In Padilla v. Kentucky, — U.S.-, 130 S.Ct. 1473, 1486, 176 L.Ed.2d 284 (2010), the Supreme Court held that an attorney provides ineffective assistance of counsel by failing to inform a client that a guilty plea carries a risk of deportation. The district court concluded that Padilla did not announce a new rule under the framework set forth in Teague v. Lane, 489 U.S. 288, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989), and consequently applied its holding to Petitioner Roselva Chaidez’s collateral appeal. Because we conclude that Padilla announced a new rule that does not fall within either of Teague’s exceptions, we reverse the judgment of the district court.1
I. Background
Chaidez entered the United States from her native Mexico in 1971, and became a lawful permanent resident in 1977. In June 2003, Chaidez was indicted on three counts of mail fraud in connection with a staged accident insurance scheme in which the loss to the victims exceeded $10,000. On the advice of counsel, Chaidez pled guilty to two counts on December 3, 2003. She was sentenced to four years’ probation on April 1, 2004, and judgment was entered in her case on April 8, 2004. Chaidez did not appeal.
Federal law provides that an alien who is “convicted of an aggravated felony at any time after admission is deportable.” 8 U.S.C. § 1227(a)(2)(A)(iii). Chaidez’s plea of guilty to a fraud involving a loss in excess of $10,000 rendered her eligible for removal from the United States as an aggravated felon. See 8 U.S.C. § 1101 (a) (43) (M) (i). The government initiated removal proceedings in 2009, after Chaidez unsuccessfully filed an application for U.S. citizenship.
In an effort to avoid removal, Chaidez sought to have her conviction overturned. To that end, she filed a motion for a writ of coram nobis in her criminal case on January 25, 2010. She alleges ineffective assistance of counsel in connection with her decision to plead guilty, claiming that her defense attorney failed to inform her that a guilty plea could lead to removal. Chaidez maintains that she would not have pled guilty if she had been made aware of the immigration consequences of such a plea.
On March 31, 2010, while Chaidez’s motion was pending before the district court, the Supreme Court issued its decision in Padilla. In a thoughtful opinion, Judge Gottschall acknowledged that this case presents a close call. She concluded that Padilla did not announce a new rule for Teague purposes, but rather was an application of the Court’s holding in Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). Having concluded that Padilla applied to Chaidez’s case, the district court considered the merits of her coram nobis petition. The court granted the petition and vacated Chaidez’s conviction. The government appeals the district court’s underlying ruling regarding the retroactive effect of Padilla.
*687II. Discussion
The writ of coram nobis, available under the All Writs Act, 28 U.S.C. § 1651(a), provides a method for collaterally attacking a criminal conviction when a defendant is not in custody, and thus cannot proceed under 28 U.S.C. § 2255. United States v. Folak, 865 F.2d 110, 112-13 (7th Cir.1988). The writ is an extraordinary remedy, allowed only where collateral relief is necessary to address an ongoing civil disability resulting from a conviction. Godoski v. United States, 304 F.3d 761, 762 (7th Cir.2002). Because a writ of error coram nobis affords the same general relief as a writ of habeas corpus, Howard v. United States, 962 F.2d 651, 653 (7th Cir.1992), we proceed as we would in a habeas case. See United States v. Mandanici, 205 F.3d 519, 527 (2d Cir.2000) (applying Teague in a case involving a coram nobis petition); United States v. Swindall, 107 F.3d 831, 834 (11th Cir.1997) (same). Our review is de novo.
In Padilla, the Court considered the petitioner’s claim that his counsel provided ineffective assistance by erroneously advising him that pleading guilty to a drug distribution charge would not impact his immigration status. The Kentucky Supreme Court had rejected Padilla’s claim, concluding that advice regarding the collateral consequences of a guilty plea (“i.e., those matters not within the sentencing authority of the state trial court”), including deportation, is “outside the scope of representation required by the Sixth Amendment.” 130 S.Ct. at 1481. As the Padilla Court noted, many state and federal courts had similarly concluded that a defendant’s Sixth Amendment right to effective assistance of counsel was limited to advice about the direct consequences of a guilty plea (i.e., length of imprisonment), and did not extend to information regarding collateral consequences (i.e., deportation). Id. However, in a majority opinion authored by Justice Stevens, the Padilla Court concluded that “advice regarding deportation is not categorically removed from the ambit of the Sixth Amendment right to counsel.” 130 S.Ct. at 1482. Noting that it had “never applied a distinction between direct and collateral consequences to define the scope of constitutionally ‘reasonable professional assistance’ required under Strickland,” the Court declined to consider the appropriateness of the direct/eollateral distinction generally. Id. at 1481. Rather, it found such a distinction “ill-suited to evaluating a Strickland claim concerning the specific risk of deportation.” Id. at 1481-82.
