OPINION
RONALD LEE GILMAN, Circuit Judge.In 1975, Joseph P. Dyer, III was convicted in a Tennessee state court on two counts of first-degree murder and on two counts of grand larceny. He was sentenced to death, but his sentence was commuted to life imprisonment because Tennessee’s death penalty statute was declared unconstitutional while Dyer’s direct appeal was pending. His convictions and sentence were affirmed on appeal.
When Dyer was granted his second parole hearing in 1998, the Tennessee parole board used the statutory parole standard in effect at the time of the hearing — rather than the standard in effect at the time of his offenses — to determine that Dyer was not eligible for parole. Dyer filed a petition for postconviction relief in the state-court system, claiming that the parole board members violated both the Ex Post Facto Clause of the United States Constitution and the Due Process Clause of the Fourteenth Amendment to the Constitution when they applied the parole standards enacted after his offenses. The Tennessee courts dismissed Dyer’s claims.
Dyer, acting pro se, subsequently filed a petition for a writ of habeas corpus in federal district court pursuant to 28 U.S.C. § 2254, asserting the same grounds for relief. Reviewing the decision of the Tennessee Court of Appeals under the deferential AEDPA standard, the district court denied Dyer’s petition. For the reasons set forth below, we VACATE the judgment of the district court and REMAND the case for further proceedings consistent with this opinion.
I. BACKGROUND
At Dyer’s first parole hearing in 1993, the parole board denied his request for parole because of the seriousness of his offenses. Dyer was granted a second parole hearing in 1998, but the board again denied him parole on the same basis. Following the 1998 hearing, Dyer filed a petition in the Chancery Court of Davidson County for a writ of common law certiorari to review the actions of the parole board. In his petition, Dyer claimed that the parole board committed ex post facto and due process violations when it applied the current parole standard rather than the standard in effect at the time of his convictions.
The relevant parole standard in effect at the time of Dyer’s convictions provided:
Tenn.Code Ann. § k0-361k (197k): Parole being a privilege and not a right, no prisoner shall be released on parole merely as a reward for good conduct or efficient performance of duties assigned in prison, but only if the board is of the opinion that there is reasonable probability that if such prisoner is released he will live and remain at liberty without violating the law, and that his release is not incompatible with the welfare of society. If the board shall so determine, such prisoner shall be allowed to go upon parole ....
(Emphasis added.)
In contrast, the relevant parole standard in effect at the time of Dyer’s 1998 hearing and applied by the parole board provided:
*283Tenn.Code Ann. § b0-28-117(a) (1998): Parole being a privilege and not a right, no prisoner shall be released on parole merely as a reward for good conduct or efficient performance of duties assigned in prison, but only if the board is of the opinion that there is reasonable probability that such prisoner, if released, will live and remain at liberty without violating the law, and that the prisoner’s release is not incompatible with the welfare of society. If the board so determines, such prisoner may be paroled
Tenn.Code Ann. § W-35-503(b) (1998): Release on parole is a privilege and not a right, and no inmate shall be granted parole if the board finds that:
(2) The release from custody at the time would depreciate the seriousness of the crime of which the defendant stands convicted or promote disrespect for the law.
(Emphasis added.) Dyer claimed in his petition that the application of “harsher, more severe” statutes at his 1998 parole hearing caused him to be denied parole, and that “[i]f the laws and rules which were in effect in 1974 had been used, the outcome of the parole hearing would have been different.”
