The plaintiff class, all of whom are prison inmates who have tested positive for the Human Immunodeficiency Virus (HIV), have sued officials of the Alabama Department of Corrections under § 504 of the Rehabilitation Act1 to force integration of prison recreational, religious, and educational programs. The district court denied relief, and we affirm.
I. Background
This litigation began over a decade ago in reaction to a legislatively inspired program to prevent the spread of Acquired Immune Deficiency Syndrome (AIDS) in Alabama’s prison system. Under statute,2 the Alabama prison system tests all entering inmates for infection with HIV, which causes AIDS. Harris v. Thigpen, 941 F.2d 1495, 1499 (11th Cir.1991). The Department of Corrections segregates inmates testing positive for the virus from the general inmate population in HIV-positive units, one for men at the Limestone Correctional Facility and one for women at the Julia Tutwiler Prison for Women. Id. at 1500.
So segregated, the HIV-positive inmates are unable to participate in many programs and activities with the HIV-negative, general population. Programs that are available only to the general population program at Limestone include “Double 0” squad jobs maintaining the prison grounds, jobs on the prison farm, bus squad jobs, facility maintenance jobs, trash detail jobs, kitchen jobs, runner jobs, upholstery classes, electrical technology classes, auto mechanics classes, construction trade classes, automotive body repair classes, horticulture classes, welding classes,3 staff barber jobs, inmate barber jobs, laundry jobs, gardening jobs in the prison’s vegetable gardens, the “Free by Choice” program (in which prisoners go to schools to talk to pupils about substance abuse and criminality), basketball and baseball tournaments, and “Alabama Volunteers in Corrections” meetings (to prepare prisoners for release). Programs that are available *1293separately to Limestone HIV-positive inmates include paralegal training classes (HIV-positive inmates see videotapes of live instruction); adult basic education, GED testing, Narcotics and Alcoholics Anonymous meetings, graduation ceremonies, drafting classes, haircuts, visitation, medical treatment, gymnasium and library time, chapel services, dining, and prisoner transportation.
A similar range of programs is unavailable to HIV-positive inmates at Tutwiler: data processing classes, clerical classes, cosmetology classes, sewing classes, building trades classes, automotive repair classes, welding classes, floral design classes, small business machine repair classes, quantity foods service classes, nutrition classes, concerts and talent shows, softball and volleyball games, the “Free by Choice” program, college classes, Laubach literacy training, sewing factory jobs, data processing jobs, “downtown” jobs for government agencies, community projects jobs, road squad jobs, kitchen jobs, yard jobs, maintenance jobs, housekeeping jobs, laundry jobs, trash jobs, runner jobs, and haircutting jobs. Like their male counterparts, however, the HIV-positive prisoners at Tutwiler also have access to many programs separately from the general population: chapel services, some rehabilitation programs such as substance abuse and stress management counseling, visitation, organized recreational activities such as May Day and Oktoberfest, dining, medical care, adult basic education, GED testing, library use, and prisoner transportation.
The plaintiff class challenged this practice of denying some programs to the HIV-positive, and providing other programs separately, as a violation of several constitutional rights and of § 504 of the Rehabilitation Act of 1973.4 The district court denied relief after a bench trial. The court concluded that no constitutional rights were violated. See Harris, 941 F.2d at 1521. 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. See Harris, 941 F.2d at 1523. The action was remanded to the district court for this fact-finding. This court directed the district court especially to evaluate the risk of HIV transmission in each program in order to determine whether the plaintiffs were otherwise qualified to participate in each program. In so directing the district court, 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.” Id. at 1526.
Measuring the significance of the risk of HIV transmission in a host of programs (and thus determining whether the plaintiffs were otherwise qualified to participate) was accordingly the focus of the second trial. One relevant fact was undisputed: In the state of medical knowledge and art at the time of trial, HIV infection inevitably progressed to AIDS. AIDS always led to death, often after lengthy suffering. But the parties’ evidence and arguments reflected two different approaches to the problem of measuring risk. Both sides presented substantial evidence in support of their positions.
On one hand, the plaintiffs argued that experience teaches us that the odds of HIV transmission in prison programs are remote at best. Their evidence fell into two categories. First, the plaintiffs presented expert testimony that incidents of HIV transmission in many activities are rare or virtually unknown. For instance, at the time of trial there were no reported cases of transmission as a result of lesbian sex. There was a similar lack of reported incidents of transmission from sports injuries, stabbing, or tattooing. Only “sporad*1294ic” instances of transmission from oral sex and fistfights had been reported. In short, the possibility of transmission in certain unusual circumstances (for instance in a fight if both participants bleed copiously into each other’s wounds, or during barbering if bloody razors are immediately reused) had not been realized in any commonly recurring way.
Second, however, the plaintiffs’ medical expert had to acknowledge that anal sex and needle-sharing are high-risk activities. Rather than assert the medical unlikelihood of transmission from these activities, the plaintiffs presented evidence of the rarity of such conduct in the programs in which they wish to participate. The plaintiffs’ prison security expert combed prison files for incident reports and found none describing anal sex or needle-sharing during a wide variety of prison programs at both Limestone and Tutwiler. Reported incidents being rare or unknown in the past, the plaintiffs reason, we can expect them to be rare or unknown in the future even if HIV-positive inmates participate in integrated programs. The plaintiffs also showed that many of these programs are in high demand, and that inmates therefore have every incentive to be on their best behavior; moreover, they point out, the degree of surveillance in most programs makes an occurrence of anal sex, for instance, implausible.
