concurring in part and dissenting in part:
I respectfully disagree with the majority’s resolution of five of the issues addressed in its opinion. First, § 504 of the Rehabilitation Act of 1973 does not require federal-fund grantees and their wards to live with even slight odds of contracting a painful and fatal disease. Second, prisoners’ rights under the Rehabilitation Act, like their constitutional rights, are properly limited by legitimate penological concerns such as security. Third, the district court correctly concluded that providing the additional security necessary to integrate the plaintiffs in all the programs at issue here would be an undue burden on the Department of Corrections. Fourth, the district court should not be mandated to consider programs requiring residential integration when — as the majority found — the plaintiffs have waived the argument that the district court should have addressed their rights to residential integration. I would accordingly find no error in the district court’s treatment of these issues. Finally, this action does not merit the extreme measure of reassignment to another district judge. I agree, however, with the majority’s holding that the district court’s exclusion of evidence of prisoner classification as a reasonable accommodation requires remand..1 Thus, I would vacate the judgment and remand solely for the district court to reconsider whether inmate-specific risk evaluation is a reasonable accommodation in light of the excluded evidence concerning inmate classification.
I. Background
This litigation began ten years ago in reaction to a legislatively mandated AIDS-protection program in Alabama’s prison system. Under statute,2 the Alabama prison system tests all entering inmates for infection with the Human Immunodeficiency Virus, which causes AIDS.3 Those inmates testing positive for the virus are segregated from the general inmate population in “HIV+” units, one for men at the Limestone Correctional Facility and one for women at the Julia Tutwiler Prison for Women.4 So segregated, the HIV-positive inmates are unable to participate in many programs and activities with the HIV-negative, general population.5
The plaintiff class challenged this practice as a violation of several constitutional rights and of § 504 of the Rehabilitation Act of 1973.6 The district court denied relief after a bench trial. The court concluded that no constitutional rights were violated. It further concluded that the plaintiffs were not “otherwise qualified,” as required for rights to arise under § 504, to participate in integrated programs because their participation would pose a significant HIV-transmission risk. This court affirmed judgment against the plaintiffs on the constitutional claims. This court held, however, that § 504 requires a program-by-program analysis to determine if the plaintiffs merit relief.7 The action was remanded to the district court for this fact-finding. This court directed the district *1344court especially to evaluate the risk of HIV transmission in each program. In doing so, the panel nonetheless acknowledged that “the court’s conclusion of the significance of the risk of HIV transmission with regard to each program [could] be unaltered.”8
The remaining factual dispute at the trial that followed remand (and on this appeal) is narrow. Several key facts are undisputed. In the current state of medical knowledge and art, HIV infection inevitably progresses to AIDS. AIDS always leads to death, often after lengthy suffering. There is likewise no dispute that HIV is transmitted through body fluids, and that transmission by blood is much more probable than transmission by any other means. The primary disputed issue is the likelihood of blood-to-blood contact between HIV-positive and HIV-negative inmates during the various programs in which the plaintiffs seek to participate.
The defendants presented evidence that transmission is possible in any integrated activity. They did this in two ways. First, the defendants showed that certain activities inherently raise the possibility of blood transmission.9 Second, the defendants adduced evidence that high-risk behavior (such as anal intercourse, sharing of drug needles,10 or tattooing11) or behavior that presents the possibility of transmission (such as violent bloodshed) may occur in virtually any activity in which HIV-positive and HIV-negative prisoners mingle without supervision.12 This evidence included testimony about and numerous incident reports of high-risk behavior in a variety of settings, from bathrooms to the kitchen to the library.13 Defense witnesses provided anecdotal evidence that even prisoners with strong incentives to behave have nonetheless run risks.14 The defense case also contained evidence that HIV-negative prisoners will not easily tolerate integration. This evidence included a 1988 study of Alabama prisoners concluding that HIV-negative inmates may react violently to integration15 and corroborating anecdotal evidence.16 Finally, the defendants presented circumstantial evidence that Alabama’s total segregation policy, one of only two in the nation, is superior to integrated programs in slowing HIV’s spread. The all-time seroconversion rate per thousand prisoners in Alabama’s prisons is .00006%, as compared to annual rates of .19, .33, and .41% in Nevada, Illinois, and Maryland, respectively, all states with integrated populations.17
The. plaintiffs naturally took a different tack.. While never categorically denying the possibility of transmission if a bloody fight breaks out, tattoo needles are shared, sports injuries occur, or female inmates have sex with each other, they marshaled evidence of the rarity or absence of any such incidents of transmission.18 They further presented evidence that there are no reported or wit*1345nessed incidents of anal sex, needle-sharing, or other high-risk behavior during any of the activities in which they wish to participate.19 The plaintiffs also offered evidence that the degree of surveillance in most circumstances makes such behavior implausible. Moreover, the plaintiffs’ anecdotal evidence is that HIV-positive and HIV-negative prisoners in both Limestone and Tutwiler have mingled peacefully and without HIV transmission in “welfare committee” meetings,20 fire drills,21 “Communicators’ Meetings,”22 and literacy training classes23.
Based on this record, the parties’ approach to measuring the odds of transmission is predictable. On one hand, the defendants contend that transmission could happen whenever prisoners mix because of the perpetual potential for high-risk activity or for violence — and the HIV-negative prisoners’ anti-HIV sentiments boost the odds of confrontation and bloodshed. HIV-positive participation in any of the programs, according to the defendants, thus poses a significant risk of transmission. On the other hand, the plaintiffs argue that none of the defendants’ chthonic scenarios has ever been realized in prison history, and that the defendants’ fear of clashes between HIV-positive and HIV-negative inmates is unwarranted in today’s enlightened prison society.