Rather than analyzing Dyer’s claims based on the statutory changes, the Tennessee Court of Appeals treated Dyer’s petition as if it primarily relied on Rule 1100-1-1-.06 of the Rules of the Tennessee Boards of Parole (the good-candidacy rule), an earlier parole rule that provided: “The Board operates under the presumption that each resident who is eligible for parole is a worthy candidate and thus the Board presumes that he will be released on parole when he is first eligible.” Dyer v. Tenn. Bd. of Paroles, 2001 WL 401596, *1 (Tenn.Ct.App. Apr.23, 2001) (unpublished) (quoting Tenn. Comp. R. & Regs. r. 1100-1-1.06 (superseded)). Once the Tennessee court focused its attention on the good-candidacy rule, it quickly determined that Dyer’s claims were foreclosed by Kaylor v. Bradley, 912 S.W.2d 728, 733 (Tenn.Ct.App.1995) (holding that the parole board’s failure to apply the good-candidacy rule in effect at the time of the prisoner’s offense was not an ex post facto violation). Turning to one of Dyer’s other arguments — that the retroactive application of the new parole standard, particularly the seriousness provision, increased his punishment — the court held that “the [seriousness] section does not affect parole eligibility date or denial of parole, but instead enumerates one reason which the parole board may elect, in its discretion, to deny parole.” Dyer, 2001 WL 401596, at *2.
Following this decision in the Tennessee Court of Appeals, Dyer filed a petition for a writ of habeas corpus in federal district court pursuant to 28 U.S.C. § 2254. Dyer again argued that the parole board violated the Ex Post Facto and Due Process Clauses of the Constitution when it applied current parole statutes rather than the statute in effect at the time of his offenses. The district court granted the warden’s motion for summary judgment, denied Dyer’s cross-motion for summary judgment, and dismissed Dyer’s petition. Dyer filed a timely notice of appeal. This court granted Dyer a certificate of appeal-ability as to whether the parole board violated the Ex Post Facto Clause when it retroactively applied Tennessee Code §§ 40-28-117(a) (the may/shall provision) and 40-35-503(b) (the seriousness provision) in reaching its 1998 parole decision regarding Dyer.
II. ANALYSIS
A. Standard of review
In a habeas corpus appeal, we review a district court’s legal conclusions de novo, *284but will not set aside its factual findings unless they are clearly erroneous. Lucas v. O’Dea, 179 F.3d 412, 416 (6th Cir.1999). The standard of review for state-court determinations, however, is governed by the Antiterrorism and Effective Death Penalty Act (AEDPA), codified at 28 U.S.C. § 2254(d). AEDPA provides that
[a]n application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall hot be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim
(1) resulted in a decision that 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) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.
Id.
For the purposes of AEDPA, we review the last state court decision on the merits, which in this case is the decision of the Tennessee Court of Appeals. See Howard v. Bouchard, 405 F.3d 459, 469 (6th Cir.2005). A state-court decision is considered contrary to federal law “if the state court arrives at a conclusion opposite to that reached by the [Supreme] Court on a question of law or if the state court decides a case differently than the [Supreme] Court has on a set of materially indistinguishable facts.” Williams v. Taylor, 529 U.S. 362, 413, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000). When the state court issues a decision that is contrary to federal law, we review the merits of the petitioner’s claim de novo. See Magana v. Hofbauer, 263 F.3d 542, 551 (6th Cir.2001) (citing Williams, 529 U.S. at 396-98, 120 S.Ct. 1495); see also Fulcher v. Motley, 444 F.3d 791, 799 (6th Cir.2006) (“The [state] Supreme Court’s decisional rule was contrary to clearly established federal law, therefore de novo review is appropriate.”).
The application of federal law is unreasonable where “the state court identifies the correct governing legal principle from [the Supreme] Court’s decisions but unreasonably applies that principle to the facts of the prisoner’s case.” Id. When assessing unreasonableness, “a federal habeas court may not issue the writ simply because it concludes in its independent judgment that the relevant state-court decision applied clearly established federal law erroneously or incorrectly. Rather, that application must also be unreasonable.” Id. at 411, 120 S.Ct. 1495. Factual findings made by the state court, moreover, are presumed correct in the absence of clear and convincing evidence to the contrary. 28 U.S.C. § 2254(e)(1).
When a state court fails to address the petitioner’s federal claim, we review the claim de novo. Howard, 405 F.3d at 467. If, however, a state court does not squarely address the claim but engages in what resembles the proper constitutional analysis, we review the record and the law, and will reverse only if we determine that the state court decision was contrary to, or an unreasonable application of, federal law. Filiaggi v. Bagley, 445 F.3d 851, 854 (6th Cir.2006).
