OPINION

BEA, Circuit Judge:

Rocio Brenda Henriquez-Rivas petitions for review of a decision of the Board of Immigration Appeals ("BIA") sustaining the government's appeal of an Immigration Judge's ("IJ") grant of asylum, and denying her applications for withholding of removal and protection under the Convention Against Torture. Henriquez-Rivas claims she is entitled to asylum because, as a person who testified in a criminal trial against members of a gang who killed her father in El Salvador, she is a member of a particular social group, on account of which she faces a well-founded fear of persecution if she were to return to El Salvador. For the reasons discussed below, we find that the BIA misapplied its own precedent in holding that witnesses who testify against gang members may not constitute a particular social group due to a lack of social visibility. Accordingly, we grant Henriquez-Rivas' petition for review and remand to the BIA for further proceedings.

I. Statutory Framework

Under the Immigration and Naturalization Act ("INA"), the Attorney General may grant asylum to a "refugee." 8 U.S.C. § 1158(b)(1)(A). To qualify as a refugee, an alien must prove that he is unwilling or unable to return to his country of origin "because of persecution or a well-founded fear of persecution on account of race, religion, nationality, membership in a particular social group, or political opinion." 8 U.S.C. § 1101(a)(42). "An applicant alleging past persecution has the burden of establishing that (1) his treatment rises to the level of persecution; (2) the persecution was on account of one or more protected grounds; and (3) the persecution was committed by the government, or by forces that the government was unable or unwilling to control." Baghdasaryan v. Holder, 592 F.3d 1018, 1023 (9th Cir.2010).

"If past persecution is established, a rebuttable presumption of a well-founded fear arises, 8 C.F.R. § 208.13(b)(1), and the burden shifts to the government to demonstrate that there has been a fundamental change in circumstances such that the applicant no longer has a well-founded fear." Tawadrus v. Ashcroft, 364 F.3d 1099, 1103 (9th Cir.2004) (internal quotation marks omitted).

The term "particular social group" is ambiguous. Donchev v. Mukasey, 553 F.3d 1206, 1215 (9th Cir.2009). The BIA first interpreted the term "particular social group" in Matter of Acosta, 19 I. & N. Dec. 211 (BIA 1985), overruled on other grounds by Matter of Mogharrabi, 19 I. & N. Dec. 439 (BIA 1987). In Acosta, the alien argued that he was a member of a particular social group comprising members of a taxi driver cooperative in El Salvador. 19 I. & N. Dec. at 232. The BIA defined a "particular social group" as follows:

Id. at 233. Applying that definition in Acosta, the BIA rejected the alien's arguments because the identifying characteristic of the group (working as a taxi driver) was not immutable. Taxi drivers could change jobs at any time. Id. at 234.

In the years following Acosta, the BIA found that each of the following were members of a particular social group: former members of the Salvadoran national police;1 homosexuals in Cuba who were forced to register with the government;2 young female members of a tribe in Togo who had not undergone female genital mutilation and were opposed to the practice;3 and Filipinos of mixed Filipino-Chinese ancestry.4

We adopted the Acosta definition of "particular social group" in Hernandez-Montiel v. INS, 225 F.3d 1084, 1092-93 (9th Cir.2000), overruled on other grounds by Thomas v. Gonzales, 409 F.3d 1177 (9th Cir.2005) (en banc).5 In Hernandez-Montiel, we held that a particular social group "is one united by a voluntary association, including a former association, or by an innate characteristic that is so fundamental to the identities or consciences of its members that members either cannot or should not be required to change it." Id. at 1093 (emphasis in original). Applying that framework to the facts of Hernandez-Montiel, we held that male homosexuals with female sexual identities qualified as a particular social group because they shared an immutable characteristic "so fundamental to one's identity that a person should not be required to abandon [it]." Id. at 1093-94.

In 2006, the BIA refined the Acosta standard by stating that an asylum applicant must also demonstrate that his proposed particular social group has "social visibility" and "particularity." Matter of C-A-, 23 I. & N. Dec. 951, 957, 960 (BIA 2006). In C-A-, the BIA held that a group of "noncriminal drug informants working against the [Colombian] Cali drug cartel" was not a particular social group because the group did not have "social visibility" or "particularity." Id. at 961. The BIA stated that, in so holding, it was not departing from its prior precedent: "[W]e continue to adhere to the Acosta formulation." Id. at 956.

In C-A-, the BIA discussed some groups that are "understood by others to constitute social groups," id. at 959, and other groups that are "highly visible and recognizable by others in the country in question," id. at 960. The BIA rejected the proposed social group in C-A-, noting that "the very nature of the conduct at issue is such that it is generally out of the public view." Id. at 960. The BIA said that "[r]ecognizability or visibility is limited to those informants who are discovered because they appear as witnesses or otherwise come to the attention of cartel members." Id.

BIA cases following C-A- further elaborated the meaning of the additional criteria of "social visibility" and "particularity." In Matter of S-E-G-, the petitioners were three siblings from El Salvador who were threatened by Mara Salvatrucha ("MS-13") gang members after refusing gang recruitment attempts. 24 I. & N. Dec. 579, 579-80 (BIA 2008). The BIA affirmed the IJ's denial of asylum, finding that the proposed social group of "Salvadoran youth who have been subjected to recruitment efforts by MS-13 and who have rejected or resisted membership in the gang based on their own personal, moral, and religious opposition to the gang's values and activities" did not have "particularity" or "social visibility." Id. at 581, 583. The group lacked "particularity" because the category was too "amorphous" and the group membership was not easily definable. Id. at 584-85. The group was also not "socially visible": "There is little in the background evidence of record to indicate that Salvadoran youth who are recruited by gangs but refuse to join ... would be `perceived as a group' by society...." Id. at 587; see also Matter of E-A-G-, 24 I. & N. Dec. 591, 594 (BIA 2008) (reversing IJ's grant of asylum after defining "social visibility" in terms of "social perception": "respondent does not allege that he possesses any characteristics that would cause others in Honduran society to recognize him as one who has refused gang recruitment").

Following C-A- and subsequent BIA cases, we have applied the "social visibility" requirement as one of general social "perception" rather than of on-sight visibility. In Santos-Lemus v. Mukasey, we concluded that the proposed group of "young men in El Salvador resisting gang violence" was not socially visible; there was no evidence that the petitioner would be "perceived ... to be a member of any kind of anti-gang group." 542 F.3d 738, 745-46 (9th Cir.2008) (emphasis added). We similarly held in Ramos-Lopez v. Holder that Honduran men who resisted recruitment into the MS-13 were not "socially visible" because there was no evidence that they were "generally visible to society." 563 F.3d 855, 862 (9th Cir.2009).

Most circuits have accepted the BIA's "social visibility" and "particularity" criteria. See, e.g., Gaitan v. Holder, 671 F.3d 678, 681-82 (8th Cir.2012); Rivera-Barrientos v. Holder, 666 F.3d 641, 649-52 (10th Cir.2012); Scatambuli v. Holder, 558 F.3d 53, 59-60 (1st Cir.2009). But the Third and Seventh Circuits have rejected "social visibility" as an unreasonable interpretation of the ambiguous statutory term. See Valdiviezo-Galdamez v. Att'y Gen., 663 F.3d 582, 606-07 (3d Cir.2011); Gatimi v. Holder, 578 F.3d 611, 615-16 (7th Cir.2009). The Third Circuit also rejected "particularity" as merely a "different articulation[] of the [`social visibility'] concept." Valdiviezo-Galdamez, 663 F.3d at 608.

