OPINION

Prison gangs threaten the safety and security of prisons and prisoners. California has sought to combat these threats— and punish prison-gang affiliation — by segregating prison-gang members and associates from the general prison population. To that end, California houses prison-gang members and associates in Security Housing Units (SHUs), maximum-security facilities in which prisoners are kept in solitary confinement for over 22 hours a day.

California also encourages good behavior among its prisoners with good-conduct credits that reduce prisoners’ sentences. Most prisoners earn credits on a one-to-one basis — for one day of good conduct, they earn one day of credit. So, a prisoner who behaves well can potentially cut his sentence in half. But prisoners can also lose credits, or their credit-earning status can change, based on misconduct. Until 2010, prison-gang members and associates housed in SHUs earned credits at a 'reduced three-to-one rate. But California amended its penal code in 2010 to modify the credit-earning status of prison-gang members and associates in segregated housing. Those prisoners can no longer earn any credits, regardless their conduct. The amendment thus causes prison-gang members and associates housed in SHUs to serve a longer portion of their prison sentences than they would have under the old regime, effectively increasing their sentences. The issue here is whether the 2010 amendment violates the Ex Post Fac-to Clause of the United States Constitution when applied to a prisoner whose underlying criminal offense was committed before that amendment’s enactment. We conclude it does.

I

In 2003, petitioner-appellant Antonio A. Hinojosa pleaded guilty in California superior court to first-degree robbery (to which he admitted a firearm enhancement) and participation in criminal-street-gang activity. He was sentenced to 16 years of imprisonment.

In 2009, Hinojosa was “validated” as a “prison-gang associate” and transferred to the SHU at Corcoran. Validation is the process by which inmates are classified as prison-gang members or associates. Once validated, a prison-gang member or associate “is deemed to be a severe threat to the safety of others or the security of the institution and will be placed in a SHU for an indeterminate term.” CaLCode Regs, tit. 15, § 3341.5(c)(2)(A)(2) (2009). At the time Hinojosa was validated, there were two ways validated prison-gang members and associates could get out of the SHU. The first is going “inactive.” An inactive inmate is one who has not been involved in prison-gang activity for a minimum of six years. Id. § 3378(e) (2009). Once deemed inactive, the prison’s Departmental Review Board may authorize an inmate’s transfer out of the SHU, but that decision is discretionary. See id. § 3341.5(c)(5) (2009). The Board “is authorized to retain an inactive gang member or associate in a SHU based on the inmate’s past or present level of influence in the gang, history of misconduct, history of criminal activity, or other factors indicating that the inmate poses a threat to other inmates or institutional security.” Id. The second way to get out of the SHU is to “debrief’ — what some prisoners might describe as “snitch.” The process has two steps: an interview phase and an observation phase. Id. § 3378.1(a) (2009). In the interview phase, the inmate must provide staff with “information about the gang’s structure, activities and affiliates,” as well as “a written autobiography of [his] gang involvement, which is then verified by staff for completeness and ae- curacy.” Id. § 3378.1(b) (2009). In the observation phase, inmates are observed for up to twelve months in segregated housing with other inmates undergoing the debriefing process. Id. § 3378.1(c) (2009).

Under the version of California Penal Code § 2933.6 in effect at the time of Hinojosa’s 2003 conviction and 2009 validation, he was eligible to earn good-conduct credits while housed in the SHU, albeit at a rate lower than prisoners housed in the general population. See Cal.Penal Code § 2933.6(a), (b) (2009); In re Efstathiou, 200 Cal.App.4th 725, 728 , 133 Cal.Rptr.3d 34 (2011). But effective January 25, 2010, § 2933.6 was amended to eliminate accrual of credits for inmates, such as Hinojosa, who had been transferred to the SHU upon validation as a prison-gang member or associate. See Cal-Penal Code § 2933.6(a), (b) (2010). The amendment did not revoke any credits Hinojosa earned before the effective date of the amendment; it prevented him from accruing any further custody credits while housed in the SHU. As a result of this change in credit-earning status, Hinojosa’s minimum release date was extended one year, from September 27, 2015, to September 27, 2016.