B. Requirements for proving an ex post facto violation
Two issues, both regarding the parole board’s retroactive application of the new parole provisions, are now before us: (1) Did the parole board commit an ex post facto violation when it retroactively applied the may/shall provision, and (2) did it commit an ex post facto violation when it ret*285roactively applied the seriousness provision? Because these two inquiries employ a similar method of analysis, we will discuss them together.
The Constitution prohibits states from imposing ex post facto laws. U.S. CONST, art. I, § 10, cl. 1. An ex post facto law possesses two elements: (1) “it must apply to events occurring before its enactment,” and (2) “it must disadvantage the offender affected by it.” Lynce v. Mathis, 519 U.S. 433, 441, 117 S.Ct. 891, 137 L.Ed.2d 63 (1997) (citation and quotation marks omitted) (holding that a Florida statute canceling provisional release credits violated the Ex Post Facto Clause). Retroactive application of parole provisions falls within the ex post facto prohibition if such an application creates a “sufficient risk of increasing the measure of punishment attached to the covered crimes.” Garner v. Jones, 529 U.S. 244, 250, 120 S.Ct. 1362, 146 L.Ed.2d 236 (2000) (citation and quotation marks omitted).
The Supreme Court has not articulated a precise formula for determining whether a risk is “sufficient.” See Cal. Dep’t of Corrections v. Morales, 514 U.S. 499, 509, 115 S.Ct. 1597, 131 L.Ed.2d 588 (1995) (“We have previously declined to articulate a single ‘formula’ for identifying those legislative changes that have a sufficient effect on substantive crimes or punishments to fall within the constitutional prohibition [on ex post facto laws], and we have no occasion to do so here.”). Despite its failure to proclaim a particular formula, the Supreme Court has consistently required petitioners to proffer actual evidence, rather than mere speculation, regarding the retroactive application’s disadvantageous effect. See, e.g., Morales, 514 U.S. at 509, 115 S.Ct. 1597 (“The amendment [in question] creates only the most speculative and attenuated possibility of producing the prohibited effect of increasing the measure of punishment for covered crimes, and such conjectural effects are insufficient for any threshold we might establish under the Ex Post Facto Clause.”).
The 'Supreme Court has held that the retroactive application of parole laws is unconstitutional where inmates can demonstrate with certainty that their punishment increased as a result of the retroactive application. In Lynce, for example, the Court considered whether Florida’s retroactive elimination of provisional-release credits for certain classes of inmates constituted an ex post facto violation. 519 U.S. at 435, 117 S.Ct. 891. The petitioner in Lynce was released early from prison partially as a result of his accumulation of provisional-release credits awarded by the state of Florida to alleviate prison overcrowding. Id. at 436, 117 S.Ct. 891. Shortly after he was released, the Florida legislature cancelled these credits, and a rearrest warrant was issued for Lynce. Id. The Lynce Court held that “the actual course of events makes it unnecessary to speculate about what might have happened,” and that the retroactive application “has unquestionably disadvantaged petitioner because it resulted in his rearrest and prolonged his imprisonment.” Id. at 446-47, 117 S.Ct. 891. As a result of this clear adverse effect on the petitioner, the Court held that Florida’s actions were unconstitutional. Id; see also Weaver v. Graham, 450 U.S. 24, 35-36, 101 S.Ct. 960, 67 L.Ed.2d 17 (1981) (holding that Florida violated the Ex Post Facto Clause when it retroactively applied a statute eliminating gain-time release credits because the elimination “makes more onerous the punishment for crimes committed before its enactment”).
