WALLACE, Circuit Judge

:In March 2012, prisoners in the custody of the Arizona Department of Corrections (ADC), together with the Arizona Center for Disability Law, brought a civil rights class action against senior ADC officials alleging systemic Eighth Amendment violations in Arizona’s prison system. The inmates alleged that ADC’s policies and practices governing health care delivery in ADC prisons and conditions of confinement in ADC isolation units expose them to a substantial risk of serious harm to which Defendants are deliberately indifferent. On the eve of trial, the parties signed a settlement agreement (Stipulation) by which Defendants agreed to comply with more than 100 "performance measures" designed to improve the ADC health care system and reduce the harmful effects of prisoner isolation. Since the action settled, the parties have engaged in several disputes over Defendants’ alleged non-compliance with the performance measures, which has required the assigned magistrate judge to issue various rulings interpreting and enforcing the Stipulation. These rulings are the subject of the consolidated appeals now before us.

I.

The Stipulation went into effect on February 25, 2015, the date on which Magistrate Judge David Duncan granted final approval. Consistent with the district court’s earlier class certification order, Parsons v. Ryan , 289 F.R.D. 513 (D. Ariz. 2013), aff’d , 754 F.3d 657 (9th Cir. 2014), the Stipulation defines one class and one subclass. The class is defined as "[a]ll prisoners who are now, or will in the future be, subjected to the medical, mental health, and dental care policies and practices of the ADC." Stipulation ¶ 3. This covers approximately 33,000 inmates in 10 state-operated prisons. The subclass is defined as "[a]ll prisoners who are now, or will in the future be, subjected by the ADC to isolation, defined as confinement in a cell for 22 hours or more each day or confinement in [five enumerated] housing units." Id. This isolation subclass covers the approximately 3,000 inmates in ADC custody classified as "maximum custody."

The Stipulation requires Defendants to comply with 103 health care performance measures at each of the 10 state-operated prisons. The performance measures obligate Defendants to adopt certain standards and practices across a wide spectrum of health care categories, including diagnostic services, preventative services, mental health, and access to care. For example, Performance Measure 13 provides that "[c]hronic care and psychotropic medication renewals will be completed in a manner such that there is no interruption or lapse in medication." Performance Measure 33 mandates that "[a]ll inmates will receive a health screening by an LPN [licensed practical nurse] or RN [registered nurse] within one day of arrival at the intake facility." Defendants are required to measure and report their compliance with the health care performance measures on a monthly basis.The Stipulation also requires Defendants to comply with nine performance measures specific to "maximum custody" inmates. For example, pursuant to Maximum Custody Performance Measure 1, all maximum custody inmates housed at the ADC’s maximum custody facilities must be offered a minimum number of hours of out-of-cell time per week. As with the health care performance measures, Defendants must measure and report their compliance with the maximum custody performance measures on a monthly basis.

The performance measures require Defendants to meet or exceed a certain threshold rate of compliance based upon how long the Stipulation has been in effect. For example, for the first 12 months after the Stipulation went into effect, Defendants were required to meet or exceed a 75 percent rate of compliance. Stipulation ¶¶ 10, 20. For the second 12 months, the required threshold increased to 80 percent. Id. Defendants’ duty to measure and report on a particular performance measure terminates if (1) the performance measure meets the required compliance threshold for 18 months out of a 24-month period and (2) the performance measure has not been out of compliance for three or more consecutive months within the previous 18-month period.

The Stipulation also provides the process by which the parties resolve disputes over compliance. In the event Plaintiffs believe Defendants are in non-compliance with one or more of the performance measures, Plaintiffs must first provide Defendants a written statement describing the alleged non-compliance, to which Defendants must provide a written response. Stipulation ¶ 30. The parties must then meet and confer in an attempt to resolve the dispute informally and, if informal efforts fail, participate in formal mediation. Id. ¶¶ 30, 31. If the dispute is not resolved through formal mediation, either party may file a motion to enforce the Stipulation in the district court. Id. ¶ 31.

Finally, the Stipulation explains the nature and scope of the magistrate judge’s authority to resolve disputes arising out of the Stipulation. The relevant provision, Paragraph 36, provides as follows:

In the event the Court finds that Defendants have not complied with the Stipulation, it shall in the first instance require Defendants to submit a plan approved by the Court to remedy the deficiencies identified by the Court. In the event the Court subsequently determines that the Defendants’ plan did not remedy the deficiencies, the Court shall retain the power to enforce this Stipulation through all remedies provided by law, except that the Court shall not have the authority to order Defendants to construct a new prison or to hire a specific number or type of staff unless Defendants propose to do so as part of a plan to remedy a failure to comply with any provision of this Stipulation. In determining the subsequent remedies the Court shall consider whether to require Defendants to submit a revised plan.

Stipulation ¶ 36.

The appeals now before us are from various rulings of Magistrate Judge Duncan (acting on behalf of the district court) interpreting and enforcing the Stipulation. The first appeal involves Plaintiffs’ challenge to the district court’s ruling that the Stipulation precludes the court from ordering Defendants to develop a general staffing plan as a remedy for Defendants’ non-compliance. The second appeal concerns Defendants’ challenge to the magistrate judge’s order dated November 10, 2016, in which he ordered Defendants to use "all available community health care services" to meet their obligations under the Stipulation. The final appeal concerns Defendants’ challenge to the magistrate judge’s interpretation of the Stipulation’s subclass to include inmates classified as "close custody." For the reasons set forth below, we affirm the district court’s November 10, 2016 order, but reverse the other two rulings.