After exhausting his administrative remedies, Hinojosa filed a petition for writ of habeas corpus in the Superior Court of California challenging the application of amended § 2933.6 to change his credit-earning status. As recounted by the superior court, Hinojosa presented two claims:

The superior court addressed these two claims separately, providing “separate and independent grounds” for denying each. As to the first claim, the superior court held that Hinojosa’s plea agreement did not “contain an express promise or guarantee” regarding his credit-earning status and thus that the application of amended § 2933.6 to Hinojosa did not violate the terms of his plea agreement or violate due process. As to the ex post facto claim, the superior court denied it “on grounds [Hi-nojosa] ha[d] not sought review of his claim of error in the proper judicial venue.” The superior court denied his petition.

Hinojosa petitioned the California Court of Appeal and then the Supreme Court of California for a writ of habeas corpus, raising the same claims. Both courts denied his petitions without opinions.

Hinojosa then filed pro se a petition for writ of habeas corpus in the United States District Court for the Central District of California. In his petition, he claimed (1) application of amended § 2933.6 to change his credit earning status violated the Ex Post Facto Clause, and (2) ineffective assistance of counsel for failing to inform him of the chance he would lose his credit-earning status. In his report and recommendation, Magistrate Judge Michael R. Wilner analyzed those claims under the Antiterrorism and Effective Death Penalty Act (AEDPA), 28 U.S.C. § 2254 (d), and recommended denial of Hinojosa’s petition. As to the ex post facto claim, Magistrate Judge Wilner concluded the California Superior Court had not unreasonably applied federal law in denying Hinojosa’s claim

Magistrate Judge Wilner rejected Hinojo-sa’s ineffective-assistance-of-counsel claim because “no reasonable criminal defense attorney could be faulted for failing to anticipate and advise a client about a future change in the law governing prison credits.”

District Judge Gary A. Feess adopted Magistrate Judge Wilner’s report and recommendation in full, denied Hinojosa’s petition, and dismissed the action with prejudice. Judge Feess denied Hinojosa’s request for a certificate of appealability as to either of his claims. Hinojosa timely petitioned us for a certificate of appeala-bility, which we granted only as to “whether the 2010 amendment to California Penal Code § 2933.6, which deprives a prisoner of a future opportunity to earn an earlier release, violates the Ex Post Facto Clause.” This appeal followed. We have jurisdiction pursuant to 28 U.S.C. §§ 1291 and 2253(c).

II

We review de novo the district court’s denial of a petition for a writ of habeas corpus. Lambert v. Blodgett, 393 F.3d 943, 964 (9th Cir.2004). And we review de novo the district court’s determination that AEDPA applies to a petitioner’s claim. See id. at 965 .

III

Hinojosa is not the first California prisoner to challenge amended § 2933.6 under the Ex Post Facto Clause. In Nevarez v. Barnes, 749 F.3d 1124 (9th Cir.), cert. denied sub nom. Nevarez v. Ducart, — U.S. -, 135 S.Ct. 295 , 190 L.Ed.2d 215 (2014), we considered this same question on similar facts. There, we held that AEDPA barred us from granting Nevarez relief because the California courts’ denial of his ex post facto claim was not “an objectively unreasonable application of clearly established federal law” as determined by the Supreme Court of the United States. 749 F.3d at 1128 ; see 28 U.S.C. § 2254 (d)(1). If AEDPA applies here, we are bound by our decision in Nevarez and must affirm the district court’s denial of Hinojosa’s petition. See Miller v. Gammie, 335 F.3d 889, 899 (9th Cir.2003) (en banc). If not, we must consider the merits of his petition.

AEDPA bars us from granting a writ of habeas corpus “with respect to any claim that was adjudicated on the merits in State court proceedings” unless the state court’s decision (1) “was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States;” or (2) “was based on an unreasonable determination of the facts.” 28 U.S.C. § 2254 (d). Hinojosa contends that no state court decided his ex post facto claim “on the merits” and thus that AED-PA does not apply. We agree.

