EDWARD C. PRADO, Circuit Judge:

Plaintiff-Appellee David Rasheed Ali is an observant Muslim and in the custody of the Texas Department of Criminal Justice ("TDCJ"). This appeal concerns his suit seeking permission to grow a "fist-length" (i.e., four-inch) beard and wear a kufi, a knit skullcap, as required by his religious beliefs. Ali alleges that, as applied to him, TDCJ's grooming policy, which bans four-inch beards, and religious headwear policy, which prohibits kufis to be worn outside of an inmate's cell or religious services, violate the Religious Land Use and Institutionalized Persons Act ("RLUIPA"), 42 U.S.C. § 2000cc et seq. After a five-day bench trial, the trial court granted declaratory and injunctive relief enabling Ali to grow a four-inch beard and wear his kufi throughout TDCJ's facilities. Defendant-Appellant William Stephens, in his capacity as TDCJ Director, appealed. Finding no reversible error, we AFFIRM.

I. BACKGROUND

Ali is confined to TDCJ's Michael Unit, a maximum security prison. He is a "trusty" inmate, which is the lowest security level classification, and lives in a dormitory outside of the Michael Unit's fence line. Ali is also an observant Muslim. According to Ali, his faith requires him to have a beard that is not shorter than a fist's length, which is approximately four inches, and to wear his kufi at all times.

A. Procedural Background

In March 2009, Ali brought this suit, proceeding pro se, against the Director of TDCJ.1 Ali asserted that TDCJ's grooming and religious headwear policies violated RLUIPA to the extent they prevented him from growing a fist-length beard and wearing his kufi in accordance with his religion.2 At the time he filed suit, TDCJ's grooming policy required male inmates to be clean shaven. The sole exception was for inmates who had been diagnosed with certain dermatological conditions. This medical exemption allowed an inmate to shave with clippers rather than a razor and, depending on the nature of the condition, grow a quarter-inch beard. TDCJ did not provide any exemption to its grooming policy for religious reasons. Inmates that violated the grooming policy were subject to disciplinary action. In addition, TDCJ's religious headwear policy permitted male inmates to wear religious caps, such as kufis, only when they were within their housing area, such as a cell or dormitory, or at religious services.

In his suit, Ali sought declaratory judgment, as well as preliminary and permanent injunctions requiring TDCJ to exempt Ali from its beard and kufi restrictions. In 2010, the district court denied Ali's motion for a preliminary injunction and dismissed his complaint for failing to state a claim pursuant to 28 U.S.C. § 1915A. See Ali v. Quarterman, 434 Fed.Appx. 322, 324 (5th Cir.2011). This Court, however, vacated the dismissal of Ali's RLUIPA claims concerning both the grooming and headwear policies and remanded for further proceedings. Id. at 325-26. It also vacated the denial of the preliminary injunction as to the grooming policy but held that Ali had abandoned his appeal of the denial of the preliminary injunction as to the headwear policy. Id. at 326.

In February 2014, the trial court3 granted in part a preliminary injunction that allowed Ali to grow a quarter-inch beard, relying on our intervening decision in Garner v. Kennedy, 713 F.3d 237 (5th Cir.2013). See Ali v. Stephens, No. 9:09-CV-52, 2014 WL 495162, at *2 (E.D.Tex. Feb. 4, 2014). In Garner, a Muslim inmate brought a RLUIPA challenge seeking to grow a quarter-inch beard and wear his kufi while traveling to and from services. 713 F.3d at 241. After a bench trial, the district court in that case granted an injunction permitting the inmate to grow a quarter-inch beard but denied his kufi request. Id. TDCJ appealed, and we affirmed.4 Id. at 240.

In this case, the trial court cited Garner to explain why it was denying Ali a preliminary injunction to grow a fist-length beard. As it noted, "the record in this case is different than the record in Garner" in part because Ali was requesting a longer beard than the one sought in Garner. Ali, 2014 WL 495162, at *3. The court also explained that TDCJ had submitted evidence that it had not in Garner, such as estimates regarding the cost of changing its grooming policy. Id.

In July 2014, the trial court held a five-day bench trial. Ali called three expert witnesses, including George Sullivan and Roy Timothy Gravette, who between them had over 60 years of experience working for and auditing correctional facilities. They testified about the impact of beards and kufis based on their experience with prisons that already permitted them. TDCJ's expert witness, Ronald Angelone, testified primarily about his experience with beards in the prison systems in which he had served as the director. Robert Eason, TDCJ's Deputy Director, testified about TDCJ's security interests associated with Ali's requested exemptions and his findings related to his tours of other prisons that allow inmates to have beards and wear kufis throughout their facilities.

In September 2014, the trial court granted an injunction allowing Ali to have a beard not to exceed four inches and to wear his kufi throughout TDCJ's facilities. See Ali v. Stephens, 69 F.Supp.3d 633, 654-55 (E.D.Tex.2014). Among its findings of fact, it concluded that Ali's expert witnesses were "more credible" than TDCJ's witnesses because Ali's witnesses "both have significant experience working in prisons where beards are allowed and [kufis] are allowed to be worn at all times." Id. at 642. TDCJ timely appealed.

B. Post-Trial Developments

While this appeal was pending, the Supreme Court decided Holt v. Hobbs, ___ U.S. ___, 135 S.Ct. 853, 190 L.Ed.2d 747 (2015). The Court in Holt held that the grooming policy of Arkansas's prison system violated RLUIPA to the extent it prevented a Muslim inmate from growing a half-inch beard in accordance with his religious beliefs. Id. at 859. The policy at issue — like TDCJ's policy at the time of trial — banned inmates from growing beards, with the sole exception that inmates with dermatological needs could grow facial hair no longer than a quarter-inch. Id. at 860. In response to Holt, TDCJ moved to stay its appeal while it developed a new grooming policy. We denied the stay, and TDCJ implemented its new policy prior to oral argument. Under its current grooming policy, inmates "with religious belief who want to grow a beard" are permitted, subject to TDCJ's approval, to have a beard that is not longer than "one-half (½) inch in length."5 There is no evidence that TDCJ has changed its religious headwear policy in any pertinent respect.