II.

We review de novo the district court’s interpretation of a stipulation of settlement. See Jeff D. v. Andrus , 899 F.2d 753, 759 (9th Cir. 1989). "[W]e defer to any factual findings made by the district court in interpreting the settlement agreement unless they are clearly erroneous." City of Emeryville v. Robinson , 621 F.3d 1251, 1261 (9th Cir. 2010).

We review the district court’s enforcement of a settlement agreement for abuse of discretion. Wilcox v. Arpaio , 753 F.3d 872, 875 (9th Cir. 2014). Under abuse-of-discretion review, we will reverse only if the district court made an error of law, or reached a result that was illogical, implausible, or without support in the record. United States v. Hinkson , 585 F.3d 1247, 1261–63 (9th Cir. 2009).

III.

Before turning the merits, we consider first the issue of subject matter jurisdiction over all three appeals. See Munoz v. Mabus , 630 F.3d 856, 860 (9th Cir. 2010). After three and a half years of litigating this case, Defendants move to dismiss the appeals on the ground that Magistrate Judge Duncan did not have jurisdiction to enter the orders at issue. "We review de novo whether a magistrate judge has jurisdiction," Wilhelm v. Rotman , 680 F.3d 1113, 1118 (9th Cir. 2012), recognizing that "our appellate jurisdiction depends on the proper exercise of magistrate judge jurisdiction," Anderson v. Woodcreek Venture Ltd. , 351 F.3d 911, 911 (9th Cir. 2003).

The Federal Magistrates Act, 28 U.S.C. §§ 631 – 39, governs the jurisdiction and authority of federal magistrate judges. United States v. Reyna-Tapia , 328 F.3d 1114, 1118 (9th Cir. 2003). The Act provides that "[u]pon consent of the parties, a full-time United States magistrate judge ... may conduct any or all proceedings in a jury or nonjury civil matter and order the entry of judgment in the case, when specially designated to exercise such jurisdiction by the district court." 28 U.S.C. § 636(c)(1). Thus, two requirements must be met before a magistrate judge may properly exercise civil jurisdiction: (1) the parties must consent to the magistrate judge’s authority and (2) the district court must "specially designate[ ]" the magistrate judge to exercise jurisdiction. Columbia Record Prods. v. Hot Wax Records, Inc. , 966 F.2d 515, 516 (9th Cir. 1992). We conclude both of these requirements were satisfied here.

First, Defendants do not dispute they voluntarily consented to the magistrate judge’s jurisdiction. After settling the case, the parties filed a joint motion to refer the case to Magistrate Judge Duncan in which they stated "[p]ursuant to 28 U.S.C. § 636(c) and Fed. R. Civ. P. 73, the parties hereby consent to have Magistrate Judge David Duncan conduct all further proceedings in this case." This is sufficient to demonstrate Defendants’ explicit, voluntary consent to the magistrate judge’s jurisdiction. See Anderson , 351 F.3d at 915 ; Gomez v. Vernon , 255 F.3d 1118, 1125–26 (9th Cir. 2001).

Second, the district court "specially designated" Magistrate Judge Duncan to exercise jurisdiction. In our decision in Columbia Record Productions , we suggested that designation generally derives from an "individual district judge." 966 F.2d at 516–17 ; see also Hill v. City of Seven Points , 230 F.3d 167, 168–69 (5th Cir. 2000) (equating "special designation" to "[t]he district court’s order of reference"). That is what occurred here. On October 22, 2014, District Judge Diane Humetewa entered a written order referring the case to Magistrate Judge Duncan and directing the clerk of court to reassign the case accordingly. Thus, Magistrate Judge’s Duncan designation was effective, and he had jurisdiction to enter the orders from which the parties appeal.

Defendants contend Magistrate Judge Duncan lacked jurisdiction because the parties "hand-picked" him, thereby disregarding the district court’s case assignment procedures. Citing the Seventh Circuit’s decision in Hatcher v. Consolidated City of Indianapolis , 323 F.3d 513 (7th Cir. 2003), Defendants argue that "magistrate judge assignment is a matter for the court to decide, not the parties," and therefore the district judge’s referral of the case to the parties’ hand-picked choice was invalid. Id. at 518.

Hatcher does not control the outcome here. In Hatcher , the parties entered into a settlement agreement by which they agreed to refer an unresolved legal fees issue to a named magistrate judge. Id. at 514–15. The Seventh Circuit concluded that the parties’ referral to a specific magistrate judge via settlement agreement could not be carried out because it disregarded the district court’s procedures for assigning magistrate judges. Id. at 517–19. Here, by contrast, it was the district court itself that referred the case to Magistrate Judge Duncan, not the parties. Although the parties specifically requested referral to Magistrate Judge Duncan, they did not proceed on the authority of their own "referral" as in Hatcher . Rather, they proceeded based on the district court’s designation by written order. This judicial designation validates the referral here, and differentiates it from the invalid referral in Hatcher . Therefore, Defendants’ reliance on Hatcher is unavailing.

