, 1182-84 . Were we faced with the same facts, we, too, would have to turn to the descriptive aspects of the definitions to determine whether the actions at issue in Sunoco Inc. and Public Service Co. were remedial or removal. For the same reasons the result in Navistar is not in conflict with our holding, then, these Tenth Circuit cases do not conflict, either.'

In sum, we conclude that the “initiation of physical on-site construction of the remedial action” can only occur after the final remedial action plan is adopted, and that, in this case, the statute of limitations, therefore, could not have begun to run until the final remedial action was approved on May 8,1995. The Department’s suit was brought within six years of the approval of the remedial action plan and is not, thus, barred by the statute of limitations.

II. Neville’s Defenses on the Merits

Neville raised an affirmative defense — waiver and estoppel — in the district court. The argument was that Neville cannot be liable under CERCLA for the costs of overseeing the clean-up incurred by the Department because the Department had promised that it would not sue Neville for full recovery costs if Neville conducted the research, planning, and clean-up of the. site. The district court ruled that Neville could not assert equitable defenses to a CERCLA recovery action. We review the grant of summary judgment de novo. United States v. Chapman, 146 F.3d 1166, 1169 (9th Cir.1998).

CERCLA section 107(a) and (b), 42 U.S.C. § 9607 (a) and (b), allow for only three defenses to CERCLA liability. A covered person is liable under the statute “subject only to the defenses set forth in subsection (b) of this section.” 42 U.S.C. § 9607 (a). Subsection (b) lists three defenses “(1) an act of God; (2) an act of war; [and] (3) an act or omission of a third party....” 42 U.S.C. § 9607 (b)(1)-(3). In Levin Metals Corp. v. Parr-Richmond Terminal Co., 799 F.2d 1312, 1316-17 (9th Cir.1986), we suggested that these defenses were exclusive.

Id. (emphasis added) (internal quotation marks and citations omitted).

Every court of appeals that has considered the precise question whether § 9607 permits equitable defenses has concluded that it does not, as the statutory defenses are exclusive. See Gen. Elec. Co. v. Litton Indus. Automation Sys., Inc., 920 F.2d 1415, 1418 (8th Cir.1990) (holding that CERCLA does not provide an “unclean hands” defense) (questioned on other grounds in Key Tronic Corp. v. United States, 511 U.S. 809 , 114 S.Ct. 1960 , 128 L.Ed.2d 797 (1994)); see also Blasland, Bouck & Lee, Inc. v. City of North Miami, 283 F.3d 1286, 1304 (11th Cir.2002) (holding that CERCLA bars equitable defenses); Velsicol Chem. Corp. v. Enenco, Inc., 9 F.3d 524 , 530 (6th Cir.1993) (same); Town of Munster v. Sherwin-Williams Co., 27 F.3d 1268, 1270 (7th Cir.1994) (same). Following the implication of Lev-in Metals and the express holdings of these cases from other circuits, we conclude that the three statutory defenses are the only ones available, and that traditional equitable defenses are not. The district court was correct, therefore, in holding that Neville could not raise equitable defenses to liability under CERCLA.

Neville argues separately that equitable defenses are at least relevant to the amount of recovery that the Department receives. Even if it is liable for some of the oversight costs, Neville asserts, that amount is limited by equitable consideration that the Department’s implied promise to pursue recovery of only a limited “activity fee,” rather than the Department’s actual recovery-oversight costs. Neville cites as support for this argument numerous cases in which courts considered equitable factors in allocating costs in suits for contribution. See, e.g., Alcan-Toyo Am., Inc., v. N. Ill. Gas Co., 881 F.Supp. 342, 346-47 (N.D.Ill.1995); Akzo Coatings, Inc. v. Aigner Corp., 909 F.Supp. 1154, 1161-62 (N.D.Ind.1995); New York v. Almy Bros., 971 F.Supp. 69, 73 (N.D.N.Y.1997).

Suits for contribution, however, are entirely distinct under the statute from suits for recovery of costs. The former is governed by 42 U.S.C. § 9613 (f)(1), which explicitly states, “In resolving contribution claims, the court may allocate response costs among liable parties using such equitable factors as the court determines are appropriate.” The provisions of CERCLA governing suits for recovery of costs, 42 U.S.C. §§ 9607 (a) and 9613(g)(2), make no such reference to equitable factors. Also, “the critical distinction between [suits for contributions and suits for cost recovery] is that under § 107 [ 42 U.S.C. § 9607 (a)], the court merely determines whether the party is jointly and severally liable, without regard to the amount of fault; but under § 113 [ 42 U.S.C. § 9618 (f)(1)], the court also divides the fault of the parties, using equitable factors.” Catellus Dev. Corp. v. L.D. McFarland Co., 910 F.Supp. 1509, 1514 (D.Or.1995). California is not bringing suit here for contribution, so the specific language allowing the court to consider equitable factors when apportioning contribution is inapplicable.

