concurring in part and dissenting in part:
I concur in the judgment on parts I and II of the majority opinion, although as I explain below would hold we have jurisdiction over Keating’s § 2254 petition for different reasons. I dissent from the due process/harmless error analysis the majority employs in parts III and IV, and would instead reverse the district court’s decision to grant the petition.
I
A
In my view the majority’s jurisdictional gymnastics are neither necessary nor appropriate. Instead of treating Keating’s instant habeas petition as an amendment or resubmittal of his first petition, and instructing the district court to deem it as having been so filed (apparently as of the date the original petition was filed), I would simply say that the petition was filed prematurely (ie., before our mandate issued), but since nothing substantive happened in the meantime, once the mandate did issue Keating was then “in custody” and his habeas action became ripe. No one suggests that Keating would not have been “in custody” (literally) on February 6, and Hensley v. Municipal Court, 411 U.S. 345, 352, 93 S.Ct. 1571, 36 L.Ed.2d 294 (1973) provides authority for this approach. In Hensley, the Court held that a convicted state petitioner released on his own recognizance pending appeal was “in custody” for purposes of habeas jurisdiction. In addition to the fact that Hensley was subject to statutory restrictions and was at large only by virtue of stays, the Court noted that its
conclusion that the petitioner is presently in custody does not interfere with any significant interest of the State. Indeed, even if we were to accept respondent’s argument that petitioner is not in custody, that result would do no more than postpone this habeas corpus action until petitioner had begun service of his sentence. It would still remain open to the District Court to order petitioner’s release pending consideration of his habe-as corpus claim. Even if petitioner remained in jail only long enough to have his petition filed in the District Court, his release by order of the District Court would not jeopardize his “custody” for purposes of a habeas corpus action. Plainly, we would badly serve the purposes and the history of the writ to hold that under these circumstances the petitioner’s failure to spend even 10 minutes in jail is enough to deprive the District Court of power to hear his constitutional claim.
*1066Hensley, 411 U.S. at 352-53, 93 S.Ct. 1571 (internal footnote and citations omitted).
This approach is also consistent with how we handle the flip side, when notices of appeal are filed after a decision in the district court but before judgment is entered. In that circumstance we simply take the pragmatic view that dismissing for lack of jurisdiction and forcing the appellant to refile at the appropriate time is silly; the triggering event will have passed and nothing of moment will have occurred in the interim. See Fed.R.App.P. 4(a)(2). Put differently, the court to which the matter is going lacks jurisdiction when the filing is made because the filing is premature, but has jurisdiction by the time it matters. The premature filing is treated as having been made at the time jurisdiction actually attaches. Here, the situation is quite similar. Even if the State is correct that Keating was not “in custody” on January 22, 1998, there is no dispute that he would have been taken into custody and would have begun to complete serving his sentence on February 6 (following issuance of this court’s mandate on February 5).1 To hold that he was not in custody on January 22 and to dismiss for lack of jurisdiction for that reason “would do no more than postpone this habeas corpus action until petitioner had begun service of his sentence.” See Hensley, 411 U.S. at 352, 93 S.Ct. 1571. In this case, that would have been for no more than fifteen days. Nothing happened in the district court, except a hearing date (after the mandate was expected to issue) was set. For this reason it makes sense to hold that whether or not the district court had jurisdiction on January 22, Keating’s January 22 filing-although premature-became effective as of the time this court’s mandate issued returning him to custody because no one questions that the court had jurisdiction as of February 5.
B
The majority’s answer to the jurisdictional question is troubling for a number of reasons. First, Keating never asked the district court to construe his second habe-as petition as an amendment to his first petition, nor did he try to “resubmit” his first petition without the unexhausted claim. Rather, he deliberately filed his second petition as a separate action under a separate case number. He never argued in district court or here that he should somehow be relieved in this respect.
Second, the bottom line of this court’s order with respect to Keating’s first habe-as petition was “DISMISSED WITHOUT PREJUDICE.” See Keating v. Hood (“Keating 7”), 133 F.3d 1240, 1242 (9th Cir.1998). As we explained in Henry v. Lungren, 164 F.3d 1240, 1241 (9th Cir.1999), dismissal of an original petition for failure to exhaust state remedies, without retaining jurisdiction, terminates the litigation. This means that once this court’s mandate issued in Keating I, there was no petition to be “amended” or “resubmitted.” Contrast, for example, this court’s order in Reutter v. Crandel, 109 F.3d 575, 578 (9th Cir.1997), where we also held that the petition was mixed and required dismissal but expressly stated: “Reutter may strike the unexhausted claim and resubmit his petition to the district court.” This court’s order in Keating I provides no such option. Thus, on January 22 when Keating’s second habeas petition was filed, the district court no longer had jurisdiction to do anything with respect to his first petition since it was still on appeal; even if it had been asked to do so, it could not have construed the second petition as an amendment or resubmittal because it couldn’t do anything with respect to the first petition on January 22. It got jurisdiction back when our mandate issued on February 5, but at that time the first petition no longer existed because our dismissal expressly terminated that litigation. Thus, the district *1067court could not have construed the second petition as an amendment or resubmittal of the first petition (on either January 22 or February 6) because there was no first petition to amend or resubmit. And there still isn’t.