Defendants also argue the district judge was precluded from referring the case specifically to Magistrate Judge Duncan because the District of Arizona’s Local Rules require that magistrate judges be assigned by automated random selection. But although the Local Rules provide for magistrate judge jurisdiction "when the case is ... randomly assigned by the Clerk to a Magistrate Judge upon the filing of the case," the Rules also allow for magistrate judge jurisdiction "when a case is initially assigned to a District Judge and thereafter the case is reassigned to a Magistrate Judge with the District Judge’s approval." LRCiv 72.2(a)(13). There is nothing in the Rules that requires "reassign[ment] to a Magistrate Judge with the District Judge’s approval" to occur by automated random selection. Rather, the phrase "with the District Judge’s approval" implies that the reassignment decision is one of discretion, not random assignment. The broader structure of the Rules confirms this reading. See LRCiv 3.7(a)(1) (stating that the Clerk of Court must initially assign civil cases by automated random selection and in a manner that does not permit the parties to choose a particular judge "[u]nless otherwise provided in these Rules or ordered by the Court"); LRCiv 73.1(d) (stating, in part, that cases assigned to a magistrate judge by random automated selection "shall remain with the Magistrate Judge to whom assigned unless otherwise ordered by the Court"). Therefore, we reject the argument that the district court’s referral of the case to Magistrate Judge Duncan violated the Local Rules.We conclude Magistrate Judge Duncan’s jurisdiction was proper. Defendants’ motion to dismiss the appeals for lack of jurisdiction is denied.

IV.

We turn now to Plaintiffs’ appeal from the district court’s February 3, 2017 order in which the court concluded that the Stipulation precluded it from ordering Defendants to develop a plan to increase staffing. The district court reasoned that such a plan would violate the Stipulation’s provision "that the Court shall not have the authority to order Defendants ... to hire a specific number or type of staff." Plaintiffs contend this interpretation violates the plain language of the Stipulation and runs contrary to principles of contract interpretation.

A.

As a preliminary matter, we address briefly Defendants’ jurisdictional challenge to this appeal. Defendants argue this appeal is untimely because Plaintiffs filed it more than 30 days after the district court stated during a September 2016 status hearing that the Stipulation bars the court from issuing a general staffing order. This argument is groundless. The main purpose of the September 2016 status hearing was to evaluate the effectiveness of Defendants’ remediation plan, not to resolve definitively a dispute about whether the Stipulation allows the district court to issue a general staffing order. The magistrate judge did not purport to resolve this issue conclusively until the parties briefed it, after which he issued a written order on February 3, 2017, denying Plaintiffs’ motion for an order requiring Defendants to develop a staffing plan. It is this order that is the relevant decision for starting the appeals clock. See Campbell Indus., Inc. v. Offshore Logistics Int’l Inc. , 816 F.2d 1401, 1404 (9th Cir. 1987) ("Only when a judge acts in a manner which clearly indicates an intention that the act be final, and a notation of that act has been entered on the docket, does the time for appeal begin to run."). Plaintiffs filed their notice of appeal on February 17, 2017, well within 30 days of the February 3 order. Therefore, the appeal is timely. See Fed. R. App. P. 4(a)(1)(A).

B.

We proceed now to the merits. Our interpretation of the Stipulation is governed by "principles of local law which apply to interpretations of contracts generally." Jeff D. , 899 F.2d at 759. Here, we apply Arizona contract law because the parties entered into the Stipulation in Arizona, Defendants are senior officials of the Arizona Department of Corrections, and the Stipulation concerns the policies and practices of the Arizona prison system. See Kelly v. Wengler , 822 F.3d 1085, 1095 (9th Cir. 2016) ; Jeff D. , 899 F.2d at 759–60.

"The purpose of contract interpretation is to determine the parties’ intent and enforce that intent." Grosvenor Holdings, L.C. v. Figueroa , 222 Ariz. 588, 593, 218 P.3d 1045 (Ct. App. 2009). To determine the parties’ intent, we "look to the plain meaning of the words as viewed in the context of the contract as a whole." Earle Invs., LLC v. S. Desert Med. Ctr. Partners , 242 Ariz. 252, 255, 394 P.3d 1089 (Ct. App. 2017). "If the contractual language is clear, we will afford it its plain and ordinary meaning and apply it as written." Liberty Ins. Underwriters, Inc. v. Weitz Co., LLC , 215 Ariz. 80, 83, 158 P.3d 209 (Ct. App. 2007).

Here, the Stipulation is clear on the limits of the district court’s authority to enforce the Stipulation. The relevant provision, Paragraph 36, provides that "the Court shall retain the power to enforce this Stipulation through all remedies provided by law, except that the Court shall not have the authority to order Defendants to ... hire a specific number or type of staff unless Defendants propose to do so as part of a plan to remedy a failure to comply with any provision of this Stipulation." Stipulation ¶ 36. Under this provision, the district court could not, for example, order Defendants to hire 20 additional employees at the Yuma facility or 10 additional registered nurses at the Tucson facility.

However, Paragraph 36 does not, by its plain language, preclude the district court from ordering Defendants to develop and implement a plan to increase staffing in general. Such a general staffing order would not, without more, violate the Stipulation because Defendants would retain discretion over the specific number and type of personnel to hire pursuant to such an order. For example, Defendants could develop a plan that relied on a small number of new hires, while emphasizing structural reforms to the prison health care delivery system. Or, Defendants could develop a plan that relied on significant increases in hiring in one specific job category, while leaving other staffing levels in place. Regardless of Defendants’ specific decisions, the key is that a general staffing order would not bind Defendants to "hire a specific number or type of staff" dictated by the district court. That decision would remain within Defendants’ discretion. Therefore, we conclude the plain language of the Stipulation permits the district court to order Defendants to develop a general staffing plan. The district court’s contrary conclusion was error.

