Opinion by Judge Bennett; Partial Concurrence and Partial Dissent by Judge Rawlinson.
OPINION
BENNETT, District Judge:
Plaintiffs seek damages from the United States for injuries to a child allegedly caused by exposure to the toxic heavy metal thallium from soil dumped into a landfill adjacent to the child’s residence and school. The child, by her guardian ad litem, appeals a decision of the district court finding that the United States acted “reasonably” and did not breach any duty in conducting the soil remediation project. The district court also found that it did not have subject matter jurisdiction, because the “discretionary function” exception to tort liability of the United States applies in this case. We reverse and remand for further proceedings.
I. BACKGROUND
A. Factual Background
Whether or not the district court’s findings of fact are clearly erroneous depends upon “the entire evidence” in the record. See United States v. Hinkson, 585 F.3d 1247, 1260 (9th Cir.2009) (en banc). Therefore, this statement of the factual background identifies both the district court’s findings and other evidence in the record that is relevant to the review of the district court’s findings.
[1024]*1024As the district court found, in 1989, the Environmental Protection Agency (EPA) placed the United States Marine Corps Base at Camp Pendleton on the “National Priorities List” of sites requiring environmental cleanup. The Department of the Navy entered into a comprehensive environmental cleanup plan for Camp Pendleton, known as a Federal Facility Agreement (FFA), with the concurrence of the EPA, the California Department of Toxic Substances Control (DTSC), and the San Diego Regional Water Quality Control Board (RWQCB).
Although the district court did not refer to such a provision in its findings, the FFA required, among other things, that the Navy,1 as the party responsible for the cleanup, designate a Quality Assurance Officer (QAO) as follows:
20.1 In order to provide quality assurance and maintain quality control regarding all field work and sample collection performed pursuant to this Agreement, the Marine Corps agrees to designate a Quality Control Officer (QAO) who will ensure that all work is performed in accordance with approved work plans, sampling plans and QAPPS [Quality Assurance Project Plans]. The QAO shall maintain for inspection a log of quality assurance field activities and provide a copy to the Parties upon request.
FFA, ¶ 20.1. The district court did not note in its findings that the Naval Facilities Engineering Command (NAVFA-CENGCOM) also uses a Safety and Health Program Manual (the Manual) for all environmental cleanup operations. The Manual specified, in pertinent part, that “[e]ach NAVFACENGCOM activity shall ensure that plans are reviewed and accepted prior to issuing the Notice to Proceed.” Manual, ¶ 0407.b. The Manual also provided as follows:
c. Reviews. All HASPs [ (health and safety plans) ] shall be reviewed prior to initiating site work by a competent person. Competent person shall mean a certified industrial hygienist [ (CIH) ] or equivalent by training and/or experience. In addition, an EFD/EFA Construction Safety Manager or designated representative who has sufficient knowledge and authority to review and accept construction safety procedures shall review HASPs for construction safety requirements.
Manual, ¶ 0407.C.
Sites at Camp Pendleton requiring cleanup were divided into “operable units.” “Operable Unit 3” (OU-3) consisted of five contaminated areas where wastes had been burned. Tests of the soil in all five areas indicated contamination with toxic substances, but two, known as Sites 1A and 2A, showed elevated levels of thallium.2 As to the sites showing elevated [1025]*1025levels of thallium, the district court found as follows:
Of the 154 samples [taken at Site 1A], only one sample exceeded the safe standard for thallium — and not by much. At Site 2A, 99 soil samples were taken. ... Of the samples, only two samples exceeded] the safe standard for thallium. One sample was only a little high. The other sample was high, but appeared to be an unreliable test result. The Defendant’s contractor used the Inductively Couple Plasma — Atomic Emission Spectroscopy method to identify thallium in the soil samples. That method sometimes produces false positives where there are metals in the sample. In the soil at site 2A, there were high concentrations of other metals such as zinc and manganese, which likely caused the false positive test for thallium. Lead was much more prevalent than thallium in the soil moved to Box Canyon and was considered to be the primary risk to human health.
Although the district court found that lead was the primary risk to human health, the Navy’s “Record of Decision” (ROD), which set forth its plan to clean up or “remediate” the sites in OU-3, actually states, “[t]he primary contributors to the HI [hazard index] [for Site 1A] are arsenic, copper.” Lead is not mentioned in that list, although concentrations of lead above established safety levels were also noted. Similarly, the ROD actually states, “The primary hazard contributors [for Site 2A] are manganese, thallium, and zinc.” Again, lead is not mentioned in that list, although concentrations of lead above established safety levels were again noted.
The HI for both Site 1A and Site 2A indicated that there was a potentially “unacceptable” risk to human health requiring cleanup. However, the district court found that the thallium at these sites was believed to be in low concentrations in the contaminated soil.
The Navy’s plan to clean up or “remediate” the sites in OU-3 involved excavation of contaminated soils from the OU-3 sites, transportation of those soils across the base by dump truck, and dumping of those soils into a landfill known as Box Canyon Landfill or Site 7. The Box Canyon Landfill was adjacent to the Wire Mountain Family Housing area of the Marine Corps base at Camp Pendleton and near an elementary school.