The majority based that conclusion on “the unique nature of deportation” — specifically, its severity as a penalty and its close relationship to the criminal process. Id. at 1481. The Court noted that recent changes in federal immigration law, including the Immigration Act of 1990 and the Illegal Immigration Reform and Immigrant Responsibility Act of 1996 (IIRIRA), had served to further “enmesh[] criminal convictions and the penalty of deportation,” by making “removal nearly an automatic result for a broad class of noncitizen offenders.” Id. at 1478-81. Those changes convinced the Court that “deportation is an integral part ... of the penalty that may be imposed on noncitizen defendants who plead guilty to specified crimes,” and cannot be “divorce[d] ... from the conviction.” Id. at 1480-81. As a result, the Court concluded that Strickland applied to Padilla’s ineffective assistance claim. 130 S.Ct. at 1482.
The Court went on to consider the first Strickland prong — whether Padilla had established that his counsel’s representation fell below an objective standard of reasonableness. In order to determine what constituted reasonable representation under *688the circumstances, the Court looked to prevailing professional norms set forth by the American Bar Association and numerous other authorities. Id. at 1482, 1485. The Court found that, dating back to the mid-1990s, those authorities had been in agreement that counsel must advise his or her client regarding the risk of deportation. Id. Thus, the Court held that defense counsel provides constitutionally deficient representation by failing to inform a defendant that a guilty plea carries a risk of deportation. Id. at 1486.
Chaidez seeks to have Padilla applied to her case on collateral review, despite the fact that the criminal case against her was final on direct review when Padilla was decided. Teague governs our analysis. Whorton v. Bockting, 549 U.S. 406, 416, 127 S.Ct. 1173, 167 L.Ed.2d 1 (2007). Under Teague, a constitutional rule of criminal procedure applies to all cases on direct and collateral review if it is not a new rule, but rather an old rule applied to new facts. Id. A new rule applies only to cases that still are on direct review, unless one of two exceptions applies. Id. In particular, a new rule applies retroactively on collateral review if (1) it is substantive or (2) it is a “ ‘watershed rul[e] of criminal procedure’ implicating the fundamental fairness and accuracy of the criminal proceeding.” Id. (citations omitted).
The parties agree that if Padilla announced a new rule neither exception to non-retroactivity applies. Therefore, whether Padilla announced a new constitutional rule of criminal procedure is the sole issue before us. The district courts that have addressed that issue — including those in this circuit — are split. See United States v. Diaz-Palmerin, 2011 WL 1337326 (N.D.Ill. April 5, 2011) (Padilla did not announce a new rule); Martin v. United States, 2010 WL 3463949 (C.D.Ill. Aug. 25, 2010) (same); United States v. Chavarria, 2011 WL 1336565 (N.D.Ind. April 7, 2011) (same); United States v. Laguna, 2011 WL 1357538 (N.D.Ill. April 11, 2011) (Padilla announced a new rule); United States v. Aceves, 2011 WL 976706, at *3 (D.Hawai’i March 17, 2011) (collecting cases). The Third Circuit recently became the first of our sister circuits to weigh in, holding that Padilla simply applied the old Strickland rule, such that it is retroactively applicable on collateral review. United States v. Orocio, 645 F.3d 630, 640-42 (3d Cir.2011).