The superior court did not decide Hinojosa’s ex post facto claim on the merits. It denied the claim because Hinojosa filed it “in the [imjproper judicial venue.” The state conceded as much at oral argument. But the state cites Harrington v. Richter, 562 U.S. 86 , 131 S.Ct. 770 , 178 L.Ed.2d 624 (2011), and Johnson v. Williams, 568 U.S. -, 133 S.Ct. 1088 , 185 L.Ed.2d 105 (2013), for the proposition that we must presume the California Supreme Court decided Hinojosa’s ex post facto claim on the merits when it summarily denied his petition. That argument fails to comprehend the relationship between Richter , whereby we must presume state courts decide federal claims on the merits, see 562 U.S. at 99-100 , 131 S.Ct. 770 , and Ylst v. Nunnemaker, 501 U.S. 797 , 111 S.Ct. 2590 , 115 L.Ed.2d 706 (1991), which directs us to consider the last reasoned decision of the state courts, see id. at 806 , 111 S.Ct. 2590 . Where the last reasoned state-court decision rejects a federal claim solely on procedural grounds, any presumption that a subsequent summary denial decided the claim on the merits is rebutted. See James v. Ryan, 733 F.3d 911, 915-16 (9th Cir.2013) cert. denied, 572 U.S. -, 134 S.Ct. 2697 , 189 L.Ed.2d 740 (2014); see also Williams, 133 S.Ct. at 1094 n. 1 (“Consistent with our decision in Ylst v. Nunnemaker, 501 U.S. 797, 806 , 111 S.Ct. 2590 , 115 L.Ed.2d 706 (1991), the Ninth Circuit ‘look[ed] through’ the California Supreme Court’s summary , denial of Williams’ petition for review and examined the California Court of Appeal’s opinion, the last reasoned state-court decision-”). Here, the last reasoned decision is that of the superior court, which denied Hinojosa’s ex post facto claim solely on the ground of improper venue. That determination is not a determination “on the merits.” 28 U.S.C. § 2254 (d). So, we are not bound by AEDPA.

Nonetheless, if a state court dismisses a federal claim on an independent state procedural ground that is firmly established and regularly followed, we normally will not consider the claim. Beard v. Kindler, 558 U.S. 53, 55, 60-61 , 130 S.Ct. 612 , 175 L.Ed.2d 417 (2009). But the state has not raised a state procedural ground as a defense at any stage of Hinojosa’s federal proceedings. The defense is thus waived. See Trest v. Cain, 522 U.S. 87, 89 , 118 S.Ct. 478 , 139 L.Ed.2d 444 (1997) (“[Procedural default is normally a defense that the State is obligated to raise and preserve if it is not to lose the right to assert the defense thereafter.” (internal quotation marks and brackets omitted)). And although we may raise procedural default sua sponte, Windham v. Merkle, 163 F.3d 1092, 1100-01 (9th Cir.1998), we decline to do so here. Hinojosa raises a serious question about whether the superi- or court’s dismissal of his claim for improper venue is, in fact, a firmly established and regularly followed rule. See In re Oluwa, 207 Cal.App.3d 439 , 255 Cal.Rptr. 35, 37 (1989) (holding habeas petition challenging denial of custody credits “is not related to the conditions of ... confinement” and was properly brought in the district of conviction); Griggs v. Superior Court, 16 Cal.3d 341, 347 , 128 Cal.Rptr. 223 , 546 P.2d 727 (1976) (holding habeas petition should be transferred, not dismissed, for improper venue). We will not make the state’s arguments for it, even only to rebut them. So, we turn to the merits.

IV

“No State shall ... pass any ... ex post facto Law....” U.S. Const. art. I, § 10, cl. 1. “To fall within the ex post facto prohibition, a law must be retrospective— that is, ‘it must apply to events occurring before its enactment’ — and it ‘must disadvantage the offender affected by it’ by altering the definition of criminal conduct or increasing the punishment for the crime.” Lynce v. Mathis, 519 U.S. 433, 441 , 117 S.Ct. 891 , 137 L.Ed.2d 63 (1997) (citations omitted). We address these two prongs in turn.