II. Factual Background and Proceedings Below

With this framework in mind, we turn now to the facts of the case. Rocio Brenda Henriquez-Rivas is a native and citizen of El Salvador. In 1998, when Henriquez-Rivas was twelve years old, her father was murdered by four members of the M-18 street gang. Two of the men were known as "Chimbera" and "Popo." Henriquez-Rivas saw the men enter her house and assault her father. They told her father to ask for forgiveness, which he did. Thinking the men were going to attack her, Henriquez-Rivas fled. As she was running away from the house, Henriquez-Rivas heard four gun shots. She did not see who fired the gun, but her sister Mirabel told her that Chimbera shot her father. When the police arrived, they told Henriquez-Rivas that her father was dead.

Henriquez-Rivas identified two of the suspects in a lineup behind protective glass. She also testified against them in court. Both Chimbera and Popo were present in court while Henriquez-Rivas testified, and both were convicted. Chimbera was sentenced to 7 years in prison because he was a minor and Popo was sentenced to 25 to 30 years in prison.

After her father's death, Henriquez-Rivas lived with her half-sister, Olga. In 2000, when Henriquez-Rivas returned to her father's house to get some paperwork, a man by the name of Julio told her not to return to the house. Julio explained that some men had been at her house and mentioned to Julio that they killed her father at the direction of someone else.

At her asylum hearing, Henriquez-Rivas testified that her family members saw Chimbera, who had been released from prison, about two times between 1999 and 2004. In 1999, one of Henriquez-Rivas' sisters saw Chimbera a block from her house and called the police. The police arrived and arrested Chimbera. Another time, Mirabel, Henriquez-Rivas' younger sister who also witnessed the murder, saw Chimbera on the bus working as a fare collector. Chimbera saw Mirabel and stared at her. For this reason, Mirabel later came to the United States. Henriquez-Rivas did not personally encounter Chimbera.

In 2005, a man came to Henriquez-Rivas' school and asked if anyone knew "Rocio Henriquez." She thought it was strange that the man would ask for her, so she denied knowing a Rocio Henriquez. Henriquez-Rivas then decided to leave El Salvador. She believed that if she remained in El Salvador the gang members would harm her for testifying against them and because they were required to pay restitution to her family. After Henriquez-Rivas left for the United States, she learned that Chimbera had come to her town asking for her.

Henriquez-Rivas entered the United States, without inspection, on January 16, 2006. The Department of Homeland Security initiated removal proceedings against her on the ground that she was "an alien present in the United States without being admitted or paroled." 8 U.S.C. § 1182(a)(6)(A)(i). She conceded removability, and filed an application for asylum, withholding of removal, and protection under the Convention Against Torture.

After a hearing, the IJ found Henriquez-Rivas' testimony credible. The IJ concluded that Henriquez-Rivas did not establish she was persecuted on account of political opinion, but concluded that Henriquez-Rivas did establish she was a member of the particular social group of "people testifying against or otherwise oppos[ing] gang members." The IJ found she had suffered past persecution in El Salvador because her father was murdered, the gang members tried to kill her when they killed her father, and she was threatened by gang members after testifying against them in court. The IJ also concluded that Henriquez-Rivas had established a well-founded fear of future persecution, as there was "a reasonable possibility of suffering such persecution [if] she were to return to El Salvador." The IJ further found that the Salvadoran government is unable to control gang violence, that the government did not prove changed country conditions, and that internal relocation would not be reasonable. The IJ granted asylum on that basis.

The BIA reversed the IJ's decision. The BIA reversed the determination that the group of "people testifying against or otherwise [opposing] gang members" constitutes a particular social group. The BIA concluded that the proposed group lacked the requisite "social visibility" to qualify as a particular social group. Henriquez-Rivas petitioned for review. A three-judge panel of our court initially denied Henriquez-Rivas' petition for review. We granted rehearing en banc.

III. Standards of Review

We review questions of law de novo. Santos-Lemus, 542 F.3d at 742. We review the BIA's factual findings for substantial evidence. Id. The BIA's construction of ambiguous statutory terms in the INA through case-by-case adjudication is entitled to deference under Chevron U.S.A., Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 844, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). INS v. Aguirre-Aguirre, 526 U.S. 415, 425, 119 S.Ct. 1439, 143 L.Ed.2d 590 (1999). If the BIA's construction is reasonable, we must accept that construction under Chevron, even if we believe the agency's reading is not the best statutory interpretation. See Nat'l Cable & Telecomms. Ass'n v. Brand X Internet Servs., 545 U.S. 967, 980, 125 S.Ct. 2688, 162 L.Ed.2d 820 (2005). The standard of review is not necessarily "more searching" if the BIA's decision represents a change from prior agency policy. FCC v. Fox Television Stations, Inc., 556 U.S. 502, 514, 129 S.Ct. 1800, 173 L.Ed.2d 738 (2009).

IV. Discussion

A.

Much of the inter-circuit disagreement over the "social visibility" requirement relates to an ambiguity in the BIA's use of the term. Does it mean that the proposed particular social group is "understood by others to constitute [a] social group[]"? C-A-, 23 I. & N. Dec. at 959 (emphasis added). Or, does it mean a bystander can literally see the difference that makes that person a member of the group (so-called "on-sight" or ocular visibility)? Cf. In re A-M-E & J-G-U-, 24 I. & N. Dec. 69, 74 (BIA 2007) (citing C-A-, 23 I. & N. Dec. at 956) (in denying particular social group status to the proposed group "affluent Guatemalans," the BIA stated that "the shared characteristic of the group should generally be recognizable by others in the community") (emphasis added). As Judge Posner stated when writing for the Seventh Circuit in Gatimi:

578 F.3d at 615. The Third Circuit similarly criticized any notion that "social visibility" was limited to persons who could be identified as members of a group just by their appearance:

Valdiviezo-Galdamez, 663 F.3d at 606-07 (alteration in original) (footnote omitted).

We agree that a requirement of "on-sight" visibility would be inconsistent with previous BIA decisions and likely impermissible under the statute. However, we do not read C-A- and subsequent cases to require "on-sight" visibility. To be sure, it is difficult to articulate precisely what the BIA meant by "social visibility" in C-A-. The BIA's elaboration in case-by-case adjudication subsequent to C-A- is somewhat inconsistent, as discussed below. But an "on-sight" visibility requirement would not make sense when coupled with the discussion in C-A- of previous BIA decisions. See C-A-, 23 I. & N. Dec. at 959-60.6 Referencing Acosta's examples of "former military leadership or land ownership" during its discussion of "social visibility," the BIA called them "easily recognizable traits." Id. Those traits would not be "easily recognizable" if the "social visibility" criterion required "on-sight" visibility, since former military officers do not always wear epaulets, nor do landowners wear T-shirts mapping their holdings. Instead, the key in these older BIA cases, as well as in C-A-, is whether the social groups are "understood by others to constitute social groups." Id. at 959 (emphasis added).