In contrast, the Supreme Court has refused to hold the retroactive application of parole laws unconstitutional where inmates’ claims have been based on specula*286tion or conjecture. In Morales, for example, the inmate challenged the retroactive application of a law that maintained the date of his initial parole hearing, but granted the California parole board the discretion to increase the interval between parole hearings from one year to three years if the board determined that there was no reasonable probability that the inmate would be suitable for parole in the interim period. 514 U.S. at 504, 115 S.Ct. 1597. The Court emphasized that the statute applied only to inmates whose chances of release on parole were “quite remote,” and it focused on the fact that the law in question created “only the most speculative and attenuated possibility of producing the prohibited effect of increasing the measure of punishment for covered crimes.” Id. at 509, 115 S.Ct. 1597. In response to the inmate’s claim that there is “no principled way to determine how significant a risk of enhanced confinement is to be tolerated,” the Court stated that “[o]ur cases have never accepted this expansive view of the Ex Post Facto Clause, and we will not endorse it here.” Id. Without evidence beyond pure conjecture of a sufficient risk of increased punishment, the Court was unwilling to hold that this procedural change in California’s parole laws violated the Ex Post Facto Clause. See Garner, 529 U.S. at 250, 120 S.Ct. 1362 (classifying Morales as a case involving a purely procedural change to California’s parole laws).
The Supreme Court considered a similar scenario in Gamer, where the parole board in Georgia retroactively applied a law that increased the interval between subsequent parole hearings from three years to eight years for inmates serving life sentences who had previously been denied parole. 529 U.S. at 248, 120 S.Ct. 1362. Although the Gamer Court acknowledged that the requisite risk could be inherent in the statute, it also reached the following conclusion:
When the rule does not by its own terms show a significant risk, the [inmate] must demonstrate, by evidence drawn from the rule’s practical implementation by the agency charged with exercising discretion, that its retroactive application will result in a longer period of incarceration than under the earlier rule.
Id. at 255, 120 S.Ct. 1362. Because the lower courts had failed to ascertain beyond mere speculation whether retroactive application of the Georgia rule created a significant risk for the inmate in question, the Gamer Court held that the record did not support a holding that the state had committed an ex post facto violation. Id. at 256-57, 120 S.Ct. 1362.
The relevant Supreme Court precedents, at least from a factual perspective, provide us with little guidance in this case. At one extreme are Lynce and Weaver involving near-certain increased risks of punishment, whereas Morales and Gamer are at the other end of the spectrum in which the inmates provided only speculation regarding such risks. Dyer’s case, as analyzed below, falls somewhere in the middle. Intuitively, the retroactive application of new parole statutes that reduce the level of discretion afforded the parole board might effectuate a sufficient risk of increased punishment, but the ultimate result depends upon how the parole board actually exercises its discretion. Regardless of the factual circumstances, however, the Supreme Court has made clear that in order for us to conduct the necessary ex post facto inquiry, we must determine whether Dyer has produced specific evidence of a sufficient risk of increased punishment. See, e.g., Garner, 529 U.S. at 255-57, 120 S.Ct. 1362 (holding that the Eleventh Circuit erred in basing its decision on nothing *287more than speculation of increased punishment).
C. Dyer’s claims
Dyer’s first claim alleges that the retroactive application of the may/shall provision created a sufficient risk of increased punishment in violation of the Ex Post Facto Clause. This provision is nearly identical to the provision in effect at the time of Dyer’s offenses, except that the 1974 version of the statute compelled the parole board to parole an inmate if two conditions were met, whereas the 1998 version allows the parole board discretion to parole even if the inmate meets the same two conditions. Compare TenmCode Ann. § 40-3614 (1974) with TenmCode Ann. § 40-28-117(a) (1998) (containing essentially the same language except for “shall” in the 1974 version and “may” in the 1998 version). In other words, Dyer claims that the 1998 statute gives the parole board discretion where it once had none.
Dyer also alleges, in addition to the may/shall disparity, that the retroactive application of the seriousness provision— the statute with no parallel in 1974 — constituted an ex post facto violation. See Tenn.Code. Ann. § 40-35-503(b) (1998). He contends that he has no chance of parole because he was convicted of “especially serious offenses.” In support of this proposition, Dyer notes that the three parole board members who voted to deny him parole in 1998 based their decision on the seriousness provision. For this reason, Dyer alleges that the risk of increased punishment is not purely speculative or attenuated.
