dissenting:
I respectfully dissent. I would reverse the district court’s dismissal of Stewart’s habeas petition.
In my view, Stewart was entitled to statutory tolling of the Antiterrorism and Effective Death Penalty Act’s (“AEDPA”) one-year limitations period during the 100-day interval between his successive state habeas petitions. Alternatively, I would remand with instructions to hold an evi-dentiary hearing on Stewart’s claim for an equitable exception to AEDPA’s filing deadline under Schlup v. Delo, 513 U.S. 298, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995), as applied in McQuiggin v. Perkins, — U.S. -, 133 S.Ct. 1924, 185 L.Ed.2d 1019 (2013), to timeliness issues.1 Stewart has presented post-conviction evidence that, if credible, demonstrates this case to be the sort of extraordinary one that, under Schlup, requires reaching the merits of his otherwise barred federal habeas claims.
In holding otherwise, the majority treats the applicable standard of review as unsettled, when that is not so-there is no doubt a de novo standard is applicable. Additionally, the majority seriously misapprehends the record it reviews; on the current record, considered properly, Stewart should be accorded an evidentiary hearing as to his Schlup claim.
I.
A.
AEDPA’s one-year limitations period is tolled for “[t]he time during which a properly filed application for State post-conviction or other collateral review ... is pending.” 28 U.S.C. § 2244(d)(2). A petition is “pending” during the interval between a lower state court’s adverse decision and the filing of an appeal only when the appeal is timely. Evans v. Chavis, 546 U.S. 189, 191, 126 S.Ct. 846, 163 L.Ed.2d 684 (2006). Under California’s unusual collateral review system, petitioners are required to file successive petitions for habe-as corpus to each higher court in turn, each successive petition serving as the equivalent to a notice of appeal. See Walker v. Martin, — U.S. -, 131 S.Ct. 1120, 1125-26, 179 L.Ed.2d 62 (2011) (describing California’s system of post-conviction review). This “equivalent of a notice of appeal is timely if filed within a ‘reasonable time.’ ” Chavis, 546 U.S. at 192, 126 S.Ct. 846 (quoting In re Harris, 5 Cal.4th 813, 828 n. 7, 21 Cal.Rptr.2d 373, 855 P.2d *944391 (1993)). Under this indeterminate system, there is no firm upper limit on the length of time that California courts consider reasonable for interval tolling between successive habeas petitions. See, e.g., In re Reno, 55 Cal.4th 428, 146 Cal. Rptr.3d 297, 283 P.3d 1181, 1208 (2012). By retaining discretion, California courts may “home in on case-specific consideration and ... avoid the harsh results that sometimes attend consistent application of an unyielding rule.” Martin, 131 S.Ct. at 1130.
Chavis permitted interval tolling under California’s system of collateral review on the assumption “that California’s ‘reasonable time’ standard would not lead to filing delays substantially longer than those in States with determinate timeliness rules.” Id. at 199-200, 126 S.Ct. 846 (emphasis added) (citing Carey v. Saffold, 536 U.S. 214, 222-23, 122 S.Ct. 2134, 153 L.Ed.2d 260 (2002)); accord Velasquez v. Kirkland, 639 F.3d 964, 967 (9th Cir.2011). The Court looked to the “30 to 60 days[ ] that most States provide for filing an appeal to the state supreme court”' as a point of comparison for assessing whether the interval between the filing of successive state habeas petitions conforms to California’s “reasonable time” requirement. Chavis, 546 U.S. at 201, 126 S.Ct. 846.
But Chavis did not hold that a delay longer than 60 days could never be reasonable. It held unreasonable only the unexplained six-month delay before the Court. Absent justification, that six-month interval was “far longer than the ‘short period[s] of time’ that most States provide for filing an appeal to the state supreme court.” Id. (alteration in original).2
Whether California courts would consider a given filing delay reasonable is inherently a fact- and circumstance-specific question, Martin, 131 S.Ct. at 1130; the shortest period deemed “unreasonable” under California law in one case does not necessarily mark the longest delay permissible in another. Accordingly, while we have held that multiple unexplained delays of 80 to 115 days between successive petitions are “unreasonable” under California law, see Velasquez, 639 F.3d at 968; Chaffer II, 592 F.3d at 1048, that does not mean that the single period of 100 days at issue in Stewart’s case could not be reasonable, in light of his circumstances. In particular, the petitioner in Velasquez was “represented by counsel at all stages.” 639 F.3d at 968. Stewart, in contrast, filed his petition to the California Supreme Court pro se, a difference California courts often emphasize in holding delayed filings reasonable. See, e.g., In re Saunders, 2 Cal.3d 1033, 88 Cal.Rptr. 633, 472 P.2d 921, 925-26 (1970); In re Lucero, 200 Cal. App.4th 38, 44-45, 132 Cal.Rptr.3d 499 (2011); In re Burdan, 169 Cal.App.4th 18, 31, 86 Cal.Rptr.3d 549 (2008).
In a recent case, we left open the possibility that a four-and-a-half-month delay may be reasonable under California law. *945We remanded to the district court to determine in the first instance whether the prisoner’s petition was timely, noting that the interval “may be within the range of reasonableness if [the petitioner’s] explanation for the delay is adequate under California law.” Noble v. Adams, 676 F.3d 1180, 1184 (9th Cir.2012). Noble emphasized that “the California Court of Appeal has excused delays of several months where the petitioner offered an adequate explanation for the delay,” particularly in pro se cases. Id. (citing In re Burdan, 169 Cal.App.4th at 30-31, 86 Cal.Rptr.3d 549 (excusing a delay of ten months for a pro se petitioner whose attorney said he would handle the appeal but did not do so); In re Crockett, 159 Cal.App.4th 751, 757-58, 71 Cal.Rptr.3d 632 (2008) (excusing a delay of approximately five months where the attorney “had no prior experience with appellate writs and could not obtain the assistance of experienced appellate counsel”)).
The majority’s attempt to cabin the holding of In re Burdan, one of the cases cited in Noble, to the context of petitions challenging parole decisions is unconvincing. To be sure, In re Burdan explained that the rationale for requiring prompt filing of habeas petitions challenging convictions in the capital context does not apply “where a life prisoner challenges a parole decision.” 169 Cal.App.4th at 31, 86 Cal.Rptr.3d 549. But the court went on to state in a separate paragraph, “[a]t any rate” — in others words, the policy considerations underlying the timeliness rules aside — “[w]e do not find a delay of 10 months for an unrepresented prison inmate to file a petition for writ of habeas corpus in the Court of Appeal, after denial of a similar petition in the superior court; to be unreasonable.” Id.
In addition to the cases cited in Noble, several other California cases have permitted the filing of a habeas petition more than 100 days after a lower court’s denial of the preceding petition. See In re Bower, 38 Cal.3d 865, 873 n. 3, 215 Cal.Rptr. 267, 700 P.2d 1269 (1985); In re Moss, 175 Cal.App.3d 913, 921-22, 221 Cal.Rptr. 645 (1985); In re Spears, 157 Cal.App.3d 1203, 1208, 204 Cal.Rptr. 333 (1984). The majority finds none of these cases persuasive, but only because it reads them selectively.