While not completely contradicting any of this testimony, the defendants stressed other facts. They introduced testimony that HIV transmission is theoretically possible, even if not documented, in sports accidents and during fights — wherever there is a large exchange of blood between an infected person and an uninfected one. And the defendants introduced evidence that concededly high-risk activity abounds in prison. This evidence included a six-inch high stack of incident reports from the past few years describing hidden hypodermic needles, homosexual acts, and bloody fights. The defendants also showed that a 1991 outbreak of syphilis in the HIV-positive population at Limestone, in which 86 inmates were ultimately treated, was traceable to a single inmate. Finally, the defendants offered studies showing that residentially integrated prison systems in Maryland, Nevada, and Illinois have experienced seroconversions at annual rates of .41%, .19%, and .33%, respectively.5 In Alabama, by contrast, over the entire roughly eight-year period of entrance and exit HIV-testing before the time of trial, (see R.31 at 167), only two inmates out of the at least 30,000 tested had seroconverted — an all-time rate of .0067%. (See id. at 106 (at least 30,000 inmates tested); id. at 166 (two total sero-converted).) Had the Alabama serocon-version rates resembled those in Illinois, in the neighborhood of 5 inmates per year would have contracted HIV at Limestone alone, perhaps 40 over an eight-year period.6
Neither side’s evidence was iron-clad on the points they emphasized. The plaintiffs undermined the defendants’ evidence of widespread high-risk behavior with anecdotal evidence of risk-free mixing. The defendants, on the other hand, called into question the plaintiffs’ reliance on the absence of incident reports by offering testimony that much high-risk activity goes unreported, even when discovered, and that inmates have developed techniques such as “hot railing” — posting a lookout— to avoid detection by roving guards. The defendants showed that a runner, who as such held a highly coveted job, forfeited it by engaging in a high-risk act in the HIV-positive unit. They also offered evidence of a single, sensational incident of high-risk behavior committed by two Limestone kitchen workers who engaged in homosexual sex in the kitchen bakery beside a mixing bowl of peanut butter and jelly.
*1295The district court generally took the defendants’ view of the facts. The court made two significant findings of fact, neither of which the plaintiffs have challenged as clearly erroneous. 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 unpredictable — 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.
From these facts, the court concluded that the transmission risk is significant in all programs. 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 cellmates, who in turn spread it farther, as was the case with the 1991 syphilis outbreak at Limestone. Given this degree of harm, even slim odds of transmission make the risk significant. As the court put it in words echoed throughout its 476-page opinion, “elimination of high risk behavior is impossible.... Because the DefendanfiPrison 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 qualified.’ ” (R.6-532 at 313-314 (emphasis added); accord, e.g., id. at 82, 277-78.)
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’ penological concerns, including the danger of violence that might arise from inmate prejudice toward and fear of HIV-positive prisoners. The court looked to Turner v. Safley, which permits infringement of prisoners’ First Amendment rights provided that the prison regulation “is reasonably related to legitimate penological interests,”7 as a guide to evaluating the Department of Corrections’ 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. In one program, interstate prisoner exchange,8 the court found that penological concerns alone — the cost of medical care for the HIV-positive — sufficed to excuse the exclusion of HIV-positive inmates.
The district court continued its analysis, as directed by this court,9 by asking if reasonable accommodations would make the plaintiffs otherwise qualified. Although the court found that in many programs the plaintiffs had already been reasonably accommodated by provision of identical, but segregated programs,10 the court found as to other programs that the only reasonable accommodation would be additional guards to prevent high-risk be*1296havior. 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.
On appeal, the plaintiffs attacked every step of the analysis. On the “otherwise qualified” issue, they contended 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 pe-nological objectives into account. Furthermore, they argued that the district court erroneously considered the burden on the Department of Corrections of accommodating all of the plaintiffs’ demands, rather than considering whether hiring the few guards necessary to integrate each program would individually overburden the entire prison system.
A panel of this court agreed with the plaintiffs on most issues. See Onishea v. Hopper, 126 F.3d 1323 (11th Cir.1997). The en banc court vacated the panel opinion, 133 F.3d 1377 (11th Cir.1998), and now revisits the case with new briefing. The plaintiffs’ contentions remain essentially the same and are addressed in turn below.11 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. Kennedy v. Herring, 54 F.3d 678, 682 (11th Cir.1995).
II. Discussion
A. What is a “significant risk”?
While not challenging the district court’s findings of fact, the plaintiffs argue that the district court misapplied the law and this court’s mandate in determining whether they qualify for relief under the Rehabilitation Act of 1973. In particular, they contend that the court wrongly interpreted a “significant risk” of HIV transmission to mean “any risk.” We conclude that the district court’s conclusion, based on its unchallenged findings of fact, is substantially correct.