Regardless, I do not see how we can now direct the district court to amend Keating’s first petition because (even if it were still alive) the State had filed a responsive pleading to the first petition. Therefore Keating himself could not have amended his first petition or resubmitted an amended petition without leave of court under Fed.R.Civ.P. 15(a). That, of course, is a discretionary decision. Keating’s case is therefore doubly distinguished from Calderon v. United States Dist. Ct. (“Thomas”), 144 F.3d 618 (9th Cir.1998), and Calderon v. United States Dist. Ct. (“Taylor”), 134 F.3d 981 (9th Cir.1998), upon which the proposed opinion relies, see ante at 1059, for in both Taylor, 134 F.3d at 986, and (presumably) Thomas, 144 F.3d at 620, the petitioner sought leave to amend his original petition to delete an unexhausted claim and the State had not yet filed a responsive pleading when the petitioner did so.
In any event, Henry precludes determining whether Keating is “in custody” by reference to the date when he filed his first petition. Like Keating, Henry filed his original petition when he was clearly in custody and like Keating I, the original petition was dismissed without prejudice for failure to exhaust. Henry filed his second petition after his release from prison and discharge from parole, but he argued that he was in actual custody on the theory that the date he filed the earlier habeas was the operative date-not the date he filed the present petition. We disagreed, noting that “[t]he district court’s dismissal of Henry’s original petition for failure to exhaust state remedies ‘terminated the litigation.’ Farmer v. McDaniel, 98 F.3d 1548, 1552 (9th Cir.1996), cert. denied, 520 U.S. 1188, 117 S.Ct. 1474, 137 L.Ed.2d 686 (1997). The relevant date is the date on which Henry filed the present, second petition.” Henry, 164 F.3d at 1241. We held that the filing of the present, second habeas petition, following dismissal without prejudice of the first petition, does not relate back to the date of the first petition because the district court did not retain jurisdiction over Henry’s original petition when the court dismissed for failure to exhaust. This is exactly what happened here, and it seems that the majority’s opinion and order effectively makes an end-run around Henry simply because Henry went back to state court instead of to the Ninth Circuit. See ante at 1060 n. 9. No reason occurs to me why this court’s “DISMISSAL WITHOUT PREJUDICE” (without allowing for new filings or retaining jurisdiction) did not “take effect” just the same as the district court’s similar dismissal in Henry.
Finally, I don’t see how the majority’s nunc pro tunc order could deem Keating’s “amended” or “resubmitted” habeas petition filed as of any date other than the date on which his second petition was actually filed (January 22, 1998). To the extent that the proposed opinion relies on Calderon v. United States Dist. Ct. (“Kelly ”), 163 F.3d 530 (9th Cir.1998), for the proposition (asserted in the parenthetical following the cite to Kelly, see ante at 1059-60) that the “district court could exercise nunc pro tunc power to deem recently filed habeas petitions to have been filed as of the filing date of two earlier petitions,” Kelly does not in fact say this. Instead, what Kelly says is that the district court could, in its discretion pursuant to a motion under Rule 60(b)(6), set aside its earlier dismissals and permit or deem the later habeas petitions “to be filed in the earlier-filed [1992 and 1993] cases nunc pro tunc as of the date they were filed in the 1998 cases.” Id. at 540 (emphasis added). Thus, even if there were a live first petition, and even if Keating had sought leave to amend or resubmit, and even if the district court had allowed amendment or resubmittal, it would not follow that the filing would relate back to August 3', 1995 (when Keating originally filed his first petition). I do not see how it could, under *1068Henry or otherwise. Nor would it follow, see ante at 1060, that “[b]eeause Keating’s first petition was filed before the effective date of the Antiterrorism and Effective Death Penalty Act, the provisions of the AEDPA do not govern resolution of this case.” Rather, we would be back to square one: a filing (however it is rechar-acterized) that occurred-even if deemed to have been made in Keating’s first petition-on January 22, a date on which the district court had no jurisdiction with respect to the first petition and over which it lost jurisdiction at the same time it gained it back (when this court’s mandate dismissed the petition effective February 5).
For these reasons, I would simply hold that whatever jurisdictional defect may have existed on January 22 was cured once our mandate issued February 5.