The Navy contracted with IT/OHM to perform much of the work on the OU-3 project. The Navy’s contract with IT/ OHM required IT/OHM to prepare a HASP (referred to in the contract as a “Site Health and Safety Plan” or SHSP), and IT/OHM did so. Under the HASP, IT/OHM’s industrial hygienist (or “Health and Safety Officer”) had on-site responsibility and authority to modify or stop work if working conditions presented a risk to health and safety. The HASP included requirements for monitoring ambient air and airborne contaminants, including levels for “total dust” that should have required work stoppages. If dust levels from monitors placed near the housing area or the school exceeded a certain level, then the contractor was to determine the source of the dust, increase the dust control efforts in the landfill, and if increased dust control measures were “ineffective,” stop soil moving activities in the landfill. The district court found that the air monitors employed by IT/OHM were appropriate for the task. However, the district court failed to note that the air monitoring device specified did not monitor “total dust,” but only dust particles smaller than a certain size. In addition to air monitoring requirements, the HASP required wetting of newly dumped soil to suppress dust [1026]*1026and covering of contaminated soil with layers of uncontaminated soil.
The Manual, ¶ 0407.C, required the Navy to approve the HASP before work was to begin. Nevertheless, there is no evidence that either of the Navy’s CIHs ever approved the HASP for the OU-3 project. One of the Navy’s CIHs, Janet Corbett, testified that she did not review the HASP. The Navy’s other CIH, Andrew Bryson, testified that he had no record showing that he had reviewed the HASP for the OU-3 project and that he did not recall doing so.
During the summer and early fall of 1999, in the course of executing the OU-3 soil remediation plan, 240,000 cubic yards of contaminated soil from four polluted sites were transported to and disposed of in the Box Canyon Landfill, including soil from the two sites contaminated with thallium. During the project, Navy personnel monitoring the project met regularly with personnel from IT/OHM. The district court found that there were “occasions” in the course of the project when the air monitoring equipment registered dust levels in excess of the levels that were supposed to require work stoppages (called “exceedences” by the parties), but did not note that record evidence showed that such exceedences occurred more than 200 times. The district court found that action levels were set so low that there “were exceedences where there was no visible dust.” It is undisputed that the work was never stopped because of these exceedences. The district court did not find, but the undisputed record evidence shows, that the Navy’s QAO for the project, Nars Ancog, never looked at the air monitoring data collected by IT/OHM, nor did anyone else from the Navy. Residents in the nearby Wire Mountain Family Housing area testified that, at times during the remediation project in 1999, visible clouds of dust blew from the landfill. The district court found that “[a]ny visible dust was likely from uncontaminated soil.”
Myers’s house in the family housing area of Camp Pendleton was adjacent to the Box Canyon Landfill. Indeed, the fence line for and access road to the landfill were only about 50 feet from her backyard, where Myers often played. The landfill was also only about 200 feet from an elementary school, where Myers played and later attended school. Soon after the dumping of contaminated soil into the adjacent landfill, Myers became ill. She suffered, and she continues to suffer, from gastrointestinal distress, peripheral neuropathy (a kind of nerve damage), cognitive deficits, and alopecia (loss of body hair), all of which are known side-effects of exposure to thallium.
Some analyses of Myers’s urine in March 2000 indicated concentrations of thallium well in excess of (as much as ten times) those expected in the urine of persons who had not been exposed to thallium. Subsequent tests purportedly showed no concentrations of thallium in excess of those expected in a non-exposed person. The raw data for those later tests was destroyed after this litigation commenced, however, so that there is no way to determine the reliability of those tests. The district court described the evidence about whether Myers’s urine samples revealed normal thallium levels for humans as “mixed.” District Court Decision at 8. The district court found that, even if one assumed that exposure to thallium caused Myers’s medical condition, it was not obvious how or when she was exposed to the toxic metal.
The Navy or its contractors collected post-project samples in 2000, consisting of over 100 site soil samples, 30 wipe or “swipe” (dust) samples, ten bulk samples, and even hair samples from Myers’s dog. Consistent with the district court’s find[1027]*1027ings, the Navy asserts that only one of these samples, taken from the elementary school, showed thallium, but the level shown was actually below the naturally-occurring background level. In contrast, Myers asserts that these samples were taken months after the end of the OU-3 project and after an intervening winter of wind and rain. She also contends that, contrary to the Navy’s assertion, while the final report of sampling showed results of less than one microgram of thallium in some of the samples (identified as a “non-detect”), the raw data actually indicated results that exceeded one microgram, the detection limit used for the tests, so that they should have been construed as “positives.” Myers also contends that her expert’s review of the raw data and actual test results indicated that eight wipe or “swipe” samples from air conditioning ducts in Myers’s house and other adjacent houses were positive for thallium.