C. The Statutory Scheme

Section 3 of RLUIPA, which concerns institutionalized persons, states:

42 U.S.C. § 2000cc-1(a). RLUIPA, which provides a private cause of action, id. § 2000cc-2(a), implements a burden-shifting framework, Chance v. Tex. Dep't of Criminal Justice, 730 F.3d 404, 410 (5th Cir.2013). The plaintiff's initial burden is two-fold: he or she must show that (1) the relevant religious exercise is "grounded in a sincerely held religious belief" and (2) the government's action or policy "substantially burden[s] that exercise" by, for example, forcing the plaintiff "to `engage in conduct that seriously violates [his or her] religious beliefs.'" Holt, 135 S. Ct at 862 (quoting Burwell v. Hobby Lobby Stores, Inc., ___ U.S. ___, 134 S.Ct. 2751, 2775, 189 L.Ed.2d 675 (2014)). If the plaintiff carries this burden, the government bears the burden of proof to show that its action or policy (1) is in furtherance of a compelling governmental interest and (2) is the least restrictive means of furthering that interest. 42 U.S.C. § 2000cc-1(a); Holt, 135 S.Ct. at 863.

The Supreme Court recently emphasized that "[s]everal provisions of RLUIPA underscore its expansive protection for religious liberty." Holt, 135 S.Ct. at 860. Courts must construe RLUIPA "in favor of a broad protection of religious exercise, to the maximum extent permitted by the terms of this chapter and the Constitution." Id. (quoting 42 U.S.C. § 2000cc-3(g)). In addition, RLUIPA "may in some circumstances require [a][g]overnment to expend additional funds to accommodate [inmates'] religious beliefs." Hobby Lobby, 134 S.Ct. at 2781 (citing 42 U.S.C. § 2000cc-3(c)); see also Holt, 135 S.Ct. at 860. Finally, the law defines "`religious exercise' capaciously to include `any exercise of religion, whether or not compelled by, or central to, a system of religious belief.'" Holt, 135 S.Ct. at 860 (quoting 42 U.S.C. § 2000cc-5(7)(A)).

Although RLUIPA subjects governmental action to exacting scrutiny, "it also affords prison officials ample ability to maintain security." Id. at 866. When applying RLUIPA, "courts should not blind themselves to the fact that the analysis is conducted in the prison setting." Id. In particular, we must recognize that "[p]rison officials are experts in running prisons and evaluating the likely effects of altering prison rules." Id. at 864. Yet our deference is not unyielding: courts are not "bound to defer" to a prison system's assertions. Id. "[I]t is the obligation of the courts to consider whether exceptions are required under the test set forth by Congress." Id. (quoting Gonzales v. O Centro Espirita Beneficente Uniao do Vegetal, 546 U.S. 418, 434, 126 S.Ct. 1211, 163 L.Ed.2d 1017 (2006)). Thus, while we "should respect" the prison officials' expertise, we cannot abandon "the responsibility, conferred by Congress, to apply RLUIPA's rigorous standard." Id. Even before Holt clarified the deference owed to prison officials under RLUIPA, we observed that "[r]ather than deferring to the prison's general policy regarding a matter, we have consistently tested the prison's asserted interests with regard to the risks and costs of the specific accommodation being sought." Chance, 730 F.3d at 418; see also id. at 419 (emphasizing that the deference owed to "TDCJ's expertise in prison administration and security ... does have limits" (internal citation omitted)).

II. STANDARD OF REVIEW

After a bench trial, we review a trial court's findings of fact for clear error and its conclusions of law de novo. Garner, 713 F.3d at 242. Under clear error review, if the trial court's factual findings are "plausible in light of the record viewed in its entirety, we must accept them, even though we might have weighed the evidence differently if we had been sitting as a trier of fact." Anderson v. Sch. Bd. of Madison Cty., 517 F.3d 292, 296 (5th Cir. 2008) (quoting Price v. Austin Indep. Sch. Dist., 945 F.2d 1307, 1312 (5th Cir.1991)); see also Ogden v. C.I.R., 244 F.3d 970, 971 (5th Cir.2001) (per curiam) ("Clear error exists when this court is left with the definite and firm conviction that a mistake has been made."). "When reviewing a district court's factual findings, this court may not second-guess the district court's resolution of conflicting testimony or its choice of which experts to believe." Grilletta v. Lexington Ins. Co., 558 F.3d 359, 365 (5th Cir.2009) (per curiam). Credibility determinations are "peculiarly within the province of the district court." Id. (quoting League of United Latin Am. Citizens No. 4552 v. Roscoe Indep. Sch. Dist., 123 F.3d 843, 846 (5th Cir.1997)). Accordingly, "the clearly erroneous standard of review following a bench trial requires even `greater deference to the trial court's findings when they are based upon determinations of credibility.'" Guzman v. Hacienda Records & Recording Studio, Inc., 808 F.3d 1031, 1036 (5th Cir.2015) (quoting Luhr Bros., Inc. v. Shepp (In re Luhr Bros., Inc.), 157 F.3d 333, 338 (5th Cir. 1998)).

In the context of RLUIPA, determining whether a prison system has satisfied its statutory burden is "highly dependent on a number of underlying factual issues" and, as such, is "best characterized as a mixed question of fact and law, which is subject to de novo review." Garner, 713 F.3d at 242. Thus, although we review the court's factual findings for clear error, we review de novo its application of those findings in determining whether the challenged government action is in furtherance of a compelling governmental interest and is the least restrictive means to advancing that interest. Id.

III. DISCUSSION

On appeal, TDCJ does not challenge the trial court's holding that its grooming and religious headwear policies substantially burden Ali's religious exercise. We therefore decline to address this issue. See Garner, 713 F.3d at 244. TDCJ instead contends the trial court erred by holding that its policies violate RLUIPA as applied to Ali because they (1) do not further any compelling governmental interests and (2) are not the least restrictive means. We first address whether the grooming policy complies with RLUIPA.