A

A law is retrospective if it “applfis] to events occurring before its enactment.” Id. at 441 , 117 S.Ct. 891 (citation omitted). The Supreme Court has instructed that the “critical question is whether the law changes the legal consequences of acts completed before its effective date.” Weaver v. Graham, 450 U.S. 24, 31 , 101 S.Ct. 960 , 67 L.Ed.2d 17 (1981). But which acts? Hinojosa argues that the relevant conduct is the criminal conduct for which he is incarcerated. The state contends it is Hinojosa’s continued prison-gang association.

Our precedent supports Hinojosa’s position. Altering a prisoner’s ability to earn credits affects the length of his prison term and therefore affects the measure of punishment attached to the original crime. See United States v. Paskow, 11 F.3d 873, 879 (9th Cir.1993) (“[A statute], which forfeited good-time credits upon revocation of parole, violated the prohibition on ex post facto laws because it constituted ‘a sanction that extends the time remaining on petitioner’s original sentence’ rather than a punishment for ‘the second offense.’ ” (quoting Beebe v. Phelps, 650 F.2d 774, 776 (5th Cir. Unit A July 1981))); see also Weaver, 450 U.S. at 32 , 101 S.Ct. 960 (“[A] prisoner’s eligibility for reduced imprisonment is a significant factor entering into both the defendant’s decision to plea bargain and the judge’s calculation of the sentence to be imposed.”). For that reason, we have consistently looked to the prisoner’s underlying criminal conduct for the purpose of determining whether a law is retrospective. See Paskow, 11 F.3d at 877 (“These two factors must be assessed in connection with the date of the defen dant’s offense, not of his conviction or sentencing.”); Watson v. Estelle, 886 F.2d 1093, 1096 (9th Cir.1989) (“The key ex post facto inquiry is the actual state of the law at the time the defendant perpetrated the offense.”); see also Weaver, 450 U.S. at 31 , 101 S.Ct. 960 (asking whether the statute at issue “applies to prisoners convicted for acts committed before the provision’s effective date”). That analysis holds true even where the prisoner commits some intervening misconduct that triggers a change in his credit-earning status. See Paskow, 11 F.3d at 878 -79 (citing Greenfield v. Scafati 277 F.Supp. 644, 644-45 (D.Mass.1967), aff'd, 390 U.S. 713 , 88 S.Ct. 1409 , 20 L.Ed.2d 250 (1968) (per curiam)).

Greenfield , which we have adopted as controlling authority, illustrates this principle. As we described that case in Pas-kow:

Paskow, 11 F.3d at 878-79 (citations omitted). The panel thus concluded: “Because parole eligibility is part of the sentence for the underlying offense, its terms and conditions are fixed at the moment the underlying offense is complete. Therefore, like the length of a term of incarceration, the conditions affecting parole eligibility cannot be retrospectively altered.” Id. at 879 .

The state contends that our decision in Hunter v. Ayers, 336 F.3d 1007 (9th Cir.2003), establishes that the relevant conduct here is Hinojosa’s in-prison misconduct, not his underlying criminal conduct. We disagree. The panel in Hunter assumed without deciding that the relevant conduct for the purposes of retrospectivity was Hunter’s in-prison misconduct. See 336 F.3d at 1012-13 . But the panel neither raised nor answered that question. Nor did it have to: regardless whether the relevant date was the date of his in-prison misconduct or the date of his underlying criminal offense, the regulation violated the Ex Post Facto Clause as applied to Hunter. As such, it is no surprise that Hunter did not distinguish or even cite any of the cases relevant to that question. To the extent Hunter’s dicta does identify the in-prison infraction as the relevant conduct, we hold that dicta is not “well-reasoned”— indeed, the opinion provides no reasoning — and we decline to follow it. Cf. United States v. Johnson, 256 F.3d 895, 914 (9th Cir.2001) (en banc). Rather, we are bound by the express holdings of Paskow and Weaver . A prisoner’s eligibility for early release — whether by means of good-conduct credits or parole — is part of his underlying criminal sentence. See Weaver, 450 U.S. at 31-32 , 101 S.Ct. 960 ; Paskow, 11 F.3d at 879 . And where a prisoner’s sentence is effectively increased by new regulations that alter his credit-earning status, that alteration “changes the legal consequences” of his underlying criminal conduct. Weaver, 450 U.S. at 31 , 101 S.Ct. 960 . We therefore hold, in accord with Paskow and Weaver , that the date relevant to our retrospectivity analysis is the date of the prisoner’s underlying criminal conduct. See Paskow, 11 F.3d at 878-79 ; Weaver, 450 U.S. at 30-32 , 101 S.Ct. 960 .