II
On the merits, I am more troubled by the harmless error analysis than with the result itself. In a nutshell: I do not see how we can decide this case without dealing with California v. Roy, 519 U.S. 2, 117 S.Ct. 337, 136 L.Ed.2d 266 (1996), which the majority opinion largely ignores. Roy involved Beeman2 error-failure to instruct on mens rea-in a felony-murder case. This is essentially a carbon copy of the instructional error in Keating. The difference is that in Keating, there were two theories of liability (direct perpetrator and aiding and abetting) whereas in Roy, there was only one. If anything, it seems to me, this makes Roy’s analysis more pertinent, not less.
It is difficult to understand how Stromberg, Yates, Griffin, Suniga, Qualls and Ficklin3 square with Roy. But I also do not see how we can avoid the conundrum.4
The Stromberg line of cases says that when the jury is given instructions on two theories of liability, one of which is constitutionally deficient, the court must conclude with absolute certainty “that the jury did not and could not have relied on the faulty instruction to convict petitioner” in order to hold the error harmless. See Ficklin, 177 F.3d at 1150. The deficiency in the Stromberg cases has to do with constitutionally protected, or otherwise noncriminal, conduct.
Roy, on the other hand, speaks directly to a missing element instruction — specifically, a missing intent element. It makes clear that we must apply Brecht to determine whether the type of error that occurred here-failing to instruct on intent-is harmless. See Roy, 519 U.S. at 4-5, 117 S.Ct. 337 (citing Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993)). This requires us to review the record to determine the error’s effect. The question a court has to answer in this context is: Did the failure to include intent [in the direct perpetrator] instruction have a “substantial and injurious effect or influence in determining the jury’s verdict.” Roy v. Gomez (Roy II), 108 F.3d 242, 243 (9th Cir.1997) (en banc) (adopting analysis, reasoning and conclusions stated in dissent to en banc decision in Roy v. Gomez (Roy I), 81 F.3d 863, 870-71 (9th Cir.1996) (en banc) (Wallace, J. dissenting)). It is only if, at the end of this exercise, two of us have a “grave doubt” that the error (conceded in this case) has a substantial and injurious effect that O’Neal’s “tie-breaker” rule applies. See O’Neal v. McAninch, 513 *1069U.S. 432, 444-45, 115 S.Ct. 992, 130 L.Ed.2d 947 (1995).
One way to rationalize Roy with the Stromberg line of authority is to start with the erroneous instruction (here, direct perpetrator) and determine whether a verdict of guilty under it is harmless error. If not (or if two of us are in “equipoise”), then O’Neal and Stromberg apply with full force and relief must be granted. If, on the other hand, an error would be harmless under Roy, there is no basis for Stromberg reversal.
Since Roy tells us how to analyze harmless error in a failure-to-instruct-on-intent case, we cannot play ostrich to the record- or dismiss the California Court of Appeal opinion quite so easily-as the majority does. Stated differently, where Strom-berg-type error is predicated on a Roy-type error, I doubt that the Stromberg-typé error obviates the need first to examine the record to determine whether the underlying Roy-type error is harmless.
From the record it appears that the error is harmless, particularly in light of our obligation to give due deference to the California Court of Appeal’s factual findings. See 28 U.S.C. § 2254(e)(1) (“ ... [A] determination of a factual issue made by a State court shall be presumed to be correct. The applicant shall have the burden of rebutting the presumption of correctness by clear and convincing evidence.”).5 The California Court of Appeal made quite extensive factual findings on Keating’s criminal conduct and intent. First, it discussed evidence of his criminal intent throughout the years in question:
He personally instigated the bond sales. He continued to demand bond sales even when he knew they were literally worthless because of the deteriorating condition of ACC’s net worth. He was personally warned in November 1986 (the month Keating ordered Symes and Fidel to initiate the bond sales program) by Patricarca of the FHLB of 14 specific problems, including Lincoln’s failure to comply with the direct investment regulation and not meeting its net worth requirement. In December 1986 Patri-carca personally pointed out the risky nature of the investments in Arizona real estate. In 1987 Keating sought the assistance of several United States senators to shore up his support with the FHLB Board even in the face of negative coverage in Forbes Magazine. Keating ordered the sale of one-year bonds to generate capital in October. In February 1988 Dochow met privately with Keating concerning the funneling of Lincoln assets to cover ACC’s deteriorating financial condition. The third-quarter 1988 loss was such a disaster for ACC that it was quite, clear ACC could not repay the $94.8 million tax advance from Lincoln, much less any individual investor’s principal, yet Keating brought the entire bond sales force to Phoenix where he personally exhorted greater bond sales.