Section 504 of the Rehabilitation Act, under which the plaintiffs seek relief,12 prohibits discrimination against “otherwise qualified individual[s] with a disability.” 29 U.S.C. § 794(a). An “individual with a disability” does not include “an individual who has a currently contagious *1297disease or infection and who, by reason of such disease or infection, would constitute a direct threat to the health or safety of other individuals.” 29 U.S.C. § 705(20)(D). This restriction adopts the Supreme Court’s interpretation of § 504 in School Board v. Arline13. Bragdon v. Abbott, 524 U.S. 624, 118 S.Ct. 2196, 2210, 141 L.Ed.2d 540 (1998). Under Arline, the “basic factors” for determining if the carrier of a contagious disease is entitled to § 504’s protection 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.”
Arline, 480 U.S. at 288, 107 S.Ct. at 1131 (quoting Br. for American Med. Ass’n as Amicus Curiae at 19). The Court further directed the courts to “defer to the reasonable medical judgments of public health officials.” Id.; see also Bragdon, 118 S.Ct. at 2211 (public health authorities’ determinations entitled to “special weight and authority”). In short, “[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.” Arline, 480 U.S. at 287 n. 16, 107 S.Ct. at 1131 n. 16.
We can infer from Arline’s language 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 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.” Webster’s Third International Dictionary 2116 (1986). It does not just mean “big.” See id. 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.
Thus, when the adverse event is the contraction of a fatal disease, the risk of transmission can be significant even if the probability of transmission is low: death itself makes the risk “significant.” But federal courts disagree about how low the odds may be, and how much evidence it takes to prove a significant risk.14 On one hand, the Fourth, Fifth, and Sixth Circuits have implicitly followed a cautious rule. For these courts, a showing of a specific and theoretically sound means of possible transmission was enough to justify summary judgment against an HIV-positive plaintiff on the ground that the infection posed a “significant risk” to others in the workplace, even though reported incidents of transmission were few or nonexistent, and the odds of transmission were admittedly small. See Bradley v. University of Tex. M.D. Anderson Cancer Ctr., 3 F.3d 922, 924 (5th Cir.1993) (affirming summary judgment in favor of a hospital that refused to permit an HIV-positive surgical assistant to assist surgeries, even though risk was “small”); see Doe v. University of Md. Med. Sys. Corp., 50 F.3d 1261, 1264-65 (4th Cir.1995) (affirming summary judg*1298ment against HIV-positive physician even though there were no documented cases of surgeon-to-patient transmission, and hospital itself described risk as “minimal but nevertheless ascertainable”); Estate of Mauro ex rel. Mauro v. Borgess Med. Ctr., 137 F.3d 398, 405, 407 (6th Cir.) (affirming summary judgment against HIV-positive surgical technician even though Centers for Disease Control calculated odds of HIV transmission during a surgery as between 1 in 42,000 and 1 in 420,000), cert. denied, — U.S. -, 119 S.Ct. 51, 142 L.Ed.2d 39 (1998).
The First Circuit, on the other hand, has construed the phrase “significant risk” to mean that not only must the danger be theoretically justifiable, it must also have been realized in at least several cases. See Abbott v. Bragdon, 107 F.3d 934, 948 (1st Cir.1997) (affirming summary judgment in favor of the HIV-positive plaintiff whose dentist denied her services, notwithstanding evidence of seven documented cases of patient-to-dentist HIV transmission; “Dr. Bragdon is not entitled to demand absolute safety”), aff'd in part, vacated and remanded in part, 524 U.S. 624, 118 S.Ct. 2196, 141 L.Ed.2d 540 (1998). This circuit’s panel precedent arguably anticipates the First Circuit’s view, see Martinez v. School Bd., 861 F.2d 1502 (11th Cir.1988) (“remote theoretical possibility” of HIV transmission through tears, saliva, and urine not significant risk; remand for determination of possibility of blood-to-blood transmission), although limited fact-findings make it difficult to determine that case’s standard. The Ninth Circuit arguably anticipated the First Circuit’s position, as well. See Chalk v. United States Dist. Ct., 840 F.2d 701, 709 (9th Cir.1988) (“[I]t was error to require that every theoretical possibility of harm be disproved.”)
The Supreme Court has reviewed the First Circuit’s judgment in Abbott v. Bragdon, but the Court’s opinion does not resolve the conflict on this question. The Court stated no rule, and it neither affirmed nor reversed the First Circuit’s conclusion on this issue. Rather, the Court explained that, “[f]or the most part,” the First Circuit’s analysis was correct. Bragdon v. Abbott, 524 U.S. 624, 118 S.Ct. 2196, 2211, 141 L.Ed.2d 540 (1998). The Court then offered the First Circuit some helpful hints for revisiting the significant-risk issue on remand.
From these helpful hints, we know that the relevant scientific knowledge is that at the time of the discrimination, id.; that an unreasonable offer of accommodation merits no weight, see id.; that public health statements recommending certain precautions are of scant value because they do “not assess the level of risk,” id.; that professional organizations’ opinions may be too intertwined with other matters (such as ethics) to give objective medical evaluations of risk, see id. at 2211-12; that the testimony of health experts may be of some value, see id.; that inconclusive scientific studies deserve little weight, see id.; that evidence of seven cases of patient-to-doctor transmission is not necessarily sufficient, “standing alone,” to show the risk to be significant, id.; but that all of these observations are qualified by the fact that the parties did not brief the issue, see id. at 2213. These helpful hints, however, do not shed much light on the amount or nature of the evidence that is necessary to prove a risk to be “significant.”