Defendants make two arguments for why the district court’s interpretation of the Stipulation is correct. First, Defendants advance the position the district court accepted below: that an order to develop a plan to increase staffing in general is the "functional equivalent" of an order requiring a specific number and type of staff. We disagree. A general staffing order by the district court would not intrude upon Defendants’ discretion to determine the "specific number or type of staff" they believe is appropriate. As explained earlier, Defendants could develop a plan that places more emphasis on structural changes than on new hires, or a plan that limits new hires to a specific job category. Although it is true that a general staffing order would require Defendants to make staff hiring part of the solution, it would preserve Defendants’ discretion to determine the number and type of staff to hire as part of that solution. Accordingly, we reject the argument that a general staffing order is the "functional equivalent" of an order to hire a specific number and type of staff.

Second, Defendants contend we must defer to the district court’s interpretation because that interpretation was based on the district court’s first-hand understanding of the parties’ intent. But Defendants’ reference to the district court’s understanding of the parties’ intent is to a statement the court made during a status hearing months before the court’s written order. We do not review oral statements from the bench on a matter later committed to writing; we review instead the written order entered by the district court. Playmakers LLC v. ESPN, Inc. , 376 F.3d 894, 896 (9th Cir. 2004) ("Where the record includes both oral and written rulings on the same matter, ‘we review the written opinion and not the oral statements.’ ") (internal quotation marks omitted). In his written order, the magistrate judge concluded he could not enter a general staffing order on the ground that such an order would necessarily involve ordering a specific number or type of staff. This order—which followed briefing by the parties—makes no mention of the district court’s understanding of the parties’ intent. Under these circumstances, we will not treat the district court’s earlier oral remarks as a basis for its later written decision. Ellison v. Shell Oil Co. , 882 F.2d 349, 352 (9th Cir. 1989) ("Oral responses from the bench may fail to convey the judge’s ultimate evaluation. Subsequent consideration may cause the district judge to modify his or her views."). Therefore, we reject Defendants’ attempt to invoke the district court’s first-hand understanding of the parties’ intent as a basis for its interpretation.

In sum, we conclude the district court erred in interpreting the Stipulation as precluding it from ordering Defendants to develop and implement a plan to increase staffing in general. We therefore reverse the district court’s February 3, 2017 order. Consistent with our ruling, the district court may, in future proceedings, consider whether a general staffing order that does not require Defendants to hire a specific number or type of staff is an appropriate remedy for Defendants’ non-compliance.

V.

We turn now to Defendants’ appeal from the district court’s November 10, 2016 order requiring Defendants to "use all available community healthcare services" to ensure compliance with certain performance measures ("Outside Provider Order" or "OPO").

The district court entered the Outside Provider Order to remedy Defendants’ non-compliance with certain performance measures that require inmates to receive health care services within prescribed time frames. For example, Performance Measure 37 provides: "Sick call inmates will be seen by an RN within 24 hours after an HNR [Health Needs Request form] is received (or immediately if identified with an emergent need, or on the same day if identified as having an urgent need)." Defendants’ compliance rates with this and related performance measures were as low as 34 percent at the time the district court entered the OPO.

After giving Defendants an opportunity to remedy their non-compliance under their own remediation plan, the district court entered the OPO. The OPO provides:

Defendants shall use all available community health care services including, but not limited to, commercial pharmacies, community-based practitioners, urgent care facilities, and hospitals (collectively, "Outside Providers") to provide the health care services required in the Stipulation’s Performance Measures. This shall happen immediately following the expiration of the time-frame detailed in each PM. For example, if a PM requires Defendants to provide an inmate with a specific type of care within 24 hours (or 14 days), then Defendants shall have this inmate seen by an appropriate Outside Provider in hour 25 (or day 15).

The Court notes that these requirements only apply when Defendants are not able to comply with the Stipulation’s Performance Measures using the procedures detailed in their remediation plan. In other words, if Defendants can comply with the Stipulation without using Outside Providers, then they are under no obligation to use Outside Providers.

The district court did not abuse its discretion in issuing the Outside Provider Order. After finding Defendants in substantial non-compliance with certain performance measures, the district court properly ordered Defendants to submit a remediation plan, and then approved that plan despite the court’s skepticism that it represented a serious solution. The district court then gave Defendants three months to demonstrate compliance, and later granted them additional time to comply even as the data indicated "a serious failure to be even close on a number of the performance measures." Finally, only after the latest data showed that Defendants remained in substantial non-compliance did the district court issue the OPO. In light of the district court’s strict adherence to the dispute resolution procedure outlined in the Stipulation and careful consideration of the record, we conclude the district court did not abuse its discretion in issuing the OPO.

Defendants’ challenge the OPO on several grounds. First, Defendants argue the OPO effectively re-writes the Stipulation to require 100 percent compliance with the performance measures, rather than 80 percent. As an example, Defendants point to the OPO’s impact on Performance Measure 37, which requires sick call inmates to be seen by an RN within 24 hours of submitting a health needs request form. Pursuant to the OPO, Defendants must ensure sick call inmates not seen by an RN within 24 hours are seen within 25 hours. In Defendants’ view, the difficulty of tracking inmate-provider contact after hour 24 effectively forces Defendants to ensure all inmates are seen within 24 hours, lest they risk violating the OPO.