Further, the statement in C-A- that "the very nature of the conduct at issue is such that it is generally out of the public view" should be understood in the context of societal understanding, not "on-sight" visibility. Id. at 960. It is true that informants against criminal cartels generally take pains to stay out of the public view. Those informants who are discovered, however, are socially visible under C-A-: "Recognizability or visibility is limited to those informants who are discovered because they appear as witnesses or otherwise come to the attention of cartel members." Id. Thus, anti-cartel informants, who might not be recognizable on-sight as members of that group, would be socially visible — particularly to revenge-seeking cartel members — if their identity were discovered because they testified in court, as Henriquez-Rivas did here.7

Subsequent BIA cases do not interpret C-A- as imposing an "on-sight" visibility requirement. See S-E-G-, 24 I. & N. Dec. at 587 (asking whether the proposed social group of Salvadoran youth who had resisted recruitment efforts from the MS-13 gang would be "perceived as a group" by society); E-A-G-, 24 I. & N. Dec. at 594 (considering whether society would recognize respondent, a Honduran youth, as having resisted gang recruitment after defining "social visibility" as "the extent to which members of a society perceive those with the characteristic in question as members of a social group").8

Our own case law following C-A- has similarly declined to impose an "on-sight" visibility requirement. Instead, we have required that the shared characteristic "`generally be recognizable'" by other members of the community, or evidence that members of the proposed group would be "`perceived as a group' by society." Santos-Lemus, 542 F.3d at 746 (quoting S-E-G-, 24 I. & N. Dec. at 586-87).

Absent a requirement of on-sight visibility, "social visibility" as detailed in C-A- is consistent with BIA precedent prior to C-A-. It defines "social visibility" in terms of perception by a society, not ocular recognition. So construed, C-A- was merely a refinement of Acosta. So long as the "social visibility" and "particularity" criteria are applied in a way that did not directly conflict with prior agency precedent, we would be hard-pressed to reject the new criteria as unreasonable under Chevron. See Marmolejo-Campos v. Holder, 558 F.3d 903, 914 (9th Cir.2009) (en banc).

Concluding that social visibility refers to "perception" rather than "on-sight" visibility does not fully clarify the requirement. Neither we nor the BIA has clearly specified whose perspectives are most indicative of society's perception of a particular social group: the Petitioner herself? Her social circle? Her native country as a whole? The United States? The global community?9 Different audiences will be more or less likely to consider a collection of individuals as a social group depending on their own history, course of interactions with the group, and the overall context. Although we leave it to the BIA to decide this issue in the first instance, we think that some observations may be in order as it considers the issue anew.

Looking to the text of the statute, in the context of persecution, we believe that the perception of the persecutors may matter the most. Under the INA, a petitioner's belief that she has been persecuted does not alone prove persecution; rather, she must show persecution or a well-founded fear of future persecution on account of a protected ground. 8 U.S.C. § 1101(a)(42)(A). The petitioner is persecuted precisely because the persecutor recognizes the object of his persecution. Further, the petitioner's awareness of her own group status is not a baseline requirement — for example, an infant may not be aware of race, sex, or religion. Society in general may also not be aware of a particular religious sect in a remote region. However, a group may be persecuted because of the persecutor's perceptions of the existence of those groups. Cf. Sanchez-Trujillo, 801 F.2d at 1576 (holding that the proposed group, "young, urban, working class males of military age who had never served in the military," did not constitute a particular social group, we suggested that "a persecutor's perception of a segment of a society as a `social group'" could be relevant to the particular social group analysis).10 We do not mean to imply that an alien should be required in every case to prove that his persecutors perceived his social group to be socially visible. When there is evidence that a social group is visible to society, there is no need to prove that the petitioner's persecutors perceived that group as visible. See Matter of R-A-, 22 I. & N. Dec. 906, 918 (BIA 2001) (en banc) (noting that a "showing of how the characteristic is understood in the alien's society ... may [help us to] understand that the potential persecutors in fact see persons sharing the characteristic as warranting suppression or the infliction of harm.").11 We mean only to suggest that evidence of perceptions in society as a whole is not the exclusive means of demonstrating social visibility. When a particular social group is not visible to society in general (as with a characteristic that is geographically limited, or that individuals may make efforts to hide), social visibility may be demonstrated by looking to the perceptions of persecutors. Such perceptions may be highly relevant to, or even potentially dispositive of, the question of social visibility. Cf. Sanchez-Trujullo, 801 F.2d at 1576 & n. 7.

By highlighting the perception of the persecutor, other demographic divisions would become less relevant. One would ask whether, as far as the persecutor is concerned, there is a particular characteristic (such as male homosexuals with female sexual identities, see Hernandez-Montiel v. INS, 225 F.3d 1084, 1093 (9th Cir.2000)), that defines a finite collection of individuals as a group. If the answer is yes, the fact that those individuals may have a variety of other characteristics, and belong to various other groups, would not be a bar to potential relief.

We next consider the "particularity" requirement. Admittedly, both BIA and our own precedent have blended the "social visibility" and "particularity" analysis: "social visibility" has been determined based on perception, as discussed above, and "particularity" too has been based on society's perception whether a group has delimitable boundaries. See, e.g., Ramos-Lopez v. Holder, 563 F.3d 855, 861 (9th Cir. 2009) (affirming the BIA's determination that petitioner, asserting membership in a particular social group comprised of Honduran males who had resisted MS-13 recruitment efforts, failed the particularity requirement because there was no indication that there was "any perception that the males in question were members of a class") (internal quotation marks omitted); S-E-G-, 24 I. & N. Dec. at 584 ("The essence of the `particularity' requirement... is whether the proposed group can accurately be described in a manner sufficiently distinct that the group would be recognized, in the society in question, as a discrete class of persons." (emphasis added)). Consideration of the "plethora of different lifestyles, varying interests, diverse cultures, and contrary political leanings" among the members of the group also seems to pertain to "social visibility" rather than "particularity": the point is that the diverse backgrounds of members of the purported group may prevent them from being perceived as belonging to the same group. Soriano v. Holder, 569 F.3d 1162, 1166 (9th Cir.2009) (quoting Sanchez-Trujillo, 801 F.2d at 1577) (internal quotation marks omitted).12 It is therefore unsurprising that, given the way the BIA has applied the term, the Third Circuit has concluded that "`[p]articularity' appears to be little more than a reworked definition of `social visibility.'" Valdiviezo-Galdamez, 663 F.3d at 608.

We will not go quite so far. The "particularity" requirement is separate, and it is relevant in considering whether a group's boundaries are so amorphous that, in practice, the persecutor does not consider it a group. The ultimate question is whether a group "can accurately be described in a manner sufficiently distinct that the group would be recognized, in the society in question, as a discrete class of persons." S-E-G-, 24 I. & N. Dec. at 584. If a persecutor does not actually rely on specific boundaries or definitions to identify the group, it may be more difficult to believe that a collection of individuals is in fact perceived as a group. Ultimately, the "particularity" consideration is merely one factor as to whether a collection of individuals is considered to be a particular social group in practice.

We clarify the "social visibility" and "particularity" criteria without reaching the ultimate question of whether the criteria themselves are valid. The BIA could find that Henriquez-Rivas' proposed social group is cognizable under either the Acosta immutability standard or the newer standard that considers "social visibility" and "particularity." Thus, we need not decide, in this case, at this time, whether we should align ourselves with the Third and Seventh Circuits and invalidate these requirements.

B.