Applying that principle here, we conclude that amended § 2933.6 is retrospective as applied to Hinojosa. To borrow Paskow’s language: “At the time [Hinojo-sa] committed his underlying crime ... [validated gang associates housed in a SHU] could accumulate good-conduct credits from the beginning of their incarceration.” Paskow, 11 F.3d at 878 . Amended § 2933.6 “was adopted after [Hinojosa] committed his underlying crime” and “prevented [Hinojosa] from being released as early as he might have been had he been permitted to amass good-conduct credits under the statute in effect at the time he committed the underlying crime.” Id. (emphasis omitted). It thus “changes the legal consequences of acts completed be fore its effective date.” Weaver, 450 U.S. at 31 , 101 S.Ct. 960 .

In its answering brief, the state attempts to distinguish this case from Pas-kow and Weaver on the ground that amended § 2933.6

See also In re Sampson, 197 Cal.App.4th at 1242, 130 Cal.Rptr.3d 39 (“[Petitioner's ineligibility for conduct credit accrual is not punishment for the offense of which he was convicted. ... It is punishment for gang-related conduct that continued after January 25, 2010.”).

We do not question whether California can punish prison misconduct, including prison-gang-related misconduct, through administrative disciplinary procedures. See, e.g., Wilkinson v. Austin, 545 U.S. 209, 222-23 , 125 S.Ct. 2384 , 162 L.Ed.2d 174 (2005) (holding that administrative punishment does not implicate a liberty interest protected by the Fifth Amendment unless the punishment “imposes atypical and significant hardship on the inmate in relation to the ordinary incidents of prison life.” (quoting Sandin v. Conner, 515 U.S. 472, 484 , 115 S.Ct. 2293 , 132 L.Ed.2d 418 (1995))). And we acknowledge that most administrative punish•ments — for example, segregated housing, loss of visitation privileges, restricted (and unpleasant) diets, and reduced exercise or social time — generally will not implicate the Ex Post Facto Clause. So long as an administrative punishment is in place before a prisoner commits the punishable prison misconduct, imposition of such punishment does not change the legal consequences of any prior acts. But administrative punishments that effectively extend a prisoner’s sentence — such as revocation of good-conduct credits or change in credit-earning status — are another story. A prisoner’s term of imprisonment is punishment for his underlying criminal conduct. So, an administrative punishment that effectively extends a prisoner’s sentence goes beyond punishing prison misconduct. It changes the legal consequences of his underlying criminal conduct. If that conduct was committed before the administrative punishment was enacted, the punishment is retrospective. See Weaver, 450 U.S. at 32-33 , 101 S.Ct. 960 .

The state is correct: Hinojosa’s “gang-related misconduct” occurred after, and is separate from, his underlying crimes. But in punishing Hinojosa for his in-prison gang-related misconduct, the state has effectively increased his prison sentence for his underlying crimes. And it has done so by means of a regulation that was enacted after Hinojosa committed those crimes. Amended § 2933.6 is thus retrospective as applied to Hinojosa.