B. Procedural Background
On July 10, 2002, Myers’s guardian ad litem filed suit pursuant to the Federal Tort Claims Act (FTCA), 28 U.S.C. § 2671 et seq., against the United States, as the appropriate party in a FTCA case for claims against the Navy.3 Myers’s claims were for negligence, nuisance, trespass, strict liability for ultra-hazardous activity, and battery. The trial judge trifurcated the bench trial into “breach of duty,” “actual and proximate causation,” and “damages” phases.
During the first phase of the trial, Myers presented evidence in support of her allegations that the Navy breached its duty of care in the following ways: (1) the Navy violated its own established policies set forth in the Manual by failing to have a CIH review the HASP for the Camp Pendleton project before the work of disposing of contaminated soils in the landfill began; and (2) the Navy’s designated QAO acted negligently by failing to oversee the air (dust) monitoring plan carried out by IT/ OHM. Post-trial briefs were submitted in May 2006. The case then languished for the next three years.
When the district court finally entered its Decision with Findings of Fact and Conclusions of Law (Decision) on the first phase of the bench trial, it offered no explanation for the long delay. The Decision did, however, state that the court found “that the Government did act reasonably,” leading the court to find in favor of the Navy and against the plaintiff. In reaching this conclusion, however, the district court relied, in part, on the determination of a “causation” issue — whether Myers was exposed to thallium from the OU-3 project, a question on which Myers had not been fully heard — in what was supposed to be the “breach of duty” phase of the trial. The Decision also stated that, “[mjoreover, Defendant’s actions fall within the discretionary function exception.” This ruling was a reversal, without explanation, of the district court’s denial of the Navy’s motion to dismiss, which had raised the “discretionary function” exception as a bar to Myers’s suit. The district court’s Decision on phase one of the bench trial obviated the need for phases two and three.
This appeal followed.
II. LEGAL ANALYSIS
This appeal presents two primary issues: (1) Did the district court commit reversible error in finding that Myers’s claims were barred by the “discretionary function” ex[1028]*1028ception? and (2) Did the district court commit reversible error in finding that the United States acted “reasonably” in fulfilling its duty to ensure that IT/OHM used proper safety precautions during the soil remediation project?
A. Applicability Of The “Discretionary Function” Exception
1. Standard of review
We review de novo the dismissal of a FTCA suit for lack of subject matter jurisdiction under the “discretionary function” exception. See Terbush v. United States, 516 F.3d 1125, 1128 (9th Cir.2008). We review determinations of underlying facts for clear error. Autery v. United States, 424 F.3d 944, 956 (9th Cir.2005).
2. The “discretionary function” exception
The FTCA waives the government’s sovereign immunity for tort claims arising out of the negligent conduct of government employees acting within the scope of their employment. Among the exceptions to that waiver is the “discretionary function exception,” which provides immunity from suit for any claim “based upon the exercise or performance or the failure to exercise or perform a discretionary function or duty on the part of a federal agency or an employee of the Government, whether or not the discretion involved be abused.” 28 U.S.C. § 2680(a). The “discretionary function” exception insulates certain governmental decisionmaking from “‘judicial “second guessing” of legislative and administrative decisions grounded in social, economic, and political policy through the medium of an action in tort.’ ” Terbush, 516 F.3d at 1129 (quoting United States v. S.A. Empresa de Viacao Aerea Rio Grandense (Varig Airlines), 467 U.S. 797, 814, 104 S.Ct. 2755, 81 L.Ed.2d 660 (1984)). “The government bears the burden of proving that the discretionary function exception applies.” GATX/Airlog Co. v. United States, 286 F.3d 1168, 1174 (9th Cir.2002); see also Whisnant v. United States, 400 F.3d 1177, 1181 (9th Cir.2005) (same).
The two-prong test to determine the applicability of the exception requires the court to determine (1) whether challenged actions involve an element of judgment or choice; and (2) if a specific course of action is not specified, whether the discretion left to the government is of the kind that the discretionary function exception was designed to shield, namely, actions and decisions based on considerations of public policy. Terbush, 516 F.3d at 1129 (citing Berkovitz v. United States, 486 U.S. 531, 536-37, 108 S.Ct. 1954, 100 L.Ed.2d 531 (1988)). The first-prong inquiry “looks at the ‘nature of the conduct, rather than the status of the actor’ and the discretionary element is not met where ‘a federal statute, regulation, or policy specifically prescribes a course of action for an employee to follow.’ ” Id. (quoting Berkovitz, 486 U.S. at 536, 108 S.Ct. 1954). “If there is such a statute or policy directing mandatory and specific action, the inquiry comes to an end because there can be no element of discretion when an employee ‘has no rightful option but to adhere to the directive.’ ” Id. On the other hand, if there is no statute or policy directing mandatory and specific action, the court must continue to the second prong of the analysis. Id. The second prong requires the court to determine whether the discretion left to the government is the kind of discretion protected by “public policy,” which is “understood to include decisions grounded in social, economic, or political policy.” Id. (internal quotation marks omitted). “Even if the decision is an abuse of the discretion granted, the exception will apply.” Id.