We now turn to reviewing the BIA's application of its particular social group precedent to the facts of this case. The BIA concluded that the proposed social group of people who testified against gang members "lacks the requisite `social visibility' to be considered a particular social group within the meaning of the Act." In so doing, the BIA did not fully explain its position but cited many of the cases discussed above, including Santos-Lemus v. Mukasey, Matter of E-A-G-, Matter of S-E-G-, and Matter of C-A-. From E-A-G-, the BIA derived the principle that the evidence must "establish that members of society, or even gang members themselves, would perceive those opposed to gang membership as members of a social group."

In denying Henriquez-Rivas asylum because of a lack of "social visibility," the BIA failed to follow its own precedent as stated in C-A- and its progeny.13 This case clearly falls within the language in C-A- holding that those who testify against cartel members are socially visible: "[V]isibility is limited to those informants who are discovered because they appear as witnesses or otherwise come to the attention of cartel members." 23 I. & N. Dec. at 960 (emphasis added). Here, Henriquez-Rivas testified in open court against the gang members who killed her father.14 What is more, Chimbera and Popo were present in the courtroom while Henriquez-Rivas testified. Further, Henriquez-Rivas testified that Chimbera or other supposed MS members came to her home looking for Henriquez-Rivas and her sister. Thus, under the terms of C-A-, the proposed social group of those who testified in court against gang members "involve[s] characteristics that [are] highly visible and recognizable by others in the country in question." Id. Because of C-A-'s clear language about the "social visibility" of those informants who testify in court, there is no substantial evidence to support the BIA's conclusion that Henriquez-Rivas' proposed particular social group lacks "social visibility."

Because the BIA erroneously assumed that the proposed social group was not cognizable under its precedent, it failed to consider significant evidence that Salvadoran society recognizes the unique vulnerability of people who testify against gang members in criminal proceedings, because gang members are likely to target these individuals as a group. See id. at 959 (considering whether the proposed group is "generally easily recognizable and understood by others to constitute [a] social group[]"). Notably, as Henriquez-Rivas cited in her briefing before the BIA as well as in her opening brief on petition for review, the Salvadoran legislature enacted a special witness protection law in 2006 to protect people who testify against violent criminal elements, such as MS, in Salvadoran court. Decreto No. 1029/2006, Ley Especial para la Protección de Víctimas y Testigos ["Special Law for Victim and Witness Protection"], (May 11, 2006).15 See U.S. Dep't of State, Country Report on Human Rights Practices: El Salvador at 6 (2006); see also MCA, Inc. v. United States, 685 F.2d 1099, 1103 n. 12 (9th Cir.1982) (noting that this court may take judicial notice of foreign laws under Fed. R.Civ.P. 44.1 if the parties give written notice of their intent to raise an issue of foreign law). It is difficult to imagine better evidence that a society recognizes a particular class of individuals as uniquely vulnerable, because of their group perception by gang members, than that a special witness protection law has been tailored to its characteristics.

Our prior cases denying asylum to those opposed to gangs are distinguishable. Several of our previous cases considered proposed social groups of those generally opposed to gangs or resistant to gang recruitment. See Santos-Lemus, 542 F.3d at 746; Ramos-Lopez, 563 F.3d at 861. Those cases did not involve the very specific situation of testifying against gang members in court, and considered only generalized opposition to gangs and gang recruitment. The "opposition to gangs" group might not be "socially visible" if the society in question does not perceive those with such views as constituting a distinct group of persons. But for those who have publicly testified against gang members, their "social visibility" is apparent. To the extent that Santos-Lemus, Ramos-Lopez, and related cases mischaracterized the "social visibility" requirement by requiring "on-sight" visibility, they are no longer good law.

It is not clear whether the BIA in this case held that Henriquez-Rivas' proposed particular social group lacked "particularity." The BIA merely stated, "defining the group as persons opposing gang members is too amorphous." This was an incorrect statement of Henriquez-Rivas' proposed social group, which referred to those who had testified against M-18 gang members in open court, and thus, "can accurately be described in a manner sufficiently distinct that the group would be recognized, in the society in question, as a discrete class of persons." S-E-G-, 24 I. & N. Dec. at 584. Membership in Henriquez-Rivas' proposed group can be easily verified — and thus delimited — through court records documenting group members' testimony. The IJ found Henriquez-Rivas' testimony credible, and the government does not dispute that Henriquez-Rivas actually testified in court in El Salvador against members of the MS gang.

Our previous cases rejecting as a "particular social group" those acting as government informants are arguably in conflict with our holding today insofar as they require an additional element of shared birth, racial or ethnic origin, or some other innate aspect of homogeneity for the group to qualify as a "particular social group." In Soriano v. Holder, petitioner contended that he had a well-founded fear of future persecution in the Philippines because he had acted as a police informant against a Filipino criminal gang while living in Los Angeles. 569 F.3d at 1163. We denied his petition for review, thus affirming the BIA's denial of asylum for lack of "particularity":

Id. at 1166 (internal citation, quotation marks, and brackets omitted); see also Velasco-Cervantes v. Holder, 593 F.3d 975, 978 (9th Cir.2010) (rejecting a proposed group of former material witnesses for the United States government for lack of "particularity" because "any person of any origin can be involuntarily placed in that role in any type of legal proceeding" (emphasis added)).

These cases reflect the confusion between the "particularity" and "social visibility" requirements. The diversity of "lifestyles" and "origin" to which these cases refer did not concern the "particularity" requirement per se, nor are they relevant to our analysis, as we explain above. Accordingly, to the extent that Soriano and Velasco-Cervantes make considerations of diversity of lifestyle and origin the sine qua non of "particularity" analysis, they are overruled.

Because we grant the petition on the basis that the BIA erred in applying its own precedents in deciding whether Henriquez-Rivas was a member of a particular social group, we need not reach the question whether the BIA erred when it failed to consider Henriquez-Rivas' argument that she was persecuted on account of political opinion. Nor do we reach the issues of withholding of removal or protection under the Convention Against Torture.

The petition for review is GRANTED, the BIA's decision is VACATED, and the case is REMANDED for further proceedings.

McKEOWN, Circuit Judge, concurring:

I concur in the result and the opinion, except to the extent the majority counsels that the perception of the persecutor "may matter the most" in analyzing social visibility or claims that the persecutor's view is "potentially dispositive" of the question. On this point, Chief Judge Kozinski has the better argument. See Matter of E-A-G-, 24 I. & N. Dec. 591, 594 (BIA 2008) (describing "social visibility" as "the extent to which members of a society perceive those with the characteristic in question as members of a social group") (emphasis added); In re A-M-E-, 24 I. & N. Dec. 69, 74 (BIA 2007) (noting that the 2002 guidelines of the United Nations High Commissioner for Refugees "endorse an approach in which an important factor is whether the members of the group are `perceived as a group by society'") (emphasis added). Consistent with using society's perspective as a baseline, training materials for asylum officers — who make the first determination on eligibility for applicants affirmatively seeking asylum — instruct that the social visibility "requirement can be met by showing that members of the group possess a trait or traits that make the members recognizable or distinct in the society in question."1

Defining social visibility from the perspective of society better comports with the case law; perhaps just as importantly, it also makes common sense. As the Chief Judge points out, "[d]efining a social group in terms of the perception of the persecutor risks finding that a group exists consisting of a persecutor's enemies list." See also Mendez-Barrera v. Holder, 602 F.3d 21, 27 (1st Cir.2010) ("The relevant inquiry is whether the social group is visible in the society, not whether the alien herself is visible to the alleged persecutors."). To the extent the BIA's prior decisions are ambiguous as to whose perspective is critical in assessing social visibility, we should — as the majority recognizes — leave that determination to the BIA in the first instance. The BIA is not in need of our advisory opinion on the subject.