B

Not all retrospective laws are unconstitutional. A retrospective law does not violate the Ex Post Facto Clause unless it “ ‘disadvantage^] the offender affected by it’ by altering the definition of criminal conduct or increasing the punishment for the crime.” Lynce, 519 U.S. at 441 , 117 S.Ct. 891 (citation omitted). But an increase in punishment need not be an increase in the maximum term of imprisonment. As the Supreme Court explained in Weaver , a new regulation that changes an inmate’s ability to earn good-conduct credits increases his punishment if the “new provision constricts the inmate’s opportunity to earn early release.” Weaver, 450 U.S. at 35-36 , 101 S.Ct. 960 .

Citing California Department of Corrections v. Morales, 514 U.S. 499 , 115 S.Ct. 1597 , 131 L.Ed.2d 588 (1995), the state argues that Hinojosa’s risk of an increased sentence is “too attenuated” to rise to an ex post facto violation. 514 U.S. at 514 , 115 S.Ct. 1597 . We disagree and distinguish Morales. Under the law in place when petitioner Morales murdered two people and pleaded guilty to those crimes, parole-eligible inmates were entitled to yearly parole-board hearings. Id at 502-03, 115 S.Ct. 1597 . But while Morales was incarcerated, the state changed the law to “authorize!] the Board [of Parole Hearings] to defer subsequent suitability hearings for up to three years ... if the Board ‘finds that it is not reasonable to expect that parole would be granted at a hearing during the following years and states the bases for the finding.’ ” Id at 503, 115 S.Ct. 1597 (citation omitted). At his first parole hearing in 1989, the board issued a reasoned decision finding it was not reasonable to expect that parole would be granted in the following years. Id at 502-03, 115 S.Ct. 1597 . It scheduled Morales’s next hearing for 1992. Id at 503, 115 S.Ct. 1597 .

Morales filed a petition for habeas corpus in federal district court, which the district court denied. Id at 504, 115 S.Ct. 1597 . We reversed, holding that “any retrospective law making parole hearings less accessible would effectively increase the [prisoner’s] sentence and violate the ex post facto clause.” Id (citation omitted). The Supreme Court reversed us, drawing a clear distinction between cases like Weaver , where a retrospective law directly results in an increased prison sentence, and cases like Morales , where the amended statute “creates only the most speculative and attenuated risk of increasing the measure of punishment attached to the covered crimes.” Id at 514, 115 S.Ct. 1597 .

This case falls under Weaver , not Morales . Whereas the amended statute in Morales did not change the “substantive formula” for reducing the statutory sentencing range or the standards for determining parole suitability, id at 507, 115 S.Ct. 1597 , amended § 2933.6 expressly alters the “substantive formula” for awarding good-conduct credits to prisoners, like Hinojosa, who have been validated as prison-gang members or associates. That change is not speculative; it has effectively increased Hinojosa’s prison time by one year.

The state argues alternatively that Hi-nojosa is not disadvantaged by amended § 2933.6 because he “could always choose to opt out of a prison gang.” If he did, the state suggests, he would be out of the SHU and back in the general population, earning credits at the same rate as everyone else. But it is not so easy. One does not simply “opt out” of a prison gang. Hinojosa cannot stop being a prison-gang associate in the eyes of the state unless he. waits six years or debriefs. And aside from the fact that a prisoner who debriefs may claim to face death or serious injury at the hands of his former compatriots, the entire debriefing process can take well over a year. See In re Sampson, 197 Cal.App.4th at 1240 , 130 Cal.Rptr.3d 39 . The state has made no representation that Hinojosa would necessarily regain his pri- or credit-earning status upon beginning— or even completing — the .process. Nor is there any provision by which Hinojosa could have the credits he is denied while debriefing reinstated once he completes the process.