For example, the majority dismisses the relevance of In re Spears and In re Moss on the ground that they predate the California Supreme Court’s express use of the “reasonable time” language from In re Harris, 5 Cal.4th at 828 n. 7, 21 Cal. Rptr.2d 373, 855 P.2d 391 (citing In re Clark, 5 Cal.4th 750, 21 Cal.Rptr.2d 509, 855 P.2d 729, 751-52 (1993)), and therefore contain no discussion of the operative timeliness standard. See Maj. Op. at 936-37 & n.9. But Harris relied on In re Clark, which, by its own terms, did not “modify the timeliness requirements applicable to all habeas corpus petitions,” except for certain exceptions not pertinent here. In re Clark, 21 Cal.Rptr.2d 509, 855 P.2d at 751; accord In re Sanders, 21 Cal.4th 697, 87 Cal.Rptr.2d 899, 981 P.2d 1038, 1047 (1999). In any case, the majority overlooks the content of the Spears and Moss: Spears unequivocally states that “18 months is not a significant delay.” 157 Cal.App.3d at 1208, 204 Cal.Rptr. 333 (citing In re Hancock, 67 Cal.App.3d 943, 945 n. 1, 136 Cal.Rptr. 901 (1972)). And Moss held that, given the petitioner’s inability to secure appellate counsel, “a delay of nine months ... is not a significant delay.” 175 Cal.App.3d at 922, 221 Cal.Rptr. 645.
The majority’s dismissal of In re Bower as insufficiently reasoned is equally unconvincing. See Maj. Op. at 936-37. Rather, the cursory character of Bower’s analysis indicates that the Supreme Court of California considered it unnecessary to expound on why a delay of less than one year *946is not “undue.” See 38 Cal.3d at 873 n. 3, 215 Cal.Rptr. 267, 700 P.2d 1269. Presumably, a considerably shorter delay, 100 days, would require even less explanation.
In short, the single 100-day delay between the California Court of Appeal’s denial of Stewart’s petition on May 23, 2003 and the filing of his subsequent pro se petition in the California Supreme Court on August 31, 2013 was not necessarily unreasonable under California law.
B.
Moreover, unlike the petitioner in Velasquez, 639 F.3d at 968, and like the petitioners in Spears, Moss, Bower, and Burdan, Stewart offered an explanation for the delay, one that California courts would likely find adequate. In his petition to the Supreme Court of California, Stewart wrote: “Due to [sic] state of emergency status declared June 23, 2003 at Salinas Valley State Prison, and the transferring of petitioner out of Salinas Valley State Prison to Calipatria State Prison, the writ is brought in a timely period.” The district court did not acknowledge this explanation, let alone assess its adequacy under California law. The majority, in turn, dismisses the state of emergency and prison transfer as an “implausible” excuse for the filing delay, on the ground that no research was required in the preparation of the petition to the California Supreme Court because it presented no new evidence or claims. See Maj. Op. at 937.
In sharp contrast to the majority’s dismissive approach to Stewart’s explanation, the Supreme Court in Chavis evaluated a California habeas petitioner’s explanation for his delay by “viewing every disputed issue most favorably to [him].” 546 U.S. at 201, 126 S.Ct. 846. Viewed through the Chavis prism, rather than that of the majority, Stewart’s explanation for the delay in filing his California Supreme Court petition passes muster. The need for ongoing investigation or research is but “one example of good cause,” not the only potential justification for delay during prison lockdowns and transfers. In re Robbins, 18 Cal.4th 770, 805, 77 Cal.Rptr.2d 153, 959 P.2d 311 (1998). There are other impediments that a prison’s emergency status may pose to the timely filing of a prisoner’s pro se petition. While Stewart’s opening brief on appeal mentioned his inability to “research his petition” due to “a lock-down” and “subsequent transfer,” it also cited, more generally, “limited access to resources due to the prison emergency and subsequent transfer.”
The mechanics of re-drafting and filing a petition, including obtaining the papers previously filed, copying them by accessing means of duplication, procuring suitable packaging material for the documents, and determining the location of the appropriate court, can all be onerous for a pro se prisoner, lacking both legal training and resources. Stewart’s assertions permit an inference that the restrictions on his movements imposed during the state of emergency and changes in his custodial conditions made it difficult for him to accomplish such tasks. Thus, while the fact that the claims and supporting evidence presented in successive habeas petitions are nearly identical undercuts excusing a lawyer’s delay in filing a petition, see Velasquez, 639 F.3d at 968, the multiple difficulties prisoners proceeding pro se encounter while in lockdown and during transfers are indeed pertinent to excusing delay, even where the successive petitions are similar or identical.
Finally, that Stewart’s “explanation” appeared in the “Supporting Facts” section of his pro se petition to the California Supreme Court rather than the designated portion of the form, see Maj. Op. at 937, has no bearing on whether it may ade*947quately justify the delay. I have found no indication — and the majority cites none— that California courts would consider such a technical factor dispositive and so refuse to consider an explanation that does provide good cause for delay.
I conclude that, in light of the explanation Stewart offered, California courts would not consider the 100-day period between the state Court of Appeal disposition and the filing of Stewart’s habeas petition in the California Supreme Court unreasonable. The majority’s contrary conclusion generates precisely the sort of “harsh results” California has sought to avoid, by adopting a discretionary system. Martin, 131 S.Ct. at 1130. I would therefore reverse the district court’s dismissal of Stewart’s petition as untimely.
II.
Even if Stewart is not entitled to statutory tolling and his federal petition would therefore be time-barred under AEDPA, he may be entitled to an equitable exception to § 2244(d)(1). In my view, he has presented evidence of actual innocence that, if credible, is sufficient to pass through the Schlup gateway. See Perkins, 133 S.Ct. at 1931; Larsen v. Soto, 742 F.3d 1083, 1095 (9th Cir.2013); Lee v. Lampert, 653 F.3d 929, 934 (9th Cir.2011) (en banc). Reviewing Stewart’s Schlup showing de novo, and in light of all available evidence, I would hold that an evidentiary hearing is necessary to determine whether he has shown that it is more likely than not that no reasonable juror would have convicted him on the supplemented record, and thus that he is entitled to have his “otherwise time-barred claims heard on the merits.” Lee, 653 F.3d at 932, 937. I would therefore reverse and remand for the district court to hold a Schlup hearing.
A.
The majority suggests that the case law is ambiguous concerning the correct standard of review — de novo or abuse of discretion — for Schlup actual innocence claims. That characterization is plain wrong. By reopening a settled issue, the majority has, with no justification, muddied our jurisprudence. Moreover, Schlup claims arise with some frequency, and, as I believe may be true here, the correct, de novo standard of review can often make a difference.