[1029]*1029
3. Analysis
a. Direction of “mandatory and specific” action
Myers argues that there are two sources of mandatory and specific action in this case: The Manual provision requiring the Navy to review HASPs and the FFA provision requiring the project QAO to ensure that all work is performed in accordance with approved work plans, sampling plans, and QAPPS. The Navy argues that neither source imposed sufficiently specific requirements to divest the Navy of its discretion.
i. Manual provisions. On appeal, Myers asserts that the Manual required that the HASP be reviewed by the Navy’s CIH or similar competent person, not just by an employee of the contractor. The Navy argues that the provision of the Manual on which Myers relies does not specify how any review of the HASP was to be conducted and was not specific enough to remove discretion. The Navy also contends that the HASP was reviewed by the contractor’s CIH. The district court made no findings on this issue.
The relevant provision in the Manual, ¶ 0407.C, uses the unambiguously mandatory “shall” in stating the requirement for review of HASPs “by a competent person.” A competent person is defined as a CIH or “equivalent by training and/or experience.” The preceding provision in the Manual specifies that “[e]ach NAVFACENGCOM activity shall ensure that plans are reviewed and accepted prior to issuing the Notice to Proceed.” Manual, ¶ 0407.b. Because the NAVFACENGCOM is the Naval Facilities Engineering Command, and this provision is also cast in the unambiguously mandatory terms “shall ensure,” this provision imposed upon the Navy itself a “mandatory and specific” duty to ensure that plans were reviewed and accepted. See Terbush, 516 F.3d at 1129. In short, read in conjunction, paragraphs 0407.b and 0407.C of the Manual required review of HASPs by the Navy’s “competent person.” No meaningful review — and certainly no meaningful review by the Navy — would be accomplished by having a contractor’s CIH review the contractor’s own HASP, particularly if the contractor’s CIH is also the author of the contractor’s HASP, as is the case here.
This “federal ... policy specifically prescribes a course of action for an employee to follow,” review by the Navy of a contractor’s HASP by a competent person, such that “the employee has no rightful option but to adhere to the directive.” Berkovitz, 486 U.S. at 536, 108 S.Ct. 1954. Thus, the Manual “specifically prescribes a course of conduct,” leaving nothing to the Navy’s discretion. See Kelly v. United States, 241 F.3d 755, 761 (9th Cir.2001). The provision is not so uncertain in its definition of the requisite training of the person conducting the review that it is “discretionary.” Rather, the provision expressly requires review by “a certified industrial hygienist [(CIH)] or equivalent by training and/or experience.” Manual, ¶ 0407.C. (emphasis added). This specification of “equivalence” to a CIH “by training and/or experience” is sufficiently specific to define the requisite training of the person conducting the review. Although some professional judgment might be involved in deciding whether or not a particular person actually is the “equivalent” of a CIH by training and/or experience, that professional judgment is not the same as “discretion.”
Moreover, in Bolt v. United States, 509 F.3d 1028 (9th Cir.2007), this court found that a comparable policy provision was sufficiently mandatory and specific to make the “discretionary function” exception inapplicable. In Bolt, this court held that the Army’s Snow Removal Policy was sufficiently “specific and mandatory” to [1030]*1030avoid application of the “discretionary function” exception, where it required that snow be removed from family housing parking areas “once per year in late February or March.” Bolt, 509 F.3d at 1032-33. This court found that this provision “expressly impose[d] a specific and mandatory duty to clear Family Housing Parking Areas of snow and ice once a year, before the end of March,” so that the Army had failed its burden under the first prong of the “discretionary function” analysis. Id at 1033. This was so, even though the policy did not specify how the snow was to be removed or the training or qualifications of the person to perform the snow removal. See also Vickers v. United States, 228 F.3d 944 (9th Cir.2000). Even supposing that the Navy had some discretion in the fulfillment of its duty to review HASPs, it had no discretion under the policy expressed in the Manual about whether or not to review the HASP at all and no discretion for such a review to be performed by anyone other than a Navy CIH or other competent person.
Therefore, upon de novo review, we hold that the district court erred in determining that the Navy had met its burden on the first prong of the “discretionary function” analysis, because the Manual did impose “mandatory and specific” requirements for review by the Navy of the contractor’s HASP. Terbush, 516 F.3d at 1129.4 Because the Manual directed mandatory and specific action, “the inquiry comes to an end because there can be no element of discretion when an employee ‘has no rightful option but to adhere to the directive,’ ” and the “discretionary function” exception is inapplicable. Id (quoting Berkovitz, 486 U.S. at 536, 108 S.Ct. 1954).5
ii. FFA provisions. Myers also argues that the FFA included mandatory and specific provisions regarding the Navy’s obligation to ensure that safety procedures were followed. Specifically, she relies on the provision of the FFA requiring the Navy to designate a QAO to oversee field work and to ensure compliance with work plans and sampling plans. FFA, ¶ 20.1. The Navy contends that this provision did not specifically describe how the QAO was to ensure compliance with work or safety plans, nor did it remove discretion to delegate certain functions to the contractor. The district court did address this dispute, at least in its 2004 ruling on the Navy’s motion to dismiss: The district court held that the FFA did not create any mandatory duty, because it did not specify how the Navy would carry out its duty to supervise. In its ruling on the first phase of the bench trial, the district court did not specifically address the FFA provisions at issue. The district court did conclude, however, that “the evidence showed that Defendant made policy-based decisions regarding discretionary questions of whether to do the remediation work at all, and whether to do the work itself — or select a contractor.”