The district court generally took the defendants’ view, concluding that the risk of transmission in every program was significant. While the court’s reasoning was not wholly uniform for every program, the court generally reached this result by measuring the odds based on two background findings, rather than extrapolating past incidents of transmission. The first finding is that sex, intravenous drug use, and bloodshed are a perpetual possibility in prison whenever a security guard trained to stop it is not watching, and that prison life is inherently unpredictable24 — especially when large-scale mixing of HIV-positive and HIV-negative prisoners in Alabama is untried. The second finding is that HIV is transmitted by sex, intravenous drug use, and blood-to-blood contact. If these activities can spread HIV, and these activities can occur between HIV-positive and HIV-negative inmates, the court reasoned, then HIV transmission is more than a theoretical possibility, even if we have no examples.25
From this conclusion, the court easily found this risk to be significant. After all, each case of transmission, however rare, claims at least one life. More lives could follow if the infected general-population inmate spreads the virus to his dormmates. To support this inference of harm, the court pointed to an investigation of a 1991 syphilis outbreak at Limestone in which medical officials located 86 prisoners suspected of sexual contact with a single infected inmate.26 Given this degree of harm, even slim odds of transmission make the risk significant. As the court put it in words echoed throughout its 475-page opinion, “elimination of high risk behavior is impossible.... Because the Defendant/Prison system has decided that such conduct is likely, and because of the catastrophic severity of the consequences if such conduct does occur, this Court holds that integrating the [program under discussion] would present a significant risk of transmitting the deadly HIV virus. Accordingly, the HIV + inmates are not ‘otherwise quali*1346fied.’ ”27
The court took this otherwise-qualified analysis one step further. As part of its evaluation of the plaintiffs’ qualifications to participate in the programs, it weighed the Department of Corrections’s penological concerns, including the danger of violence that might arise from inmate prejudice toward and fear of HIV-positive prisoners.28 The court looked to Turner v. Safley, which permits infringement of prisoners’ First Amendment rights provided that the prison regulation “is reasonably related to a legitimate penological interest,” 29 as a guide to evaluating the Department of Corrections’s interests. In almost every program, the court concluded that the Department of Corrections could legitimately seek to prevent violence and epidemic HIV by the measures it has taken.
The court continued its analysis, as directed by this court,30 by asking if reasonable accommodations would make the plaintiffs qualified. Although the court found that in many programs the plaintiffs had already been reasonably accommodated by provision of identical, but segregated programs,31 the court found as to other programs that the only effective accommodation would be additional guards to prevent high-risk behavior. The court concluded, however, that hiring the dozens of guards necessary to integrate all the programs safely would place an undue financial burden on the Department of Corrections.
The plaintiffs have attacked every step of the analysis. On the “otherwise qualified” issue, they contend that the district court weighed too heavily the gravity of the harm that could result if these small risks are realized, and that the district court improperly took legitimate penological objectives into account. Furthermore, they argue that the district court erroneously considered the burden on the Department of Corrections of accommodating all of the plaintiffs’ demands, rather than myopically considering whether hiring the few guards necessary to integrate each program would individually overburden the entire prison system. These arguments present issues of law and mixed issues of law and fact; both are reviewed de novo, although findings of fact stand unless clearly erroneous.32
* * *
The plaintiffs raise a final issue unrelated to the merits. They argue that Judge Varner should have recused himself from the retrial of this case. Three circumstances, the plaintiffs contend, make recusal mandatory: the judge’s alleged consultation with his son (who is a physician) concerning the facts of this case; the father-son relationship between Judge Varner’s law clerk (who did not work on this case) and defendants’ lead counsel; and finally Judge Varner’s asserted bias against the plaintiffs, as evidenced by a few of the Judge’s remarks on the record during the trial. Because none of these circumstances warrant recusal, I concur in the majority’s conclusion on this issue. I would refuse, however, to undermine that conclusion by reassigning the action based on the very same facts.
II. Discussion
I break with the majority’s analysis of the merits of the claims asserted in this action in three ways.33 These differences correspond *1347to the three different stages of the district court’s analysis', first, its determination that the plaintiffs were not otherwise qualified to participate in the relevant programs; second, the court’s inquiry into the effect of legitimate penological interests on the prisoners’ Rehabilitation Act rights; and finally its consideration of reasonable accommodations, a step demanded by the finding that the prisoners were not otherwise qualified. The majority holds that the district court applied the wrong standard on the first step, improperly added the second step, and erroneously found on the last step that the plaintiffs’ proposed reasonable accommodations, considered together rather than program by program, are unduly burdensome.
