The appeal in this case presents the question whether a defendant who committed a crime and was sentenced prior to[*109] United States v. Booker, — U.S. ——, 125 S.Ct. 738, 160 L.Ed.2d 621 (2005), can be given a sentence that is within the statutory maximum but higher than the sentence that he would normally have received (absent departures) under the mandatory guideline regime. We conclude, without difficulty, that the ex post facto clause of the Constitution is not offended by this result; and, on the present facts, no due process -objection to the higher sentence can be maintained. For other reasons, a remand for resentencing is justified.

The facts are uncomplicated. On November 12, 2002, James Lata robbed Citizens Bank in Nashua, New Hampshire, informing the manager that he (Lata) had a gun and a bomb. Caught in 2003, Lata was tried and convicted of bank robbery by force and violence in federal district court in May 2004. 18 U.S.C. § 2113(a) (2000). On August 2, 2004, Lata was sentenced to 8 years in prison which is well within the statutory maximum sentence of 20 years. Id.

However, the sentence was greater than the guideline maximum that would normally have been imposed, absent a departure upward, under the guidelines applicable either at the time the crime was committed or at the time the sentence was passed. Under those guidelines, Lata’s base offense level for the robbery was 20, which, with adjustments prescribed by the guidelines, produced an adjusted offense level of 25. ^[1] Prior convictions gave Lata two criminal history points but, since the present offense was committed while he was on probation, two more points were added, placing him in category III. U.S.S.G. § 4A1.1. An offense level of 25 and a eate-gory III criminal history created a guideline range of 70-87 months, id. ch. 5, pt. A, so the 96-month sentence was above the range.

Although Booker had not been decided at the time of the sentencing, the district court deemed the mandatory regime unconstitutional based on the Apprendi/Blakely line of decisions, Apprendi v. New Jersey, 530 U.S. 466, 120 S.Ct. 2348, 147 L.Ed.2d 435 (2000); Blakely v. Washington, 542 U.S. 296, 124 S.Ct. 2531, 159 L.Ed.2d 403 (2004), and it did not therefore regard itself as bound by the guidelines. The court so ruled after Lata objected that the guidelines were unconstitutional under Apprendi/Blakely, insofar as they permitted enhancement based upon judge-found facts.

The district court’s decision to exceed tfie guideline sentence was based on Lata’s extensive criminal record and the fear induced by the threat of the bomb and the gun at the robbery. Lata had been the subject of arrests, charges and convictions from his juvenile days onward and, with time out for a lengthy prison sentence for bank robbery in the 1970s, had a fairly dismal record of continuing involvement with violent crime. For á variety of reasons, much of this was not reflected in the raw points awarded for criminal history.

Lata appealed to this court. In his initial brief, Lata argued that no jury having passed on the facts underlying the enhancements to his guideline sentence (five levels and the two probation-related criminal history points), the maximum possible sentence for him was within the 37-46 month range; this is the range that would have resulted if he had been sen[*110] tenced without enhancements, using an offense level of 20 and a criminal history-category of II. This argument is defeated by Booker, which permits enhancements based on judge-found facts with advisory guidelines, Booker, 125 S.Ct. at 749-50, 764; United States v. Antonakopoulos, 399 F.3d 68, 75 (1st Cir.2005), and need not be further considered.

However, as a fallback argument, Lata argued that the maximum sentence that could properly be imposed was the 70-87 month range that resulted from an ordinary application of the guidelines, including the enhancements already described. A sentence exceeding 87 months, in Lata’s view, violates both the ex post facto clause of the Constitution, U.S. Const. art. I, § 9, cl. 3, and the due process clause variant, U.S. Const. amend. V, that may apply even where the ex post facto clause is inapplicable.

Lata’s initial appellate brief was filed prior to the decision in Booker itself. Accordingly, we invited both sides to supplement their earlier briefs in light of Booker and also asked Lata whether as a further alternative he wished to argue for a remand on the ground that the district judge might have reached a different sentence under the post-Booker guideline regime now in place. Lata has now made such a request but without prejudice to his main constitutional claims based on ex post facto precedents, which he renews.

The ex post facto clause argument is readily answered. The ex post facto clause forbids not only legislative creation of new criminal liability after the event but also a legislative increase in punishment after the event, but it does not apply of its own force to changes worked by judicial decisions. Rogers v. Tennessee, 532 U.S. 451, 460, 121 S.Ct. 1693, 149 L.Ed.2d 697 (2001); Marks v. United States, 430 U.S. 188, 191, 97 S.Ct. 990, 51 L.Ed.2d 260 (1977). In this instance, the change in the guideline regime from mandatory to advisory was worked entirely by judicial interpretation, based on the Sixth Amendment and severance analysis in Booker.