A rule is said to be new when it was not “dictated by precedent existing at the time the defendant’s conviction became final.” Teague, 489 U.S. at 301, 109 S.Ct. 1060 (emphasis in original). That definition of what constitutes a new rule reflects the fact that Teague was developed in the context of federal habeas, which is designed “to ensure that state convictions comply with the federal law in existence at the time the conviction became final, and not to provide a mechanism for the continuing reexamination of final judgments based upon later emerging legal doctrine.” Sawyer v. Smith, 497 U.S. 227, 235, 110 S.Ct. 2822, 111 L.Ed.2d 193 (1990). See also Danforth v. Minnesota, 552 U.S. 264, 280, 128 S.Ct. 1029, 169 L.Ed.2d 859 (2008) (describing Teague as a “limit[ation on] the authority of federal courts to overturn state convictions”). Thus, the Court has explained that Teague “validates reasonable, good-faith interpretations of existing precedents made by state courts even though they are shown to be contrary to later decisions.” Saffle v. Parks, 494 U.S. 484, 488, 110 S.Ct. 1257, 108 L.Ed.2d 415 (1990) (quoting Butler v. McKellar, 494 U.S. 407, 414, 110 S.Ct. 1212, 108 L.Ed.2d 347 (1990)). The pertinent inquiry here is whether Padilla’s outcome was “susceptible to debate among reasonable minds.” Butler, 494 U.S. at 415, 110 S.Ct. 1212. *689Put differently, “our task is to determine whether a ... court considering [Chaidez’s] claim at the time [her] conviction became final” — pre-Padilla—“would have felt compelled by existing precedent to conclude that [Padilla ] was required by the Constitution.” Saffle, 494 U.S. at 488, 110 S.Ct. 1257.
That task is a “difficult” one where, as here, the decision at issue “extends the reasoning of ... prior cases,” as opposed to “explicit[ly] overruling ... an earlier holding.” Id. However, the Court’s retro-activity jurisprudence provides guidance. In assessing whether the outcome of a case was susceptible to reasonable debate, the Court has looked to both the views expressed in the opinion itself and lower court decisions. Lack of unanimity on the Court in deciding a particular case supports the conclusion that the case announced a new rule. See Beard v. Banks, 542 U.S. 406, 414-15, 124 S.Ct. 2504, 159 L.Ed.2d 494 (2004) (concluding that a rule was new where, in the case announcing the rule, four Justices dissented, expressing the view that the Court’s outcome was not controlled by precedent); Sawyer v. Smith, 497 U.S. 227, 236-37, 110 S.Ct. 2822, 111 L.Ed.2d 193 (1990) (concluding that Caldwell v. Mississippi 472 U.S. 320, 105 S.Ct. 2633, 86 L.Ed.2d 231 (1985), announced a new rule, in part based on the views of the Caldwell dissenters). Similarly, if the lower courts were split on the issue, the Court has concluded that the outcome of the case was susceptible to reasonable debate. See Butler, 494 U.S. at 415, 110 S.Ct. 1212 (“That the outcome in [Arizona v.] Roberson [, 486 U.S. 675, 108 S.Ct. 2093, 100 L.Ed.2d 704 (1988) ] was susceptible to debate among reasonable minds is evidenced further by the differing positions taken by the judges of the Courts of Appeals for the Fourth and Seventh Circuits”); O’Dell v. Netherlands 521 U.S. 151, 166 n. 3, 117 S.Ct. 1969, 138 L.Ed.2d 351 (1997) (finding support for its conclusion that a case announced a new rule “in the decisions of the state courts and the lower federal courts,” none of which previously had adopted the rule). These considerations convince us that Padilla announced a new rule.