The majority’s first holding requires courts to treat all transmissible illnesses equally, regardless of their consequences; this does not comport with the law. The second holding requires courts to apply identical standards to classrooms, workplaces, and prisons, and thus does not observe both the unique importance of security in the prison environment and limited judicial competence in the penal realm. Finally, the majority’s third holding fails to focus the undue burden analysis on the unique circumstances of a case— which here means a case in which dozens of correctional officers would be needed to accommodate the plaintiffs in all programs, even if one officer would reasonably accommodate them in one.34
A. “Significant Risk,” Not “High Odds”
Section 504 of the Rehabilitation Act of 1973,35 under which the plaintiffs seek relief, requires a federal-fund grantee reasonably to accommodate a disability only if the disabled person is “otherwise qualified” for the activity.36 As the Supreme Court interpreted the statute in School Board v. Arline,37 the “basic factors” for determining if the carrier of a contagious disease is otherwise qualified include ,
[findings of] facts, based on reasonable medical judgments given the state of medical knowledge, about (a) the nature of the risk (how the disease is transmitted), (b) the duration of the risk (how long is the carrier infectious), (c) the severity of the risk (what is the potential harm to third parties) and (d) the probabilities the disease will be transmitted and will cause varying degrees of harm.38
The Court further directed the courts to “defer to the reasonable medical judgments of public health officials.”39 The point, as the Court put it, is that “[a] person who poses a significant risk of communicating an infectious disease to others in the workplace will not be otherwise qualified for his or her job . if reasonable accommodation will not eliminate that risk.”40
Contrary to the majority’s holding, this standard does not seek simply to identify a certain percentage possibility of transmission above which the plaintiff is not otherwise qualified. Arline’s selection of factors and *1348its language show that the significance of a risk is a product of the odds that transmission will occur and the severity of the consequences. First, Arline’s four factors include both “the severity of the risk” and the “the probabilities the disease will be transmitted,” not just “the odds the virus will spread.” This suggests that each must interplay with the other in the otherwise-qualified inquiry. Second, significance by itself connotes more than size: “significant” means “deserving to be considered,” “important,” “weighty,” or “notable.”41 It does not just mean “big.”42 And it is the potential gravity of the harm that imbues certain odds of an event with significance. This is indeed common sense: to borrow an analogy from the district court’s opinion, we are far more likely to consider walking a tightrope to pose a significant risk if the rope is fifty feet high than if it is one foot off the ground. This is so even if the odds of losing our balance are the same however far we have to fall.
In the context of HIV transmission, Arline ’s standard thus means that something less than high odds of transmission will make the plaintiffs not otherwise qualified. It has been uncontroverted at every stage of this case that HIV is inevitably, and painfully, fatal. These horrendous consequences render even very low odds of transmission enough to make a risk significant. The other circuits that have confronted this issue directly have indeed so interpreted Arline’s standard. The Fourth Circuit has affirmed a district court’s conclusion that an HIV-positive neurosurgery resident could be barred from surgery, notwithstanding the fact that there were no documented cases of surgeon-to-patient HIV transmittal.43 Similarly, the Fifth Circuit has held that a “small” chance of blood-to-blood contact between an HIV-positive surgical assistant and his patients nonetheless constitutes a “significant” risk: “A cognizable risk of permanent duration with lethal consequences suffice[d]” to disqualify the assistant.44
Based on Arline’s standard thus interpreted, the district court’s conclusion was correct. First, the district court finding of a cognizable possibility of transmission has record support. The court found as fact the following: that blood-to-blood contact likely transmits HIV; that violence, intravenous drug use, and sex may cause blood contact and occur in prisons in the most unlikely and unexpected places; and that it is impossible to know or surveil much of what goes on. These subsidiary findings, which are supported by expert and lay testimony, legitimately support the conclusion that mixing prisoners raises a cognizable possibility of spreading HIV. “Conjecture,” as the majority calls it, it may be, but the district court has no other option in this case. After all, Arline’s standard requires the court to measure risk. Measuring risk is predicting the future. Unless the judge is a seer, he must predict the future by drawing inferences from both current conditions and past events. It may be preferable to extrapolate past events, but in this ease that would, in effect, impose a somebody-has-to-die-first stan*1349dard — the risk is not cognizable until someone dies from the conditions posing the risk. This court should reject such a standard.
Next, the district court properly concluded that this risk was significant. Arline’s sliding-scale standard authorized the court to cleave to caution and find the risk significant in a case such as this when death is a certain result of the possible event. The district court’s conclusion also comported with Arline in another significant respect: Arline urges circumspection when a finding may contravene the judgment of public health professionals.45 Prison officials charged with protecting prisoners’ health have determined that this policy is justified, and the court properly hesitated before rejecting their judgment. The district court’s conclusion that the plaintiffs’ participation in the programs poses a significant risk to others’ health, and that the plaintiffs thus are not otherwise qualified, should be upheld.
B. Security Matters
In its otherwise-qualified inquiry, the district court also reviewed the four factors suggested in Turner v. Safley46 to determine whether the defendants’ security concerns disqualified the plaintiffs from participating in the programs.47 The trial court concluded that the likelihood of inmate resentment and fear of the HIV-positive prisoners made violenee a reasonable worry, and that the wardens were therefore justified in separating the HIV-positive prisoners from the HIV-negative. Turner is not a Rehabilitation Act case, and the propriety of using its factors in considering a § 504 plaintiffs qualifications is new to this court.
I must first dispose of a threshold matter: the prior panel’s silence on this issue is not law of the case. The law-of-the-case doctrine “comes into play only with respect to issues previously determined.”48 An appeals court may impliedly resolve an issue.49 But for that resolution to govern all subsequent proceedings, the issue must necessarily have been decided.50 The doctrine does not embrace all issues that could have been addressed,51 even if addressing them would have been appropriate.52
An implied holding is at best what we have from the first panel’s opinion, which is silent on Turner’s application. And we should not infer any holding at all on Turner from the panel’s silence. After all, the issue was not before the court. Following the first trial, the district court denied the plaintiffs relief on their Rehabilitation Act claim on the finding that the plaintiffs were not otherwise qualified because “the probability of transmission, in the prison environment, is significant” in all programs, without an individuated assessment.53 The plaintiffs challenged *1350this conclusion on one ground: that the district court could not have found the plaintiffs to be not otherwise qualified without individualized findings as to the risk of transmission in every program for every member of the class.54 Thus, because the Rehabilitation Act Turner issue was not raised in the briefs, the panel may have decided not to address an issue that was not properly presented. Many panels so decide.55 Furthermore, whether Turner’s factors may weigh into otherwise-qualified analysis was not an issue the court had to resolve before holding that the Rehabilitation Act requires a program-by-program qualifications assessment. There is, therefore, no law of the case relating to this issue.