In this absence of any clear Supreme Court direction,15 we opt for the position of the Fourth, Fifth, and Sixth Circuits. Their cautious approach balances two arguably conflicting statutory policies. First, requiring evidence that the asserted risk of transmission has a sound theoretical basis prevents the Act from overlooking the unfounded fears and prejudices that Congress meant to uproot. See Arline, 480 U.S. at 285-87, 107 S.Ct. at 1129-31. Second, the more cautious rule protects federal funds recipients from well-*1299founded, worries that deaths can result from a ruling that an HIV-positive plaintiff is otherwise qualified, and thus it avoids the absurd conclusion that Congress has decreed even a few painful deaths in service of the Act’s noble goals. See id. at 287 & n. 16, 107 S.Ct. at 1131 & n. 16; Southeastern Community College v. Davis, 442 U.S. 397, 406, 99 S.Ct. 2361, 2367, 60 L.Ed.2d 980 (1979) (reversing judgment in favor of a would-be nursing student because her hearing problems could impede her performance in emergency situations, thus endangering patients). Accepting proof of a theoretically possible method of transmission as sufficient recognizes — unlike the First Circuit’s approach — that saying a risk of an event is small does not mean that it will not happen. And each time the event occurs, it is real people — not cold ciphers — that suffer the consequences. Saying that only one in 42,000 persons will die if exposed to a risk, see Mauro, 137 F.3d at 407, does not make that one’s death insignificant.
We thus hold that when transmitting a disease inevitably entails death, the evidence supports a finding of “significant risk” if it shows both (1) that a certain event can occur and (2) that according to reliable medical opinion the event can transmit the disease. This is not an “any risk” standard: the asserted danger of transfer must be rooted in sound medical opinion and not be speculative or fanciful. But this is not a “somebody has to die first” standard, either: evidence of actual transmission of the fatal disease in the relevant context is not necessary to a finding of significant risk.
Based on Arline’s standard thus interpreted, the district court’s conclusion was correct. Two of the district court’s unchallenged factual findings support its result. First, the district court found that violence, intravenous drug use, and sex may cause blood-to-blood contact and happen in prisons in the most unlikely and unexpected places and that it is impossible to know or watch much of what goes on. Second, the court found (based on adequate expert testimony) that blood-to-blood contact, such as that resulting from anal sex or needle-sharing, likely transmits HIV, and that violent exchanges of blood raise the specter of transmission. These two findings were enough for the court, sitting as fact-finder, to conclude that the risk was significant in any program in which prisoners participate.16
B. May legitimate penal concerns figure into whether an inmate is otherwise qualified to participate in a program?
The plaintiffs contend that the district court improperly engaged in the analysis prescribed by Turner v. Safley, 482 U.S. 78, 89-90, 107 S.Ct. 2254, 2261-62, 96 L.Ed.2d 64 (1987), to adapt constitutional rights to the prison context. Turner provides four factors to guide a court’s determination of whether a prison regulation is “reasonably related to legitimate penological interests”; if the regulation is, then it escapes constitutional scrutiny. Id. at 89, 107 S.Ct. at 2261. Those four factors 17 served as makeweights to the district court’s conclusions here for all the programs, and for one program (the interstate prisoner exchange program), the *1300Turner analysis served as the sole basis for denying relief.
We can agree with the plaintiffs' primary argument that Turner does not, by its terms, apply to statutory rights. That does not mean, however, that the district court's use of the Turner factors requires vacating its judgment. Section 504's provisions extend only to "otherwise qualified individual[s] with a disability." 29 U.S.C. § 794(a). "An otherwise qualified person is one who is able to meet all of a program's requirements in spite of his handicap." Southeastern Community College v. Davis, 442 U.S. 397, 406, 99 S.Ct. 2361, 2367, 60 L.Ed.2d 980 (1979). What the "requirements" of a program are is an issue of fact. See, e.g., Majors v. Housing Auth., 652 F.2d 454, 458 (5th Cir. Unit B Aug.3, 1981) (vacating summary judgment because of genuine issue of material fact).
The district court was entitled to find on this record-it indeed seems obvious-that the requirements for participation in prison programs are determined in part by the same "legitimate penological interests" that Turner respects in the First Amendment context. Security is one such legitimate interest. Turner and its antecedents indeed so recognize. See Turner, 482 U.S. at 91, 107 S.Ct. at 2262-63; Bell v. Wolfish, 441 U.S. 520, 546, 99 S.Ct. 1861, 1878, 60 L.Ed.2d 447 (1979) ("[Mjaintaining institutional security and preserving internal order and discipline are essential. . . Jones v. North Carolina Prisoners' Labor Union, Inc., 433 U.S. 119, 132, 97 S.Ct. 2532, 2541, 53 L.Ed.2d 629 (1977) ("The interest in preserving order and authority in the prisons is self-evident."); Pell v. Procunier, 417 U.S. 817, 823, 94 S.Ct. 2800, 2804, 41 L.Ed.2d 495 (1974) ("[CIen-tral to all other corrections goals is the institutional consideration of internal security within the corrections facilities themselves."); Procunier v. Martinez, 416 U.S. 396, 412-13, 94 S.Ct. 1800, 1811, 40 L.Ed.2d 224 (1974) ("the legitimate governmental interest in the order and security of penal institutions justifies the imposition of certain restraints"). Penological concerns such as security and cost are legitimate, and the evidence in this case shows that these are in fact the concerns behind the program requirements that participating prisoners neither create a threat of disorder or unreasonable costs. Thus, even if the district court's importation of Tii,rner `s standards into the Reha-bifitation Act was not precisely correct as a matter of legal theory, determining whether penological concerns impose requirements for program participation is not error. See Crawford v. Indiana Dep't of Corrections, 115 F.3d 481, 487 (7th Cir.1997).