Neville makes one last defensive argument: The Department may not sue for its recovery costs, Neville contends, because those costs were not consistent with the national contingency plan. Whether a party can recover certain costs under § 9607 depends on whether or not those costs were incurred consistently with the “national contingency plan.” 42 U.S.C. § 9607 (a)(4)(A) (providing that a covered person who violates CERCLA “shall be liable for ... all costs of removal or remedial action incurred by the ... State ... not inconsistent tvith the national contingency plan”) (emphasis added). The national contingency plan is promulgated by the EPA and “provide[s] the organizational structure and procedures for preparing and responding to ... releases of hazardous substances.” 40 C.F.R. § 300.1 . See also Wash. St. Dep’t of Transp. v. Wash. Natural Gas Co., 59 F.3d 793 , 799 (9th Cir.1995) (“WSDOT”). To show that the Department’s actions were inconsistent with the national contingency plan, the burden is on Neville to show that the Department acted in an arbitrary and capricious manner in choosing a particular response action. See id. at 802 (citing United States v. Hardage, 982 F.2d 1436, 1442 (10th Cir.1992)). When a state is seeking recovery of response costs, consistency with the national contingency plan is presumed. Id. at 799-800.

Neville has provided no evidence that the Department acted “arbitrarily and capriciously in choosing a particular re sponse action to respond to a hazardous waste site.” Hardage, 982 F.2d at 1442 (emphasis added). Accord WSDOT, 59 F.3d at 802 (“To prove that a response action of the EPA was inconsistent with the NCP, a defendant must prove that the EPA’s response action was arbitrary and capricious.”) (emphasis added). In fact, Neville does not challenge any response action taken by the Department. Neville challenges instead the Department’s attempt to recover the full oversight costs after suggesting that, should Neville cooperate and conduct the clean-up itself, the Department would only require Neville to pay an “activity fee.” This change in policy and pursuit of the full costs of oversight cannot be “inconsistent with” the national contingency plan, as the national contingency plan does not direct the state to limit its recovery of response costs in any way. See 40 C.F.R. §§ 300.1 et seq. The district court, therefore, did not err by finding that Neville was responsible for all the Department’s response costs.

III. Motion for Leave to Amend

Finally, Neville appeals the district court’s denial of its motion for leave to amend its counterclaim. We review the district court’s denial of Neville’s motion for leave to amend for an abuse of discretion. See Griggs v. Pace Am. Group, Inc., 170 F.3d 877 , 879 (9th Cir.1999).

Generally, leave to amend pleadings “shall be freely given when justice so requires.” Fed.R.Civ.P. 15(a). “[T]he grant or denial of an opportunity to amend is within the discretion of the District Court,” and denial of leave to amend is appropriate if the amendment would be futile. Foman v. Davis, 371 U.S. 178, 182 , 83 S.Ct. 227 , 9 L.Ed.2d 222 (1962). “[F]utility includes the inevitability of a claim’s defeat on summary judgment.” Johnson v. Am. Airlines, Inc., 834 F.2d 721 , 724 (9th Cir.1987).

Neville moved to amend its complaint to allege a violation of due process and equal protection under the California Constitution. The district court denied Neville’s motion for leave to amend because the court determined that amendment would be futile. The district court held that, even if Neville were to amend its complaint and allege a violation of equal protection, Neville could not point to a triable issue of material fact to support such an allegation. On appeal, Neville argues only that the district court erred by (1) holding that the California Constitution requires a showing of “invidious discrimination” to prove selective prosecution, and (2) refusing to compel discovery on the issue of selective prosecution.

The district court did not abuse its discretion. First, the court was correct in its interpretation of California constitutional law in holding that Neville had to allege discrimination based on an “invidious” criterion. Baluyut v. Superior Court, 12 Cal.4th 826, 50 Cal.Rptr.2d 101 , 911 P.2d 1, 5 (1996), on which Neville exclusively relies, holds that a defendant must show that “he has been deliberately singled out for prosecution on the basis of some invidious criterion” in order to prove discriminatory prosecution. Id. (citation and internal quotation marks omitted). That case goes on to define “invidious” as “unrelated to legitimate law enforcement objectives.” Id. at 6. Additionally, Baluyut states that “[ujnequal treatment which results simply from laxity of enforcement or which reflects a nonarbitrary basis for selective enforcement of a statute does not deny equal protection and is not constitutionally prohibited discriminatory enforcement.” Id. at 5.

Neville fails to allege, first, that other similarly situated parties were excused from paying the full oversight costs and thus that it was “deliberately singled out for prosecution.” Even if Neville could show this, it would have to allege also that the reason for this discrepancy was not simply laxity of enforcement, but was in fact a result of invidious discrimination, i.e., unrelated to law enforcement purposes. As the record now stands, Neville has shown that the Department changed its policy about collecting oversight costs. However, this change was explained by the Department: the agency determined that the non-enforcement policy was inconsistent with state statutes. Thus, the Department has provided a non-arbitrary, law enforcement rationale for the change in policy. On the basis of this record, the district court did not abuse its discretion by denying leave to amend, or by denying discovery when no actionable injury was alleged.