But even if Hinojosa could easily opt out of his prison gang, the amended statute would still disadvantage him. We look at the effect amended § 2933.6 has on Hinojosa now, all other things being equal. The question is: if Hinojosa does not change his conduct — if he continues doing what he was doing before § 2933.6 was amended — is his prison time effectively lengthened? The answer is yes. Amended § 2933.6 thus works to his disadvantage. Were we to hold otherwise, the state could impose any manner of new requirements upon prisoners, who would have to comply simply to retain the same credit-earning status they enjoyed before the new requirements were enacted. See Weaver, 450 U.S. at 34-35 , 101 S.Ct. 960 (rejecting the state’s argument that Weaver could make up for his change in credit-earning status by performing “special behavior” to earn credits). Such a result would be irreconcilable with the Ex Post Facto Clause’s protection against “the lack of fair notice and governmental restraint when the legislature increases punishment beyond what was prescribed when the crime was consummated.” Id. at 30 , 101 S.Ct. 960 . For just as retroactively altering “a prisoner’s eligibility for reduced imprisonment” can disadvantage a prisoner, id. at 32 , 101 S.Ct. 960 , so can new conditions placed bn that eligibility.

In conclusion, we emphasize what we hold today — and what we do not. We do not question whether the state can enact a new statute punishing in-prison misconduct. Nor do we question here whether the state can apply that new statute to prisoners whose underlying criminal conduct predates the statute’s enactment. ■ But the state cannot use such a statute retroactively to effect an increase in prison time. The Ex Post Facto Clause forbids it.

V

Amended § 2933.6 violates the Ex Post Facto Clause as applied to prisoners, like Hinojosa, who committed their underlying criminal conduct before the amendment’s enactment. Accordingly, we REVERSE the judgment of the district court and REMAND with instructions to GRANT the writ of habeas corpus. The writ will direct the state to release Hinojosa on the date he would have been released under the version of § 2933.6 that was in place prior to January 25, 2010. See Weaver, 450 U.S. at 36 n. 22, 101 S.Ct. 960 .

No petition for rehearing will be entertained and mandate shall issue forthwith. See Fed. R.App. P. 2.

REVERSED and REMANDED.