While not conceding that Turner-like factors are properly considered at all, the plaintiffs also argue that the district court misapplied Turner first by making a clearly erroneous finding that integrating programs would risk violence, and second by concluding that the Department of Corrections' segregation policy is not an exaggerated response to the problem of prison security and slowing the virus's spread.
We reject both arguments. The district court's finding as to the risk of HIV-related violence was, as the plaintiffs point out, based on conflicting evidence. The defendants offered a study of inmate opinion that was both off-point (it asked for views on residential integration, not program integration) and six years old at the time of trial. But the plaintiffs did not do much better; they offered only anecdotal evidence of peaceful intermingling between HW-positive and HIlT-negative inmates. If two findings were possible here-risk of violence from intermingling or no risk of violence from intermingling-both required drawing a series of inferences from the parties' scant evidence. The district court chose to infer from the defendants' study (1) that residential and program integration would be equally objectionable to certain inmates and (2) that inmate opinion had remained substantially unchanged over a six-year period. To make the finding plaintiffs urged, the district court would have had to infer at least (1) that the HW-negative inmates involved *1301in peaceful intermingling represent the unanimous sentiment among HIV-negative inmates and (2) that special circumstances surrounding the intermingling did not contribute to the peacefulness. The district court’s choice of the former inferences over the latter ones is not clearly erroneous.
The court’s conclusion that the response is not exaggerated, moreover, properly follows Turner’s guidance. (The plaintiffs do not contend that Turner’s “exaggerated response” analysis in particular is inappropriate for Rehabilitation Act otherwise-qualified analysis.) Turner compels a conclusion that a response is exaggerated only when the inmate produces evidence of “easy alternatives” that come at a de minimis cost to valid penological interests. Turner, 482 U.S. at 91, 107 S.Ct. at 2262. The plaintiffs’ proposed “easy alternative” is to exclude from programs the prisoners who might react violently to integration. The record does not support the ease of this alternative, which would apparently require the Department of Corrections to evaluate the attitudes and predict the behavior of every program participant. The district court could correctly conclude that segregated programs are not an exaggerated response to the problem of violence.
In sum, then, we conclude that the district court could properly use factors such as Turner’s to determine whether the plaintiffs were otherwise qualified to participate in the programs. The district court’s analysis of those factors, moreover, was substantially correct.
C. Did the district court appropriately evaluate “reasonable accommodations”?
The Harris panel directed the district court on remand to consider whether reasonable accommodations would permit the plaintiffs to participate in any of the programs at issue. See Harris, 941 F.2d at 1527. The district court did so and concluded in most programs that the proposed accommodations, which mostly consisted of hiring additional guards to prevent high-risk behavior, would impose an undue burden on the prison programs. The plaintiffs challenge the district court’s conclusions on four grounds.
Before we get to the specific challenges, however, we have to point out that we are not construing any law whose source we may clearly identify. This is because the reasonable-accommodation principle (that a disabled person is entitled to participate in a program, notwithstanding lack of qualification, if a reasonable accommodation would qualify him) does not appear in § 504. See 29 U.S.C. §§ 706, 794. The principle appears, rather, in agency regulations promulgated under the statute’s authority. See, e.g., 28 C.F.R. § 42.511(a) (Department of Justice); 29 C.F.R. § 32.13 (Department of Labor). The applicability of these regulations as a general rule follows federal funding from an agency. See, e.g., 28 C.F.R. § 42.502; 29 C.F.R. § 32.2(a). Unfortunately, the district court denied the plaintiffs discovery of the source of the Department of Corrections’ federal funding, and the plaintiffs did not pursue questioning on this subject at trial. Thus, with one exception,18 we do not know what regulations apply to the Department of Corrections. And this matters: Different agency regulations lay out the contours of the reasonable-accommodation principle differently. For instance, Department of Justice regulations require no accommodation at all in nonemployment programs, see 28 C.F.R. § 42.540(0(2), while Department of Labor regulations require reasonable accommodation in employment training programs, see 29 C.F.R. § 32.13(a).
*1302Our ignorance of what law controls has not, however, been troubling to any party; the parties briefed the question only upon order of this court. Nor was it troubling to the Harris panel. See Harris, 941 F.2d at 1527. Rather than disturb what have become the parties’ settled expectations, therefore, we elect to apply some sort of least-common-denominator reasonable-accommodation principle, and we assume for present purposes (notwithstanding some arguably contrary regulations) that the reasonable-accommodation principle applies to all the programs under discussion. Cf. Alexander v. Choate, 469 U.S. 287, 300-04, 105 S.Ct. 712, 719-20, 83 L.Ed.2d 661 (1985) (construing the statute without the aid of apparently applicable regulations). But in doing so, we keep in mind that regulations issued under statute control the interpretation of the statute as long as the interpretation is permissible. See Chevron, U.S.A., Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 843, 104 S.Ct. 2778, 2782, 81 L.Ed.2d 694 (1984). We thus look to some regulations for guidance, but our interpretation here would not trump any agency regulation that interprets § 504 differently.