As the majority acknowledges, see Maj. Op. at 933-34, we generally review the denial of a § 2254 petition de novo, whether that denial turns on procedural grounds or on the merits of the petition. See Stancle v. Clay, 692 F.3d 948, 952-53 (9th Cir.2012) (reviewing de novo the denial of a petition on timeliness grounds); Waldron-Ramsey v. Pacholke, 556 F.3d 1008, 1011 (9th Cir.2009) (reviewing de novo whether the statute of limitations should be equitably tolled). The same standard applies to the district court’s determination that a petitioner has failed to meet Schlup’s actual innocence standard. Neither precedent nor policy supports any contrary conclusion.
i.
Precedent makes clear beyond dispute that de novo review is proper. Most notably, recent en banc decisions have evaluated Schlup claims by independently and exhaustively reviewing the supplemented records that such cases present. In Lee v. Lampert, for example, an en banc panel of this Court intensively canvassed the supplemented record to make its oum “probabilistic determination about what reasonable, properly instructed jurors would do” if faced with the new evidence, as the Schlup standard demands. 653 F.3d at 938 (internal quotation marks omitted); see also Smith v. Baldwin, 510 F.3d 1127, 1139-46 (9th Cir.2007) (en banc) (canvassing the supplemented record exhaustively *948to assess a Schlup claim). In Sistrunk v. Armenakis, to name another example, “[t]he sole issue on appeal” was a Schlup claim, yet the en banc court still observed that the denial of a habeas petition was properly reviewed de novo. 292 F.3d 669, 672-77 (9th Cir.2002) (en banc). Three judge panels of this Court have conducted similarly comprehensive and independent reviews of the supplemented records petitioners have presented to them. See, e.g., Gandarela v. Johnson, 286 F.3d 1080, 1085-87 (9th Cir.2001); Wood v. Hall, 130 F.3d 373, 379 (9th Cir.1997). Most recently, a threejudge panel of this Court observed that satisfaction of “the Schlup standard is a legal question that we review de novo.” Larsen, 742 F.3d at 1086 n. 6. Although none of these cases explained why they conducted de novo review of Schlup claims, that is exactly what they did. And the circuits that have confronted the issue expressly have all concluded that de novo review of Schlup claims is appropriate. See Munchinski v. Wilson, 694 F.3d 308, 337 (3d Cir.2012); Doe v. Menefee, 391 F.3d 147, 163 (2d Cir.2004); United States ex rel. Bell v. Pierson, 267 F.3d 544, 552 (7th Cir.2001); Beeman v. Iowa, 108 F.3d 181, 184 (8th Cir.1997); O’Dell v. Netherland, 95 F.3d 1214, 1250 (4th Cir. 1996) (en banc), aff'd, 521 U.S. 151, 117 S.Ct. 1969, 138 L.Ed.2d 351 (1997).
True, as the majority points out, Parad-is v. Arave reversed the denial of a habeas petition alleging actual innocence under Schlup on the ground that, because “the district court applied an erroneous [legal] standard for the showing of actual innocence here, it abused its discretion.” 130 F.3d 385, 396 (9th Cir.1997). But that decision preceded the en banc cases I have cited. See Ortega-Mendez v. Gonzales, 450 F.3d 1010, 1019 (9th Cir.2006) (explaining that three-judge panels are not bound by the decision of a prior three-judge panel when it is irreconcilable with a subsequent en banc decision). Further, Paradis mentioned the standard of review in passing; it had no occasion to consider, and did not, whether a more stringent standard was appropriate, as the district court’s decision flunked even the most deferential review. Further, where the challenge is to the legal standard applied, abuse of discretion and de novo review reach the same result; it is with respect to mixed questions of fact and law that the difference between the two standards matters most, cf. United States v. Hinkson, 585 F.3d 1247, 1259-60 (9th Cir.2009) (en banc), and such a question was not at issue in Parad-is.
The majority’s invocation of Justice O’Connor’s concurrence in Schlup — which supposedly supports abuse-of-discretion review — fares no better. As Justice O’Connor recognized in her Schlup concurrence, the majority opinion in that case did not “speak to the standard of appellate review.” 513 U.S. at 334, 115 S.Ct. 851. It did not need to do so, because the underlying decision in that case was — just as in Paradis — based “on an erroneous view of the law,” and thus failed even under deferential review. Id. at 333, 115 S.Ct. 851. A concurrence of one justice as to an issue concededly not decided by the majority is not precedent.
Moreover, a Supreme Court case subsequent to Schlup, House v. Bell, 547 U.S. 518, 126 S.Ct. 2064, 165 L.Ed.2d 1 (2006), essentially applied de novo review to a Schlup claim. That case observed that “[djeference is given to a trial court’s assessment of evidence presented to it in the first instance.” Id. at 539, 126 S.Ct. 2064. But House then explained that “the Schlup inquiry ... requires a holistic judgment about ‘all the evidence,’ ” an inquiry which “does not turn on discrete findings regarding disputed points of fact, and ‘[i]t is not the district court’s independent judgment as to whether reasonable doubt exists that *949the standard addresses.’” Id. at 539-40, 126 S.Ct. 2064 (quoting Schlup, 513 U.S. at 328, 329, 115 S.Ct. 851) (emphases added). The majority recognizes, correctly, that the standard in House “approximates de novo review.” Maj. Op. at 938.
Here — unlike in House — there has been no evidentiary hearing, and the trial court purported to accept all the evidence “presented to it in the first instance” as credible. There was therefore no occasion to make factual findings as to that evidence in isolation. House thus confirms that the application of the Schlup standard on the entire record is one as to which the district court neither finds facts nor exercises discretion. De novo review is therefore appropriate.3
In short, my colleagues read the relevant precedents incorrectly. The applicable case law is in no way indeterminate as to the applicable standard of review for Schlup issues. De novo consideration is the proper standard. The majority’s equivocation upends settled law, sowing doubt where there should be none. “Our task to clarify the law — not to muddy the waters.” United States v. Virginia, 518 U.S. 515, 574, 116 S.Ct. 2264, 135 L.Ed.2d 735 (1996) (Scalia, J., dissenting). The majority inverts that eommonsense maxim.
ii.