We disagree. Although the OPO requires Defendants to use outside providers if Defendants cannot otherwise treat inmates within the prescribed time frame, it does not, in fact, change the threshold for substantial compliance. The threshold for substantial compliance remains 80 percent. In other words, the OPO is simply a remedy to address Defendants’ non-compliance, it does not change what constitutes compliance for purposes of avoiding judicial enforcement. So long as Defendants meet or exceed the 80 percent benchmark provided in the Stipulation, the OPO has no effect. Therefore, we disagree with the notion that the OPO effectively requires 100 percent compliance.

Second, Defendants argue that the district court abused its discretion by entering the OPO without considering alternative remedies. Not so. Not only does the record indicate the district court considered alternatives on its own accord, see, e.g. , Transcript of November 9, 2016 status hearing, District Ct. Dkt. 1765 at 9 (stating that the OPO is "the only [remedy] that I have been able to conclude that could work"), the court also stressed that Defendants should identify alternatives as soon as it became clear their preferred plan was not working. Defendants did not do so. Although Defendants now point out they have developed an "open-clinic concept" that has led to increased compliance with one of the performance measures, Defendants made no mention of this plan until after the district court issued the OPO. We will not fault the district court for failing to adopt a partial solution that Defendants did not timely propose.

Third, Defendants contend the OPO creates an "unprecedented" security risk by requiring Defendants to transport "hundreds of inmates on a daily basis" to outside medical facilities. We reject this argument because it relies on a premise not supported by the record. Although the OPO requires Defendants to use outside providers if Defendants cannot otherwise comply with the performance measures, the OPO does not require Defendants to transport a specific number (or any number) of inmates to outside facilities. As the district court pointed out, Defendants can avoid transporting inmates offsite by bringing outside providers to the prisons, or by simply hiring more healthcare providers to work within the prison system. Defendants can also avoid transporting inmates offsite by making greater use of information technology to provide clinical care remotely, or by adopting internal changes—such as the open clinic concept Defendants are currently implementing—that ensure compliance with the Stipulation. In light of the considerable discretion Defendants have in deciding how to connect inmates with outside providers, the presumption that the OPO requires large-scale transportation of inmates offsite is unwarranted.

Fourth, Defendants argue the OPO imposes an "impossible" logistical burden because the potential volume of inmate transports would require vehicles and staff beyond Defendants’ current resources. This argument is similar to the argument regarding "security risks" addressed above, and fails for the same reason—the OPO does not require any specific number of inmates to be transported offsite. As explained above, if Defendants prefer not to transport inmates offsite, they have alternatives for ensuring inmates receive the care to which they are entitled. Thus, we reject Defendants’ argument that the OPO is excessively burdensome. Cf. Armstrong v. Schwarzenegger , 622 F.3d 1058, 1071 (9th Cir. 2010) ("A demonstration that an order [issued to vindicate the federal rights of prisoners] is burdensome does nothing to prove that it was overly intrusive.").

Finally, Defendants argue that the district court erred in certifying the OPO as compliant with the Prisoner Litigation Reform Act (PLRA). Under the PLRA, a court may not order "any prospective relief [with respect to prison conditions] unless the court finds that such relief is narrowly drawn, extends no further than necessary to correct the violation of the Federal right, and is the least intrusive means necessary to correct the violation of the Federal right." 18 U.S.C. § 3626(a)(1)(A). Defendants assert the OPO does not comply with the PLRA because the district court never determined that a constitutional violation occurred.

Defendants are incorrect. In approving the Stipulation, the district court held "[b]ased upon the entire record in this case and the parties’ Stipulation" that the Stipulation was "necessary to correct the violations of the Federal right of the Plaintiffs." This conclusion necessarily required a finding of a constitutional violation—that is, if there were no violation of a federal right, there would be nothing for the Stipulation to "correct." Therefore, the district court found the requisite constitutional violation in granting the initial prospective relief in this case.

Nor do we accept Defendants’ suggestion that the district court was required to make new findings of a constitutional violation before entering the OPO. The district court issued the OPO to enforce compliance with the Stipulation (which the parties agreed was necessary to correct violations of Plaintiffs’ federal rights); it did not issue the OPO as prospective relief in response to new violations of federal rights. That is, the same constitutional violations upon which the Stipulation rests are the same violations the OPO is intended to remedy. Accordingly, the district court was not required to make new findings of a constitutional violation before enforcing the Stipulation with the OPO.

In sum, we conclude the district court did not abuse its discretion in issuing the Outside Provider Order.

VI.

The final issue before us involves the district court’s December 23, 2016 order ("Close Custody Order"), in which the court interpreted the subclass to include all close custody inmates not otherwise participating in a prison jobs program. Defendants contend the district court erred in adopting this interpretation because the amount of out-of-cell time offered to close custody inmates places them outside the definition of the subclass.

The Stipulation defines the subclass as follows:

All prisoners who are now, or will in the future be, subjected by the ADC to isolation, defined as confinement in a cell for 22 hours or more each day or confinement in the following housing units: Eyman–SMU I; Eyman–Browning Unit; Florence–Central Unit; Florence–Kasson Unit; or Perryville–Lumley Special Management Area.

Stipulation ¶ 3.

At the time the Stipulation went into effect, the five housing units in the subclass definition made up the entirety of Arizona’s maximum custody prison facilities.