A. TDCJ's Grooming Policy

TDCJ first argues that a four-inch beard constitutes "long hair" and that Fifth Circuit precedent establishes that TDCJ's grooming policy complies with RLUIPA as a matter of law to the extent it bans long hair. In support, TDCJ relies on Longoria v. Dretke, 507 F.3d 898 (5th Cir.2007) (per curiam). In that case, a Texas inmate alleged that his religion barred him from cutting his head hair and sought an exemption from TDCJ's short-hair policy. Id. at 900. This Court dismissed his RLUIPA claim as frivolous, explaining that it was bound by Diaz v. Collins, 114 F.3d 69 (5th Cir.1997). See Longoria, 507 F.3d at 904. The Longoria court noted that in Diaz, after an evidentiary hearing, the district court had found that "long hair... facilitates the transfer of contraband and weapons into and around TDCJ institutions" and "requiring prisoners to have short hair makes it more difficult for escaped prisoners to alter their appearance." Longoria, 507 F.3d at 904 (alteration in original) (quoting Diaz, 114 F.3d at 72-73). This Court then concluded that the evidentiary showing in Diaz was "sufficient to preclude [the inmate's] RLUIPA claim" to grow long, unshorn head hair. Longoria, 507 F.3d at 904.

Longoria, however, does not foreclose Ali's request for a four-inch beard. As we observed, RLUIPA compels a "fact-intensive inquiry" into the particular costs and risks that the requested exemption engenders. Chance, 730 F.3d at 418 (quoting Moussazadeh v. Tex. Dep't of Criminal Justice, 703 F.3d 781, 795-96 (5th Cir. 2012)). We, in turn, have repeatedly conducted "case-specific inquiries" when addressing a RLUIPA claim. Id. at 411 (citing Garner, 713 F.3d at 245-46 and Moussazadeh, 703 F.3d at 795-96); see id. ("[O]ur RLUIPA analysis requires a careful consideration of each case's specific facts...."). For instance, we have even recognized that a holding against an inmate that assembled a "thin" record does not "foreclose" another inmate from subsequently demonstrating less restrictive means are available. Moussazadeh, 703 F.3d at 795; see also Yellowbear v. Lampert, 741 F.3d 48, 62 (10th Cir.2014) ("[T]he feasibility of requested exceptions usually should be assessed on a `case-by-case' basis, taking each request as it comes." (quoting O Centro, 546 U.S. at 436, 126 S.Ct. 1211)). Therefore, in assessing Ali's request for a four-inch beard, we focus on the record before us to analyze whether TDCJ has "not merely ... explain[ed] why it denied the exemption [to its grooming policy] but ... prove[d] that denying the exemption is the least restrictive means of furthering a compelling governmental interest."6 Holt, 135 S.Ct. at 864. For the reasons provided, we conclude that TDCJ has not met this burden.

1. The "Compelling Interest" Test

In deciding whether TDCJ has stated a compelling interest, the court does not ask if the challenged policy, in general, furthers a compelling governmental interest in security and costs. Id. at 863; see also Chance, 730 F.3d at 418. Instead, the government must show that "the compelling[-]interest test is satisfied through application of the challenged law `to the person' — the particular claimant whose sincere exercise of religion is being substantially burdened." Holt, 135 S.Ct. at 863 (quoting Hobby Lobby, 134 S.Ct. at 2779). This requires "scrutiniz[ing] the asserted harm of granting specific exemptions to particular religious claimants" and "`look[ing] to the marginal interest in enforcing' the challenged government action in that particular context." Id. (quoting Hobby Lobby, 134 S.Ct. at 2779). Applied in this case, we assess TDCJ's interests in preventing Ali from having a four-inch beard.

Moreover, determining whether TDCJ's policy is "substantially underinclusive" may "implicate the RLUIPA analysis." Id. at 865. As the Tenth Circuit elaborated in the RLUIPA context, "[a] law's underinclusiveness — its failure to cover significant tracts of conduct implicating the law's animating and putatively compelling interest — can raise with it the inference that the government's claimed interest isn't actually so compelling after all." Yellowbear, 741 F.3d at 60; see also Williams-Yulee v. Fla. Bar, ___ U.S. ___, 135 S.Ct. 1656, 1668, 191 L.Ed.2d 570 (2015) ("Underinclusiveness can ... reveal that a law does not actually advance a compelling interest."). We have similarly observed that a prison system's justification for denying an inmate's requested privilege is "dampened" where it affords other inmates a similar privilege. Moussazadeh, 703 F.3d at 795-96.

The Supreme Court's analysis in Holt is instructive. The Court found the prison system's grooming policy "substantially underinclusive" in two respects. Holt, 135 S.Ct. at 865. First, although the prison system did not allow inmates to grow half-inch beards as the plaintiff requested, it "permit[ted] inmates to grow more than a [half]-inch of hair on their heads." Id. Yet the prison system's policy did not require inmates to "go about bald" even though head hair is "a more plausible place to hide contraband than a [half]-inch beard." Id. at 866. An inmate's clothing and shoes similarly were better hiding places for contraband yet inmates were not required to be "barefoot[ ] or naked." Id. Second, in light of the fact that the prison already permitted quarter-inch beards for inmates with dermatological conditions, it had failed to establish "that a [quarter]-inch difference in beard length poses a meaningful increase in security risk." Id.

2. The "Least Restrictive Means" Test

The least-restrictive-means test "`is exceptionally demanding,' and it requires the government to `sho[w] that it lacks other means of achieving its desired goal without imposing a substantial burden on the exercise of religion by the objecting part[y].'" Id. at 864 (alterations in original) (quoting Hobby Lobby, 134 S.Ct. at 2780). The challenged policy cannot stand if "available, effective alternatives" are less restrictive of the inmate's religious exercise. Moussazadeh, 703 F.3d at 795 (quoting Ashcroft v. Am. Civil Liberties Union, 542 U.S. 656, 666, 124 S.Ct. 2783, 159 L.Ed.2d 690 (2004)). Moreover, courts "must not `assume a plausible, less restrictive alternative would be ineffective.'" Holt, 135 S.Ct. at 866 (quoting United States v. Playboy Entm't Grp., 529 U.S. 803, 824, 120 S.Ct. 1878, 146 L.Ed.2d 865 (2000)). The state's burden is not to show that it considered the claimant's proposed alternatives but rather to demonstrate those alternatives are ineffective. See Yellowbear, 741 F.3d at 63.