If the proper standard of review were an open question — which it is not — we should be answering that question, not leaving it open to perplex future litigants and panels. And were we to approach the issue afresh, we could only conclude that de novo review is proper for Schlup claims.
selecting an appropriate standard of review often requires assessing whether “one judicial actor ... [is] better positioned than another to decide the issue in question.” Cooter & Gell v. Hartmarx Corp., 496 U.S. 384, 403, 110 S.Ct. 2447, 110 L.Ed.2d 359 (1990) (internal quotation marks omitted). De novo review is especially appropriate where, as here, relief turns on application of “‘fluid concepts that take their substantive content from the particular contexts in which the standards are being assessed.’ ” Cooper Indus., Inc. v. Leatherman Tool Grp., Inc., 532 U.S. 424, 436, 121 S.Ct. 1678, 149 L.Ed.2d 674 (2001) (quoting Ornelas v. United States, 517 U.S. 690, 696, 116 S.Ct. 1657, 134 L.Ed.2d 911 (1996)); accord Harte-Hanks Commc’ns, Inc. v. Connaughton, 491 U.S. 657, 686, 109 S.Ct. 2678, 105 L.Ed.2d 562 (1989). Allocating authority over such concepts to appellate courts with supervisory power facilitates the uniform development of the law and enhances predictability. Cooper Indus., 532 U.S. at 436, 121 S.Ct. 1678; Ornelas, 517 U.S. at 697, 116 S.Ct. 1657; Salve Regina College v. Russell, 499 U.S. 225, 231, 111 S.Ct. 1217, 113 L.Ed.2d 190 (1991). Such coherence and uniformity are especially necessary where “the stakes — in terms of impact on future cases and future conduct— are too great to entrust them finally to the judgment of the trier of fact.” Bose Corp. v. Consumers Union of the United States, Inc., 466 U.S. 485, 501 n. 17, 104 S.Ct. 1949, 80 L.Ed.2d 502 (1984).
All these concerns apply here. The Schlup standard is, in its very nature, “fluid.” Much like probable cause or reasonable suspicion, which the Court has *950characterized as “fluid concepts” subject to de novo review, Ornelas, 517 U.S. at 696, 116 S.Ct. 1657, the Schlup standard describes a probabilistic inquiry that emphasizes differences of degree rather than kind. Retaining appellate control over that- standard is essential for two, interrelated reasons. First, the interest Schlup protects — avoiding conviction of an actually innocent person — reflects a concern that . “has long been at the core of our criminal justice system.” 513 U.S. at 325, 115 S.Ct. 851. Second, and conversely, appellate courts must guard against excessive application of the Schlup standard, “[t]o ensure that the fundamental miscarriage of justice exception ... remain[s] ‘rare’ and [is] only applied in the ‘extraordinary case.’ ” Id. at 321, 324, 115 S.Ct. 851. Preserving the narrow scope of such a vague but important standard thus requires substantial appellate oversight.
By contrast, none of the usual justifications for deference apply here. “When, for example, the issue involves the credibility of witnesses and therefore turns largely on an evaluation of demeanor, there are compelling and familiar justifications for leaving the process of applying law to fact to the trial court and according its determinations presumptive weight.” Miller, 474 U.S. at 114, 106 S.Ct. 445. At the pre-evidentiary hearing stage, this consideration has no application whatever (and only a limited application after an eviden-tiary hearing, as House explains, 547 U.S. at 539-40, 126 S.Ct. 2064).
Similarly, “it is ‘especially common’ for issues involving supervision of litigation to be reviewed for abuse of discretion.” Salve Regina, 499 U.S. at 233, 111 S.Ct. 1217 (quoting Pierce v. Underwood, 487 U.S. 552, 558 n. 1, 108 S.Ct. 2541, 101 L.Ed.2d 490 (1988)). This deference recognizes the trial judge’s traditional control over the conduct of a trial, and his need for “broad power to cope with the complexities and contingencies inherent in the adversary process.” Geders v. United States, 425 U.S. 80, 86, 96 S.Ct. 1330, 47 L.Ed.2d 592 (1976). The Schlup standard, however, concerns matters never presented at trial, and thus does not trespass on the trial judge’s traditional bailiwick. Indeed, a court evaluating a Schlup claim is not “bound by the rules of admissibility that would govern at trial” and may include in its assessment evidence that “was either excluded or unavailable at trial.” Schlup, 513 U.S. at 327-28, 115 S.Ct. 851; accord Lee, 653 F.3d at 938 (quoting House, 547 U.S. at 538, 126 S.Ct. 2064).
Ultimately, then, pertinent policy considerations compel the conclusion that precedent supports: This Court must review Schlup claims de novo, not for abuse of discretion.
iii.
Analogy to parallel bodies of law further confirms this conclusion. Consider, for example, review of claims of insufficient evidence under Jackson v. Virginia, 443 U.S. 307, 99 S.Ct. 2781, 61 L.Ed.2d 560 (1979). The Schlup standard is structurally similar to — although it is substantively distinct from- — the Jackson standard. Whereas Jackson asks whether a rational trier of fact could convict on the record evidence, Schlup asks whether a reasonable trier of fact would more likely than not convict on a supplemented record. See Schlup, 513 U.S. at 330, 115 S.Ct. 851; Lee, 653 F.3d at 938 n.13.
If anything, Schlup claims are more amenable to de novo review than Jackson claims: Jackson entails assessment of the trier of fact’s actual findings, usually after the presentation of live testimony, and thus treads more heavily on the factfin-der’s traditional prerogatives. See Schlup, 513 U.S. at 330, 115 S.Ct. 851. By contrast, the Schlup standard “focuses the *951inquiry on the likely behavior of the trier of fact,” before the trier of fact has had any opportunity to consider evidence contained in the supplemented record. Id. Nevertheless, this Court reviews de novo a district court’s assessment of the evidence under Jackson. See, e.g., United States v. Garrido, 713 F.3d 985, 999 (9th Cir.2013). A fortiori, it ought to review Schlup claims under the same standard.
Harmless error determinations offer a second useful analogy. Harmless error standards vary widely by doctrinal context. See, e.g., Brecht v. Abrahamson, 507 U.S. 619, 637, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993); Strickland v. Washington, 466 U.S. 668, 694, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984); Chapman v. California, 386 U.S. 18, 24, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967). Despite this variety, harmless error analysis, much like the Schlup standard, always concerns a hypothetical question — whether the trier of fact would have arrived at a different conclusion but for the error. By no means does such coun-terfactual analysis require appellate courts to act as “a second jury to determine whether the defendant is guilty.” Neder v. United States, 527 U.S. 1, 19, 119 S.Ct. 1827, 144 L.Ed.2d 35 (1999) (internal quotation marks omitted). Instead, the court reviews the entirety of the record “‘in typical appellate-courtfashion ’ ” to predict the outcome absent the error. United States v. Rodrigues, 678 F.3d 693, 696 (9th Cir.2012) (emphasis added) (quoting Neder, 527 U.S. at 19, 119 S.Ct. 1827). Such review is de novo. Id. So, too, here: Schlup poses a hypothetical question answered by surveying the record, and we appropriately review the district court’s counterfactual conclusion de novo.
iv.