We note at the outset that the Tennessee Court of Appeals failed to address Dyer’s claims individually. It did not cite the specific provisions challenged by Dyer or analyze their effect — choosing instead to rely on the trial court’s characterization of the provisions as purely procedural. See Dyer v. Tenn. Bd. of Paroles, 2001 WL 401596, at * 2 (Tenn.Ct.App. Apr.23, 2001). Addressing Dyer’s claims collectively, the state appellate court found persuasive the trial court’s holding that “[t]he current statutes do not affect the length of the petitioner’s life sentences, do not change the application of sentence reduction credits to parole eligibility data and do not affect eligibility for parole consideration in any manner.” Dyer v. Tenn. Bd. of Paroles, 2001 WL 401596, at * 2 (Tenn.Ct.App. Apr.23, 2001) (quoting from Dyer v. Tenn. Bd. of Paroles, No. 99 — 86—III, slip, op. at 3 (Tenn. Ch. Ct. Sept. 20, 1999). Further reasoning that the seriousness provision “enumerates one reason for which the Board may elect, in its discretion, deny parole,” the Tennessee Court of Appeals held that retroactive application of this provision “does not affect parole eligibility date or denial of parole.” Dyer v. Tenn Bd. of Paroles, 2001 WL 401596, at *2 (Tenn.Ct.App. Apr. 23, 2001).
The district court held, with little analysis, that the state court’s classification of the posW.974 changes as “procedural in nature” was neither contrary to, nor involved an unreasonable application of, clearly established federal law. See Dyer v. Bowlen, slip op. at 5-6 (quoting from Dyer v. Tenn Bd. of Paroles, No. 99-86-III, slip. op. at 3 (Tenn. Ch. Ct. Sept. 20, 1999); see also Morales, 514 U.S. at 508, 115 S.Ct. 1597 (holding that the retroactive application of procedural changes does not implicate the Ex Post Facto Clause). We respectfully disagree.
Although the Tennessee Court of Appeals failed to address Dyer’s claims individually, its decision bears enough of a resemblance to an ex post facto analysis so as to require us to review the decision under the deferential lens of AEDPA. We will thus not reverse unless the decision *288was contrary to, or involved an unreasonable application of, federal ex post facto law. See Filiaggi, 445 F.3d at 854. The Supreme Court held in Morales that a successful ex post facto claim requires only that an inmate demonstrate a sufficient risk of increased punishment. 514 U.S. at 509, 115 S.Ct. 1597. Gamer further defines the framework for determining the requisite risk by instructing lower courts to first consider the risk inherent in the wording of the statute itself and then, alternatively, to explore the evidence of the statute’s practical implementation. 529 U.S. at 255, 120 S.Ct. 1362. Although Dyer’s claims are factually distinguishable from those in Morales and Gamer, the rule of constitutional law pronounced in those cases remains the proper standard by which to measure an ex post facto violation. See, e.g., Morales, 514 U.S. at 509, 115 S.Ct. 1597; Garner, 529 U.S. at 255, 120 S.Ct. 1362.
The Tennessee Court of Appeals, however, required that an inmate show an actual increase in punishment. Dyer v. Tenn Bd. of Paroles, No. 99 — 86—III, slip, op. at 3 (Tenn. Ch. Ct. Sept. 20, 1999) (“The change [] is not imposition of a greater or more serious punishment than was proscribed by law at the time of the offense.”). This approach runs contrary to the constitutional standard set forth in Morales and Gamer because it placed too great a burden on Dyer. Rather than requiring Dyer to prove that the retroactive application of the parole statutes created a sufficient risk of increased punishment, the state court demanded that Dyer prove he actually received a more serious punishment. In doing so, the state court subjected Dyer to a more exacting standard— a standard that is contrary to the clearly established law as stated in Morales and Gamer. Cf. Magana, 263 F.3d at 550 (holding that the state court’s decision was contrary to clearly established Supreme Court precedent when it required a petitioner to demonstrate with absolute certainty, rather than reasonable probability, that the ineffective assistance of his counsel prejudiced him).