3. Analysis

TDCJ argues that a blanket prohibition on four-inch beards is the least restrictive means of furthering its compelling interest in (1) preventing the transfer of contraband within prison; (2) facilitating identification of inmates within prison and in the event an inmate escapes; and (3) controlling costs and, relatedly, maintaining orderly prison administration. The trial court rejected each of these arguments, holding that TDCJ did not prove that any of its asserted interests satisfied either the compelling-interest or least-restrictive-means test as applied to Ali. We analyze each interest in turn.

a. Preventing contraband

TDCJ contends its grooming policy is the least restrictive means of furthering its compelling interest in combatting the transfer of contraband within its facilities. The trial court found that "[p]ossession of contraband by inmates is one of the largest security issues in TDCJ." It also found that, based on testimony from both Ali's and TDCJ's witnesses, contraband has been discovered in inmates' beards at prisons that permit longer beards, specifically, the Federal Bureau of Prisons ("BOP") and the California Department of Corrections and Rehabilitation ("CDCR"). The court observed, however, that TDCJ had failed to introduce documentary evidence in support of its position and held that banning Ali's four-inch beard would not further a compelling interest in preventing contraband.

We disagree with the trial court's application of its factual findings to the compelling-interest test in this case. TDCJ clearly has "a compelling interest in staunching the flow of contraband into and within its facilities." Holt, 135 S.Ct. at 863. While "prison officials' mere say-so" may be insufficient to satisfy RLUIPA, id. at 866, the trial court erred by overemphasizing the lack of documentary evidence, particularly given the fact that Ali's own expert, Tim Gravette, testified that contraband had been found in beards in BOP facilities where he has worked. More importantly, the record indicates that a fist-length beard poses a greater risk with regard to contraband than a half-inch beard, which is the length Holt sanctioned and TDCJ permits under its current policy. In Holt, none of the witnesses testified to "any instances ... in Arkansas or elsewhere" in which an inmate hid contraband in a half-inch beard. Id. at 861. The Supreme Court, in turn, rejected the prison system's argument that banning such beards furthered a compelling interest in rooting out contraband. Id. at 863. In contrast, the record here indicates that inmates can and do hide contraband in longer beards. Thus, the difference in length between the beard permitted under TDCJ's current policy and the beard requested here poses a meaningful increase in security risks vis-à-vis the threat of contraband smuggling.

Ali responds that TDCJ's grooming policy is underinclusive because it permits an entire class of persons — female inmates — to have hair that is "much longer and thicker than a fist-length beard." TDCJ's female inmates are permitted to grow long hair, which must be neatly groomed, yet TDCJ did not introduce any evidence of finding contraband in a female inmate's hair. According to Ali, the underinclusiveness of TDCJ's grooming policy is substantial because of the trial court's finding that female inmates commit the same type of disciplinary infractions as men, although at a slightly lower rate on a per capita basis.

Even though TDCJ's policy concerning its female inmates is relevant to our analysis, we find that TDCJ has an adequate explanation for its differential treatment. As the Tenth Circuit has noted, a government can rebut a claim that its policy is underinclusive "by showing that it hasn't acted in a logically inconsistent way — by (say) identifying a qualitative or quantitative difference between the particular religious exemption requested and other ... exceptions already tolerated." Yellowbear, 741 F.3d at 61. At trial, TDCJ introduced evidence indicating that the contraband threat posed by male inmates is qualitatively different than that of female inmates. TDCJ's Senior Warden, Elizabeth Bailey, testified that the type of contraband female inmates smuggle is a lesser security concern because they tend to be non-dangerous items such as eyeliner or lipstick whereas men are more likely to smuggle cell phones or weapons. Further, as the trial court found, there are fewer correctional officers ("COs") per prisoner for its male prisons than its female prisons. In light of the record, we cannot say that TDCJ's stricter hair-length policy for male inmates is so inconsistent with its asserted interest in security that the challenged policy is substantially underinclusive.

Consequently, we hold the trial court erred in concluding that TDCJ's ban on four-inch beards did not satisfy the compelling-interest test. Our inquiry, however, does not end here. TDCJ must also prove that its current grooming policy is the least restrictive means, a burden the trial court concluded TDCJ did not meet. We agree given the record before us.

The trial court found that when searching male inmates, TDCJ's procedure is to have COs visually inspect short hair and "require inmates with longer hair to shake out their own hair with their fingers." It also found that TDCJ policy is to deny an inmate a religious devotional item if an inmate misuses that item or "present[s] a security risk based on documented behavior." The court then held that an effective alternative to banning all four-inch beards would be to have the CO perform the same search of a beard "as is done [for] searches of hair": the CO can visually inspect the beard and, if necessary, have the inmate run his fingers through his beard. The court also noted that, in conjunction with these searches, TDCJ could revoke an inmate's beard privilege if he abused it or refused to comply with the searches.

The trial court did not err in light of the record. Ali's expert witness, George Sullivan, testified that, based on his experience auditing prisons that allow longer beards and personally conducting searches as a CO, a visual inspection accompanied by having the inmate shake his own beard, if needed, effectively reveals contraband. Tim Gravette, Ali's other expert, similarly testified that COs can search long beards by having an inmate shake out his beard hair, which is the technique used by BOP.7 Finally, Holt bolsters the court's conclusions. In that case, the Supreme Court found that a less restrictive alternative to prohibiting beards would be to require inmates to conduct a self-search, albeit with a comb rather than his fingers, and that an institution could revoke an accommodation should an inmate abuse it. Holt, 135 S.Ct. at 864, 867.