In October 2016, Plaintiffs sought records for two inmates housed in Florence-Central and Eyman-SMU I to assess compliance with the maximum custody performance measures. In response, Defendants informed the district court that although the inmates in question were housed at those units, they were classified as "close custody," rather than maximum custody. Defendants explained that close custody inmates are offered at least 15.5 hours of out-of-cell time per week, placing them outside the definition of the subclass and therefore outside the coverage of the maximum custody performance measures. The district court found to the contrary, concluding that "close custody inmates are subject to substantially similar conditions as maximum custody inmates and, therefore, are part of the Subclass." Defendants appealed.

A.

As a threshold matter, we address Plaintiffs’ argument that the district court’s Close Custody Order was insufficiently "final" and therefore not appealable under 28 U.S.C. § 1291.

We generally have jurisdiction over only final decisions of the district courts. 28 U.S.C. § 1291. A "final decision" is typically "one which ends the litigation on the merits and leaves nothing for the court to do but execute the judgment." United States v. Washington , 761 F.2d 1404, 1406 (9th Cir. 1985) (citation omitted). Under the collateral order doctrine, however, an order that does not strictly end the litigation may nonetheless be considered sufficiently final when it is "too important to be denied review and too independent of the merits of the case to require deferral of review." Plata v. Brown , 754 F.3d 1070, 1075 (9th Cir. 2014) (citation and internal quotation marks omitted). "To warrant review under the collateral order doctrine, the order must ‘(1) conclusively determine the disputed question, (2) resolve an important issue completely separate from the merits of the action, and (3) be effectively unreviewable on appeal from a final judgment.’ " Id. (quoting Will v. Hallock , 546 U.S. 345, 349, 126 S.Ct. 952, 163 L.Ed.2d 836 (2006) ).

We have jurisdiction to review the Close Custody Order under the collateral order doctrine. The Order is conclusive in that it is the district court’s final determination of whether close custody inmates are part of the isolation subclass. The Order involves "an important issue completely separate from the merits" because it decides a question of law not connected to the merits of Defendants’ liability for Eighth Amendment violations. Finally, the Order is "effectively unreviewable on appeal" because Defendants’ good-faith compliance with the Order would, in effect, deprive Plaintiffs of the opportunity to challenge it. If Defendants comply with the Close Custody Order, Plaintiffs would have no reason to move to enforce it, which would close off the most likely avenue for appeal indefinitely. Accordingly, we conclude we have jurisdiction over Defendants’ appeal from the Close Custody Order.

Plaintiffs’ first argument to the contrary is that the Close Custody Order is not "final" because it does not require Defendants "to take, or refrain from taking, any action whatsoever." That is incorrect. The district court’s ruling requires Defendants to apply the maximum custody performance measures to close custody inmates when, prior to the Close Custody Order, those performance measures applied only to maximum custody inmates. Thus, by extending the application of the maximum custody requirements to close custody inmates, the Close Custody Order does, in fact, require Defendants to take action.

Plaintiffs also argue the Close Custody Order is not "final" because it only asks the parties to take the intermediate step of developing a plan for implementing the court’s interpretation of the subclass. This argument misreads the Order. The district court’s request that the parties develop a plan for implementation was in reference to the classification of a subset of inmates who work 20 hours per week as part of a prison jobs program—inmates the court found were not members of the subclass. The district court did not ask the parties to develop a plan concerning its ruling that all close custody inmates not in the jobs program are part of the subclass. It is this latter ruling that Defendants challenge. Therefore, Plaintiffs’ reliance on the district court’s request that the parties develop a plan does not make the Close Custody Order any less "final" for purposes of our jurisdiction.

B.

We turn now to the merits. Defendants argue the district court erred in interpreting the subclass to include close custody inmates offered 15.5 hours or more out-of-cell time each week.

In interpreting the subclass to include close custody inmates, the district court concluded that "close custody inmates are subject to substantially similar conditions as maximum custody inmates, and therefore, are part of the Subclass." This was error. The touchstone for inclusion in the subclass is not "substantially similar conditions" but rather the amount of isolation experienced by inmates. The subclass is defined as inmates who are confined in a cell for 22 hours or more each day (i.e., inmates who receive less than 14 hours of out-of-cell time each week). Therefore, by concluding that inmates offered 15.5 hours of out-of-cell time each week fall within the subclass, the district court effectively rewrote the subclass definition. The parties set the benchmark for inclusion in the subclass at 14 hours; the district court cannot unilaterally move that benchmark to 15.5 hours. See Isaak v. Mass. Indem. Life Ins. Co. , 127 Ariz. 581, 623 P.2d 11, 14 (1981) ("It is not within the power of [a] court to ‘revise, modify, alter, extend, or remake’ a contract to include terms not agreed upon by the parties.") (citation omitted).

Plaintiffs contend the district court did not err in interpreting the subclass because Defendants did not prove that inmates who are offered 15.5 hours or more out-of-cell time per week actually take the time offered. For example, Plaintiffs assert that out-of-cell activities offered to close custody inmates include visitation and religious services, but that the record does not show that all such inmates receive visitors or participate in religious services. On this ground, Plaintiffs argue the district court was correct to conclude that out-of-cell time offered to close custody inmates is merely "theoretical," and therefore an insufficient basis for treating close custody inmates differently than maximum custody inmates.