The majority opinion in Padilla drew a concurrence authored by Justice Alito and joined by Chief Justice Roberts, as well as a dissenting opinion authored by Justice Scalia and joined by Justice Thomas. That the members of the Padilla Court expressed such an “array of views” indicates that Padilla was not dictated by precedent. O’Dell, 521 U.S. at 159, 117 S.Ct. 1969. Moreover, the views expressed in each of the opinions support that conclusion. Statements in the concurrence leave no doubt that Justice Alito and Chief Justice Roberts considered Padilla to be ground-breaking. See 130 S.Ct. at 1488, 1491, 1492 (referring to the majority’s holding as a “dramatic departure from precedent,” “a major upheaval in Sixth Amendment law,” and a “dramatic expansion of the scope of criminal defense counsel’s duties under the Sixth Amendment”). And the two dissenting Justices, who expressed the view that the majority’s extension of the Court’s Sixth Amendment jurisprudence lacked “basis in text or in principle,” certainly did not see Padilla as dictated by precedent. 130 S.Ct. at 1495 (Scalia, J., dissenting). See also Sawyer, 497 U.S. at 236-37, 110 S.Ct. 2822. Even the majority suggested that the rule it announced was not dictated by precedent, stating that while Padilla’s claim “followed] from” its decision applying Strickland to advice regarding guilty pleas in Hill v. Lockhart, 474 U.S. 52, 106 S.Ct. 366, 88 L.Ed.2d 203 (1985), Hill “does not control the question before us.” Id. at 1485 n. 12. It seems evident from Supreme Court precedent that Padilla cannot be an old rule simply because existing case law *690“inform[ed], or even controlled] or governed],” the analysis. Saffle, 494 U.S. at 488, 110 S.Ct. 1257. Nor will the rule of Padilla be deemed old because precedent lent “general support” to the rule it established, Sawyer, 497 U.S. at 236, 110 S.Ct. 2822, or because it represents “the most reasonable ... interpretation of general law,” Lambrix v. Singletary, 520 U.S. 518, 538, 117 S.Ct. 1517, 137 L.Ed.2d 771 (1997). Padilla can only be considered an old rule if Supreme Court precedent “compelled]” the result. Saffle, 494 U.S. at 490, 110 S.Ct. 1257. The majority’s characterization of Hill suggests that it did not understand the rule set forth in Padilla to be dictated by precedent.
Our conclusion that Padilla announced a new rule finds additional support in prePadilla decisions by state and federal courts. Prior to Padilla, the lower federal courts, including at least nine Courts of Appeals, had uniformly held that the Sixth Amendment did not require counsel to provide advice concerning any collateral (as opposed to direct) consequences of a guilty plea. See Padilla, 130 S.Ct. at 1487 (Alito, J., concurring in judgment) (“Until today, the longstanding and unanimous position of the federal courts was that reasonable defense counsel generally need only advise a client about the direct consequences of a criminal conviction,” not collateral consequences such as deportation); Chin & Holmes, Effective Assistance of Counsel and the Consequences of Guilty Pleas, 87 Cornell L.Rev. 697, 699, 703 (2002) (stating that “virtually all jurisdictions” to have considered the issue had held that “defense lawyers must explain the direct consequences of a plea, such as length of imprisonment and amount of fine, but need not explain ‘collateral consequences,’ such as ... deportation”); Commonwealth v. Clarke, 460 Mass. 30, 35-36, 949 N.E.2d 892 (Mass.2011) (“Padilla effectively changed the law in the nine circuit courts of the United States Court of Appeals that had previously addressed the issue” of whether “defense counsel was ineffective by failing to advise her client of the virtually automatic deportation consequences of his guilty plea”). Courts in at least thirty states and the District of Columbia had reached the same conclusion. 87 Cornell L.Rev. at 699. Such rare unanimity among the lower courts is compelling evidence that reasonable jurists reading the Supreme Court’s precedents in April 2004 could have disagreed about the outcome of Padilla. See Lambrix, 520 U.S. at 538, 117 S.Ct. 1517 (finding it “plain ... that a jurist considering all the relevant material ... could reasonably have reached a conclusion contrary to our holding in” Espinosa v. Florida, 505 U.S. 1079, 112 S.Ct. 2926, 120 L.Ed.2d 854 (1992) (per curiam), where “both before and after [petitioner’s] conviction became final, every court decision we are aware of did so”).
In concluding that Padilla did not announce a new rule, the Third Circuit downplayed the significance of the contrary lower court decisions, reasoning that they generally pre-dated the adoption of the professional norms relied on by the Padilla Court. Orocio, 645 F.3d at 639-40. Not so. While Justice Alito cited primarily pre-1995 cases in his concurrence, in the years preceding Padilla, the lower federal courts consistently reaffirmed that deportation is a collateral consequence of a criminal conviction and that the Sixth Amendment does not require advice regarding collateral consequences. See United States v. Fry, 322 F.3d 1198, 1200 (9th Cir.2003); Jimenez v. United States, 154 Fed.Appx. 540, 541 (7th Cir.2005) (unpublished). In doing so, three Courts of Appeals explicitly rejected the argument that the enactment of the IIRIRA altered the calculus. See United States v. Gonzalez, 202 F.3d 20, 28 (1st Cir.2000) (holding that deportation remained a collateral consequence of conviction after the passage of *691the IIRIRA, and reaffirming its prior conclusion that an attorney’s failure to advise a client of his plea’s immigration consequences does not give rise to a cognizable ineffective assistance claim); Santos-Sanchez v. United States, 548 F.3d 327, 336-37 (5th Cir.2008) (concluding that “neither IIRIRA nor AEDPA has so altered the nature of deportation as to render it a direct consequence of a guilty plea,” and reaffirming that “counsel’s failure to inform a defendant of the collateral consequences of a guilty plea is never deficient performance under Strickland ”); Broomes v. Ashcroft, 358 F.3d 1251, 1257 (10th Cir.2004) (concluding that even under the IIRIRA and AEDPA, “deportation remains a collateral consequence of a criminal conviction, and counsel’s failure to advise a criminal defendant of its possibility does not result in a Sixth Amendment deprivation”).