Chief Judge KOZINSKI, with whom Judge BYBEE joins, dissenting:

In summarily reversing us just six years ago, the Supreme Court held that it's the BIA, not we, who must decide whether a petitioner is a member of a particular social group for purposes of asylum. Gonzales v. Thomas, 547 U.S. 183, 126 S.Ct. 1613, 164 L.Ed.2d 358 (2006) (per curiam). The Court quoted approvingly the Solicitor General's cert petition for the proposition that "a court's role in an immigration case is typically one of `review, not of first view.'" Id. at 185, 126 S.Ct. 1613 (internal quotation marks omitted). The majority today forgets this admonition and engages in a good deal of first viewing, in clear contravention of Thomas and INS v. Orlando Ventura, 537 U.S. 12, 16-17, 123 S.Ct. 353, 154 L.Ed.2d 272 (2002). Along the way, it deepens a circuit conflict on an issue where national uniformity is vital, and sows uncertainty into our circuit law where previously there was clarity. The far wiser course would be for us to vacate the order taking the case en banc as improvidently granted and reinstate the three-judge panel's disposition.

1. Congress has given the Attorney General discretion to grant asylum to certain limited classes of aliens — those who have a well-founded fear of persecution in their home countries "on account of race, religion, nationality, membership in a particular social group, or political opinion." 8 U.S.C. §§ 1101(a)(42), 1158(b)(1)(A). Of these, persecution on account of membership in a particular social group is by far the most amorphous. See Maj. op. at 1083; Stanley D. Radtke, Defining a Core Zone of Protection in Asylum Law, 10 J.L. & Soc. Challenges 22, 29 (2008). Race, nationality, religious affiliation and political opinion can usually be determined by objective evidence, but what constitutes a social group requires a judgment about shared norms and perceptions in the society where the individual is living.

The asylum statute uses the term "particular social group" but gives no clue as to what it means. There are no committee reports, hearing transcripts or floor statements that shed light on that obscure phrase. See Maryellen Fullerton, A Comparative Look at Refugee Status Based on Persecution Due to Membership in a Particular Social Group, 26 Cornell Int'l L.J. 505, 513-14 (1993). What we do know is that the language came from the 1951 United Nations Convention Relating to the Status of Refugees, July 28, 1951, 189 U.N.T.S. 137, by way of the 1967 United Nations Protocol Relating to the Status of Refugees, Jan. 31, 1967, 19 U.S.T. 6223, 606 U.N.T.S. 267, which the United States ratified in 1968. See Sanchez-Trujillo v. INS, 801 F.2d 1571, 1575 (9th Cir.1986). Congress intended that the Act's refugee definition "be interpreted in conformance with the [1967] Protocol's definition." See INS v. Cardoza-Fonseca, 480 U.S. 421, 436-37, 107 S.Ct. 1207, 94 L.Ed.2d 434 (1987). But, the term "social group" was added to the Convention last minute and without discussion, so there is little extrinsic evidence as to what it means. Radtke, supra, at 32. Advice from the United Nations has come only ex-post. See, e.g., U.N. High Comm'r for Refugees, Guidelines on International Protection: "Membership of a particular social group" within the context of Article 1A(2) of the 1951 Convention and/or its 1967 Protocol relating to the Status of Refugees, U.N. Doc. HCR/GIP/02/02 (May 7, 2002).

Because of this indeterminacy in the drafting process, the United States, along with other developed countries, has had to struggle to give meaning to a term that has little pedigree of its own. The process has, of necessity, involved case-by-case adjudication and called for periodic adjustment as our understanding of the term has evolved in light of experience.

What agglomeration of people a society recognizes as a group is seldom written down and must typically be deduced from inconclusive evidence as to how individuals sharing common characteristics are viewed by those around them. Take as an example a society we know something about — the United States. I'd guess most people consider the following to be identifiable social groups: Vietnam veterans, male homosexuals, college students, lawyers, Masons, cancer survivors, blind people, Cajuns, practitioners of Falun Gong and hippies. And the following groups of people who have something in common would, nevertheless, probably not be viewed as social groups: left-handed people, high school dropouts, blondes, crime victims, disabled people, dog owners, second-born children and haters of broccoli.

This leads to several observations: First, it's debatable whether some of the proposed groups should be on the first or second list. Crime victims may be too broad to define a social group, but what about victims of rape or domestic violence, or families of persons who were killed or maimed by drunk drivers? Second, the appropriate granularity of any group definition is subject to debate. Are Vietnam veterans a group on their own or are they merely part of the larger group of combat veterans? Do male homosexuals make up a separate social group or are they part of a larger social group that includes female homosexuals and transgendered persons? Third, many — perhaps most — people in some social groups may not identify themselves as part of any such group. Masons choose to be Masons, but not all homosexuals identify themselves by their sexual orientation, and some may wish to conceal their sexual identity. Nevertheless, if society sees homosexuals as a distinct social group, individuals with that sexual orientation will be treated as involuntary members, if and when their sexual orientation is discovered.

As these examples illustrate, determining whether someone is a member of a social group, even within our own society, is no mean task. The matter is further complicated when the question is whether another society, with a culture and language different from our own, considers some subset of its population to be a social group. The question is quite important, however, because any number of people sharing a characteristic could be considered a social group. The group may be as small as two and as large as a majority of the population, paving the way for huge numbers of people to obtain political asylum. Yet, Congress surely didn't mean to open the immigration floodgates to everyone in the world who is oppressed. Indeed, the Guidelines to the U.N. Protocol state quite clearly that "the social group category was not meant to be a `catch all' applicable to all persons fearing persecution." In re C-A-, 23 I. & N. Dec. 951, 960 (BIA 2006) (quoting U.N. High Comm'r for Refugees, Guidelines on International Protection, supra).

The United States received 74,000 asylum applications last year, half again as many as any other industrialized nation. See U.N. High Comm'r for Refugees, Asylum Levels and Trends in Industrialized Countries 3 (2011), available at http://www.unhcr.org/4e9beaa19.html. Confronted with the difficult and sensitive task of determining whether individuals seeking asylum are persecuted on the basis of membership in a large variety of proposed social groups,1 the BIA has announced and refined the standard for dealing with such issues. In In re Acosta, 19 I. & N. Dec. 211 (BIA 1985), the BIA took a first pass at the issue by holding that persecution on account of membership in a particular social group means "persecution that is directed toward an individual who is a member of a group of persons all of whom share a common, immutable characteristic.... [This characteristic] must be one that the members of the group either cannot change, or should not be required to change because it is fundamental to their individual identities or consciences." Id. at 233. Acosta made it clear that "[t]he particular kind of group characteristic that will qualify under this construction remains to be determined on a case-by-case basis." Id. The BIA's decisions on this issue are entitled to Chevron deference, and more. See INS v. Aguirre-Aguirre, 526 U.S. 415, 424-25, 119 S.Ct. 1439, 143 L.Ed.2d 590 (1999).