Finally, neither the lengthiness nor the commendable exhaustiveness of the district court’s discussion of the evidence can alter the de novo nature of our review. To be sure, the district court’s thorough analysis of the record greatly eases our work, by blazing a trail through the evidence and pointing to the factual intersections most important to our ultimate inquiry. It remains our task, however, to predict from the supplemented record in its entirety, not what actually happened, but what a reasonable and properly instructed jury would have concluded, had it been faced with the new evidence as well as that presented at trial.4
As to the district court’s decision not to hold an evidentiary hearing before ruling on Stewart’s Schlup claim, that ruling, as the majority correctly notes, is reviewed for abuse of discretion. See Smith, 510 F.3d at 1137. But if we are unable, without an evidentiary hearing, to determine the “likely impact” of the petitioner’s new evidence of innocence “on reasonable jurors,” Lee, 653 F.3d at 945, as Schlup requires, then the district court’s denial of Stewart’s request for such a hearing constitutes an abuse of discretion. In other words, if our probabilistic determination as to whether jurors would convict Stewart necessarily depends upon the relative credibility and accuracy of the new witnesses as compared to those who gave contrary testimony at trial, then the refus*952al to hold an evidentiary hearing is an abuse of discretion.
B.
Applying the proper standard of review to Stewart’s claim, I conclude that he has met the Schlup standard sufficiently to require an evidentiary hearing.
The Schlup standard is “exacting”; it “ ‘permits review only in the “extraordinary” case.’ ” Lee, 653 F.3d at 938 (quoting House, 547 U.S. at 538, 126 S.Ct. 2064). “[T]enable actual-innocence gateway pleas are rare: ‘[A] petitioner does not meet the threshold requirement unless he persuades the district court that, in light of the new evidence, no juror, acting reasonably, would have voted to find him guilty beyond a reasonable doubt.’ ” Perkins, 133 S.Ct. at 1928 (alteration in original) (quoting Schlup, 513 U.S. at 329, 115 S.Ct. 851). This “demanding” standard is “seldom met.” Id.
Still, Schlup “does not require absolute certainty about the petitioner’s guilt or innocence.” House, 547 U.S. at 538, 126 S.Ct. 2064. “[W]here post-conviction evidence casts doubt on the conviction by undercutting the reliability of the proof of guilt, but not by affirmatively proving innocence, that can be enough to pass through the Schlup gateway to allow consideration of otherwise barred claims.” Lee, 653 F.3d at 938, 943 (quoting Sistrunk, 292 F.3d at 673).
In my view, the record reveals this case may well be an “extraordinary” one of the type contemplated by Schlup and Perkins, see Perkins, 133 S.Ct. at 1933 — that is, one “fall[ing] within the narrow class of cases implicating a fundamental miscarriage of justice.” Majoy v. Roe, 296 F.3d 770, 776 (9th Cir.2002). “This is not a case of conclusive exoneration. Some aspects of the State’s evidence ... still support an inference of guilt. Yet the central ... proof connecting [Stewart] to the crime”— here the victims’ eyewitness identification of their assailants — “has been called into question, and [Stewart] has put forward substantial evidence pointing to a different suspect” — namely, Darnell Jackson. House, 547 U.S. at 553-54, 126 S.Ct. 2064. Absent an evidentiary hearing at which both the credibility of the post-trial witnesses and the accuracy of their statements are probed, the district court could not properly determine, and we cannot properly review, whether, after considering all of the conflicting testimony, it is “more likely than not [that], in light of the new evidence, ... any reasonable juror would have reasonable doubt” as to Stewart’s guilt. Id. at 538, 126 S.Ct. 2064 (emphasis added).
I would remand for the district court to determine, after a Schlup hearing, whether this is such a “rare case.” Id. at 554, 126 S.Ct. 2064. To explain why requires examining the facts of this case in some detail.
i.
Stewart was convicted, after a joint trial, of two counts of attempted murder. He was identified at trial as the driver of a car from which a passenger, Stewart’s co-defendant, Richard Lee, shot Mark and Michael Parish (“the Parish brothers”) in a nearby vehicle, injuring both non-fatally. The prosecution’s case against both defendants rested chiefly on the eyewitness testimony of the victims, who insisted that they had seen clearly the faces of their assailants — both the shooter, who was physically closer to them (according to Mark Parish’s trial testimony, approximately eight feet away), and the driver— and could also identify the car involved. In closing arguments, the prosecution underscored the Parish brothers’ certainty of their identification, repeating the Parish’s words: “I’ll never forget the people who *953shot me. I’m not mistaken.” And, to further emphasize the victims’ reliability, the government explained the risks the Parish brothers faced in testifying, because they would be branded as “snitches” in their gang milieu.
The prosecution’s case against Stewart also relied heavily on descriptions of the car from which the victims were shot. As I detail in Part II.B.ii those descriptions, unlike the identification of the assailants, were wildly contradictory, both for each individual witness and among witnesses, and so worth little as evidence against Stewart.
The prosecution’s additional evidence against Stewart included: Stewart’s affiliation with the Skyline Pirus, which was hostile to the Parish brothers’ gang; police testimony that Stewart was found, hours after the shooting, cleaning his mother’s BMW; evidence of gun-shot residue detected on Stewart’s vehicle the night of the shooting; contradictory accounts, including from Stewart and his mother, regarding Stewart’s whereabouts on the evening of the shooting; and testimony of an acquaintance of the Parish brothers, Kevin Brown, that Stewart had gloated sometime after the shooting, “I killed your homeboy,” which Brown assumed was a reference to the shooting of the Parish brothers (even though there were two of them and they were not killed).
In 2000, four years after the trial, Lee’s conviction was overturned based on a statement from a third party informant and friend of Lee’s, Darnell Jackson, that directly contradicted the victims’ testimony as to the identity of the shooter. Jackson named Arnold Adkins, recently deceased, as the shooter of the Parish brothers, while asserting that Stewart was the driver of the car from which they were shot. It does not appear from the record that Jackson’s statement was made under oath. Nonetheless, the government did not oppose Lee’s habeas petition, conceding that “there appears to be newly discovered evidence which is sufficiently credible to cast doubt on the integrity of [Lee’s] convictions.”
Because his conviction was based largely on the very same victim accounts called into question by -the post-trial evidence exculpating Lee, Stewart filed his own state habeas petition, asserting his innocence. See In re Weber, 11 Cal.3d 703, 724, 114 Cal.Rptr. 429, 523 P.2d 229 (1974). The basis for his petition was evidence acquired through his attorney’s investigation of the information provided in Lee’s habeas case — namely, a letter sent by the District Attorney to Lee’s counsel describing Jackson’s statement.5 Stewart contended that Jackson’s statement gravely impeached the Parish brothers’ account of the crime, and that other evidence indicated Jackson, not Stewart, had been the driver. In support of his petition, Stewart submitted declarations from three acquaintances of Jackson’s — Roy Vinson, Arnold Johnson, and Tatianna Daniels — all of whom implicated Jackson in the Parish brothers’ shooting; a statement from Maurice League, a friend of both Mark Parish and Stewart; a statement from Stewart’s co-defendant, Lee; and a statement from Stewart’s attorney.