The FFA provision at issue, like the Manual provision discussed above, uses mandatory language: The QAO “will era[1031]*1031sure that all work is performed in accordance with approved work plans, sampling plans and QAPPS” and “shall maintain for inspection a log of quality assurance field activities and provide a copy to the Parties upon request.” FFA, ¶ 20.1 (emphasis added). Again, these provisions “specifically prescribe[ ] a course of action for an employee [the QAO] to follow,” such that “the employee [the QAO] has no rightful option but to adhere to the directive.” Berkovitz, 486 U.S. at 586, 108 S.Ct. 1954. Unlike the provisions of the Manual discussed above, however, the first provision, at least, does not “indicate what the [Navy] must do to comply” with the QAO’s duty to “ensure” that all work is performed in accordance with work and sampling plans. Thus, it does leave that aspect to the government’s discretion. The second provision, which undisputably requires that the QAO “maintain” a log, also is not sufficiently specific as to what must be logged as a “quality assurance field activity.”
We affirm the district court’s conclusion that the cited provisions of the FFA are “discretionary.” Thus, at least as to the FFA provisions upon which Myers relies, the court must proceed to the second prong of the “discretionary function” inquiry. Terbush, 516 F.3d at 1129.
b. Decisions based on public policy
Even if neither the Manual provisions or the FFA provisions on which Myers relies mandates a specific course of action, that is not the end of the “discretionary function” analysis. Instead, the court must then consider, in the second prong of the “discretionary function” inquiry, whether the judgment left to the agency is of the kind that the “discretionary function” exception was designed to shield, that is, governmental actions and decisions based on considerations of public policy. Terbush, 516 F.3d at 1129.
The district court held that “the evidence showed that Defendant made policy-based decisions regarding discretionary questions of whether to do the remediation work at all, and whether to do the work itself — or select a contractor.” The district court apparently also concluded that the Navy made policy-based decisions about requirements for experience and training of key employees of the contractor, the frequency and degree of oversight by Navy employees, and the involvement of other federal agencies.
Myers argues that, once the Navy undertook responsibility for the safety of the project, the execution of that responsibility was not subject to the “discretionary function” exception, because execution of safety standards is not susceptible to a policy analysis. The Navy argues that the government’s discretionary oversight of a contractor, even of the contractor’s compliance with safety standards, is immune from tort suit by virtue of the “discretionary function” exception and that the Navy never took on the responsibility of approving every step of every action of its remediation contractor, but instead relied on its contractor’s expertise.
This court must determine whether the work of remediation of contaminated soil would involve protected policy judgments. Terbush, 516 F.3d at 1133. Specifically, “[t]he focus of our inquiry is ‘on the nature of the actions taken and on whether they are susceptible to policy analysis.’” Id. (quoting United States v. Gaubert, 499 U.S. 315, 325, 111 S.Ct. 1267, 113 L.Ed.2d 335 (1991)). “[I]t is therefore ‘insufficient for the government to show merely that some choice was involved in the decision-making process. The balancing of policy considerations is a necessary prerequisite.’” Bolt, 509 F.3d at 1033 (quoting ARA Leisure Servs. v. United States, 831 F.2d 193, 194 (9th Cir.1987), [1032]*1032with alterations and internal quotation marks omitted by the Bolt court). As explained in Terbush, “[t]he decision to adopt safety precautions may be based in policy considerations, but the implementation of those precautions is not....” Terbush, 516 F.3d at 1133 (internal quotation marks omitted). There is a recognized exception: “The implementation of a government policy is shielded where the implementation itself implicates policy concerns, such as where government officials must consider competing firefighter safety and public safety considerations in deciding how to fight a forest fire.” Id. (internal quotation marks omitted). This case falls within the general rule, not the exception.
The district court’s conclusion that the Navy’s decision about whether or not to pursue the remediation project at all was a discretionary one informed by public policy considerations misses the point, even if it were correct, where Camp Pendleton was on the EPA’s “National Priorities List” of sites requiring environmental cleanup. Because “[t]he focus of our inquiry is on the nature of the actions taken and on whether they are susceptible to policy analysis,” id. at 1133 (internal quotation marks omitted), we look at the nature of the actions in conducting the remediation project, not the decision to undertake the remediation project.
With the focus properly on the conduct of the remediation project, it is well to remember that “matters of scientific and professional judgment — particularly judgments concerning safety — are rarely considered to be susceptible to social, economic, or political policy.” Whisnant, 400 F.3d at 1181; Bear Medicine, 241 F.3d at 1214. More specifically, this court has held that “implementation” of a course of action is not a discretionary function. Id. Thus, while the Navy contends that its determinations about how much safety oversight was required were susceptible to policy considerations, those determinations properly fell within the scope of professional judgments about implementation of the safety plan that were not susceptible to public policy considerations.