TDCJ responds that the trial court committed reversible error because it did not "afford any level of deference" to the testimony of its witnesses. Specifically, it argues that the trial court should have deferred to two TDCJ officials, Director Robert Eason and Warden Todd Foxworth, who it claims testified that having inmates shake out their own beards would be unworkable because an inmate can manipulate the self-search in a way that avoids revealing contraband.8 We find that the trial court did not so err. To begin, contrary to TDCJ's characterization of Director Eason's testimony, his testimony is consistent with that of Ali's experts. In discussing how he would search fist-length beards, he testified that he would have inmates "run their fingers through that beard." Importantly, he did not express any concern that such searches would fail to reveal contraband, although he averred that self-searches would "take a little more time" than visual inspections. Such testimony — while relevant to TDCJ's separate interest in administrative costs, which we address below — does not support TDCJ's argument that self-searches would not uncover contraband.9

As for Warden Foxworth, he testified that having inmates run their fingers through their beards was not "feasible ... simply because of the consistency of beards" in that "[s]ome people have very thick beards." He also testified that some inmates "can't really grow a beard" and that "each [inmate] is going to be different." We cannot say that the trial court failed to adequately defer to this testimony. In its holding, the court found that beard hair can be searched in the same way as head hair — which does not have a prescribed length limit under TDCJ's policy so long as it is "neatly cut." Although we must respect a prison official's expertise, the trial court in this case did not exceed its prerogative as a fact finder in resolving competing testimony in Ali's favor where, as here, its finding was supported by testimony from both Ali's experts and TDCJ's own witness. Cf. Knight v. Thompson, 797 F.3d 934, 945 (11th Cir.2015) (noting that, in the RLUIPA context, the trial court "as the finder of fact, remain[s] free to reject" witnesses' testimony that is contradicted).10

b. Inmate identification

TDCJ contends that its grooming policy is necessary to further a compelling interest by aiding in the identification of inmates within the prison and inmates who escape. As to within-prison identification, the evidence introduced at trial indicated that inmates are provided an identification card containing their photograph and biographical information. Director Eason's testimony was that inmates "are identified by their identification cards [eight] times each day at count and several other times throughout the day." TDCJ, citing this policy, claims that its ban on four-inch prevents an inmate from being able to shave his beard and thereby no longer resembling the picture on his card. It also contends that it is especially reliant on identification cards because, according to Director Eason, officers are often rotated within their units to prevent the staff from becoming "overly familiar" with inmates and "complacent." Further, TDCJ introduced evidence that beards hinder identification because they can cover identifying marks and facial tattoos.

The trial court rejected TDCJ's arguments that banning Ali from having a four-inch beard satisfies the compelling-interest test with respect to inmate identification. In light of the governing case law and the record below, we agree. In Garner, we considered a similar contention with respect to quarter-inch beards. 713 F.3d at 247. Rejecting this argument, we reasoned that even though TDCJ "presented evidence that allowing inmates to have beards hinders inmate identification," it failed to carry its burden because "TDCJ allows inmates to shave their heads, and there was testimony that shaved heads pose just as many identification problems as allowing prisoners to grow and shave beards." Id. In Holt, the Supreme Court rejected a similar argument concerning the risk that inmates would shave to disguise themselves and even swap identification cards with each other. 135 S.Ct. at 865. The Court responded that the prison system "failed to establish why the risk that a prisoner will shave a [half]-inch beard to disguise himself is so great that [half]-inch beards cannot be allowed, even though prisoners are allowed to grow mustaches, head hair, or [quarter]-inch beards for medical reasons." Id. These other kinds of hair "could also be shaved off at a moment's notice, but the [the prison system] apparently does not think that this possibility raises a serious security concern." Id.

The reasoning of Garner and Holt apply with equal force based on the record here. The parties' evidence establish that an inmate can alter his appearance in many ways under TDCJ's current policy. An inmate, for instance, could shave his head, shave his quarter-inch beard (if he is permitted to grow one for medical reasons), or change his hairstyle. Further, Ali's expert, Tim Gravette, disagreed with TDCJ's witnesses who testified that permitting inmates to wear beards would create problems with identification, stating that there are many ways an inmate could alter his appearance. Because of the various ways an inmate can permissibly change his appearance, TDCJ has not shown that denying Ali's request for a four-inch beard furthers a compelling interest as to within-prison inmate identification. Cf. Schlemm v. Wall, 784 F.3d 362, 366 (7th Cir.2015) (observing in response to a prison system's asserted interest in suppressing gang identification that "it is difficult to depict as `compelling' a desire to cut out one potential means of [gang] identification" where other means of identification were "widely available already").

TDCJ's change in grooming policy — which now permits inmates to grow half-inch beards for religious reasons — also undermines its position. See Moussazadeh, 703 F.3d at 795-96 (accounting for TDCJ's change in policy in assessing a RLUIPA challenge). As noted, the compelling-interest test focuses on TDCJ's "marginal interest" in denying the accommodation. Holt, 135 S.Ct. at 863. We abided by this principle in Garner: in analyzing an inmate's request for a quarter-inch beard, we looked to whether there was "evidence that TDCJ would encounter greater or added difficulty if it enforced a one-quarter-inch as opposed to a clean-shaven rule." 713 F.3d at 246 (emphasis added). Here, we focus on the additional risk of permitting a four-inch beard instead of enforcing a half-inch limit. TDCJ's arguments concern how beards in general hinder identification, namely, they cover face tattoos and allow an inmate to change his appearance by shaving. Yet the testimony of TDCJ's officials indicated that half-inch beards, which TDCJ presently allows, also pose such risks. TDCJ in fact concedes that "easy identification of an inmate with a facial tattoo would be hampered whether there was a short beard or a long beard." Therefore, with respect to its interest in within-prison identification, TDCJ fails to adequately address the risks associated with the three-and-a-half-inch difference in beard length.11 See Holt, 135 S.Ct. at 866.