Over the years, the BIA has worked diligently and thoughtfully to refine the definition of membership in a social group. Along the way, it's announced a series of criteria and limitations, all designed to give the term "social group" a concrete and consistent meaning that doesn't give a free pass into the United States to anyone and everyone who is persecuted in his home country. These criteria include the following:

• A past experience is an immutable characteristic because "it has already occurred and cannot be undone." C-A-, 23 I. & N. Dec. at 958. But not every kind of "past experience that may be shared by others suffices to define a particular social group for asylum purposes." Id.; In re S-E-G-, 24 I. & N. Dec. 579, 584 (BIA 2008).

• The characteristic common to the proposed group must be such that the society in which it's situated recognizes individuals having that characteristic as constituting a distinct social group. See C-A-, 23 I. & N. Dec. at 959-60; S-E-G-, 24 I. & N. Dec. at 586. The BIA sometimes refers to this as the "social visibility" requirement. See C-A-, 23 I. & N. Dec. at 959.

• "[A] social group cannot be defined exclusively by the fact that its members have been subjected to harm, ... [but] this may be a relevant factor in considering the group's visibility in society." A-M-E, 24 I. & N. Dec. at 74.

• It isn't necessary to show a "voluntary associational relationship among the group members ... [nor] an element of cohesiveness or homogeneity among group members." C-A-, 23 I. & N. Dec. at 956-57 (internal quotation marks omitted).

• "Whether a proposed group has a shared characteristic with the requisite `social visibility' must be considered in the context of the country of concern and the persecution feared." A-M-E, 24 I. & N. Dec. at 74. The BIA has "considered as a relevant factor the extent to which members of a society perceive those with the characteristic in question as members of a social group." C-A-, 23 I. & N. Dec. at 957.

• The risk faced by the group in question must be specific to the group and not one shared by society at large, see A-M-E, 24 I. & N. Dec. at 75, or by anyone who stands in the way of the persecutors. See C-A-, 23 I. & N. Dec. at 960-61; S-E-G-, 24 I. & N. Dec. at 587 ("However, such gangs have directed harm against anyone and everyone perceived to have interfered with, or who might present a threat to, their criminal enterprises and territorial power. The respondents are therefore not in a substantially different situation from anyone who has crossed the gang, or who is perceived to be a threat to the gang's interests.").

• "[M]embership in a purported social group requires that the group have particular and well-defined boundaries...." S-E-G-, 24 I. & N. Dec. at 582. "The essence of th[is] `particularity' requirement... is whether the proposed group can accurately be described in a manner sufficiently distinct that the group would be recognized, in the society in question, as a discrete class of persons." Id. at 584.

• The persecution that petitioners seek to escape must be motivated by their membership in the group in question and not by factors that are "quite apart" from such membership. Id. at 585. The BIA explains the difference by giving this example:

These requirements can be summarized as follows: In order to establish eligibility for asylum based on persecution for membership in a social group, a petitioner must show that he's part of a group that's well-defined by a characteristic other than the fact that its members have been subjected to harm; is recognized within the society as a distinct group; is described with sufficient particularity so that it's possible to determine with reasonable certainty who's included in the group; and whose members are targeted for persecution because of their membership in the group, not on account of some other, perhaps closely associated, trait.

2. We approved the BIA's approach in Ramos-Lopez v. Holder, 563 F.3d 855 (9th Cir.2009). Petitioner in that case was a Honduran national who had refused recruitment by the Mara Salvatrucha gang, or MS-13, which subsequently threatened to kill him. See id. at 856. He came to the United States and claimed asylum on account of persecution on the basis of membership in a particular social group, namely young Honduran men who have been recruited by MS-13, but who refuse to join. Id. The BIA, relying on S-E-G-, determined that there was no such particular social group. Id. at 859-62.

Judge Tashima's well-reasoned opinion in Ramos-Lopez (unlike the majority today) starts with the framework set out by the Supreme Court in Thomas: "`The matter requires determining the facts and deciding whether the facts as found fall within a statutory term.'" Id. at 859 (quoting Thomas, 547 U.S. at 186, 126 S.Ct. 1613). Ramos-Lopez also gives due deference to our en banc opinion in Marmolejo-Campos v. Holder, 558 F.3d 903 (9th Cir.2009) (en banc), (as the majority here does not) and concludes that the BIA's construction of particular social group is entitled to Chevron deference. Ramos-Lopez, 563 F.3d at 858-60 & n. 4. Deferring to the BIA, we noted that prior gang recruitment was a past experience that couldn't be changed, but that such "`shared past experience [does not necessarily] suffice[] to define a particular social group for asylum purposes.'" Id. at 860 (quoting S-E-G-, 24 I. & N. Dec. at 584) (alterations in original). Hewing closely to the BIA's reasoning, we noted that "`gang violence and crime in El Salvador [and Honduras] appear to be widespread, and the risk of harm is not limited to young males who have resisted recruitment... but affects all segments of the population.'" Id. at 860-61 (quoting S-E-G-, 24 I. & N. Dec. at 587). Significantly, we relied on the BIA's observation "that those who have resisted recruitment are `not in a substantially different situation from anyone who has crossed the gang, or who is perceived to be a threat to the gang's interests.'" Id. (quoting S-E-G-, 24 I. & N. Dec. at 587).

Most other circuits have deferred to the BIA's interpretation of particular social group. See Orellana-Monson v. Holder, 685 F.3d 511, 521 (5th Cir.2012); Gaitan v. Holder, 671 F.3d 678, 680-82 (8th Cir. 2012); Rivera-Barrientos v. Holder, 666 F.3d 641, 647-53 (10th Cir.2012); Lizama v. Holder, 629 F.3d 440, 444-48 (4th Cir. 2011); Scatambuli v. Holder, 558 F.3d 53, 59-60 (1st Cir.2009); Ucelo-Gomez v. Mukasey, 509 F.3d 70, 73-74 (2d Cir.2007); Castillo-Arias v. U.S. Attorney Gen., 446 F.3d 1190, 1197 (11th Cir.2006); Castellano-Chacon v. INS, 341 F.3d 533, 546 (6th Cir.2003).

Two circuits have taken a contrary view. In an opinion that pays lip service to Thomas but, in fact, usurps the role of the BIA, the Seventh Circuit rejected the Board's social visibility requirement. Gatimi v. Holder, 578 F.3d 611, 615-16 (7th Cir.2009). The court's reasons for doing so are obscure but, best I can tell, the Seventh Circuit confused social visibility with on-sight visibility, and criticized the BIA for requiring that social groups be identifiable on sight:

Id. at 615.

This criticism is unfounded. As even the majority here recognizes, the BIA's social visibility requirement doesn't mean that the characteristics defining the group must be recognizable on sight. Maj. op. at 1087-89. The BIA has made it perfectly clear that the social visibility test is designed to determine whether the proposed group of which petitioner claims to be a member is perceived as a group by the society in question, not whether individual members of the group can be identified on sight. See C-A-, 23 I. & N. Dec. at 959-61. That's why it's called social visibility rather than just visibility.