The California Superior Court denied post-conviction relief, holding that “the evidence is not credible because of inherent inaccuracies and witness bias” and fails to *954“completely undermine the prosecution’s ease.” In the last reasoned state court decision on Stewart’s case, the California Court of Appeal agreed, concluding that “[t]he declarations at best raise an issue of credibility without providing a complete defense.” Because the new evidence does not “point unerringly to innocence,” the Court of Appeal held, it does not justify acquittal on the basis of a substantive claim of actual innocence. Neither state court held an evidentiary hearing.
The burdens of proof and standards applicable to a substantive actual innocence claim under California law and a Schlup claim of procedural actual innocence for federal habeas purposes are different. See Schlup, 513 U.S. at 313-16, 115 S.Ct. 851; In re Clark, 5 Cal.4th 750, 766, 21 Cal.Rptr.2d 509, 855 P.2d 729 (1993); In re Weber, 11 Cal.3d at 724, 114 Cal.Rptr. 429, 523 P.2d 229. To demonstrate actual innocence, California courts require a “complete defense” and evidence pointing “unerringly to innocence.” In re Weber, 11 Cal.3d at 724, 725, 114 Cal.Rptr. 429, 523 P.2d 229. In contrast, new evidence undermining the credibility of trial witnesses carries weight under Schlup, as credibility assessments are within the scope of the habeas court’s review, and affirmative proof of innocence is not necessary. See Schlup, 513 U.S. at 330, 115 S.Ct. 851. Thus, Stewart does not need to clear the same hurdles to pass through the Schlup gateway as he did to establish substantive actual innocence under California law.
ii.
Here is a summary of the new evidence presented on habeas:
First, Jackson’s ex-girlfriend, Tatianna Daniels, attested that Jackson confessed to her that he was the driver of the vehicle from which Arnold Adkins shot the Parish brothers. Her declaration states that Jackson drove “a red or reddish-brown 4-door Honda,” which he indicated was the car used in the shooting.
Second, Vinson’s declaration provides circumstantial evidence of Jackson’s involvement, although it does not directly name Jackson as the driver of the car from which the Parish brothers were shot. Vinson stated that around the time of the shooting (September 23, 1995), Adkins and Jackson came to Vinson’s house in a rust-colored Honda, seeking to rid themselves of a “hot” .357 revolver, the type of weapon used in the Parish brothers shooting. Around that same time, Vinson continued, Adkins confessed to shooting two men and said that the person with him at the time of the shooting was a “close friend.” Vinson knew that Adkins and Jackson were close friends and spent a lot of time together.
The circumstantial evidence in Vinson’s declaration implicating Jackson is corroborated by Haughen, Stewart’s attorney. Haughen’s declaration states that Vinson informed him by letter that Jackson was the driver of the vehicle from which Adkins shot the Parish brothers.
Third, Johnson’s declaration states that Adkins told him in 1996 that he (Adkins) and Jackson had been fired on while at a night club sometime in early 1995, and had subsequently “taken care of’ the people responsible. When he asked Adkins what he meant by “taken care” of them, Johnson attested, Adkins said “he and Darnell Jackson had shot and killed them.” Adkins was killed not long after this exchange with Johnson. Johnson declared that he spoke with Jackson at Adkins’ funeral, where Jackson told him he did not think that Adkins’ death had anything to do with the shootings in which he (Jackson) and Adkins had been involved, because “everyone believed that two other guys named PeeDu [Stewart’s street moniker] and Puff [Lee’s street moniker] had *955done the shooting and had been locked up for it.”6
Fourth, Maurice League stated that Mark Parish told him he was unsure who shot him. At trial, a defense witness, Kenneth Anderson, testified that Mark’s brother, Michael, did not know if Stewart and his co-defendant Lee were involved, but was^ pressured to identify them.
Far from being “duplicative” of Anderson’s testimony, as the majority asserts, Maj. Op. at 940, League’s declaration is entirely distinct from — though broadly consistent with — evidence presented at trial, if for no other reason than that it pertains to the other brother, Mark. The testimony of two different individuals to similar but distinct facts is corroborative, not cumulative, and thus bears on the perceived credibility of the Parish brothers. Moreover, even cumulative evidence can “tip the scales” for or against a defendant. See Parle v. Runnels, 505 F.3d 922, 928 (9th Cir.2007) (alterations omitted) (quoting Krulewitch v. United States, 336 U.S. 440, 444-45, 69 S.Ct. 716, 93 L.Ed. 790 (1949); Hawkins v. United States, 358 U.S. 74, 80, 79 S.Ct. 136, 3 L.Ed.2d 125 (1958)) (holding prejudicial erroneously admitted evidence, even if cumulative).
Fifth, Lee, Stewart’s former co-defendant, stated that the District Attorney contacted Lee with an offer to provide Lee with information, furnished by Jackson, that would exonerate Lee of the Parish brothers’ shooting, if Lee promised not to testify in an upcoming trial at which Jackson was a key witness. Lee’s testimony in that trial would have impeached Jackson. Lee agreed to this quid pro quo. His attorney filed a habeas petition on his behalf, using the information provided by the District Attorney.
Lee’s declaration also states that, in the course of the District Attorney’s investigation of his habeas petition, he was interviewed while housed at Salinas Valley State Prison about the Parish brothers’ incident. Lee attested that he told the investigator that neither he (Lee) nor Stewart was involved in the shooting, but that two other individuals were responsible. Lee’s declaration does not, however, name those individuals.
The district court gave little weight to Lee’s statement because he did not disclose the identity of the driver of the vehicle. But if the driver was Jackson, as Stewart’s post-trial evidence indicates, Lee’s refusal to name him could be explained by his desire not to impugn the credibility of the very individual whose statement secured his own release.
Finally, in support of his Schlup claim, Stewart requested discovery of all documents on which the District Attorney’s office based the decision not to oppose Lee’s habeas petition. After reviewing those materials in camera, the district court denied the discovery request, concluding that the evidence would not “exonerate” Stewart, and thus did not advance his Schlup claim. My own inspection of those materials, however, leads me to conclude that they would help to meet the Schlup standard — which is not exoneration — and that discovery under seal therefore should have been allowed. The sealed documents are relevant not only to the reliability of evidence presented at trial but also to the reliability of the post-trial declarations Stewart presented on habeas. They farther undermine the reliability of the Parish brothers’ testimony about the identity of their assailants and provide some circumstantial evidence that Jackson, *956not Stewart, may have been involved in the crime. The district court’s denial of Stewart’s discovery motion was therefore an abuse of discretion. See Cooper v. Brown, 510 F.3d 870, 877 (9th Cir.2007) (identifying the applicable standard of review).
If credible, the new evidence would thus make it more likely than not that no reasonable juror would convict Stewart based on the supplemented record.
The majority’s contrary conclusion rests on two assumptions. First, the majority assumes that the evidence exonerating Lee only undermines the, Parish brothers’ identification of Stewart “by implication,” if at all, because “in fact, Jackson’s confession implicated Stewart while exculpating Lee.” Maj. Op. at 940. But Jackson’s statement impeaching the Parish brothers’ identification of Lee as the shooter significantly damages the credibility of the Parish brothers’ testimony, given that: (1) the shooter was closer to the brothers than the driver; and (2) the brothers repeatedly avowed their certainty regarding the shooter’s identity, yet were disproved as to that identity by Jackson.