Regarding the seriousness provision, the state court held, with little analysis, that altering the level of discretion to which the parole board is entitled does not constitute an ex post facto violation. Dyer v. Tenn. Bd. of Paroles, 2001 WL 401596, at *2. The Supreme Court has explicitly held, however, that discretion in parole considerations does not insulate the state from ex post facto violations. See Garner, 529 U.S. at 253, 120 S.Ct. 1362 (“The presence of discretion does not displace the protections of the Ex Post Facto clause .... ”). By failing to acknowledge Gamer’s warning about discretion, and by disregarding the proper standard for demonstrating an ex post facto violation as pronounced in Morales and Gamer, the state court reached a decision that is contrary to federal law as determined by the Supreme Court of the United States. We are thus required to review Dyer’s claims de novo, see Magana, 263 F.3d at 551, analyzing whether Dyer has proven that a sufficient risk of increased punishment is inherent in the wording of the statutes themselves or, alternatively, that it results from their practical implementation. See Garner, 529 U.S. at 255, 120 S.Ct. 1362.
D. Reason for remand
Because the record in this case does not allow us to fulfill our mandate, we remand the case to the district court with instructions to conduct an evidentiary hearing on the practical effects of the statutes’ retroactive application. We do so because Gamer requires an inmate to demonstrate a sufficient risk of increased punishment that is either inherent on the *289face of the new statutes or is evidenced by the statutes’ practical implementation. 529 U.S. at 254, 120 S.Ct. 1362. Dyer, in our opinion, has been unable to show the requisite risk in the wording of the statutes themselves.
We are unable to determine the effect of the may/shall provision without evidence of how the parole board has actually exercised its discretion. See Garner, 529 U.S. at 250, 120 S.Ct. 1362 (“Whether retroactive application of a particular change in parole law respects the prohibition on ex post facto legislation is often a question of particular difficulty when the discretion vested in a parole board is taken into account.”). Similarly, even when the seriousness of the offense was but one discretionary factor for the parole board to consider, the parole board might have determined that Dyer’s offense of double-murder was so serious that his release was “incompatible with the welfare of society.” See Tenn.Code Ann. § 40-3614 (1974). But, as Dyer argues, because the nature of his offenses under the 1974 version of the statute was but one factor among many evaluated, the parole board might have released him in light of his rehabilitative progress while in prison. See Tenn.Code. Ann § 40-3614 (1974) (authorizing the parole board to consider many factors to determine if the inmate should be released). The new provision, in contrast, makes the seriousness of the crime sufficient in and of itself to justify the denial of parole. Dyer bases his claim on that fact.
The governing standard as announced in Morales, however, “requires a more rigorous analysis of the level of risk created by the change in law.” Garner, 529 U.S. at 255, 120 S.Ct. 1362; see also Morales, 514 U.S. at 506 n. 3, 115 S.Ct. 1597 (“[T]he focus of the ex post facto inquiry is not on whether a legislative change produces some ambiguous sort of disadvantage ... but on whether any such change alters the definition of criminal conduct or increases the penalty by which a crime is punishable.”) (citation and quotation marks omitted); Richardson v. Penn. Bd. of Probation and Parole, 423 F.3d 282, 292 (3rd Cir.2005) (recognizing “the intuitive force of the argument that adjudication under stricter standards is more likely to lead to an adverse result,” but holding that the inmate was required to demonstrate evidence of an actual disadvantage).