The Seventh Circuit also criticized the BIA for inconsistency with certain cases it had decided some two decades earlier. Gatimi, 578 F.3d at 615-16. It blamed the BIA for forging a new direction "without repudiating the other line of cases." Id. at 616. But, assuming the other cases were inconsistent with the more recent ones — and I'm not convinced they were — an agency is not bound to retain the same interpretation in perpetuity; it may — indeed it should — adapt its approach in light of experience. See FCC v. Fox Television Stations, Inc., 556 U.S. 502, 514-15, 129 S.Ct. 1800, 173 L.Ed.2d 738 (2009); Nat'l Cable & Telecomms. Ass'n v. Brand X Internet Servs., 545 U.S. 967, 981, 125 S.Ct. 2688, 162 L.Ed.2d 820 (2005). Nor is the agency required to recite an incantation when it modifies an existing approach or adopts a new one. It's enough that the agency announces a new policy, gives reasons for it and thereafter applies it consistently. See Fox Television, 556 U.S. at 514-15, 129 S.Ct. 1800; Brand X, 545 U.S. at 981, 125 S.Ct. 2688. It's when the agency treats some parties before it one way while treating others, similarly situated, differently that the problem of arbitrary enforcement arises. The Seventh Circuit pointed to no such arbitrary enforcement on the part of the BIA, so it's hard to understand what the Seventh Circuit was so grumpy about.

In Valdiviezo-Galdamez v. Attorney General of the United States, 663 F.3d 582 (3d Cir.2011), a divided panel struck down the social visibility requirement, largely channelling the Seventh Circuit's opinion in Gatimi. Id. at 603-07. Doing the Seventh Circuit one better, the Third Circuit also struck down the BIA's particularity requirement. Id. at 608. According to the Third Circuit majority, "the BIA's addition of the requirements of `social visibility' and `particularity' to its definition of `particular social group' is inconsistent with its prior decisions, and the BIA has not announced a `principled reason' for its adoption of those inconsistent requirements." Id. Judge Hardiman wrote separately, recognizing the reasonableness of the social visibility and particularity requirements and opining that the BIA was entitled to adopt both on remand. Id. at 612, 615 (Hardiman, J., concurring).

The Tenth Circuit considered the criticisms leveled at the BIA by the Seventh Circuit and roundly rejected them in a lucid opinion by Judge Tymkovich. Rivera-Barrientos, 666 F.3d 641. The Tenth Circuit noted that

Id. at 649; see also id. at 651-52 (expressly disagreeing with the Seventh Circuit's suggestion that social visibility requires that "the relevant trait be visually or otherwise easily identified" and giving examples from past BIA decisions). Bravo to the Tenth!

3. Our circuit has given proper deference to the BIA's social visibility and particularity requirement in Ramos-Lopez and numerous other cases. See, e.g., Donchev v. Mukasey, 553 F.3d 1206, 1216-17 (9th Cir.2009); Soriano v. Holder, 569 F.3d 1162, 1165-66 (9th Cir.2009); Santos-Lemus v. Mukasey, 542 F.3d 738, 744-46 (9th Cir.2008). The present case involves a straightforward application of the principles announced in those cases and so could have been decided in an unpublished disposition — as, in fact, it was. Henriquez-Rivas v. Holder, 449 Fed.Appx. 626 (9th Cir.2011). And there it might have remained, but for a concurrence by the author of the current majority opinion, joined by a visiting judge who happened to be a member of the panel that issued the misbegotten Seventh Circuit Gatimi opinion. The concurrence quotes Gatimi approvingly, echoes the charges of inconsistency and irrationality leveled by the Third and Seventh Circuits, and calls for en banc re-evaluation of our circuit law concerning political asylum based on membership in a particular social group. Id. at 628-33. Not surprisingly, the full court obliged, so here we are.

This brings us to today's opinion, which starts off well enough but then quickly and deeply falls into error. The majority first tackles the question of whether the social visibility requirement calls for ocular recognition or perception by society. Maj. op. at 1087-89. The opinion goes on for much longer than necessary but eventually reaches the obvious — and right — conclusion that social visibility refers to social perceptions, not to on-sight recognition. Id. Thankfully, my colleagues avoid the pitfall of Gatimi, which they quote prominently, id. at 1087, but eventually forsake.2

The panel then goes on to make some serious mistakes of its own. The first of these concerns who must perceive the petitioner as belonging to a particular social group. Maj. op. at 1088-89. Using circular reasoning, the opinion concludes that "the perception of the persecutors may matter the most." Maj. op. at 1089. That's three mistakes right there. See generally Conc. op. (McKeown, J.). First, it's not our call; it's the BIA's. See id. at 1094. If there's any doubt whether the BIA has decided the issue, the most we can do is point out the problem and remand for the BIA to resolve the question in the first instance. See Ventura, 537 U.S. at 16, 123 S.Ct. 353; Montes-Lopez v. Gonzales, 486 F.3d 1163, 1165 (9th Cir. 2007). Only after it does, can we decide whether that construction is reasonable; never do we get to decide such a question of interpretation in the first instance. By casting its conclusion as mere "observation" or "suggest[ion]," the majority attempts an end run around Ventura. Maj. op. at 1089, 1089-90. But observations and suggestions that go on for more than six hundred words and three footnotes amount to much more than random thoughts. The message to the BIA is clear: We have decided; please do us all the favor of falling in line now.

Second, the BIA has decided the question: "[W]e referred to the 2002 guidelines of the United Nations High Commissioner for Refugees, which endorse an approach in which an important factor is whether the members of the group are perceived as a group by society." A-M-E, 24 I. & N. Dec. at 74 (internal quotation marks omitted) (emphasis added); see also Conc. op. at 1094 (McKeown, J.). The majority's determination that it's the perception of the persecutor that matters is contrary to the approach of the BIA. This we may not do without first finding that the BIA's interpretation is incompatible with the language of the statute, which of course it's not.

Third, and worst of all, the majority's conclusion that it's the perception of the persecutor that matters is at loggerheads with the BIA's repeated admonition that "a social group cannot be defined exclusively by the fact that its members have been subjected to harm." A-M-E, 24 I. & N. Dec. at 74 (relying on the Guidelines of the U.N. High Commissioner for Refugees); see also Conc. op. at 1094 (McKeown, J.). Defining a social group in terms of the perception of the persecutor risks finding that a group exists consisting of a persecutor's enemies list. The BIA's approach is to determine whether petitioner has met his burden of showing that the society in question recognizes him as a member of a social group, and then to ask whether he is persecuted on account of his membership in that group. We are bound to follow the methodology adopted by the agency, not invent our own.

But it gets worse. The majority vacates the BIA's determination that Henriquez-Rivas failed to show that she was subjected to persecution on account of her membership in the group of people testifying against gang members because the proposed group lacks social visibility: "[P]eople testifying against gang members is merely a shared experience and not a particular social group within the meaning of the Act." Henriquez-Rivas, No. A098 660 718, at 1 (BIA May 1, 2009). The BIA also held that "defining the group as persons opposing gang members is too amorphous." Id.

My colleagues vacate the BIA's social visibility determination on the ground that it is, in their view, inconsistent with the BIA's "own precedent as stated in C-A- and its progeny." Maj. op. at 1091. According to the majority, "[t]his case clearly falls within the language in C-A- holding that those who testify against cartel members are socially visible: `[V]isibility is limited to those informants who are discovered because they appear as witnesses or otherwise come to the attention of cartel members.'" Id. at 1092 (quoting C-A-, 23 I. & N. Dec. at 960) (emphasis omitted). The lone phrase from C-A-, which is the fulcrum of the majority's reasoning, simply will not bear the weight.