. After this appeal was filed, the California Penal Code was amended to replace the term "prison gang" with "Security Threat Group,” bureaucratese that is otherwise known by the abbreviation "STG.” As such, Hinojosa is no longer a "prison-gang associate” but an "STG associate.” For the purposes of this opinion, however, we use the old terminology, which is more accurate here, as STGs may include groups other than prison gangs. The Code also distinguishes between prison-gang members and associates; that distinction is irrelevant to our analysis. . For an inmate to be validated as a prison-gang associate, the California Department of Corrections and Rehabilitations must recognize at least three reliable, documented bases (“independent source criteria items”) for concluding that the inmate's background, person, or belongings indicate his active association with other validated prison-gang members or associates. See Cal. Code Regs., tit. 15, § 3378.2(b). At least one of those bases must constitute a direct link to a current or former validated prison-gang member or associate. Id. The evidence presented against Hinojosa consisted of an envelope upon which was written the name of another validated prison-gang associate, a birthday card with gang symbols in it, photographs of Hinojosa's gang-related tattoos, and a report from an institutional gang investigator. Hinojosa does not challenge his validation as a prison-gang associate. .In his petition in district court, Hinojosa asserted that inmates who choose to debrief put themselves and their families in jeopardy of retaliation by other gang members. That may be, but it does not affect our analysis. . After briefing of this appeal concluded, California amended its regulations to introduce a third means by which validated prison-gang members and associates can get out of the SHU: the Step Down Program. See Cal.Code Regs, tit 15, § 3378.3(a) (2015). The Step Down Program is "an incentive based multi-step process for the management of [prison-gang] affiliates ... designed to monitor affiliates and assist with transition for return to [the] general population.” Id. Like debriefing, completing the Step Down Program is a lengthy process that does not entail immediate restoration of a prisoner’s credit-earning status. See id. §§ 3000, 3341.5(c)(5), 3378.3 (2015). The addition of the Step Down Program to the regulations does not change our analysis. . In Weaver , the governing law when petitioner Weaver committed and pleaded guilty to second degree murder permitted all prisoners to earn conduct credits at certain rates. Id. at 25-26 , 101 S.Ct. 960 . That rate was later reduced, and the new rates were applied to all prisoners regardless of when they committed their underlying offenses. Id. at 26-27 , 101 S.Ct. 960 . As a result, all prisoners’ minimum release dates were effectively increased, including Weaver’s. Id. Weaver petitioned the Supreme Court of Florida for a writ of habeas corpus, claiming the application of the new rates to him violated the Ex Post Facto Clause. Id. at 27 , 101 S.Ct. 960 . The Florida Supreme Court denied his petition. Id. at 27-28 , 101 S.Ct. 960 . The United States Supreme Court reversed. For the purposes of retrospectivity, the Court compared the date Weaver committed his underlying criminal offense and the date the new rates¿went into effect. Id. at 31-32 , 101 S.Ct. 960 ; see also id. at 32, 101 S.Ct. 960 (“[A] prisoner's eligibility for reduced imprisonment is a significant factor entering into both the defendant’s decision to plea bargain and the judge’s calculation of the sentence to be imposed.”). The Court then concluded that the new rates disadvantaged Weaver by increasing his prison sentence. Id. at 33 , 101 S.Ct. 960 . The Court concluded "the new provision constricts the inmate’s opportunity to earn early release, and thereby makes more onerous the punishment for crimes committed before its enactment. This result runs afoul of the prohibition against ex post facto laws.” Id. at 35-36 , 101 S.Ct. 960 . . Although Greenfield is a decision by a three-judge panel from the District of Massachusetts, the Supreme Court summarily affirmed the decision, and we have since adopted it as binding circuit precedent. See Paskow, 11 F.3d at 878 . As such, we are bound by Greenfield here, notwithstanding our holding in Nevarez v. Barnes that Greenfield "does not qualify as 'clearly established federal law [as determined by the Supreme Court of the United States]’ for purposes of AEDPA,” 749 F.3d at 1129 . As we held above, AEDPA does not apply to Hinojosa’s ex post facto claim. . In Hunter , petitioner Hunter challenged prison regulations that retroactively removed his ability to have conduct credits restored after an infraction. 336 F.3d at 1008-09 . Hunter was caught drinking "pruno” (prison wine), a disciplinary offense. Id. at 1008 . He was docked 120 days of good-conduct credits. Id. Under the regulations in place at the time he committed the offense, if he served six months following the offense without another disciplinary offense, half of his forfeited credits would be restored as of right. Id. at 10 ID. But after he committed the pru-no offense, the regulations were changed to eliminate restoration of forfeited credits for offenses like Hunter’s. Id. Hunter challenged the application of the new regulations to him as violating the Ex Post Facto Clause. Id. at 1011 . Hunter petitioned for a writ of habeas corpus under 28 U.S.C. § 2241 , which the district court granted. We affirmed, holding that application of the amended regulations to eliminate restoration of Hunter’s credits violated the Ex Post Facto Clause. Id. at 1013 . . But that distinction matters here. Although Hinojosa's and Hunter’s situations seem similar (Hinojosa was validated and transferred to the SHU, then the statute was amended; Hunter committed in-prison misconduct, then the regulation was amended), a key difference distinguishes them. Hunter’s pruno violation was completed when he drank the pruno. But according to the California courts, a validated prison-gang associate commits the continuing offense of associating with a prison gang until he debriefs or becomes inactive. See In re Sampson, 197 Cal.App.4th 1234, 1242-43 , 130 Cal.Rptr.3d 39 (2011). So, under California law, every moment Hinojosa goes without debriefing is a continuation of his misconduct. And we are bound by the California courts’ interpretation of California law. See Bradshaw v. Richey, 546 U.S. 74, 76 , 126 S.Ct. 602 , 163 L.Ed.2d 407 (2005) (per curiam). Accordingly, if the relevant date is the date of the in-prison misconduct, and Hinojosa was properly found to have committed in-prison misconduct by affiliating with a prison gang while in prison, § 2933.6 is not retrospective because until there is evidence that he has disaffiliated from that gang, he is continuing his in-prison misconduct. . Although the record does not contain the date of Hinojosa's underlying criminal offense, it was certainly before he pleaded guilty to that conduct in 2003. . Of course, the opposite is also true. If, at the time a prisoner commits his crime, regulations provide that prisoners may lose credit-earning status as a consequence of prison misconduct, there is no ex post facto violation.