In addition, even if we assumed that banning all beards over a half-inch furthered a compelling interest by facilitating within-prison identifications, TDCJ has not proved its policy is the least restrictive means. The trial court found that a less restrictive alternative would be to maintain two photographs of the inmate, one with the beard and one without. In Holt, the Supreme Court condoned the "dual-photo method" in which prison officials would have "a bearded and clean-shaven photo to use in making identifications." Id. at 865. The Court highlighted that Arkansas, "like many other States," already had a policy of "photographing a prisoner both when he enters an institution and when `his appearance changes at any time during [his] incarceration.'" Id. (alteration in original). Just so here. TDCJ's policy — like that of Arkansas and the other prison systems referenced in Holt — is to photograph an inmate during intake and to take a new photograph if his appearance changes while in TDCJ custody. Therefore, as we held in Garner, TDCJ's identification concerns can be "addressed by requiring an inmate to have his identification picture changed if he grows or shaves his beard" given that TDCJ already requires a new picture when an inmate alters his appearances "in any way." 713 F.3d at 247. Indeed, TDCJ has incorporated this method into its new grooming policy — TDCJ will issue a new identification card to an inmate permitted to grow a half-inch beard for religious reasons. TDCJ has not explained why it cannot use the same technique for a beard that is four inches.

As to identifying escaped prisoners, TDCJ argues that its policy prevents an inmate from significantly changing his appearance by shaving upon escape. TDCJ also claims that because it publicly releases a photograph when an inmate escapes, it would cause confusion and impede identification if TDCJ had to release multiple photographs, such as one photograph with the beard and one without.

The trial court acknowledged but ultimately rejected TDCJ's assertions and, on this record, we cannot say that it was incorrect. The evidence adduced at trial indicated that after escaping, an inmate can change his appearance in many ways, such as by growing or cutting his hair or facial hair, dyeing his hair, wearing a hat, or donning glasses. TDCJ's witness Senior Warden Bailey agreed that such steps would alter an inmate's appearance. Further, Ali's expert, George Sullivan, testified that having to release more than one picture of the inmate if he escaped would not pose a security risk. Sullivan explained that law enforcement officials already release both the most recent picture of the inmate and one that projects his potential change in appearances, and such a practice does not confuse the public. Thus, as in Garner, we are unpersuaded by TDCJ's argument regarding identifying escaped inmates where the evidence established an inmate could "chang[e] his appearance outside of the prison" in many ways. Id. Holt also bolsters this conclusion: the prison system there argued that inmates could change their appearance by shaving in order "to escape[ ] and to evade apprehension after escaping," but the Court found that the prison system did not carry its burden, emphasizing the other ways an inmate could change his appearance. 135 S.Ct. at 864.

c. Cost control and prison operations

TDCJ also argues that its policy is the least restrictive means of advancing its compelling interests in controlling costs and ensuring orderly program administration. "[C]ost reduction, as a general matter, is unquestionably a compelling interest of TDCJ." Moussazadeh, 703 F.3d at 795. TDCJ relatedly has a compelling interest in maintaining the orderly administration of its operations. See Baranowski v. Hart, 486 F.3d 112, 125 (5th Cir.2007). RLUIPA, however, "may require a government to incur expenses in its own operations to avoid imposing a substantial burden on religious exercise." Garner, 713 F.3d at 245 (quoting 42 U.S.C. § 2000cc-3(c)). In determining whether a cost is compelling, a court may need to "put th[e] amount in perspective" by measuring the projected expense against the resources devoted to that interest. Moussazadeh, 703 F.3d at 795. For instance, in Moussazadeh, the inmate requested kosher food, which TDCJ denied citing cost concerns. Id. at 794. In the subsequent RLUIPA challenge, the evidence indicated that the annual cost to provide the accommodation at most would be "about $88,000" whereas TDCJ's total food budget was $183.5 million. Id. at 795. We responded by noting our "skeptic[ism] that saving less than.05% of the food budget constitutes a compelling interest," although "we decline[d] to draw a bright-line rule." Id.

TDCJ argues that if four-inch beards were allowed, then staff would spend additional time searching those beards. The added time, according to TDCJ, would disrupt its daily schedule and impose significant costs because it would have to pay staff for the search time. Further, TDCJ asserts that because it is responsible for statewide policies, its costs must be measured on a statewide basis. It contends that its cost estimates should be based on the number of inmates statewide who belong to a faith group "that ha[ve] requested or ha[ve] a religious basis to request a beard." That number, according to TDCJ, is 131,478 inmates, which represents 94% of TDCJ's total male inmate population. TDCJ claims that if 25% of those inmates — which totals 32,870 inmates — requested and were granted permission to grow a four-inch beard, then COs would spend 182.6 hours searching beards every day.12 It claims that 182.6 hours of staff salary equals $1,110,372.34 annually and that its interest in saving that amount is compelling.13

The trial court rejected TDCJ's projections. It noted that Ali's expert, George Sullivan, testified that, in his experience in prisons that allow beards, 30 to 40% of Muslim inmates grew beards. Relying on this testimony, the court considered the time spent searching four-inch beards if 40% of Muslim inmates in the Michael Unit chose to grow such a beard and found that it would take 34 minutes each day. The court found that any additional time spent searching beards would either be absorbed by existing staff, thus costing TDCJ nothing, or, even if new staff was hired to search beards, amount to $3,445.84 each year, an insignificant fraction of TDCJ's $3 billion budget.

The trial court did not commit clear error in rejecting TDCJ's estimates regarding the number of inmates that are likely to request a fist-length beard. TDCJ officials admitted there had been no studies or surveys to determine the number of inmates that would seek to grow beards. See Garner, 713 F.3d at 245-46 (holding that the district court's "finding that any increased costs would be insignificant" was not "clearly erroneous" where TDCJ had conducted "no studies concerning the costs of allowing inmates to grow beards"). Nor was the trial court bound to accept TDCJ's predications in light of the speculative nature of the testimony of TDCJ's witnesses. TDCJ's cost estimates were based on 25% of inmates that belong to a faith group that, according to TDCJ Chaplain Billy Pierce, may have a religious basis for requesting an exemption if those inmates chose to request one. However, Chaplain Pierce's testimony concerned the number of inmates that "could possibly ask" for a beard. TDCJ did not present any evidence that many of the faith groups identified by Chaplain Pierce have in fact petitioned for a beard similar in length to Ali's request. Warden Foxworth similarly stated in a conclusory manner that if the "privilege" to grow a long beard "is out there," an inmate is "go[ing to] do it." He then "speculate[d]" that if Ali's request was granted, there would be "a lot of ... [one]-inch, inch and a half, [two]-inch beards." Such conjecture does not satisfy TDCJ's burden. See id. at 246; cf. Hobby Lobby, 134 S.Ct. at 2783 (rejecting, in the analogous context of the Religious Freedom Restoration Act of 1993, the government's argument that ruling in favor of the plaintiff "will lead to a flood of religious objections" when the government fails to "substantiate this prediction").