This case is similar in all important respects to Bear Medicine. What is at issue here, as in Bear Medicine, is not just a general statutory obligation to promote safety, but “a failure to effectuate policy choices already made” that are not protected under the discretionary function exception. See Bear Medicine, 241 F.3d at 1215 (quoting Camozzi v. Roland/Miller and Hope Consulting Group, 866 F.2d 287, 290 (9th Cir.1989)). Even if the Navy did have discretion in its monitoring of IT/ OHM’s actions, the Navy’s actions in carrying out its responsibilities were not protected policy judgments. Id. In other words, “once the[Navy] ha[d] undertaken responsibility for the safety of [the OU-3] project, the execution of that responsibility [wa]s not subject to the discretionary function exception,” and “[t]he decision to adopt safety precautions may be based in policy considerations, but the implementation of those precautions is not.” Id. Like the government’s argument in Bear Medicine, the Navy’s argument here “ ‘would essentially allow the Government to “administratively immunize itself from tort liability under applicable state law as a matter of ‘policy.’ ” ’ ” Id. (quoting McGarry v. United States, 549 F.2d 587, 591 (9th Cir.1976)).
The Navy’s attempts to distinguish Bear Medicine are unavailing. First, the Navy asserts that Bear Medicine is distinguishable, because the agency in that case retained for itself the responsibility to regularly inspect the work (the logging) to ensure adherence to basic safety practices, but the Navy did not do so here. This assertion is simply wrong, and any such finding is clearly erroneous, because it is [1033]*1033without “support in inferences that may be drawn from the facts in the record.” Hinkson, 585 F.3d at 1262 (citations and internal quotation marks omitted). The Manual retained the Navy’s responsibility to review HASPs, and the FFA required the Navy’s QAO to ensure that all work was performed in accordance with approved work and sampling plans and to maintain for inspection a log of quality assurance field activities. Thus, in this case, as in Bear Medicine, the Navy was required to ensure that the contractor complied with the safety provisions of the contract. Bear Medicine, 241 F.3d at 1217. Like the BIA’s failure in Bear Medicine, the Navy’s failure to have IT/OHM’s HASP reviewed by the Navy’s own CIH or other competent person and the failure of the Navy’s QAO to inspect any air monitoring were not policy judgments that Congress intended to protect from FTCA liability. Id.
The Navy also contends that Bear Medicine is distinguishable, because here, unlike the BIA in that case, the Navy presented evidence that policy factors influenced its conduct, including the efficient allocation of agency resources and the need to rely on the contractor’s expertise, because the Navy was not the organization with the required expertise. That argument is also unavailing, however. This court in Bear Medicine in fact rejected a contention that “limited resources” was a policy-based excuse for failure to adhere to accepted professional standards, id. at 1216-17, and it is no better as a policy-based excuse for failure to adhere to policy manual and contractual requirements.
Thus, the Navy has also failed to carry its burden on the second prong of the “discretionary function” analysis.
4. Summary
Upon de novo review, we find that the Navy failed to establish either prong of the “discretionary function” exception. Therefore, we reverse the district court’s determinations that Myers’s FTCA claim against the Navy is barred by the “discretionary function” exception and that the court lacked subject matter jurisdiction over that claim.
B. Reasonableness Of The Navy’s Conduct
Because we hold that Myers’s FTCA claims are not barred by the “discretionary function” exception, we must also consider whether the district court erred in holding that the Navy acted “reasonably.” Again, we reverse.
We review the district court’s decision as to the reasonableness of the Navy’s conduct for clear error. Hinkson, 585 F.3d at 1262.6 “The government can be [1034]*1034sued ‘under circumstances where the United States, if a private person, would be liable to the claimant in accordance with the law of the place where the act or omission occurred.’ ” Terbush, 516 F.3d at 1128-29 (quoting 28 U.S.C. § 1346(b)(1)). California law is applicable to Myers’s claim against the Navy, as California is “the law of the place where the act or omission occurred.” 28 U.S.C. § 1346(b)(1).
In FTCA cases, we have recognized that California’s “peculiar risk” doctrine, described in Restatement (Second) of Torts §§ 413 and 416, is an exception to the general rule that a principal is not liable for torts committed by an independent contractor. Yanez v. United States, 63 F.3d 870, 872 (9th Cir.1995). “[Section 416 liability has been construed as creating direct liability for the government’s nondelegable duty to ensure that the contractor employs proper safety procedures.” Id. at 873 n. 1. Thus, “under California’s nondelegable duty doctrine, the United States is directly liable for its own negligence when it fails to ensure that an independent contractor takes adequate safety precautions and the work to be performed involves special dangers.” Gardner v. United States, 780 F.2d 835, 838 (9th Cir.1986); McGarry v. United States, 549 F.2d 587, 590 (9th Cir.1976), cert. denied, 434 U.S. 922, 98 S.Ct. 398, 54 L.Ed.2d 279 (1977); Thorne v. United States, 479 F.2d 804, 808-09 (9th Cir.1973). This nondelegable duty is one of “reasonable care.” Restatement (Second) of Torts §§ 413 and 416.7
The district court stated, “Under California law, the foreseeability of harm and thus the necessary safety precautions to prevent the foreseeable harm is key to understanding Defendant’s duty.” The district court cited no authority for this “key” proposition, however. The district court found that the risk that Myers or anyone else would be exposed to thallium from the landfill project was not foreseeable.