The issue is accordingly now before this court. This court should hold, as the Ninth Circuit has,56 that Turner’s factors are relevant to Rehabilitation Act claims arising in prison. Four reasons support this conclusion. First, Turner’s concerns apply equally to statutory claims even though Turner’s holding is limited to constitutional claims. In either case, federal judges have no business sitting in the warden’s seat:
[T]he problems of prisons in America are complex and intractable, and, more to the point, they are not readily susceptible of resolution by decree____ Running a prison is an inordinately difficult undertaking that requires expertise, planning, and the commitment of resources, all of which are peculiarly within the province of the legislative and executive branches of government. Prison administration is, moreover, a task that has been committed to the responsibility of those branches, and separation of powers concerns counsel a policy of judicial restraint. Where a state penal system is involved, federal courts have, as we indicated in Martinez [Procunier v., 416 U.S. 396, 94 S.Ct. 1800, 40 L.Ed.2d 224 (1974) ], additional reason to accord deference to the appropriate prison authorities.57
These concerns are equally valid for both constitutional and statutory claims. In both cases, the judges run into the same problems of lack of experience and expertise when they begin to interfere with matters traditionally left to the legislative and executive branches. Here, moreover, Congress’s enactment of the statute does not put a legislative stamp of approval on judicial intrusion into state prison administration. Congress apparently never thought about prisons when it passed the Rehabilitation Act. The words “prison” and “prisoner” do not occur in the legislative history. Section 504 does not mention prisons. In short, “[tjhere is no indication that Congress intended the Act to apply to prison facilities irrespective of the considerations of the reasonable requirements of effective prison administration.”58
Second, federal judicial second-guessing of the legitimate penological decisions of state authorities, even when statutorily prompted, runs squarely counter to the most basic notions of comity. Prison administration is a core state function.59 “It is difficult to imagine an activity in which a State has a stronger interest, or one that is more intricately bound up with state laws, regulations, and procedures, than the administration of its prisons.”60 In this realm, federal courts should await a clear statement from Congress before construing statutes as a license to dictate state policy.61 Taking the Rehabilitation Act’s silence as to application in the prison environment to mean that its effect in *1351prison is identical to that elsewhere does not comport with this “clear statement” rule. On the other hand, viewing § 504’s “otherwise qualified” language as incorporating Turner’s factors, which carry an element of deference to prison authorities, provides a means of avoiding undue interference in state affairs without a clear congressional direction.
Third, Arline’s standard, born in an action by a teacher plaintiff, is inadequate for the prison setting. While Arline acknowledges that the Rehabilitation Act does not require the federal-fund grantee, or its wards, to run significant risks of any kind,62 its rule views those risks as solely ones to health from the disease. In a classroom setting, or in the surgical theater,63 these risks may be the only dangers that the communicable disease poses. But in prison, mixing carriers of a virus that is considered loathsome, however wrongly, with those who do not have the virus runs the risk of violence. The district court so found, and cases elsewhere show that such violence is not unknown.64 This danger — whether or not it leads to HIV transmission — warrants consideration along with health risks, and Turner’s factors offer a vehicle for taking into account hazards that are unique to the prison context.
Fourth, considering Turner’s factors in a prison Rehabilitation Act case is wholly consistent with the statute. As Arline’s holding makes clear, § 504 does not command the federal-fund grantee to endanger its wards.65 And in the employment context from which Arline sprang, Congress has embraced the principle of safety-threat limitations. This concern for safety enters into § 504 by way of § 504(d), which imports the standards under Title I of the American with Disabilities Act (ADA) into § 504 in employment cases.66 Section 103(b) of the ADA,67 as interpreted, in turn permits discrimination against those who as a result of their disability pose a “direct threat to the health or safety of other individuals in the workplace.”68 It would be anomalous to hold that although an authority may take safety of either the Rehabilitation Act plaintiff or his coworkers into account in taking adverse employment action against that plaintiff, a prison is not entitled to respect for its concern for protecting the safety of the plaintiffs here and their fellow inmates. Turner’s factors offer a means to implement these safety limitations in the prison setting.