One further caveat: we acknowledge that the Harris panel sensibly concluded that differences among the programs in which the plaintiffs seek to participate required different analyses. See Harris, 941 F.2d at 1527. In this court, however, the plaintiffs have painted their reasonable-accommodation arguments with a broad brush, and we follow their lead in explaining our rulings.
1. Was classifying high- and low-risk prisoners an accommodation open to consideration?
One reasonable accommodation that the plaintiffs proposed was for the Department of Corrections to refine existing systems of inmate classification to detect HIV-positive inmates who were highly unlikely to engage in high-risk behavior and to permit those inmates to participate in integrated programming. At trial, however, each time the defendants objected to classification evidence, the court excluded it on the reasoning that any evidence tending to be about individual inmates — even if the evidence concerned inmate classification systems in general — was outside the scope of the retrial mandated by this court. For the same reason, the court implied that it would not consider the evidence or the argument. The plaintiffs contend that both the exclusion of evidence and the refusal to consider the evidence in the bench trial was error.
We agree with the plaintiffs that consideration of this proposed accommodation (and, therefore, evidence of it) was within the mandate. The Harris panel directed the district court to consider reasonable accommodations, see Harris, 941 F.2d at 1527, and this argument and evidence was about classifying all the class members for access to particular programs. The district court should not, therefore, have excluded the plaintiffs’ evidence about classification, and it should have considered classification as a reasonable accommodation.
But this conclusion does not mean that reversal is required. The defendants did not always object to classification evidence, and throughout the trial the plaintiffs were allowed to introduce substantial evidence about the classification process at the Department of Corrections (including the Department of Corrections’ inmate Classification Manual) and about classification’s potential effectiveness.19 Thus, the *1303court’s evidentiary ruling did not prevent the plaintiffs from making a classification argument. Furthermore, while the court implied at trial that it would not consider classification as an accommodation, the court’s findings of fact make clear that it did consider classification, if only implicitly. The court found that inmates are “un-trustworth[y],” “volatile],” “unpredictable,” and subject to “great pressures from conflicting sources.” (R.6-531 at 56, 57.) Therefore, the court found, prison officials must always “prepare[] for the worst” and “expect the unexpected.” (Id. at 57, 58.) These findings are an implicit rejection of plaintiffs’ argument. If prisoners are inherently “unpredictable,” as the district court found, then classification is not a reasonable way of integrating programs, because it does not reduce the risk of HIV transmission.
Because the plaintiffs thus did in fact successfully introduce classification evidence, and the district court implicitly considered the feasibility of classification, the plaintiffs’ substantial rights were not affected by the court’s trial rulings, and no remand is required. Cf. Deviner v. Electrolux Motar, AB, 844 F.2d 769, 774 (11th Cir.1988) (exclusion of certain evidence not reversible error when similar evidence “leaked” through the exclusion).
2. May excessive cost make a hardship undue?
Reasonable accommodations are mandatory only to the point that they impose “undue hardship” on the federal funds recipient. See, e.g., 29 C.F.R. § 32.3 (definition of “reasonable accommodation”); see also Willis v. Conopco, 108 F.3d 282, 286 n. 2 (11th Cir.1997). The plaintiffs’ primary proposed accommodation was for the Department of Corrections to hire enough guards to surveil the programs and prevent high-risk behavior. The district court found that the plaintiffs’ recommended level of security staffing to be too low to prevent high-risk behavior, but implied that the Department of Corrections’ claimed necessary level of staffing — -as many as 70 additional guards all together, a 23% increase — would adequately reduce the risk. This level of staffing, the court concluded, would place an undue financial and administrative burden on the already-strapped prison system.
In challenging this conclusion, the plaintiffs complain primarily that the district court should not have considered cost at all. The plaintiffs cite Zimring v. Olmstead, 138 F.3d 893 (11th Cir.), cert. granted, — U.S. -, 119 S.Ct. 633, 142 L.Ed.2d 571 (1998), for the proposition that only expenditures that would “fundamentally alter the service” provided can impose an undue hardship on the federal fund recipient. Id. at 905 (interpreting Title II of the Americans with Disabilities Act). According to the plaintiffs, the defendants did not carry their burden of proving that the addition of 70 guards would “fundamentally alter” the nature of the programs to which the plaintiffs seek access, and that cost could not, therefore, have been an undue burden.