The California Supreme Court has defined “peculiar risk” as a risk “that is peculiar to the work to be done, arising either from the nature or the location of the work and against which a reasonable person would recognize the necessity of taking special precautions.” Privette v. Superior Court, 5 Cal.4th 689, 695, 21 Cal.Rptr.2d 72, 854 P.2d 721 (1993) (internal quotation marks omitted). Indeed, “[i]t is the foreseeability of that special risk which justifies the imposition of liability.” Holman v. State of California, 53 Cal.App.3d 317, 124 Cal.Rptr. 773, 781 (1975).
The district court was correct that “foreseeability” figures in both the determination of the alleged tortfeasor’s duty and whether the tortfeasor breached that duty, that is, acted unreasonably. California courts have explained the distinction between “foreseeability” in the “duty” context and “foreseeability” in the “breach” context, as follows: “[W]hile foreseeability with respect to duty is determined by focusing on the general character of the event and inquiring whether such event is likely enough in the setting of modern life that a reasonably thoughtful [person] would take account of it in guiding practical conduct, foreseeability in evaluating negligence and causation requires a more focused, fact-specific inquiry that takes [1035]*1035into account a particular plaintiffs injuries and the particular defendant’s conduct.” Laabs v. S. Cal. Edison Co., 175 Cal.App.4th 1260, 97 Cal.Rptr.3d 241, 251-52 (2009) (internal citations omitted).
Here, prior to the bench trial, in response to Myers’s motion in limine, the Navy had already conceded that the OU-3 project involved “peculiar risk.” Thus, “foreseeability” in the context of a determination of “duty” was no longer an issue.8 On the other hand, because the first phase of the bench trial was to determine whether the Navy breached its duty, it was appropriate for the district court to consider the “fact-specific” aspect of “foreseeability,” which “takes into account a particular plaintiffs injuries and the particular defendant’s conduct.” Id.
The district court’s clear error here, in the context of “breach of duty,” was in relying on what it perceived to be a lack of proof of causation as determinative of the foreseeability of Myers’s injuries. The district court determined that it was not foreseeable that Myers would be exposed to thallium from the OU-3 project. This conclusion was based on its finding that there was no showing that thallium somehow migrated to Myers’s yard or school, because tests purportedly showed an “absence” of thallium in her home, her school, and her parents’ bodies. The district court also relied on its finding that Myers had not shown that any thallium that harmed her came from the Navy’s Box Canyon operations, because experts purportedly doubted that thallium could be transported in fugitive dust, and there had been only minimal levels of thallium in the contaminated soil to start with. The proper question in the “foreseeability” inquiry for purposes of determining whether the Navy breached its duty, however, was not whether Myers was exposed to thallium from the OU-3 project — a “causation” question that was properly reserved for a later phase of the trial, and on which Myers had not been fully heard at the time that the district court made its findings— but whether it was foreseeable that a person exposed to thallium would suffer the kinds of injury that Myers suffered. Id. (“foreseeability” in the context of “breach of duty” “takes into account a particular plaintiffs injuries”). Thus, the district court clearly erred by applying the wrong legal standard in its determination of “foreseeability.” Hinkson, 585 F.3d at 1262.9
The district court also found that it was not foreseeable that thallium would become windborne in fugitive dust or that dangerous levels of thallium would be present in migrating dust, because of the low concentration of thallium in the contaminated soil in the first place, so that there was “no [foreseeable] danger to be [1036]*1036prevented.” Even if this “foreseeability” finding involved application of the correct legal standard, it was implausible and without support in inferences that may be drawn from the facts in the record, in light of evidence of the extreme toxicity of thallium and the provisions of the FFA and ROD establishing safety precautions, including dust suppression measures, to prevent exposure to “nearby receptors.” See ROD ¶2.5.8.1.
The Navy’s argument that this provision of the ROD was not applicable to the 1999 soil removal and dumping project, but only applicable to the later “capping” project for the landfill in 2000, is illogical and implausible. It would be illogical to be concerned with dust suppression only during a capping project, but not while contaminated soil was actually being dumped into the landfill. The provision of the ROD that the Navy contends was applicable to the 1999 soil removal and dumping project, ROD ¶2.4.6.1, which mentions only “potential risk for workers,” but not for “nearby receptors,” expressly applied only to removal of contaminated soil from the-five contaminated sites in OU-3, not to dumping of the contaminated soil in the landfill. See ROD¶ 2.4.5.