Further, we cannot disturb the trial court's finding that existing staff will absorb the time spent searching beards because we are not left with "the definite and firm conviction" that this finding is a mistake. Ogden, 244 F.3d at 971. The trial court's finding was based on its estimations regarding the time that would be spent searching beards requested by Muslim inmates in the Michael Unit. The estimated 34 minutes each day spent searching beards was compared to the roughly 74,160 minutes of CO time spent staffing Michael Unit each day. TDCJ responds that the proper scope of the cost inquiry is not the Michael Unit but rather all TDCJ's facilities. Although we agree that TDCJ must be able to consider statewide ramifications when responding to a RLUIPA challenge, the magistrate judge did not err in refusing to engage in such an analysis given the record. As it noted, "there [was] no evidence in the record for the court to determine the amount of correctional officer hours worked state-wide on a daily basis." RLUIPA does not require "unquestioning acceptance" of a prison system's assertions. Holt, 135 S.Ct. at 864. Therefore, in order for a court to evaluate whether the time spent searching statewide inmates would be absorbed by existing staff, TDCJ must provide some concrete evidence regarding its statewide resources allocated to its asserted interest.

Lastly, TDCJ's contentions regarding the costs and disruption caused by four-inch beards are undercut by its change in policy. For instance, TDCJ introduced evidence concerning the costs associated with providing a religious exemption to its no-beard policy, such as having to issue new identification cards with updated photographs, provide beard covers for kitchen workers, and process an inmate's request to grow a beard. Yet TDCJ's current grooming policy allows inmates to grow a half-inch beard for religious reasons. Therefore, TDCJ already must bear many of the administrative costs it cited at trial. TDCJ has not shown it will bear a significantly greater burden in this respect by permitting an inmate to grow a beard that is three-and-one-half-inches longer than is currently permitted.

Accordingly, based on the record before us, we conclude that TDCJ has not carried its burden under RLUIPA with respect to its denial of Ali's request for a fist-length beard not to exceed four inches.14

B. TDCJ's Religious Headwear Policy

TDCJ argues that its religious headwear policy, like its grooming policy, furthers its compelling interest in (1) preventing the spread of contraband, (2) allowing for rapid identification of inmates within prison, and (3) controlling costs and maintaining orderly operations. The trial court rejected these arguments and held that none of TDCJ's assertions satisfied either the compelling-interest or the least-restrictive-means test as applied to Ali. As explained below, we conclude that TDCJ has not satisfied its burden.

1. Analysis

TDCJ's religious headwear policy allows inmates to wear a kufi in their cells and at religious ceremonies but prohibits them from wearing them in other areas of the prison. As such, we address each of TDCJ's assertions by focusing on TDCJ's interest in enforcing its religious headwear policy to prohibit Ali from wearing his kufi outside of his cell and religious services.

a. Preventing contraband

The trial court held that, although an inmate "could hide contraband in or under a [k]ufi," TDCJ had failed to carry its burden to show that its headwear policy furthered a compelling interest in combatting contraband. In support, it noted that although TDCJ already permits inmates to wear kufis in their cell and at religious ceremonies and that some inmates are allowed to wear hats for work assignments, it failed to produce evidence of a single incident in which contraband was hidden "in or under a religious head covering, or even under a work cap." TDCJ responds that the testimony of its witnesses established that inmates will hide contraband in kufis if they are allowed to wear them throughout the facilities, despite the fact that a kufi, as TDCJ notes, "is not the easiest place to hide something."

Even assuming that TDCJ's headwear policy furthers a compelling interest in combatting contraband, TDCJ did not carry its burden to show that its current policy is the least restrictive means. The trial court found that Ali, like many Muslim inmates, are already allowed to possess a kufi, to wear them in their housing areas and at religious services, and to transport the kufis to and from religious services. Ali also owns another religious item, a prayer rug, which he carries to and from services. TDCJ's policy is to search Ali's kufi and prayer rug when he returns from services. TDCJ also frequently searches other items of clothing, such as hats or jackets. The trial court, in turn, concluded that a lesser restrictive alternative would be to search the kufis during routine inmate searches and, as it already does for religious devotional items, revoke any kufi privilege if it is abused.

TDCJ claims that searching the kufi would be ineffective because inmates will resist searches of a religious item and even threaten to sue. According to TDCJ, inmate resistance will deter COs from conducting searches, and inmates will then use kufis to smuggle contraband. We find TDCJ's argument unavailing in light of the district court's findings and the record below. TDCJ permits inmates to have kufis and prayer rugs and inmates are already required to make them available for inspection by COs. TDCJ fails to adequately explain why it can search an inmate's kufi when he is traveling with it to and from religious services but not if he was to wear it at other times. Further, TDCJ has not shown why it is impracticable to revoke kufi privileges for those inmates that resist such searches. See id. at 866-67 ("[A]n institution might be entitled to withdraw an accommodation if the claimant abuses the exemption in a manner that undermines the prison's compelling interests.").

b. Inmate identification

The trial court rejected TDCJ's assertion that its policy regarding kufis satisfies the compelling-interest test with respect to within-prison identification. It provided three reasons. First, male inmates could still alter their appearance by shaving or changing their hairstyle. Second, other types of head coverings that TDCJ permits, such as caps that are authorized for certain jobs, change how an inmate looks "as much as a[k]ufi." Third, female inmates that are Muslim are permitted to wear a hijab throughout its facilities. As the district court found, a hijab is a headscarf that is "larger" and "cover[s] more of the head" than a kufi. The court then concluded that kufis may actually help rather than hamper identification.