For these reasons, the district court did not err in taking into account the defendants’ security concerns that lead to segregating the HIV-positive prisoners in the programs at issue. Nor did the district court err in its analysis of Turner’s factors.69 A key fact-finding, which was adequately based on survey evidence and the presence of an intervenor inmate class that opposes inte*1352gration of the programs, is that there are still many inmates who would react negatively — indeed, violently — to integration. Application of Turner’s four factors to this case’s circumstances in light of this fact supports the district court’s conclusion that security concerns disqualify the plaintiffs. First, security is a classically legitimate penological concern, and the court could conclude that separating prisoners who, if mixed, would react violently, has a rational connection to preventing violence. This factor weighs against the plaintiffs. Turner’s second factor weighs in the plaintiffs’ favor: they have no other way of participating in integrated programs. The third factor, however, weighs heavily against the plaintiffs; the district court could properly conclude, based on sufficient evidence of inmate prejudice, that mixing HIV-positive and HIV-negative inmates could cause ripple effects that threatened security, either because of anti-HIV prejudice or as a result of the necessary diversion of resources from elsewhere to surveil the integrated programs. Finally, the court properly concluded that there are no “obvious, easy alternatives” here: plaintiffs’ implication that integration is an easy alternative because other prison facilities follow it does not undermine the court’s conclusion. Given its findings concerning the substantially higher seroconversion rates and incidence of violence elsewhere, the court could properly find that integration elsewhere served the interests of security and health less reasonably than the policies under attack here. Thus, the district court’s otherwise-qualified findings should be upheld.
C. Isolated Versus Global Burden of Accommodation
The Rehabilitation Act deems a disabled person qualified to participate in an activity if reasonable accommodations remove any disqualification.70 One of the plaintiffs’ primary reasonable-accommodation suggestions was the hiring of more guards to provide the surveillance necessary to prevent the plaintiffs and HIV-negative inmates from engaging in high-risk behavior. The district court found that the plaintiffs’ recommended level of security staffing would not prevent high-risk behavior, but implied that the Department of Corrections’ claimed necessary level of staffing — 60 additional guards all together — would adequately reduce the risk.71 This level of staffing, the court concluded, would place an undue financial and administrative burden on the already-strapped prison system.72 This conclusion was based on the effect system-wide of accommodating the plaintiffs in all the programs at issue: the court emphasized that it would “see ‘the forest as well as the individual trees.’ ”73 The plaintiffs claim that the district court erred either by failing to consider the burden in each program separately or, alternatively, by failing to pick particular programs to which limited resources should be allocated.74 The plaintiffs suggest that the court should have compared the cost of each individual guard to the Department of Corrections’ entire budget to determine if each guard alone would be a reasonable accommodation.
The district court did not err. Whether an accommodation imposes an undue burden on a defendant is determined by “the hardships imposed by the plaintiffs preferred accommodation in the context of the particular agency’s operations.”75 And cost is a hardship properly taken into account in assessing whether the burden on a federal-fund grantee is undue.76 Assessing that cost accommodation-by-accommodation, and not in the aggregate, effectively defeats the purpose of the agency-specific analysis: it permits impositions of burdens that could niekel-and-dime an agency to extinction. That reasoning ignores, rather than examines, the “context of the particular agency’s operations.” Thus, if *1353the plaintiffs demand to be accommodated system-wide, the total cost to the Department of Corrections is the relevant cost for determining the agency’s burden. Adding 60 guards at Tutwiler and Limestone, as the record suggests are needed to ensure security, would require a nearly 20% increase in guard staffing.77 The district court could conclude based on this record that accommodation by an increase in staffing would place an undue hardship on the Department of Corrections.
The plaintiffs further fault the district court for not considering whether the Department of Corrections in fact needed all the guards it currently has stationed where they are currently stationed, and for not choosing certain programs for integration if it is indeed true that providing all the guards necessary would be an undue burden. The plaintiffs cite no evidence in the record, however, that indicates that any guards elsewhere are unnecessary. To the contrary, the record suggests that both Tutwiler and Limestone are already staffed significantly below the optimum.78 Nor do the plaintiffs propose a basis upon which the district court should have allocated any portion of the necessary additional guards. The plaintiffs thus failed to carry their burden of proof that alternate, less burdensome reasonable accommodations exist by moving correctional officers from other posts to HIV-protection posts. The district court’s conclusion should accordingly be upheld.
D. Reassignment to Another District Judge
The plaintiffs in this case sought Judge Varner’s recusal on a variety of grounds. Judge Varner refused to recuse himself, and the plaintiffs appeal the denial of their motion and suggest, in the alternative, that the case be assigned on remand to another judge. Among the reasons for this remedy they point to Judge Varner’s remarks on the bench and in his opinion, the judge’s law clerk’s father-son relationship to the defendants’ lead counsel, and the judge’s possible learning of facts pertinent to the medical issues in this case from the - judge’s- son, a physician. The majority properly concludes that the record does not sufficiently evince bias or impropriety that required Judge Varner to recuse himself under the Supreme Court’s interpretation of 28 U.S.C. § 455.79 But the majority then adopts a “middle course” — reassigning- the action — that leaves Judge Varner stained with impropriety’s brush. This “middle course” is extreme and unnecessary in this case.
Reassignment is rare in this circuit and is generally appropriate " only when three factors weigh in favor of it: (1) whether the original judge would have had difficulty putting his previous views and findings aside; (2) whether reassignment is appropriate to preserve the appearance of. justice; and (3) whether reassignment would entail waste and duplication out of proportion to the gains realized from reassignment.80 None of these factors points to reassignment in this ease.
The first prong is satisfied only by a showing of extreme conduct — for example, when a judge has willfully refused to follow this court’s mandate,81 or when the judge has made such mistakes that in the court of appeals’ eyes the court appears unjust.82 *1354Judge Varner’s greatest offense, however, is that he disagreed with the plaintiffs. Far from ignoring this court’s mandate, his 475-page opinion meticulously reviews the individual evidence about the location, circumstances, and security concerns for dozens and dozens of separate programs. Nor is the appearance of justice imperiled here. As the majority observes, Judge Varner’s remarks do not show any bias mandating recusal,83 and the evidence of impropriety is also insufficient to warrant recusal. This court should on this record be as hesitant to direct reassignment as it is to find recusal to be mandated. Finally, the extraordinary waste of judicial resources that would result on remand if we shift judges after ten years of litigation and two trials counsels leaving the case before Judge Varner. I would therefore refuse to direct reassignment to another district judge.