This impeachment amplifies the problems that already marred the brothers’ testimony at trial. That testimony was riddled with inconsistencies, including contradictory statements regarding the color and make of the car, the driver’s hair style, and whether they had ever seen the shooter or driver before the incident. Given those flaws in their trial testimony, a demonstration that the brothers’ identification of one of the assailants — the easier one to identify, given their location — was decidedly wrong would likely tarnish the credibility of their identifications of the other assailant so severely that a reasonable juror would essentially discount them.
Furthermore, if Jackson were actually the driver, as Stewart’s post-trial witnesses indicate, then he would have a clear motive to implicate Stewart in his statements exonerating his friend, Lee. The majority reaches a contrary conclusion, presuming that reasonable jurors would believe Jackson’s statements rather than the contradictory testimony by Daniels, Vinson, or Johnson, whose declarations all implicate Jackson to one degree or another in the Parish brothers’ shooting. But that assumption lacks foundation, particularly given that Stewart’s post-trial witnesses signed sworn statements, while the record contains no declaration from Jackson under oath. Even were it otherwise, we cannot gauge the relative credibility of post-trial witnesses — i.e., compare Jackson’s credibility to Daniels’ — when none have testified.
The majority’s second assumption is that, even if the Parish brothers’ identification of Stewart were discredited, “sufficient direct and circumstantial evidence independent of their testimony placed Stewart and his car at the scene of the shooting.” Maj. Op. at 942. Under the majority’s view, reasonable jurors would still find Stewart guilty beyond a reasonable doubt, based principally on testimony indicating that his car was involved in the crime. But the testimony about the car was wildly inconsistent. No reasonable juror would have convicted Stewart based primarily on the evidence concerning the car from which the shots were fired.
First, we cannot, as the majority would have it, presume that the jury concluded “[t]he BMW involved in the shooting was identical to the BMW that Stewart drove,” or, as the district court did, that “the jury rejected the argument that a brown or rust-colored Honda was involved.” It is entirely possible that the jury convicted Stewart in spite of its uncertainty regarding the car, based on the victims’ eyewitness identification. Thus, we must evaluate the car evidence directly, rather than *957relying on the jury’s assumed findings regarding the vehicle.
Second, even if the car evidence were much stronger than it is, a reasonable juror would likely not convict based primarily on such evidence. People lend cars to one another. That a particular car was used in a shooting is not proof beyond a reasonable doubt — without reliable identification of the driver — of who was driving the vehicle.
Third, and critically, the Parish brothers’ descriptions of the car involved in the shooting were rife with inconsistencies. In their pretrial interviews, both brothers described the car as “rust-colored” or a “brownish-rust color,” but also — inconsistently — as a light silver-grayish, light blue, or bluish-gray color.
At trial, Mark Parish testified that the car from which he and his brother were shot was a “brownish-rust color,” and admitted that he had also described the car as “rust-colored” during the preliminary hearing. When Mark was shown two paint samples from a BMW dealer, he selected the sample that correctly depicted the color of Stewart’s mother’s ear, which the district court described as “light bronze.” But when asked what “rust-colored” means to him, Mark said, “like a copper,” and agreed that the water jug in the courtroom, which the judge described as “brown or copper color,” was “rust.” When further pressed on his understanding of the color “rust,” Mark asked to speak to “counsel.” One police officer testified at trial that Mark had said the vehicle was “a dark-colored BMW-type vehicle,” “possibly maroon.” Another officer testified that Mark had described the car as “light bluish or grayish.”
His brother, Michael, similarly testified that the car from which he was shot was “a light brown, tan, rust color,” and defined rust as “a light brown, a brownish color, like a tannish.” Then, when shown two paint samples from a BMW dealer, Michael, like his brother, selected the sample that correctly depicted the color of Stewart’s mother’s car. He also admitted, however, that he had described the ear to a police officer as “rust color” immediately after the shooting and later in the hospital, and acknowledged that he repeated that same account of the ear’s color at the preliminary hearing. When asked whether he subsequently changed the description to “bluish-gray” or “silver,” Michael stated that he did not recall. On cross-examination, Michael answered affirmatively when asked whether the assailants’ car was “a rust-colored BMW.”
The brothers’ accounts also varied with regard to whether the car from which they were shot had two or four doors, and whether its tires had three-star or five-star rims. The other eyewitnesses called by the prosecution, Michael Sturdivant and Michael Cole, described the car still differently. Sturdivant testified that he saw a gray or white car, which could have been a BMW or a Nissan' Sentra, swerving in and out of traffic near where the shooting occurred. Cole testified that after hearing shots from his home, he stepped outside to glimpse a “shiny, light-colored, possibly gray roof [of a car] as it disappeared” out of sight.
The evidence Stewart presented on ha-beas adds to the uncertainty regarding the car involved in the shooting. One eyewitness, whose statement was taken several months after the shooting, described seeing “a beige, dark colored car like a 79 Honda Accord or Civic driving slow” at the scene of the Parish brothers’ shooting. Other post-trial declarations implicating Jackson, rather than Stewart, as the driver and Arnold Adkins as the shooter describe Jackson’s car as a reddish-brown four-door Honda and attest to seeing Adkins and Jackson together in a rust-colored Honda *958with tinted windows on the days before the shooting. This evidence suggests, at the very least, that other gang members may have had vehicles matching one or another of the descriptions of the car involved in the Parish brothers’ shooting. Yet, a police officer involved in investigating the Parish brothers’ shooting testified at trial that no one looked through any gang files to determine whether anyone other than Stewart was known to drive a “rust-colored” BMW or similar model car.
The upshot is that no reasonable juror would find the car evidence alone a sufficient basis to convict Stewart. Aside from the inherent uncertainty as to who was driving, both the color and the make of the car used in the shooting are far from clear when the evidence is considered in its entirety.
iii.
Given the weakness of the car evidence, the other circumstantial evidence presented at trial does not add enough support to the verdict to eliminate reasonable doubt as to guilt. First, there is evidence concerning Stewart’s whereabouts at the time of the shooting. The majority points to Stewart’s mother’s testimony placing Stewart and the BMW he drove in the area of the shooting around the time of the crime. See Maj. Op. at 939-40. But her testimony on this point is, at most, indeterminate. She testified that she thought she saw the tail end of her own BMW but later determined it was not hers. A police detective testified that Stewart’s mother had stated to police that she saw her son in the BMW while she was picking up her daughter at her mother’s house on the evening of the shooting, in the vicinity of the crime. At trial, however, Stewart’s mother denied making such a statement.