We acknowledge that “the mere existence of conflicting authority does not necessarily mean a rule is new.” Williams v. Taylor, 529 U.S. 362, 410, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), quoting Wright v. West, 505 U.S. 277, 304, 112 S.Ct. 2482, 120 L.Ed.2d 225 (1992) (O’Con-nor, J., concurring). But, in our view, “an objective reading of the relevant cases” demonstrates that Padilla was not dictated by precedent. Stringer v. Black, 503 U.S. 222, 237, 112 S.Ct. 1130, 117 L.Ed.2d 367 (1992). It is true that, unlike so many lower courts, the Supreme Court has “never applied a distinction between direct and collateral consequences to define the scope of constitutionally ‘reasonable professional assistance’ required under Strickland.” Padilla, 130 S.Ct. at 1481. As such, prior to Padilla, the Court had not foreclosed the possibility that advice regarding collateral consequences of a guilty plea could be constitutionally required. But neither had the Court required defense counsel to provide advice regarding consequences collateral to the criminal prosecution at issue.2 130 S.Ct. at 1495 (Scalia, J., dissenting).
Moreover, the distinction between direct and collateral consequences was not without foundation in Supreme Court precedent. It can be traced to the Court’s jurisprudence regarding the validity of guilty pleas. To be valid, a guilty plea must be both voluntary and intelligent. Brady v. United States, 397 U.S. 742, 747, 90 S.Ct. 1463, 25 L.Ed.2d 747 (1970). The Court has long held that a plea is voluntary where the defendant is “fully aware of the direct consequences” of the plea. Id. The Court also has said that where “a defendant is represented by counsel during the plea process and enters his plea upon the advice of counsel, the voluntariness of the plea depends on whether counsel’s advice ‘was within the range of competence demanded of attorneys in criminal cases.’ ” Hill, 474 U.S. at 56, 106 S.Ct. 366 (citation omitted). At least some lower courts extrapolated from these holdings that counsel performs effectively by advising a client as to the direct consequences of conviction. See 87 Cornell L.Rev. at 726 (attributing the collateral consequences rule to “the Brady Court’s implication that a trial court need advise a defendant only of direct consequences to render a plea voluntary under the Due Process Clause”); Santos v. Kolb, 880 F.2d 941, 944 (7th Cir.1989) (concluding, based on Hill, that “the key to whether defense counsel has failed to provide effective as*692sistance is whether his shortcomings resulted in an involuntary or unintelligent plea”); United States v. Banda, 1 F.3d 354, 356 (5th Cir.1993) (concluding that counsel’s failure to warn of possible deportation did not amount to ineffective assistance of counsel, reasoning that that “conclusion squares with the Supreme Court’s observation that the accused must be ‘fully aware of the direct consequences’ of a guilty plea”) (quoting Brady, 397 U.S. at 755, 90 S.Ct. 1463).
Therefore, we “cannot say that the large majority of federal and state courts that ha[d] rejected” ineffective-assistanee-ofcounsel claims based on advice about the deportation consequences of a plea were “unreasonable” in their reading of existing Supreme Court precedent. Saffle, 494 U.S. at 490, 110 S.Ct. 1257. We consequently remain persuaded by the weight of lower court authority that, in 2004, a jurist could reasonably have reached a conclusion contrary to the holding in Padilla, such that Padilla announced a new rule for purposes of Teague.