We reject the plaintiffs’ argument both because Zimring’s standard does not apply here and because the district court’s findings are supported by sufficient evidence. Under the Rehabilitation Act, cost is relevant to the federal fund recipient’s burden when it, like any other burden, reaches the point of “undue hardship.” This is a consistent agency interpretation of the statute: Many (if not all) regulations promulgated under the Rehabilitation Act prescribe “undue hardship” as the standard for measuring when the federal funds recipient’s burden is too much. See, e.g., 7 C.F.R. § 15b.13(a) (Department of Agriculture); 28 C.F.R. § 42.511(a) (Department of Justice); 29 C.F.R. § 32.13(a) (Department of Labor); 34 C.F.R. § 104.12(a) (Department of Education); 45 C.F.R. § 84.12(a) (Department of Health and Human Services). And in all these regulations, determining “undue hardship” explicitly factors in “the nature and cost of the accommodation.” 7 C.F.R. § 15b.13(b)(3); 28 C.F.R. § 42.511(c); 29 *1304C.F.R. § 32.13(b); 34 C.F.R. § 104.12(c); 45 C.F.R. § 84.12(c).
This circuit’s precedents, moreover, show that a hardship becomes undue long before it effects a “fundamental alteration” in employment conditions or reallocation of program resources. See, e.g., Jackson v. Veterans Admin., 22 F.3d 277, 279 (11th Cir.1994) (undue hardship for hospital to schedule around a janitor’s unpredictable absences, which had occurred six times in two and a half months); Severino v. North Fort Myers Fire Control Dist., 935 F.2d 1179, 1183 (11th Cir.1991) (“Evenhanded treatment of employees in a covered program does not require an institution to lower or effect substantial modification of-standards to accommodate a handicapped person.”); Treadwell v. Alexander, 707 F.2d 473, 478 (11th Cir.1983) (undue hardship for park to accommodate heart-bypass patient by assigning demanding tasks to the two to four other park employees of the same rank who worked the same shift). Required accommodations, on the other hand, have demanded little from federal fund recipients in either money or other resources. See, e.g., Majors v. Housing Auth., 652 F.2d 454, 458 (5th Cir. Unit B Aug.3, 1981) (public housing complex could be required to permit exception to “no pet” rule to accommodate mentally unstable tenant who needed a dog); Tatro v. Texas, 625 F.2d 557, 564 n. 19 (5th Cir.1980) (school could be required to train teachers to catheterize little girl with no urinary control, when procedure took only a few minutes every three hours).
These regulations (some of which we assume apply here for the reasons explained above) and precedent do not fix a bright-line rule, but they do make clear that a financial burden may become too much when it reaches the point of “undue hardship.” And the district court did not clearly err in so finding here. Limestone’s warden estimated that he would need around 25 additional officers to guard integrated programs; Tutwiler’s warden estimated that she would need as many as 45 additional officers to integrate all programs safely.20 The record shows that the cost of these 70 additional officers for their first year would be $1,713,810. The budget for the entire Department in fiscal 1994 was $163,000,000. At the time of trial, this level of funding left Limestone 35 correctional officers short and Tutwiler 89 short. The court could conclude that in these circumstances spending an addition $1.7 million would cause undue hardship.
The plaintiffs assert to the contrary that their evidence of the Department of Corrections’ power to reallocate resources required the opposite conclusion. We disagree, because the plaintiffs’ reliance on the Department’s authority to reallocate resources rests on a legal equation of “undue hardship” with “impossibility.” As explained just above, that is not the standard. The Rehabilitation Act does not require the Department to do whatever it is legally capable of doing to accommodate the plaintiffs.
3. May a court consider the global burden of accommodations requested in different areas?
Here, the plaintiffs argue that the district court wrongly considered the burden to the Department of Corrections of accommodating the plaintiffs in every program that required additional security to operate safely. According to the plaintiffs, the district court should have picked and chosen among programs, selecting some for accommodations that, because of their smaller scale, would not be unduly burdensome. We reject this argument. While it may be true, as the plaintiffs assert, that “[n]othing in § 504 permits federal funds recipients to take this ‘all or nothing’ stand,” it was not the federal funds recipient here but the plaintiffs who insisted on having everything. The plaintiffs’ post-*1305trial brief, for instance, made this argument without specifically identifying any programs that the plaintiff preferred to integrate over others, and the Complaint asks for integration in all programs. It was incumbent on the plaintiffs, not the defendants or the court, to identify where they wanted partial relief, if full relief were not possible. Cf. Blessing v. Freestone, 520 U.S. 329, 117 S.Ct. 1353, 1360-62, 137 L.Ed.2d 569 (1997) (plaintiffs obligated to particularize claims for relief).
4. Can providing “separate but equal” programs be a reasonable accommodation?
The plaintiffs’ fourth challenge to the court’s analysis we need not discuss in detail. They complain that the district court ruled that “separate but equal” programming — those programs such as adult basic education that were afforded both HIV-positive and -negative prisoners separately — is not a proper accommodation of a disability. This argument is apparently new on appeal. Indeed, before the district court, the plaintiffs argued somewhat inconsistently that “separate but equal” accommodations were not only permissible under the Rehabilitation Act, but required by it. (R.5-526, at 161-62.) We decline to consider this newly raised issue.
III. Conclusion
For the foregoing reasons, the district court’s judgment is affirmed.21
AFFIRMED.
. 29 U.S.C. § 794.
. Ala.Code § 22-llA-17(a) (1996 Supp.)
. Correspondence in the months before the 1994 trial between the commissioner of corrections and the commissioner of post-secondary education suggest an intent to offer vocational and college classes to HIV-positive inmates separately from the general prison population. This intent, however, had not been fully realized at the time of trial.