AFFIRMED.

. The Department of Toxic Substances Control was a division of the California Department of Health Services until it became a separate department in 1991. We will refer to this entity simply as "the Department" throughout this opinion. . The statute divides actions for recovery of costs into initial and subsequent. The parties do not dispute that this is an "initial action." . In cases where private, non-governmental parties conduct the clean-up of a site without governmental or agency oversight and then pursue response costs under CERCLA, there will most likely still be a remedial action plan in place. See 40 CFR § 300.700 (providing that private parties should follow the public notice and comment procedures required of government actors). As no non-governmental response cost suit is before us, however, we do not address the limitation period applicable to such suits. . This is not to say that the Department was unable to bring any suit to recover any costs at the time Neville began excavating the extraction wells. Under 42 U.S.C. § 9613 (g)(2), "an action may be commenced under section 9607 of this title for recovery of costs at any time after such costs have been incurred." As soon as the Department expended its first dollar, it could have sued Neville for this dollar and sought a declaratory judgment of Neville's liability for future response costs. However, the availability of the option to bring suit earlier, and thereby obtain a declaratory judgment as to liability, should not confound Congress's clear intention that an initial action to recover remedial costs may be brought "within 6 years after initiation of physical on-site construction of the remedial action." A suit to recover the costs of particular remedial actions, as opposed to a suit to recover removal costs and obtain a declaratory judgment on liability, can only be brought once those remedial actions have been completed. . The legislative history of the Superfund Amendments and Reauthorization Act of 1986 (SARA), which added the statute of limitations provision to CERCLA, is consistent with our interpretation of § 9613(g)(2). When we can interpret a statute by its plain meaning, we only look to the congressional history to "ensure that there is no clearly contrary legislative intent.” Carson Harbor Village, Ltd., 270 F.3d at 884. There is none here. The House Report from the Judiciary Committee states: "The statute of limitations provided by this amendment for the initial cost recovery action for a remedial action is three years from the commencement of physical on-site construction of the remedial action, that is, after the [Remedial Investigation/Feasibility Study] and after design of the remedy." H.R. Rep. 99-253(111), reprinted in 1986 U.S.C.C.A.N. 3038, 3044 (emphasis added) (referring to one version of the predecessor bill, H.R. 2817, which provided a three-year statute of limitation but was otherwise identical to the final Act with regard to the pertinent limitations provision). The design of the remedy can only occur after the final remedy has been selected in the final remedial action plan. . Because of the temporal aspect of our interpretation of "remedial action," we do note that the government can only recover costs as "remedial” if those costs were incurred after the cause of action for such costs accrues. Any costs incurred before the remedial action plan was finally designated (such as the construction of the extraction wells in this case) may be recovered, however, as “removal" costs, and are subject to the statute of limitations for removal actions. See 42 U.S.C. § 9613 (g)(2)(A) ("An initial action for recovery of the costs referred to in section 9607 of this title must be commenced ...[,] for a removal action, within 3 years after completion of the removal action...."); 42 U.S.C. § 9613 (g)(2)(B) ("[I]f the remedial action is initiated within 3 years after the completion of the removal action, costs incurred in the removal action may be recovered in the cost recovery action brought under this subpara-graph.”). . A remedial design is a term of art in CERC-LA, and differs both substantively and temporally from a final remedial action plan. While there is no explicit definition of "remedial action plan” in either the statute or the regulations implementing it, the regulations do provide a detailed description of the process whereby a final remedy is selected by the agency. In this process, the lead agency must present a proposed plan, which fits the description of the "remedial action plan” of CERCLA’s Section 9617, to the public. 40 CFR § 300.430 (f)(2). The “proposed plan” must "briefly describe! ] the remedial alternatives analyzed by the lead agency, propose! ] a preferred remedial action alternative, and summarize!] the information relied upon to select the preferred alternative.” Id. The regulation continues, "The purposes of the proposed plan is to supplement the RI/FS and provide the public with a reasonable opportunity to comment on the preferred alternative for remedial action, as well as alternative plans under consideration, and to participate in the selection of remedial action at a site.” Id. The remedy chosen in the remedial action plan is only generally described in that docu ment, leaving for a subsequent date the actual design of the plan’s physical implementation. See 40 CFR § 300.430 (f)(1)-(6) (describing the process by which the lead agency chooses a final remedy and documents its selection in a record of decision); see also 42 U.S.C. § 9617 (providing publication requirements in the event that the remedial action differs from the adopted final remedial action plan). The final remedial design, on the other hand, while based on the remedy adopted in the RAP, is distinct: it is "the technical analysis and procedures which follow the selection of remedy for a site and result in a detailed set of plans and specifications for implementation of the remedial action.” 40 CFR § 300.5 .