ER 126-27. The court then recounted evidence of his close involvement with the fraudulent bond sales:
... The facts indicate Keating was in personal control of Lincoln, even down to the detail of which pictures could be hung on the walls; that he selected both ACC and Lincoln officers; that he was in personal contact with FHLB personnel who outlined for him the unsound business practices; that he was personally aware of the deteriorating financial prospects of ACC and of its inability to repay Lincoln the $94 million advanced for taxes; that he refused to provide negative information to anyone, going to the extent of hiring a public relations person in Phoenix to answer investors’ negative telephone calls and ordering Fidel to purchase all available copies of Forbes Magazine near Lincoln offices; that the bonds had been rated by Moody’s as below investment grade; and that, while knowing all the negative *1070trends, he still hosted the sales personnel at a posh party in Phoenix to-encourage further bond sales to unsuspecting and unsophisticated members of the public.
ER 130-31.
Finally, the court reiterated:
Keating was the chairman of ACC. He exercised the powers of an owner over both ACC and Lincoln by naming the officers and members of the boards of directors, directing policy, conducting negotiations with FHLB officials and other government regulators, setting the interest rates for Lincoln certificate of deposit accounts vis-a-vis ACC bonds, hiring personnel, conducting meetings to encourage bond sales, and even down to selecting the decor at the Lincoln headquarters in Irvine. He personally reviewed all press releases prior to their issuance. He directed policy and procedures. He was the person in control.
The individual investors ... were not given the pertinent information necessary to make an informed judgment-information about the losses being sustained by ACC in its real estate operations and cash flow, the discrepancies concerning the $94 million tax prepayment by Lincoln advanced to ACC, the concerns of the FHLB about the unsafe and unsound operations of Lincoln and of its inability to meet its cash requirements due in large part to excessive cash outflow to ACC for management fees and dividends, the FHLB concern about the enormous salaries paid to corporate executives, or information about the bond ratings made by Moody’s, which rated the bonds as being below investment grade, risky, and generally described as “junk bonds.” In most cases the individual investors were not even aware that they were investing in bonds, let alone subordinated debentures, but rather thought their investments were merely a form of certificate of deposit fully insured by the federal government.
We conclude from the facts that the prosecution has amply proven a persistent pattern consistent with criminal conduct whereby factual representations were made to the purchasers of debentures which were inaccurate and misleading-which representations Keating knew to be false and unfounded while contemporaneously failing to impart any negative information which, if known to the individual investors, would have provided them the basis for an informed decision.
ER 135-37 (emphasis added). I do not believe that all of these comments can be summarily dismissed or disregarded as “dicta,” as does the majority. See ante at 1063 n. 14. While we are not bound by the state appellate court’s legal conclusions, both pre and post-AEDPA law requires a considerable degree of federal court deference to state court factual findings.
Particularly in light of the deference accorded these findings, and based upon my independent review of the state trial court record, I believe it is fair and reasonable to conclude that the trial court’s failure to instruct the jury on intent was harmless given the absolutely overwhelming evidence demonstrating Keating’s criminal intent introduced at trial. As I see it, “there is not even a reasonable possibility,” ' see Roy I, 81 F.3d at 871 (Wallace, J. dissenting), based on this evidence that Keating did not specifically intend to “sell or offer to sell” securities containing “untrue statements of material fact” or securities which omitted material facts “necessary to make the statements made ... not misleading.” See ER 145^46 (“direct perpetrator” jury instructions). As such I haven’t the slightest doubt-much less a “grave doubt”-about the harmlessness of the instructional error in this case, and would therefore reverse the district court’s decision granting Keating’s habeas petition. But either way, I have “grave doubt” that the majority’s methodology is quite on target.
. As the Department of Corrections's February 3, 1998 letter directs, Keating was to self-surrender at Wasco State Prison at 3:00 p.m. on February 6, 1998.
. See People v. Beeman, 35 Cal.3d 547, 199 Cal.Rptr. 60, 674 P.2d 1318 (1984).
. See Stromberg v. California, 283 U.S. 359, 51 S.Ct. 532, 75 L.Ed. 1117 (1931); Yates v. United States, 354 U.S. 298, 77 S.Ct. 1064, 1 L.Ed.2d 1356 (1957), overruled on other grounds, Burles v. United States, 437 U.S. 1, 98 S.Ct. 2141, 57 L.Ed.2d 1 (1978); Griffin v. United States, 502 U.S. 46, 112 S.Ct. 466, 116 L.Ed.2d 371 (1991); Suniga v. Bunnell, 998 F.2d 664 (9th Cir.1993); United States v. Qualls, 140 F.3d 824, 829 (9th Cir.), vacated on other grounds, U.S. -, 119 S.Ct. 398, 142 L.Ed.2d 323 (1998); and Ficklin v. Hatcher, 177 F.3d 1147 (9th Cir.1999).
.When the district court granted Keating's original petition in April 1996, it did not have the benefit of our decision in Roy v. Gomez (Roy II), 108 F.3d 242 (9th Cir.1997) (en banc), on remand from the Supreme Court.
. This is so whether or not the AEDPA applies, although how the deference due is articulated differs somewhat.