We disagree. The subclass definition turns on the amount of time an inmate is "confine[d] in a cell" each day. Confinement, of course, connotes a lack of control over whether to leave a particular place. See Oxford English Dictionary (online ed. 2018) (defining "confinement" as "the fact or condition of being confined, shut up, or kept in one place"). On this understanding, an inmate given an opportunity to participate in out-of-cell activities cannot be considered "confined" in a cell during that time even if the inmate may theoretically decide not to take advantage of the opportunity. See Judith Resnik et al., Time-In-Cell: Isolation and Incarceration , 125 Yale L.J. F. 212, 219 (2016) (characterizing prisoner isolation as a condition of confinement in which opportunities for social contact, "such as out-of-cell time for exercise, visits, and programs," are restricted). For example, a close custody inmate who is offered 15 hours of out-of-cell time per week for education, but turns it down, is in a much different position than a maximum custody prisoner who does not have that option.

The broader structure of the Stipulation supports this reading: many of the provisions relating to maximum custody inmates require Defendants to offer inmates a minimum amount of out-of-cell time, not compel inmates to take that time. See Stipulation ¶¶ 22, 24–26. This framing of the out-of-cell-time requirement makes perfect sense: although Defendants can control whether to provide meaningful opportunities to inmates for out-of-cell activities, it cannot control whether an inmate’s individual preferences, family situation, or subjective motivations will lead or allow the inmate to take advantage of the time offered. Here, Defendants have shown that close custody inmates are offered meaningful opportunities for weekly out-of-cell time that far exceeds 14 hours per week, including for education, library visits, recreation, dinner, showers, religious group worship, and visitation. This is sufficient to place these inmates outside of the subclass. Plaintiffs next argue the plain language of the subclass definition supports the district court’s reading because that language refers not only to hours confined in a cell, but also to confinement in five specific housing units. Plaintiffs contend that the reference to specific housing units means any inmate housed in those units is part of the subclass regardless of how much out-of-cell time the inmate receives.

We reject this interpretation. Although the subclass refers to inmates housed in specific units, this reference merely captures what was known to the parties and the court at the time the court certified the subclass: that the enumerated housing units composed all of the maximum custody facilities in the Arizona prison system. Thus, enumeration of the maximum custody facilities in the subclass did not expand the subclass to include close custody inmates; it simply reflected the focus of the subclass on those inmates subjected to the most isolating conditions of confinement.

Adopting Plaintiffs’ interpretation would lead to absurd results. Under Plaintiffs’ reading, a close custody inmate who received 40 hours of out-of-cell time per week, but happens to be located at one of the maximum custody facilities such as Florence-Central, would nonetheless be subjected to the maximum custody performance measures. Those performance measures, however, require Defendants to offer inmates between 7.5 to 22.5 hours of out-of-cell time per week. Under Plaintiffs’ interpretation, then, a close custody inmate offered 40 hours of out-of-cell time per week would only need to receive a portion of that time for Defendants to comply with the Stipulation. Such a result would directly undercut one of the fundamental aims of the agreement—to reduce inmate isolation. We therefore reject the argument that an inmate’s mere location in a housing unit, rather than the amount of time confined in a cell, suffices to place the inmate within the subclass. See Bryceland v. Northey , 160 Ariz. 213, 216, 772 P.2d 36 (Ct. App. 1989) ("We will interpret a contract in a manner which gives a reasonable meaning to the manifested intent of the parties rather than an interpretation that would render the contract unreasonable.").

VII.

For the foregoing reasons, we affirm the district court’s November 10, 2016 Outside Provider Order; reverse the district court’s February 3, 2017 ruling that the Stipulation precludes it from issuing a general staffing order; and reverse the district court’s December 23, 2016 ruling that close custody inmates are part of the subclass. Consistent with this opinion, the district court may, in the future, consider ordering Defendants to develop and implement a plan to increase staffing in general as a remedy for Defendants’ non-compliance. In addition, offering close custody inmates 15.5 hours or more out-of-cell time per week is sufficient to place these inmates outside of the subclass for purposes of monitoring compliance with the Stipulation.

The parties shall bear their own costs on appeal. Any pending motions are DENIED.

AFFIRMED in part, REVERSED in part, and REMANDED.

THOMAS, Chief Judge, concurring in part and dissenting in part:I concur in the majority’s conclusions as to the staffing appeal and the outside providers appeal. I also concur in the majority’s conclusion that we have jurisdiction over the Close Custody Order appeal. However, I part ways from the majority in its conclusion that the Close Custody Order was an abuse of discretion. The Close Custody Order was based on factual findings that were plausible in light of the evidence presented by Defendants. I would affirm the district court.

I

Defendants appeal from the district court’s December 23, 2016 Close Custody Order and from the district court’s February 6, 2017 order denying their motion for relief under Federal Rule of Civil Procedure 60(b). We review for an abuse of discretion the district court’s enforcement of the settlement agreement in the Close Custody Order, Doi v. Halekulani Corp. , 276 F.3d 1131, 1136 (9th Cir. 2002), and the denial of the Rule 60(b) motion, Delay v. Gordon , 475 F.3d 1039, 1043 (9th Cir. 2007). Abuse of discretion review presents a high threshold for appeal on questions of fact. See Anderson v. City of Bessemer City , 470 U.S. 564, 573–74, 105 S.Ct. 1504, 84 L.Ed.2d 518 (1985) ("If the district court’s account of the evidence is plausible in light of the record viewed in its entirety, the court of appeals may not reverse it even though convinced that had it been sitting as the trier of fact, it would have weighed the evidence differently.").