III. Conclusion
For the foregoing reasons, I would vacate the district court’s judgment and remand for the limited purpose of permitting the district court to reconsider, in light of plaintiffs’ and defendants’ evidence of prisoner classification, whether classification offers the basis for a reasonable accommodation of the plaintiffs’ disability. I would otherwise affirm the district court’s resolution of the issues presented on appeal.
. See part II.B.4 of the majority opinion.
. Ala.Code § 22-11A-I7(a) (1996 Supp.)
. Harris v. Thigpen, 941 F.2d 1495, 1499 (11th Cir.1991).
. Id. at 1500.
. Id.
. Codified as amended at 29 U.S.C. § 794 (1994).
. Harris, 941 F.2d at 1523.
. Id. at 1526.
. See, e.g., Onishea v. Herring, No. 87V-1109-N, at 31 (M.D.Ala. Dec. 29, 1995) [hereinafter Op.]; R.29 at 158 (barbering with unsterilized razors); Op. at 32; R.29 at 167 (sports).
. The parties agreed that anal intercourse and sharing hypodermic needles are high-risk behaviors. Op. at 27.
. R.29 at 32.
. See, e.g., R.29 at 145-46 (“possibility” exists during anger-management and substance-abuse therapy for both violent and needle-sharing behavior); id. at 179 ("opportunity" exists during data processing classes at Tutwiler for needle-sharing); id. at 177-78 (large size of class and individual attention makes high-risk behavior possible in cosmetology class at Tutwiler).
. See, e.g., R.32 at 191 (syringes smuggled into Limestone wrapped in Kotex pads); Defs.’ Exs. F-5 (needles hidden in library), F-7 (syringe in public bathroom), F-20 (unspecified homosexual act in kitchen over a mixing bowl of peanut butter and jelly).
. See R.32 at 295 (runner — who as such held a coveted job among the prisoners — engaged in high-risk behavior in the HIV unit at Limestone).
. See R.36 at 142-43.
. For instance, HIV-negative prisoners refused to use the same drafting tools as the HIV-positive inmates, for fear of contracting HIV. R.27 at 228.
. Op. at 462.
. See, e.g., R.27 at 35 (no documented cases of transmission through stabbing); id. at 43-44 (never a reported incident of transmission via tattoo needles); id. at 36-39 (no confirmed cases of sports-related transmission); id. at 18 (only theoretical cases of sexual transmission between women have been described).
. See, e.g., R.25 at 76-77 (no sex or needle-sharing incidents reported in cosmetology class, just fistfights); id. at 110 (no incidents except fistfights in sewing factory at Tutwiler); R.32 at 258 (no incidents among inmate maintenance workers at Limestone); id. at 251 (no incidents among tractor crew at Limestone); id. at 247-48 (no incidents on the “double-0 squad,” which maintains the prison grounds); id. at 244 (no incidents in "college by cassette” classes); id. at 245 (no disciplinary incidents in the paralegal classes); id. at 246-47 (no incidents at GED and college graduation ceremonies); R.25 at 205 (no incidents in substance abuse treatment classes).
. R.28 at 94.
. R.33 at 26-27.
. Id. at 28-29. On the other hand, one of the plaintiffs’ experts also opined that "there has to be an unlearning process” at both Tutwiler and Limestone, and that certain activities such as dining should therefore not be immediately integrated. See R.25 at 93.
. R.33,at 210-12.
. See, e.g., Op. at 211.
. See id. at 48.
.' Op. at 464.
. Id. at 313-314 (emphasis added); accord, e.g., id. at 82, 277-78.
. Id. at 15-16.
. 482 U.S. 78, 89, 107 S.Ct. 2254, 2261, 96 L.Ed.2d 64 (1987).
. Harris v. Thigpen, 941 F.2d 1495, 1527 (11th Cir.1991).
. These activities included those such as visitation, classes, haircuts, medical treatment, sports, and legal research. Op. at 87 (visitation at Tutwiler); id. at 326 (paralegal training classes at Limestone); id. at. 336 (high-school and college level courses at Limestone); id. at 341 (GED examination at Limestone); id. at 383 (haircuts at Limestone); id. at 407 (visitation at Limestone); id. at 415 (medical treatment at Limestone); id. at 435 (sports at Limestone); id. at 444 (legal research at Limestone). The plaintiffs briefly challenge this conclusion, but the majority does not address it. I would affirm the court’s finding that providing materially identical activities in a separate location is a reasonable accommodation of the risks posed by the plaintiffs’ contagious disease.
. Kennedy v. Herring, 54 F.3d 678, 682 (11th Cir.1995).
. I agree, however, that this court's earlier mandate required the district court to consider evidence relating to inmate classification, as that *1347is probative of potential reasonable accommodations. I would therefore remand for the limited purpose of permitting the district court to revisit its reasonable-accommodations analysis, in light of that evidence and any counterevidence the defendants may choose to offer.