In Gamer, for example, the Eleventh Circuit Court of Appeals held that retroactive application of the new parole guideline seemed “certain to ensure that some number of inmates will find the length of their incarceration extended.” Garner, 529 U.S. at 249, 120 S.Ct. 1362 (citation and quotation marks omitted). The Supreme Court reversed and remanded, holding that “[t]he record before the Court of Appeals contained little information bearing on the level of risk created by the change in law.” Id. at 256, 120 S.Ct. 1362. Without evidence of an actual disadvantage to the inmate (or at least general information regarding the operation of the parole system), the Supreme Court was unwilling to hold that the Ex Post Facto Clause was violated. Id.; see also Morales, 514 U.S. at 508-09, 115 S.Ct. 1597 (rejecting the proposition that all changes that might “create some speculative, attenuated risk of affecting a prisoner’s actual term of confinement by making it more difficult for him to make a persuasive case for early release” constitute ex post facto violations).
The Third Circuit’s decision in Richardson, 423 F.3d at 291-94, demonstrates the necessity of evidence in performing the “significant risk” inquiry. In Richardson, an inmate convicted in 1984 argued that the retroactive application of parole *290amendments enacted in 1996 constituted an ex post facto violation. Id. at 284. The inmate demonstrated that the parole board did in fact apply the amendments retroactively, but the court held that he was required to show that the retroactive application individually prejudiced him. Id. at 292-93.
Although the Richardson court acknowledged “the intuitive force of the argument that adjudication under stricter standards is more likely to lead to an adverse result,” it held that the “evidentiary requirement of jurisprudence must be honored.” Id. at 292. The court distinguished Mickens-Thomas v. Vaughn, 321 F.3d 374 (3d Cir.2003), an earlier Third Circuit case in which the panel granted habeas in light of statistical evidence suggesting that the retroactive application of parole guidelines increased the inmate’s period of incarceration. Id. In other words, without actual evidence demonstrating that Richardson was subject to an increased risk of incarceration, the court was unwilling to hold that retroactive application of the parole statute was unconstitutional.
Given Dyer’s lack of opportunity to present relevant evidence, we are of the opinion that the proper remedy is to remand the case to the district court for an eviden-tiary hearing. The state repeatedly contends that Dyer has failed to proffer evidence of an actual disadvantage, but we find this argument unpersuasive considering that Dyer was never afforded the chance to prove otherwise. Without providing Dyer with an opportunity to access data pertaining to the effect of the retroactive application of the parole laws on his own prison term (or the terms of others similarly situated), his claims will never rise above speculation — and will ultimately fail. The state attempts to take advantage of Dyer’s quandary by acknowledging that even though the may/shall provision might have increased the parole board’s discretion, Dyer’s claims rest on speculation rather than proof.
Our decision to remand for an evidentia-ry hearing gains support from Gamer itself, where the Supreme Court noted that the courts of appeals are entitled to cure insufficient records by ordering discovery. Garner, 529 U.S. at 257, 120 S.Ct. 1362 (“Respondent claims he has not been permitted sufficient discovery to make this showing [that retroactive application created a significant risk of increased punishment]. The matter of adequate discovery is one for the Court of Appeals ....”).
Our dissenting colleague, however, would flat-out reverse the district court’s decision. The dissent makes much of the distinction “between situations where the substantive criteria for parole ... revocation are changed, on the one hand, and situations where the methods or procedures for applying those criteria are changed, on the other.” Dissenting Op. at 12. The dissent classifies Dyer’s complaint as one challenging the substantive criteria for parole and, as a result, concludes that the retroactive application of the seriousness provision violates the Ex Post Facto Clause because it “makes parole less available depending on the seriousness of the crime — a factor not part of the previous scheme.” Id. at 13. Moreover, the dissent contends that an eviden-tiary hearing is unnecessary where the change in the parole criteria is substantive. Id. at 12.
We acknowledge that language in various Supreme Court cases lends credence to the substantive-versus-procedural distinction. See e.g., Miller v. Florida, 482 U.S. 423, 433, 107 S.Ct. 2446, 96 L.Ed.2d 351 (1987) (“[N]o ex post facto violation occurs if the change in the law is merely procedural .... On the other hand, a change in the law that alters a substantial *291right can be ex post facto even if the statute takes a seemingly procedural form.”) (citation and quotation marks omitted). We nevertheless believe that simply characterizing Dyer’s complaint as substantive in nature provides an insufficient basis for concluding that he has demonstrated an ex post facto violation.