The district court relied on its findings that the Navy took care to prevent dangerous migration of thallium by having in place adequate safety measures and ensuring that the contractor was actually taking those safety measures, including regular safety meetings and site visits, selecting an experienced remediation contractor, requiring highly qualified people at important positions, and allowing other agencies to participate in the oversight and design of safety precautions. These findings of lack of foreseeability of harm and reasonableness of the Navy’s conduct, however, are clearly erroneous, in light of evidence of glaring omissions in the Navy’s safety oversight for the OU-3 project, which the district court simply ignored.
First, the district court could not have found on this record that the Navy complied with the requirements in the Manual to have the HASP for the OU-3 project reviewed by a Navy CIH or equivalent person. As this court held above, the Manual required review by a Navy CIH or equivalent person, not just by the contractor’s CIH, and there is no evidence that either of the Navy’s CIHs or any other equivalent person from the Navy ever reviewed the HASP. Indeed, Mr. Bryson, one of the Navy’s CIHs, testified that, if the HASP had been reviewed by the Navy, a record of that review would exist on the computer database that Ms. Corbett maintained, but there is no such record. Certainly, the Navy has not asserted that there is any evidence demonstrating that Navy personnel reviewed the HASP. Evidence that the personnel in question did not recall reviewing the HASP and the lack of evidence that would ordinarily indicate that the Navy had reviewed the HASP lead to no logical or reasonable inference that Navy personnel did, in fact, review the HASP. Rather, the only reasonable inference from such evidence and the lack of any other evidence that one or both of the Navy’s CIHs reviewed the HASP is that the HASP was never reviewed. Violation of the mandatory duty to review the HASP is plainly a breach of the duty to exercise reasonable care to ensure that the contractor took reasonable care to follow required safety precautions. Restatement (Second) of Torts § 416.
Second, there is no evidence that the Navy’s QAO ever took any steps to ensure that air monitoring samples were reviewed or that work was stopped if dust thresholds were exceeded, as required by the provision of the FFA stating that the QAO must “ensure that all work is performed in [1037]*1037accordance with approved work plans, sampling plans and QAPPS [Quality Assurance Project Plans].” FFA, ¶20.1. Such a failing was unreasonable, even if the Navy could delegate air monitoring and responses to excessive dust levels to the contractor. In fact, the record evidence suggests that the contractor took advantage of the lack of oversight, because the contractor never stopped work, even when its own dust detection levels were exceeded. Furthermore, Myers has pointed to evidence that the contractor’s employee responsible for air monitoring was instructed not to stop work when “exceedences” occurred. Again, violation of the mandatory duty to ensure adherence to the safety plans is plainly a breach of the duty to exercise reasonable care to ensure that the contractor took reasonable care to follow required safety precautions. Restatement (Second) of Torts § 416.
The district court’s finding that the Navy acted “reasonably” was clearly erroneous and is reversed. This matter must be remanded for phases two and three of the bench trial, “actual and proximate causation” and “damages,” respectively.
C. Reassignment On Remand
In the event that she obtains a reversal, Myers asserts that this matter should be remanded for further proceedings before a different judge, pursuant to 28 U.S.C. § 2106. The Navy did not address this issue in its briefing. We decline the invitation to reassign the case to a different judge on remand.
The factors that we consider to determine whether reassignment is appropriate are the following:
(1) whether the original judge would reasonably be expected upon remand to have substantial difficulty in putting out of his or her mind previously [] expressed views or findings determined to be erroneous based on evidence that must be rejected, (2) whether reassignment is advisable to preserve the appearance of justice, and (3) whether reassignment would entail waste and duplication out of proportion to any gain in preserving the appearance of fairness.
Mendez v. Cnty. of San Bernardino, 540 F.3d 1109, 1133 (9th Cir.2008) (quoting United States v. Sears, Roebuck & Co. Inc., 785 F.2d 777, 779 (9th Cir.1986)). We may direct reassignment of the case, even if we do not question the impartiality of the judge, in light of unusual factors indicating that a reassignment is advisable to preserve the appearance of justice. Living Designs, Inc. v. E.I. Dupont de Nemours & Co., 431 F.3d 353, 373 (9th Cir.2005)
Here, there is no real reason to question the impartiality of the trial judge, notwithstanding what we have found were numerous errors in his disposition of the case. Nor are we convinced that the errors we find are such that the trial judge would have substantial difficulty in putting out of his mind his previous findings. We are troubled by the undue — and unexplained— three-year delay between the conclusion of the first phase of the bench trial and the issuance of an opinion. We are not convinced, however, that this is of such gravity as to warrant reassignment of the case to preserve the appearance of justice.
We conclude that it is not necessary to direct that this case be reassigned to another judge upon remand. Nevertheless, we trust that proceedings on remand will proceed expeditiously.
Ill CONCLUSION
Because the district court erred in holding that the “discretionary function” exception barred the Navy’s liability on and the court’s subject matter jurisdiction over Myers’s claim, and clearly erred in finding that the Navy acted “reasonably” and not [1038]*1038in breach of its duty in conducting the remediation of contaminated soil in the project at issue here, we reverse and remand this action for further proceedings. Nevertheless, we find it unnecessary to reassign the case to a different judge on remand.
REVERSED AND REMANDED.