As the Massachusetts Supreme Judicial Court recently noted, “[t]here is no question that the holding in Padilla is an extension of the rule in Strickland,” “[n]or is there any question that the Supreme Court was applying the first prong of the Strickland standard when it concluded that the failure of counsel to provide her client with available advice about an issue like deportation was constitutionally deficient.” Clarke, 460 Mass, at 37, 949 N.E.2d 892. However, we disagree with that court’s conclusion that, because “the opinion in Padilla relies primarily on citation to Strickland itself,” Padilla was dictated by Strickland. Id. at 44, 949 N.E.2d 892. Under Teague, a rule is old only if it sets forth the sole reasonable interpretation of existing precedent. Lambrix, 520 U.S. at 538, 117 S.Ct. 1517. The fact that Padilla is an extension of Strickland says nothing about whether it was new or not. See Saffle, 494 U.S. at 488, 110 S.Ct. 1257 (“it is more difficult, however, to determine whether we announce a new rule when a decision extends the reasoning of our prior cases”); Butler, 494 U.S. at 415, 110 S.Ct. 1212 (“the fact that a court says that its decision is within the ‘logical compass’ of an earlier decision, or indeed that it is ‘controlled’ by a prior decision, is not conclusive for purposes of deciding whether the current decision is a ‘new rule’ under Teague ”); Frazer v. South Carolina, 430 F.3d 696, 720 (4th Cir.2005) (Luttig, J., dissenting) (“to establish that the Supreme Court relied exclusively on the principles of prior cases in reaching the rule of [Roe v.] Flores-Ortega [, 528 U.S. 470, 120 S.Ct. 1029, 145 L.Ed.2d 985 (2000) ] is not at all to establish that those cases dictated that rule, that is, that all reasonable jurists would have agreed that those precedents led inexorably to Flores-Ortega ”).
We recognize that the application of Strickland to unique facts generally will not produce a new rule. See Williams, 529 U.S. at 382, 120 S.Ct. 1495 (plurality) (“If the rule in question is one which of necessity requires a case-by-case examination of the evidence, then we can tolerate a number of specific applications without saying that those applications themselves create a new rule”) (quoting Wright v. West, 505 U.S. 277, 309, 112 S.Ct. 2482, 120 L.Ed.2d 225 (1992) (Kennedy, J., concurring)). However, that guiding principle is not absolute. Id. (stating that “[w]here the beginning point is a rule of ... general application, ... it will be the infrequent case that yields a result so novel that it forges a new rule, one not dictated by precedent”).3 We believe Padilla to be *693the rare exception. Before Padilla, the Court had never held that the Sixth Amendment requires a criminal defense attorney to provide advice about matters not directly related to their client’s criminal prosecution. In Padilla, the Court held that constitutionally effective assistance of counsel requires advice about a civil penalty imposed by the Executive Branch (now the Department of Homeland Security, formerly the Immigration and Naturalization Service) after the criminal case is closed. In our view, that result was sufficiently novel to qualify as a new rule. Indeed, if Padilla is considered an old rule, it is hard to imagine an application of Strickland that would qualify as a new rule. Perhaps in the future the Court will conclude, given the breadth and fact-intensive nature of the Strickland reasonableness standard, that cases extending Strickland are never new. But until that time, we are bound to apply Teague in the context of Strickland.
The specific contours of the Padilla holding further indicate that it is a new rule. Under the rule set forth in Padilla, the scope of an attorney’s duty to provide immigration-related advice varies depending on the degree of specialization required to provide such advice accurately. In particular, the Court held that “when the deportation consequence [of a guilty plea] is truly clear,” counsel has a duty to “give correct advice.” 130 S.Ct. at 1483. But “[w]hen the law is not succinct and straightforward,” such that “the deportation consequences of a particular plea are unclear or uncertain,” “a criminal defense attorney need do no more than advise a noncitizen client that pending criminal charges may carry a risk of adverse immigration consequences.” Id. That nuanced, new analysis cannot, in our view, be characterized as having been dictated by precedent.