.One further holding from which I dissent requires only brief explanation. The plaintiffs claim that the district court erroneously refused to consider programs such as work release and boot camp that required residential integration of the plaintiffs into the general population at other facilities. The majority notes correctly that the plaintiffs waived all claims for residential integration by not challenging on appeal the trial court’s refusal to consider the issue of integration. Because the plaintiffs have abandoned on appeal any claim for residential integration, this court should not direct the district court — in effect end-running the waiver — to consider all programs requiring residential integration. The court should instead, in light of the waiver, uphold the district court’s refusal to consider those programs.
. Codified as amended at 29 U.S.C. § 794 (1994).
. Harris v. Thigpen, 941 F.2d 1495, 1522 (11th Cir.1991).
. 480 U.S. 273, 107 S.Ct. 1123, 94 L.Ed.2d 307 (1987).
. Id. at 288, 107 S.Ct. at 1131 (quoting Br. for American Med. Ass’n as Amicus Curiae at 19).
. Id.
. Id. at 287 n. 16, 107 S.Ct. at 1131 n. 16.
. Webster’s Third International Dictionary 2116 (1986).
. See id.
. Doe v. University of Md. Med. Sys. Corp., 50 F.3d 1261, 1266 (4th Cir.1995). The court noted that the Centers for Disease Control estimate the risk of transmission from an HIV-positive surgeon to a single patient to be between one in 42,000 and one in 417,000. Id. at 1263.
. Bradley v. University of Tex. M.D. Anderson Cancer Ctr., 3 F.3d 922, 924 (5th Cir.1993); cf. Kohl v. Woodhaven Learning Ctr., 865 F.2d 930, 941 (8th Cir.1989) (concluding that a hepatitis B carrier posed a significant risk, because even though the risk of transmission to those treated immediately after blood-to-blood contact is 2.5%, 1% of those infected would eventually die from the disease). This circuit’s law is not to the contrary. Martinez v. School Board, 861 F.2d 1502, 1506 (11th Cir.1988), which the majority cites as requiring a high probability of transmission to render a plaintiff not otherwise qualified, is distinguishable. Unlike here, the district court in that action did not even consider certain transmission risks, such as through blood. Instead, the district court found that there was a "remote theoretical possibility” of transmission through other bodily fluids and ended the inquiry there. Id. This court held that the risks of transmission through tears, saliva, and urine in the case's classroom setting did not reach a significant level of risk, but the court remanded for further findings as to the possibility of blood transmission. Id. Here, of course, the district court based its finding of significant risk on all possible transmission methods.
. Arline, 480 U.S. at 288, 107 S.Ct. at 1131.
. 482 U.S. 78, 107 S.Ct. 2254, 96 L.Ed.2d 64 (1987).
. For one program, interstate prisoner exchange, the district court pretermitled the rest of its otherwise-qualified analysis and concluded that legitimate penological concerns alone warranted the plaintiffs exclusion. Op. at 269-73.
. Quern v. Jordan, 440 U.S. 332, 347 n. 18, 99 S.Ct. 1139, 1148 n. 18, 59 L.Ed.2d 358 (1979); accord, Burger King Corp. v. Pilgrim’s Pride Corp., 15 F.3d 166, 168 (11th Cir.1994); Hester v. International Union of Operating Eng'rs, 941 F.2d 1574, 1581 n. 9 (11th Cir.1991).
. Burger King, 15 F.3d at 168.
. See Wheeler v. City of Pleasant Grove, 746 F.2d 1437, 1440 (11th Cir.1984) (”[T]he law is clear that [the doctrine] comprehends things decided by necessary implication as well as those decided explicitly.’’) (emphasis added) (internal quotations omitted).
. 18 Charles Alan Wright et al., Federal Practice and Procedure § 4478, at 789 (1981).
. See Lawson v. Singletary, 85 F.3d 502, 512-13 (11th Cir.1996) (law-of-the-case doctrine did not apply, in part because the issue at hand — the constitutionality of prison regulations — was not briefed in the earlier appeal, which concerned the appropriate standard of constitutional review); Northeastern Fla. Chapter of Associated Gen. Contractors of America v. City of Jacksonville, 951 F.2d 1217, 1218 n. 1 (11th Cir.1992) (first panel vacated a preliminary injunction solely because of an insufficiently developed record, despite a special concurrence pointing out — sua sponte — that the plaintiff lacked standing; second panel nonetheless was entitled to affirm a denial of relief based on its independent conclusion that the plaintiffs lacked standing), rev'd on other grounds, 508 U.S. 656, 113 S.Ct. 2297, 124 L.Ed.2d 586 (1993).
. Harris v. Thigpen, 727 F.Supp. 1564, 1582 (M.D.Ala.1990).
. Appellants' Br., Harris v. Thigpen, Nos. 90-7083, 90-7100, at 42-50.
. See, e.g., United States v. Dieguimde, 119 F.3d 933, 934-35 (11th Cir.1997) (declining to address poorly briefed and possibly moot issue); Alabama Power Co. v. OSHA, 89 F.3d 740, 747 n. 7 (11th Cir.1996) (refusing to address issue not raised in trial court); Boca Ciega Hotel, Inc. v. Bouchard Transp. Co., 51 F.3d 235, 237 n. 6 (11th Cir.1995) (declining to address issue raised only in reply brief).
. See Gates v. Rowland, 39 F.3d 1439, 1447 (9th Cir.1994).
. Turner v. Safley, 482 U.S. 78, 84-85, 107 S.Ct. 2254, 2259, 96 L.Ed.2d 64 (1987) (citation omitted).
. Gates, 39 F.3d at 1447.