II

The district court did not abuse its discretion. In its November 8, 2016 order, the district court held that the subclass would thereafter consist of inmates who are subjected to isolation, defined as containment in a cell for 22 hours or more each day (i.e., less than 14 hours of out-of-cell time each week). The district court asked Defendants to present competent, admissible evidence demonstrating that the close custody inmates in enumerated housing units were not subject to these conditions.

On November 22, 2016, Defendants presented to the district court a list of activities that were "offered" to close custody inmates each week. In its December 23, 2016 Close Custody Order, the district court concluded that Defendants had "provided a theoretical explanation of what close custody inmates may experience without showing that any particular inmate actually has experienced these out-of-cell options." By contrast, the district court concluded that Defendants had provided sufficiently detailed and specific evidence that 276 close custody inmates worked in different positions for, on average, 30 hours per week. Defendants had provided job titles and the number of inmates working in each position. The district court concluded that these inmates experienced different confinement conditions than the maximum custody inmates, and thus, they would not be considered members of the subclass.

The district court’s conclusions are plausible in light of the record. As Plaintiffs argue, some of the activities in Defendants’ list of offerings may not have actually been made available to many of the close custody inmates. For example, Defendants list visitations as an offered activity, but many inmates may not have any visitors. Defendants list visits to the store to pick up purchases as an offered activity, but many inmates may not have any money to make purchases at the store (and thus may not be allowed to go pick up purchases). Some of the programming activities that Defendants list—such as "Re-Entry," "Substance use/AA," and "Cognitive Behavior"—have a limited number of slots, and thus would not be made available to all inmates. Given these limitations, the district court concluded that Defendants’ list of possible activities was not sufficient to show that any particular inmate is actually offered more than 14 hours of out-of-cell time each week. This conclusion was not implausible in light of the record. Therefore, I do not believe the district court abused its discretion in ruling that close custody inmates were part of the subclass. For this reason, I respectfully dissent from the majority’s contrary holding, but concur in all other respects.

CALLAHAN, Circuit Judge, concurring in part and dissenting in part:

I concur in the majority’s conclusion as to subject matter and appellate jurisdiction for the three appeals. I also concur in the majority’s conclusion that the district court erred in interpreting the subclass to include all close custody inmates not otherwise participating in a prison jobs program (Part VI of the majority opinion). However, I cannot agree with the majority’s disposition of the staffing appeal and the outside providers appeal (Parts IV and V, respectively). I would affirm the district court’s February 3, 2017 staffing order and reverse the district court’s November 10, 2016 outside providers order.

I.

The district court’s role in this case is purely to interpret and enforce the terms of the Stipulation. Although the Stipulation authorizes the district court to remedy non-compliance, at the bargaining table, the parties removed one particular "tool," as the district court put it, from the court’s "remedial toolbox." The Stipulation expressly prohibits the court from "order[ing] Defendants to ... hire a specific number or type of staff unless Defendants propose to do so as part of a plan to remedy a failure to comply with any provision of this Stipulation." The majority concludes that although the Stipulation prevents the district court from ordering Defendants to hire a specific number of staff, the court may order Defendants to increase staffing in general. I cannot agree. Instead, I agree with the district court’s interpretation of the Stipulation that the court may not do indirectly what the Stipulation prohibits it from doing directly.

The majority states that a general staffing order would preserve Defendants’ discretion to determine the exact number and type of staff to hire. But, assuming the court has the power to issue a general staffing order, the court presumably would not (and, arguably, could not) approve a proposed staffing plan unless it were to deem the plan adequate. Certainly, a vague statement by Defendants that they would "increase" staffing in some undisclosed way would not be deemed adequate. Rather, the adequacy of a general staffing order could not be determined without considering the number and type of staff. Additionally, under the majority’s rationale, if Defendants’ compliance were to remain unsatisfactory after an increase in staff, nothing would prevent the court from again deeming staffing inadequate and again ordering a "general" staffing increase. This process could continue until the court finally deems staffing adequate. Perhaps other than being much more costly, such a protracted process—whereby the court effectively tells Defendants to "keep trying" over-and-over until they have sufficiently increased staffing—bears no meaningful difference from directly ordering Defendants to hire a specific number of staff.

I agree with the district court that an order to develop a plan to increase staffing in general is the "functional equivalent" of an order requiring a specific number and type of staff, which the Stipulation prohibits. I would thus affirm the district court’s February 3, 2017 order.

II.

I cannot agree with the majority’s decision to affirm the outside providers order. The majority rejects Defendants’ argument that the order effectively requires 100 percent compliance, contrary to the 80 percent benchmark provided in the Stipulation. But the majority’s interpretation of the order conflicts with Judge Duncan’s own interpretation of his order. At the November 9, 2016 hearing where Judge Duncan announced his intention to order Defendants to use outside providers, he stated that he was requiring 100 percent compliance. Likewise, in his order denying Defendants’ Rule 60(b) motion, Judge Duncan characterized the outside providers order as "requir[ing] Defendants to pursue 100% compliance." I would defer to Judge Duncan’s own interpretation of his order and agree with Defendants that such a ruling erroneously modifies the Stipulation.

Its issuance of the outside providers order one day after orally announcing the intended decision also prevented the district court from adequately taking into account the security risks created by ordering Defendants to transport hundreds of inmates on a daily basis to outside medical facilities. In my view, the majority is too quick to dismiss this concern.

I would thus reverse the district court’s November 10, 2017 outside providers order.

III.

For the foregoing reasons, I would affirm the staffing order and reverse the outside providers order.¹ I otherwise concur in the majority’s opinion.