To begin with, the sentence describes a necessary condition for social visibility, but the BIA nowhere says it's sufficient. C-A- was one of the earliest cases where the BIA elucidated the social visibility requirement by considering the proposed group of non-criminal informants. In finding that the proposed group isn't socially visible, the Board disposed of the case on the ground that the members of the group weren't identifiable at all, and thus couldn't be the targets of persecution. See C-A-, 23 I. & N. Dec. at 960-61.

But that's not all the BIA said in C-A-. On the same page where it used the language on which my colleagues rely, the BIA set forth a further ground for rejecting the asylum application:

C-A-, 23 I. & N. Dec. at 960-61 (internal quotation marks omitted); see also Fatin v. INS, 12 F.3d 1233, 1240 (3d Cir.1993). This rationale speaks directly to whether a petitioner is the subject of persecution on account of his group membership. If someone is persecuted on grounds that are closely associated with group membership but also apply to many others in society, then the persecution is not on account of membership in a particular social group. The BIA has made it clear in subsequent cases that this is a limitation on social visibility. See, e.g., A-M-E, 24 I. & N. Dec. at 75.

When reviewing agency decisions for consistency, we aren't allowed to cherry-pick stray sentences from the agency's past opinions and ignore others. Nor may we insist that the agency stand by its rulings in perpetuity. An agency charged with administering an ambiguous statute is entitled to change its mind. See Fox Television, 556 U.S. at 514-15, 129 S.Ct. 1800.

Assuming that the BIA was at some point in the past bound by the stray sentence on which the majority relies, the agency has long since modified that position. In S-E-G-, for example, the BIA dealt with a proposed social group consisting of youths who resisted gang recruitment efforts. 24 I. & N. Dec. at 582. The gangs obviously knew who refused their recruitment efforts, and the fear of retaliation arose because of that knowledge. In that regard, the petitioners in S-E-G- were in precisely the same position as individuals who testified openly against gangs (like petitioner here). Yet the Board held that this wasn't sufficient to meet the social visibility requirement: "[Y]outh who have been targeted for recruitment by, and resisted, criminal gangs may have a shared past experience, which, by definition, cannot be changed. However, this does not necessarily mean that the shared past experience suffices to define a particular social group for asylum purposes." Id. at 584.3

4. Having found fault with the BIA's determination that petitioner's proposed group lacks social visibility, the majority goes on to make findings of its own. This is wrong. We've been told time and again that we aren't fact-finders; we're reviewers. See Ventura, 537 U.S. at 16, 123 S.Ct. 353. And, when it comes to the difficult, sensitive and fact-intensive question as to whether petitioner is a member of a particular social group, "[t]he matter requires determining the facts and deciding whether the facts as found fall within a statutory term." Thomas, 547 U.S. at 186, 126 S.Ct. 1613. The proper course in such circumstances is to apply the "ordinary `remand' rule," and refer the matter back to the agency in light of whatever views of the law we may express in our opinion. Id. at 187, 126 S.Ct. 1613. Yet my colleagues defy the Supreme Court on a point where we've been summarily reversed twice before. Id. (citing Ventura, 537 U.S. 12, 123 S.Ct. 353). Pretty gutsy.

Jumping head first into the fact-finding process, the majority makes a mess of it. According to the opinion, the BIA here

Maj. op. at 1092 (emphasis added) (internal citations omitted). But the BIA did explicitly consider and reject precisely this purported evidence: "We are unpersuaded by the respondent's apparent attempt to equate El Salvador's enactment of a witness protection law in that country to the definition of refugee under United States immigration law." Henriquez-Rivas, No. A098 660 718, at 1-2.

Furthermore a class of individuals can be identified as uniquely vulnerable and given legal protection, yet not constitute a social group. We have laws recognizing the unique vulnerability of children, see, e.g., Child Abuse Prevention and Treatment Act, 42 U.S.C. § 5101 et seq., yet we don't consider all children to constitute a particular social group. We have laws protecting individuals who are disabled — physically and mentally — see, e.g., Americans with Disabilities Act, 42 U.S.C. § 12101 et seq., yet we wouldn't say that blind people, people in wheelchairs, those who are hearing-impaired and those who have Down syndrome or autism all make up a social group. We have laws giving special protection to victims of crime, such as the Victim and Witness Protection Act of 1982, Pub.L. No. 97-291, 96 Stat. 1248 (1982), but all victims of crime don't make up a big social group. The list is endless.

This isn't to say that legal protection never coincides with group status; the legislature may well pass laws that protect what's otherwise recognized as a social group, such as veterans or school teachers. But whether this is the case is a question of fact. And, as my colleagues seem to have forgotten, we aren't fact-finders. When the majority says "[i]t is difficult to imagine better evidence [of social visibility]," maj. op. at 1092 (emphasis added), it's pretty much conceding that that's what it's doing — and doing it in direct contravention of the BIA's own finding on the same issue. The only way we could vacate the BIA's decision on this point is if we concluded that it's "illogical, implausible, or without support in inferences that may be drawn from facts in the record." Cf. United States v. Hinkson, 585 F.3d 1247, 1251 (9th Cir.2009) (en banc).

Can we honestly say that there is no support in the record for the BIA's finding? Hardly. As the BIA has pointed out, violence against anyone and everyone who interferes with gangs is endemic in El Salvador. Those who testify against gang members, those who resist their recruitment, those who inform against them to the government, those who cross them in any way — they all get on the gangs' enemies list and become the targets of violence, as do many others in society at large. That the government chooses to single out some targets of violence for special protection doesn't imply that the society at large or the gangs themselves view them as a particular social group. The government may well have reasons independent of any group status for giving these individuals special protection. The reason here is obvious: The Salvadoran government has a strong interest in protecting the integrity of its criminal justice system by giving witnesses in criminal cases special protection. This is true in the United States as well: We have various statutes that make it a criminal offense to intimidate or otherwise interfere with witnesses; we also have a witness protection program. But this doesn't mean that we consider witnesses in criminal cases or even those who are in the witness protection program to be members of a particular social group. The question is, at the very least, debatable and my colleagues are wrong to try to make the finding themselves. We are, as the Supreme Court has told us, reviewers, not first viewers.

While the majority may believe its ruling is narrow, the implications are actually vast. As pointed out earlier, witnesses against gang members are not in a materially different position from others who act in opposition to gangs. C-A-, 23 I. & N. Dec. at 960-61. The BIA has consistently adhered to this rationale, rejecting asylum applications in numerous cases. See, e.g., S-E-G-, 24 I. & N. Dec. at 586-87; A-M-E, 24 I. & N. Dec. at 74-75. Today's ruling casts doubt on this entire body of caselaw and puts the BIA in the untenable position of applying materially different law to asylum applicants who claim to be victims of gang violence depending on whether their cases will be appealed to our circuit or to one of the circuits that have approved the BIA's approach, such as the First, Second, Fourth, Fifth, Sixth, Eighth, Tenth and Eleventh. Today's opinion will thus force precisely the kind of inconsistency and arbitrariness in the agency's rulings that the majority now purports to correct.

* * *

This case should never have been taken en banc. The three-judge panel's unpublished disposition says all that need be said in this case. We should've left well enough alone.