. Torcasio v. Murray, 57 F.3d 1340, 1345 (4th Cir.1995), cert. denied sub nom. Torcasio v. Angelone,-U.S.-, 116 S.Ct. 772, 133 L.Ed.2d 724 (1996).
. Preiser v. Rodriguez, 411 U.S. 475, 491-92, 93 S.Ct. 1827, 1837, 36 L.Ed.2d 439 (1973).
. Torcasio, 57 F.3d at 1345 (citing Will v. Michigan Dep’t of State Police, 491 U.S. 58, 65, 109 S.Ct. 2304, 2309, 105 L.Ed.2d 45 (1989); United States v. Bass, 404 U.S. 336, 349-50, 92 S.Ct. *1351515, 523, 30 L.Ed.2d 488 (1971)). The Fourth Circuit has indeed held that the Rehabilitation Act does not apply to state prisons at all because of the Act's lack of an appropriate “clear statement.” Amos v. Maryland Dep't of Pub. Safety & Correctional Servs., 7 A.D. Cases 454 (4th Cir.1997). While the applicability of the Rehabilitation Act to state prisons is law of the case for this panel, the issue may merit en banc attention.
. Arline, 480 U.S. at 287, 107 S.Ct. at 1130-31 (describing the inquiry under its standard as "essential if § 504 is to achieve its goal of protecting handicapped individuals from deprivations based on prejudice, stereotypes, or unfounded fear, while giving appropriate weight to such legitimate concerns of grantees as avoiding exposing others to significant health and safety risks”) (emphasis added).
. See Doe v. University of Md. Med. Sys. Corp., 50 F.3d 1261, 1266 (4th Cir.1995); Bradley v. University of Tex. M.D. Anderson Cancer Ctr., 3 F.3d 922, 924 (5th Cir.1993).
. Anderson v. Romero, 72 F.3d 518, 520 (7th Cir.1995) (claim by HIV-positive inmates against prison authorities for disclosing his HIV-positive status and thus exposing him to attacks by other prisoners); Casey v. Lewis, 4 F.3d 1516, 1524 (9th Cir.1993) (‘TWJhenever inmates discover another inmate is HIV-positive ... threats are made against that inmate's life.”), rev'd on other grounds, - U.S. --, 116 S.Ct. 2174, 135 L.Ed.2d 606 (1996); Adams v. Drew, 906 F.Supp. 1050, 1058 (E.D.Va.1995) (plaintiff inmate assaulted by other inmates solely because of his HIV positive status).
. Arline, 480 U.S. at 287, 107 S.Ct. at 1131.
. 29 U.S.C. § 794(d) (1994).
. Codified at 42 U.S.C. § 12113(b) (1994).
. Id.; see E.E.O.C. v. Amego, Inc., 110 F.3d 135, 144 (1st Cir.1997).
. See Turner, 482 U.S. at 89-90, 107 S.Ct. at 2262 (naming the four factors in its test).
. Harris v. Thigpen, 941 F.2d 1495, 1525 (11th Cir.1991).
. Op. at 51-53.
. Id. at 53.
. id. at 51.
. Appellants’ Br. at 41-42.
. Willis v. Conopco, Inc., 108 F.3d 282, 286 n. 2 (11th Cir.1997) (emphasis added) (quoting Barth v. Gelb, 2 F.3d 1180, 1187 (D.C.Cir.1993)).
. See, e.g., Borkowski v. Valley Cent. School Dist., 63 F.3d 131, 138 (2d Cir.1995); Vande Zande v. State of Wis. Dep’t of Admin., 44 F.3d 538, 543 (7th Cir.1995) (both identifying undue hardship analysis as a species of cost-benefit inquiry).
. The record suggests, moreover, that the additional staffing would be more expensive. To get guards for the HIV unit, Limestone must currently pay those guards more. R.32 at 110.
. Warden Lobmiller of Tutwiler requested 179 guards in 1993 and was authorized to employ 90. R.35 at 13-14. The Limestone facility is approximately forty guards short of full staffing. R.32 at 106.
. Liteky v. United States, 510 U.S. 540, 554-56, 114 S.Ct. 1147, 1157, 127 L.Ed.2d 474 (1994).
. Torkington, 874 F.2d at 1446.
. See, e.g., United States v. Remillong, 55 F.3d 572, 577 (11th Cir.1995) (reassigning case after the district judge "stubbornly persisted” in disregard of this court's mandate); Clark v. Coats & Clark, 990 F.2d 1217, 1230 (11th Cir.1993) (reassigning case when the district judge’s opinion "contained strong language expressing dissatisfaction with this court's decision”).
. See, e.g., Chudasama v. Mazda Motor Corp., 123 F.3d 1353 (11th Cir.1997) (reassigning case from district judge that had refused to rule on a motion to dismiss, ignored discovery objections, adopted one party’s exaggerated statements verbatim, and precipitously imposed a default sanction on the defendant for inadequately complying *1354with an unintelligible discovery order drafted by opposing counsel and signed by the judge).
. Compare United States v. Microsoft, Inc., 56 F.3d 1448, 1463-64 (D.C.Cir.1995) (judge rejected consent decree without modifications that would address conduct accused by a bestseller that the judge had read, but not mentioned in the complaint; judge also permitted intervenors and amici to proceed anonymously because of his bestseller-induced suspicion of Microsoft); Torkington, 874 F.2d at 1447 (judge grants judgment of acquittal early in prosecution case, which the judge called a waste of taxpayer money, based upon a prosecution witness’s unelicited remark that violated a motion in limine).