TDCJ contends that, contrary to the trial court's conclusion, kufis will hinder rather than facilitate inmate identification. TDCJ asserts that an inmate may wear a kufi sporadically and that, in turn, kufis will impede rapid identification should an inmate choose to wear his kufi one day and then remove it the next day. However, in light of Holt and Ali's evidence, we are unpersuaded that TDCJ has met its burden on this point. The prison system in Holt asserted that identification concerns are "particularly acute" because inmates "live in barracks and work in fields." 135 S.Ct. at 865. As noted, the Court rejected this argument because of the other ways that an inmate can disguise himself "at a moment's notice," including shaving his head hair or his quarter-inch beard. Id. TDCJ's policies permit an inmate to take similar steps that would change his appearance. Additionally, the prison also permits male inmates to wear other garments that impede identification, such as hats while working in the kitchen or out-doors and jackets when it is cold.

Further, TDCJ argues that kufis hinder identification by covering tattoos on the top of an inmate's head, including tattoos that are used as gang identifiers. It cites the testimony of Robert Grant, an official in TDCJ's Security Threat Group, who stated that a kufi would potentially hide a gang-related tattoo. TDCJ argues that it is crucial that it monitor tattoos so it can identify an inmate's gang affiliation.

We acknowledge TDCJ's compelling interest in identifying inmates' gang affiliation. However, on this record, TDCJ has not shown its kufi restriction is the least restrictive means to furthering this interest or its interest in identification generally. At trial, TDCJ introduced photographs of inmates with tattoos that would be covered by a kufi. The trial court found such evidence unconvincing, explaining that for each photograph the inmate could have been identified by facial tattoos that a kufi would not cover. Equally important, the trial court also held that a less restrictive alternative would be to have the inmate remove his kufi should the CO need to identify the inmate or his gang affiliation and, if necessary, revoke the privilege if it is misused.15 TDCJ's witness, Robert Grant, agreed that he would be able to determine the necessary gang-related information by requiring the inmate to remove the kufi. Finally, the court found that TDCJ's current policy is to document whether an inmate is allowed to have a kufi. It concluded that TDCJ could track which inmates were allowed to have a kufi by issuing property slips that an inmate must carry on his person, as it already does for other personal property such as watches. We hold that the trial court's conclusions are not erroneous.16

c. Cost control and prison operations

The trial court found that the only kufi-related expense or disruption to operations arise from the additional staff time needed to search kufis. The court reasoned that if 30% of Muslim inmates at the Michael Unit wore kufis, it would take an extra 15 minutes each day to search them, which is spread across 74,160 minutes of correctional officer time each day.17 It concluded that this additional search time would be absorbed by the existing staff. Even if additional staff had to be hired, it would cost at most $1,533 each year in staff salary. Ali's expert agreed that any added costs from kufis would be insignificant.

TDCJ responds that the trial court erred in limiting its analysis only to Muslims in the Michael Unit. It argues that the time and costs associated with Ali's request must account for all male inmates statewide, specifically, if 25% of male inmates wore a religious cap at all times, then it would entail 115.53 hours of CO time to search those kufis and other religious garments, which equals $702,526.37 in CO salary annually. TDCJ argues alternatively that every Muslim inmate will wear a kufi if Ali is permitted to wear one. It claims that searching all Muslim inmates who wear kufis would take 21.48 hours each day and cost $130,658.27 annually.

The trial court, however, rejected these estimates as "pure conjecture," and its conclusion is not clearly erroneous. TDCJ's claim that all Muslim inmates will want to wear a kufi was not based on any study or survey. See Garner, 713 F.3d at 246. Further, Ali's expert, George Sullivan, who had spent much of his career in prison systems that allow kufis, testified that around 20 to 30% of Muslim inmates chose to wear a kufi. The court noted that, in contrast to Ali's witness, none of TDCJ's witnesses "had experience in correctional systems that allow [k]ufis to be worn throughout the prison." Thus, while TDCJ may have presented testimony that was inconsistent with Sullivan's testimony regarding the number of Muslim inmates that would wear a kufi, we will not second-guess the trial court's resolution of competing evidence. See Anderson, 517 F.3d at 296.

We also cannot accept TDCJ's cost estimates based on its assertion that 25% of all male inmates would choose to wear some kind of religious headwear should Ali be granted the requested accommodation. Under RLUIPA, we have found it appropriate to "tak[e] an object-specific approach to requests for religious items." Chance, 730 F.3d at 418. TDCJ, however, fails to specify what other religious headwear inmates would request or point to any evidence regarding the search procedure required for such headwear. Without any concrete evidence regarding which items will be requested, the risk these items would pose, and the methods required to search them, we cannot conclude TDCJ has met its statutory burden on this point. See, e.g., id.; Schlemm, 784 F.3d at 366 ("On this record the cost of accommodating Navajo inmates [by providing special religious meals] appears to be slight, and the costs of accommodating other inmates' requests (should any be made) can be left to future litigation.").

Finally, TDCJ has not shown it has a compelling interest in the costs associated with allowing Muslim inmates statewide to wear kufis. As of 2014, there were 6,446 male TDCJ inmates that identified as Muslim. Given the record, we find no clear error in the trial court's finding that the cost of searching 30% of Muslim inmates would be absorbed by existing staff. However, even if none of the search time was absorbed by existing staff, then under TDCJ's methodology, it would cost $39,221 per year to search all the kufis.18 The record below indicates that TDCJ's budget for staff salary and wages was $1.045 billion in 2014, which is roughly one-third of its total operating budget of $3.1 billion. TDCJ has not shown it has a compelling interest in saving less than .004% of its budget that is dedicated to CO compensation. See Moussazadeh, 703 F.3d at 795 (expressing doubt that TDCJ had a compelling interest in saving $88,000 in food-related expenses where that cost amounted to "less than .05% of the food budget").

Therefore, we conclude that, based on the record, TDCJ has not carried its burden under RLUIPA with respect to its denial of Ali's request to wear his kufi throughout TDCJ facilities.

IV. CONCLUSION

For the foregoing reasons, we AFFIRM the district court's judgment and permanent injunction.