In each of the cases relied upon by the dissent for its conclusion that a remand is unnecessary, the Supreme Court pointed to evidence, observable either on the face of the statute or as applied, showing that the petitioners were substantially disadvantaged by the retroactive application of parole laws. No such certainty exists here. The Lindsey Court, for example, found an ex post facto violation “regardless of the length of the sentence actually imposed, since the measure of punishment prescribed by the later statute is more severe than that of the earlier.” 301 U.S. at 401, 57 S.Ct. 797 (analyzing a scenario in which the new sentencing law changed the maximum sentence from discretionary to mandatory). In Weaver, the state retroactively applied a statute that, on its face, eliminated gain-time credits. 450 U.S. at 26-27, 101 S.Ct. 960. After reviewing the statute, the Court held that the petitioner was “disadvantaged by the reduced opportunity to shorten his time in prison simply through good conduct.” Id. 33-34, 101 S.Ct. 960.
The Miller Court reviewed the retroactive application of a sentencing guideline that changed the presumptive sentence for the petitioner’s offense from between three-and-a-half and four-and-a-half years to a range of five-and-a-half to seven years. 482 U.S. at 424, 107 S.Ct. 2446. Petitioner was sentenced to seven years. Id. In holding that the petitioner had demonstrated an ex post facto violation, the Miller Court focused on the effect of the statutory change. Under the prior law, if the petitioner had been sentenced to seven years, his sentence would have been reviewable on appeal. Id. at 435, 107 S.Ct. 2446. But under the new law, because the petitioner’s sentence was in the presumptive range, the sentencing court’s determination was unreviewable. Id. Such evidence of disadvantage was sufficient to demonstrate an ex post facto violation.
What these cases demonstrate, then, is that even when considering substantive changes to parole provisions, the Supreme Court has relied on evidence of actual disadvantage (or, as stated in Gamer, a sufficient risk of increased punishment). As discussed above, Dyer has been denied the opportunity to show such risk either on the face of the statute or in its practical application.
Yet another distinction between the so-called “substantive” cases and the present case is the level of discretion involved. The provisions at issue in Lindsey, Miller, and Weaver were couched in virtually mandatory terms. See, e.g., Miller, 482 U.S. at 435, 107 S.Ct. 2446 (“Nor do the revised guidelines simply provide flexible ‘guideposts’ for use in the exercise of discretion .... ”). Dyer’s fate, however, depends entirely on the discretion exercised by the parole board.
Our dissenting colleague minimizes the effect of the parole board’s discretion in the present case, going so far as to compare the seriousness provision to a hypothetical provision removing the possibility of parole for rape convicts. Dissenting Op. at 13 (“[T]he statutory provision in this case really differs [from the hypothetical rape provision] only in the possibility of the exercise of discretion on the part of the board to grant parole .... ”). We believe that the distinction cannot be so minimized. A rape convict could never be released on parole in the dissent’s hypothetical scenario. In Dyer’s case, however, we *292can only speculate as to the effect of the seriousness provision on his parole eligibility in the minds of the parole board, especially considering that the earlier provision similarly required discretionary consideration regarding the seriousness of the petitioner’s offense. See TenmCode. Ann. § 40-3614 (1974) (requiring the parole board to consider the “welfare of society” when making its determination).
We therefore remand the case for an evidentiary hearing — with instructions that discovery be limited to a class of inmates with comparable convictions and sentences — so that Dyer will have the opportunity to acquire data evidencing the practical implementation of the parole provisions in question. On remand, we leave it to the district court to work out the details of how best to comply with our instructions.
III. CONCLUSION
For all of the reasons set forth above, we VACATE the judgment of the district court and REMAND the case for further proceedings consistent with this opinion.