As to the gunshot residue found on Stewart’s mother’s BMW, the prosecution’s expert witnesses testified that such residue is easily transferred from one surface to another, can be deposited by sources other than guns, and would not likely remain adhered to the surface of a moving vehicle.
The evidence that Stewart was “cleaning” his car when the police located him on the night of the shooting is also thin. There was no testimony that Stewart was seen washing the vehicle or removing anything that could have been incriminating. Nor is there any suggestion of evidence connected to the crime that could have been “cleaned” away, such as blood. The officer who encountered Stewart testified that he did not remember what Stewart was doing that gave him the impression he was cleaning the vehicle when the officer approached him in the parking lot of Stewart’s apartment complex. According to the officer’s testimony, Stewart appeared to have been picking up papers or bottles from underneath the seat of the car, but Stewart ceased doing so at the officer’s request, and no attempt was made to seize any of the items he may have discarded. Thus, the conclusion that Stewart was cleaning the car was not only speculative but did not involve cleaning up anything that was even arguably evidence of the crime. A reasonable juror would be unlikely to give weight to the officer’s vague observations as proof of Stewart’s guilt.
Brown’s testimony at trial that Stewart bragged about his involvement in the Parish brothers’ shooting provides little additional support for Stewart’s conviction. A post-trial witness attested that Stewart never came to the school where he allegedly spoke with Brown about the shooting. More importantly, the reported statement was about a killing, not a non-fatal shooting; it referred to one person, rather than two; and it did not mention the Parish brothers by name. Neither of the Parish brothers died. Another member of the *959Parish brothers’ gang was murdered on the same night the brothers were shot. Stewart has not been charged with that murder, so any evidence concerning it is irrelevant.
III.
Under the Schlup gateway standard, when “the newly presented evidence ... call[s] into question the credibility of the witnesses presented at trial[,] ... the ha-beas court may have to make some credibility assessments.” Schlup, 513 U.S. at 330, 115 S.Ct. 851. As the majority notes, Schlup did not identify whether and when that assessment requires an evidentiary hearing. See Maj. Op. at 941 n. 12. Further, although the district court here accepted all the post-trial evidence as credible — in the sense that the witnesses were telling the truth as they perceived or remembered it — it went on to decide on a cold record the accuracy of that testimony, as compared to the accuracy of the incul-patory testimony at trial.
In my view, Stewart’s new evidence, if credible, combined with the exoneration of his co-defendant and the weakness of the trial evidence, sufficiently indicate that Jackson was the driver of the vehicle from which the Parish brothers were shot, so as to entitle Stewart to override the AEDPA time limit for filing a habeas claim. See Perkins, 133 S.Ct. at 1932. That a declar-ant’s testimony might be impeachable because of inconsistencies with other evidence is reason to hold an evidentiary hearing, not to refrain from holding one. Unless proffered testimony is far-fetched or disproven by physical or documentary evidence, a court cannot reasonably determine the credibility and reliability of witnesses other than through an in-person hearing, with cross-examination. In particular, there is no sensible way to resolve the Schlup issue here without a chance to cross-examine Jackson on his assertion that Stewart was the driver, in light of substantial post-trial evidence that Jackson was the driver. Because we cannot evaluate Stewart’s Schlup claim without an evidentiary hearing, the district court’s failure to conduct such a hearing is an abuse of discretion. See supra Part II. A.iv.
The key question in assessing whether the district court abused its discretion concerning holding an evidentiary hearing is whether such a hearing “would produce evidence more reliable or more probative” with regard to Stewart’s assertion of actual innocence than the declarations before the district court district. Griffin v. Johnson, 350 F.3d 956, 966 (9th Cir.2003). Here, there is no doubt that it would, by enabling the court to assess the basis of the affiants’ statements; probe the reasons for their delay in coming forward; and explore other factors bearing on how reasonable jurors are likely to perceive the overall reliability of various witnesses’ testimony. The district court’s dismissal of Stewart’s Schlup claim in absence of such a hearing was an abuse of discretion.
Conclusion
For the foregoing reasons, I would reverse the district court’s dismissal on timeliness grounds, or, in the alternative, remand for an evidentiary hearing on Stewart’s Schlup claim. Neither of those dispositions would resolve whether Stewart’s petition presents federal claims- cognizable under § 2254, and I express no opinion as to the merits of his petition.
. For convenience, I refer in this dissent to both the Schlup and McQuiggin exceptions as Schlup claims.
. Since Chavis, "California law has not clarified whether a filing delay greater than 60 days necessarily qualifies as ‘substantial.’" Chaffer v. Prosper, 542 F.3d 662, 666 (9th Cir.2008) (“Chaffer I") (emphasis added). The Supreme Court of California denied our request for certification of that question in Chaffer I, and has not answered it since. See Chaffer v. Prosper, 592 F.3d 1046, 1048 n. 1 (9th Cir.2010) (per curiam) ("Chaffer II") ("California has not provided any guidance as to what constitutes a timely non-capital habe-as petition.’’); accord. Velasquez, 639 F.3d at 967. We have characterized “California’s reasonableness standard [as] commensurate with the limitations of other states, which are 30 or 45 days.” Cross v. Sisto, 676 F.3d 1172, 1176, 1179 (9th Cir.2012) (emphasis added) (citing Carey, 536 U.S. at 222, 122 S.Ct. 2134). But neither the Supreme Court of the United States nor California’s high court has held that California’s indeterminate "reasonable time” period is uniformly coextensive with other states’ limitations.
. McQuiggin noted, in passing, that on remand "the District Court’s appraisal of Perkins’ petition as insufficient to meet Schlup's actual-innocence standard should be disposi-tive, absent cause_” 133 S.Ct. at 1936 (emphasis added). This language does not describe any of the usual standards of review— e.g., de novo, clear error, abuse of discretion. “Absent cause” could mean "unless incorrect”; it certainly does not mean "abuse of discretion.”
. We are to consider a petitioner’s diligence in pursuing his petition as a factor in our assessment of whether actual innocence has been convincingly shown. See Perkins, 133 S.Ct. at 1935-36. That consideration, however, weighs very little here, because: (1) the exoneration of Richard Lee, Stewart’s co-defendant, through information provided by Jackson, itself occurred years post-trial, yet that information, also of importance to Stewart’s Schlup claim, was considered reliable enough to overturn Lee's conviction; and (2) Stewart missed the AEDPA filing deadline by very little, as I explained in Part I.
. Stewart sought access to both Jackson’s statement itself and to the files related to the state’s investigation of Lee's habeas petition, but was provided with neither. Those files were later submitted to the district court for in camera review, and were subsequently filed under seal, after the district court determined that they did not "exonerate” Stewart, and thus were not subject to discovery. I discuss the impact of those files later.
. According to testimony at trial, police gang files list Stewart’s street name as "Pee-Du,” “P-Du” or "P-Dew” and Lee’s as "Pufferoo” or "Puff.” References to "P-Du” and "Puffe-roo” appear in the trial transcript as well as the post-trial record and the parties’ briefs.