The district court relied on the fact that Padilla itself was before the Court on a motion for post-conviction relief for its conclusion that the Court intended for Padilla to apply retroactively to cases on collateral appeal. In light of the fact that Kentucky did not raise Teague as a defense in Padilla, we do not assign the significance to Padilla’s procedural posture that the district court did. While “[r]etroactivity is properly treated as a threshold question,” Teague “is not ‘jurisdictional’ in the sense that [the] Court ... must raise and decide the issue sua sponte.” Collins v. Youngblood, 497 U.S. 37, 40-41, 110 S.Ct. 2715, 111 L.Ed.2d 30 (1990) (emphasis in original). Therefore, if a State does not rely on Teague, the Court has no obligation to address it, and can consider the merits of the claim. Caspari *694v. Bohlen, 510 U.S. 383, 389, 114 S.Ct. 948, 127 L.Ed.2d 236 (1994). We believe it is more likely that the Court considered Teague to be waived, than that it silently engaged in a retroactivity analysis.
Finally, the district court reasoned that the best way to make sense of the Padilla Court’s discussion (and dismissal) of concerns that its ruling would undermine the finality of plea-based convictions was to conclude that the majority intended Padilla to apply retroactively. 130 S.Ct. at 1484-85. The Third Circuit reached a similar conclusion. See also Orocio, 645 F.3d at 640-42. That is a reasonable reading, and certainly is the most compelling argument that Padilla is an old rule. However, we are hesitant to depart from our application of the test set forth in Teague and its progeny — which points clearly in the direction of new rule — based on inferences from indirect language. Moreover, to the extent that we attempt to discern whether members of the Court understood Padilla to be a new rule, we find the clearest indications in the concurrence and dissent, which leave no doubt that at least four Justices view Padilla as new.
III. Conclusion
The Supreme Court has defined the concept of an old rule under Teague narrowly, limiting it to those holdings so compelled by precedent that any contrary conclusion must be deemed unreasonable. While determining whether a rule is new can be challenging, and this case provides no exception, we conclude that the narrow definition of what constitutes an old rule tips the scales in favor of finding that Padilla announced a new rule. Moreover, that numerous courts had failed to anticipate the holding in Padilla, though not dispositive, is strong evidence that reasonable jurists could have debated the outcome. For the foregoing reasons, we Reverse the judgment of the district court and Remand the case for further proceedings.
. This opinion and the dissent have been circulated among all judges of this court in regular active service pursuant to Circuit Rule 40(e). A majority of the judges did not favor rehearing en banc. Circuit Judges Rovner, Wood, Williams, and Hamilton voted to rehear the case en banc.
. In Hill, the Court considered whether a criminal defendant's Sixth Amendment right to counsel was violated when his counsel misinformed him about his eligibility for parole, a collateral consequence of conviction. However, the Court found it “unnecessary to determine whether there may be circumstances under which erroneous advice by counsel as to parole eligibility may be deemed constitutionally ineffective assistance of counsel.” 474 U.S. at 60, 106 S.Ct. 366.
. The Williams Court further staled that: "It is true that while the Strickland test provides *693sufficient guidance for resolving virtually all ineffective-assistance-of-counsel claims, there are situations in which the overriding focus on fundamental fairness may affect the analysis.” 529 U.S. at 391, 120 S.Ct. 1495. Some courts, and the dissent, appear to have read the first phrase in that sentence to mean that, in effect, no case applying the Strickland test announces a new rule. See Clarke, 460 Mass, at 39, 949 N.E.2d 892; Lewis v. Johnson, 359 F.3d 646, 655 (3d Cir.2004). We believe the context in which the Court made that assertion undermines that interpretation. In 'Williams, the Court held that the "Virginia Supreme Court erred in holding that ... Lockhart v. Fretwell, 506 U.S. 364, 113 S.Ct. 838, 122 L.Ed.2d 180 (1993), modified or in some way supplanted the rule set down in Strickland." 529 U.S. at 391, 120 S.Ct. 1495. The Court simply was explaining that Strickland remains the test for analyzing ineffective-assistance-of-counsel claims, "virtually all” of which can be resolved without inquiring into "fundamental fairness,” as the Court had in Lockhart. Id. at 391-93, 120 S.Ct. 1495. See also Frazer, 430 F.3d at 723 (Luttig, J., dissenting) for a similar analysis. Moreover, the fact that "the Strickland test provides sufficient guidance for resolving virtually all ineffective-assistance-of-counsel claims,” does not mean it dictates the resolution of all such claims. Id.