. Codified as amended at 29 XJ.S.C. § 794 (1994).
. A "seroconversion" has occurred when an inmate who tested negative for HIV upon entering prison later tests positive.
. At the time of trial, one of plaintiffs' experts put Limestone's population at about 1600.
. 482 U.S. 78, 89, 107 S.Ct. 2254, 2261, 96 L.Ed.2d 64 (1987).
. In this program, based on an interstate compact, inmates may serve their terms in another state (for personal reasons, for instance) if that state agrees to send one of its inmates to Alabama.
. 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. (R.6-532 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).)
. In their en banc briefs — for the first time in the ten-year history of this lawsuit — the defendants claimed Eleventh Amendment immunity. The defendants conceded at oral argument, however, that the relief the plaintiffs seek plants this case within the fiction of Ex parte Young, 209 U.S. 123, 28 S.Ct. 441, 52 L.Ed. 714 (1908), and that the Eleventh Amendment is therefore not an issue in this case.
Two other issues can be disposed of in footnote. First, we asked the parties to brief the issue whether the Rehabilitation Act applies to prisons at all. The Supreme Court since decided that the Americans with Disabilities Act applies to prisons. See Pennsylvania Dep’t of Corrections v. Yeskey, 524 U.S. 206, 118 S.Ct. 1952, 141 L.Ed.2d 215 (1998). We find Yeskey indistinguishable and conclude that the Rehabilitation Act applies to prisons, as well.
Second, the plaintiffs complain that the district court excluded evidence concerning programs that would require residential integration. Residential integration — even for program purposes — was a matter beyond the scope of the second trial, which was to concern integration of programs. See Harris, 941 F.2d at 1527. We note further that it was the plaintiffs, not the defendants, who objected to introduction of this evidence at trial. (See R.36 at 92 ("Work release is no longer a part of this case.”).)
. As we address more thoroughly in the reasonable-accommodation discussion below, there are potentially applicable agency regulations promulgated under § 504. See, e.g., 28 C.F.R. subpt. G (Department of Justice regulations). Each agency's regulations, however, generally apply only to recipients of funds from that agency. See, e.g., 28 C.F.R. § 45.202 (restricting application of Department of Justice regulations to recipients of financial assistance from the Department of Justice). We cannot tell from the record where the Department of Corrections gets its federal funds and thus what regulations apply. We assume for this discussion that any applicable regulations would not contradict our analysis of the statute.
. 480 U.S. 273, 107 S.Ct. 1123, 94 L.Ed.2d 307 (1987).
. The cases cited here concerned suits under both the Rehabilitation Act and the Americans with Disabilities Act. The Acts' standards for carriers of contagious diseases are the same. See Estate of Mauro ex rel. Mauro v. Borgess Medical Ctr., 137 F.3d 398, 402 (6th Cir.), cert. denied, — U.S. -, 119 S.Ct. 51, 142 L.Ed.2d 39 (1998).
. On remand, the First Circuit has simply reaffirmed its earlier conclusions. See Abbott v. Bragdon, 163 F.3d 87 (1st Cir.1998).
. The plaintiffs also complain that the district court ignored this court’s earlier mandate to assess each program individually. We reject this argument. It is true that the court reached the same conclusion for a host of different programs; but nothing in the earlier mandate required the court to conclude that the programs were different. See Harris v. Thigpen, 941 F.2d 1495, 1526 (11th Cir.1991) ("[lit may turn out that the court’s conclusion of the significance of the risk of HIV transmission with regard to each program will be unaltered.”). The district court's 476-page opinion, moreover, doggedly reviews the independent evidence for each program.
. The four factors are (1) the connection between the prison regulation and a legitimate, neutral government interest, (2) the presence or absence of alternatives for the prisoners to exercise their claimed right, (3) the effect of unfettered exercise of the right on other inmates, guards, and the allocation of prison resources, and (4) the presence or absence of ready alternatives. See Turner, 482 U.S. at 90, 107 S.Ct. at 2262.
. We know that the Department of Labor contributes funds to the Department of Corrections' adult basic education programs. (See R.26 at 223.) Those programs, therefore, fall under Labor regulations, but the regulations by their own terms do not extend beyond that program. See 29 C.F.R. § 32.2(a) ("This part applies to each recipient of Federal financial assistance from the Department of Labor ... but is limited to the particular program for which Federal financial assistance is provided.”)
. (See, e.g., R.25 at 67-69; id. at 74; id. at 143-45 (Department of Corrections Classification Manual admitted as Plaintiffs’ Exhibit 1); R.27 at 127 ("cherry-picking” inmates is sound penological strategy); R.32 at 229 ("job board” classification system); R.34 at 210-213 (classification procedure); R.35 at 208-230 (great detail concerning classification process).)
. There was testimony that because of vacation and sick time it takes 1.73 officers to fill one eight-hour post; it is not clear if the wardens’ estimates were of posts or actual officers.
. In addition to their other arguments, the plaintiffs accuse Judge Varner, the district judge here, of impropriety and assert that Judge Varner should have recused himself. In the alternative, they argue that the case should be reassigned on remand. We conclude that the evidence of impropriety is insufficient to warrant recusal and vacatur of the judgment; we need not consider reassignment because we do not remand.