In this respect, the change from mandatory to advisory guidelines differs importantly from changes in the guidelines’ content worked by ordinary amendments adopted by the Commission and submitted to Congress. For ex post facto purposes, the federal courts have assumed that those changes in content should be viewed as the equivalent of statutory changes — indeed, in some cases they are formally directed by Congress. See, e.g., United States v. Jordan, 162 F.3d 1, 2 (1st Cir.1998) (referring to Congress’s enactment of a guidelines amendment). Based on this equation of guideline changes with statutes, the circuits have normally approved use of the guideline edition in force at the time of the crime if later amendments increased the sentences.

That the shift to advisory guidelines stemmed from judicial decision may seem a formal distinction but the ex post facto clause is mechanical and, from the standpoint of protecting reliance, over-inclusive. The .clause applies to changes in the definition of a crime and in the maximum sentence even in situations where there is no possibility that the defendant in fact relied on the earlier version of the statute in committing the crime. And, so far as ex post facto principles rely on concepts of fair warning, they have been absorbed into the due process clause, which underpins Lata’s only serious argument.

An after-the-offense enlargement of the contours of the crime or maximum sentence by judicial construction can raise due process objections based on lack of[*111] fair warning but only where the alteration is “unexpected and indefensible” by reference to the case law that had been expressed prior to the offense. Rogers, 532 U.S. at 461, 121 S.Ct. 1693; Bouie v. City of Columbia, 378 U.S. 347, 354, 84 S.Ct. 1697, 12 L.Ed.2d 894 (1964). This is an imprecise formula, as the conflicting opinions in Rogers reádily show, so it is of some benefit to understand the underlying policy dilemma.

The underlying problem is one of reconciling the continuing mutability of judicial doctrine with concerns about reliance and notice. Glosses on statutes or refinements in common law doctrine regularly evolve between the time of the crime and a trial that may occur years later. At least in sentencing, the criminal’s reliance on earlier glosses is usually an imaginary concern: rarely is the decision to commit the crime affected by close attention to the statutory penalties, let alone the more obscure intermediate step of guideline calculations. Lata certainly does not claim that he in fact relied on the mandatory guidelines when he robbed the bank.

But the Supreme Court’s concern with fair notice goes beyond actual reliance. Under Bouie, Rogers and Marks, some court-made changes in criminal law may be so surprising and troubling (“unexpected and indefensible”) as to offend a sense of fair warning even if the defendant probably paid no attention to the case law. In Bouie, for example, the Supreme Court deemed the state court’s expansive, non-literal reading of its trespassing statute to be unfair as applied to civil rights sit-in demonstrators who lacked fair warning that their conduct was criminal. Bouie, 378 U.S. at 355, 84 S.Ct. 1697.

In this case, the “unexpected and indefensible” test could be difficult to apply if the focus were solely upon the shift from mandatory to advisory guidelines. Booker ' was not an isolated event but the end of a multi-year evolution developed through a host of cases; how “unexpected” Booker appears depends largely on one’s time horizon. Ten years ago, Booker would have seemed improbable; gradually with Ap-prendi followed by Blakely, some major change in guideline status or operations seemed possible; and in the Booker case itself the government in fact urged a similar result to that reached by the Supreme Court. See Booker, 125 S.Ct. at 768 (characterizing the government’s remedial suggestion as “coincidfing] significantly with [the “Court’s] own”)”.

Also, the Booker majority would hardly call its own decision “indefensible” in the sense of the Bouie case (where it was almost a polite synonym for “wrong”). But the meaning of the term is no longer quite clear, compare Rogers, 532 U.S. at 461, 121 S.Ct. 1693 (using “indefensible” to mean “unjustified,” “arbitrary,” or “vindictive” breaks with prior law), with Marks v. United States, 430 U.S. 188, 97 S.Ct. 990, 51 L.Ed.2d 260 (1977) (ignoring “unjustified” test but stressing First Amendment concerns). See also Rogers, 532 U.S. at 468, 480, 121 S.Ct. 1693 (Scalia, J., dissenting) (using “indefensible” to mean inconsistent with prior law and presuming that the majority used the term to mean “unreasonable”).

However, our own case becomes easy if one looks to the underlying concern of fair warning and asks whether — measured by what Lata' could objectively know at the time he planned to rob the bank — Lata’s later sentence so far disappoints reasonable expectations as to raise due process concerns. Booker, after all, does not lay down a new fixed rule of primary conduct or redefine elements of a crime (cf. Bouie; Marks) but is only part of a mechanism or framework by which an ultimate sentence is developed. Lata’s own sentencing ex[*112] pectations when he committed the crime, if any, surely related to results. ^[2]

Before committing the crime, Lata would have known only one thing for certain, namely, the 20-year maximum statutory sentence for bank robbery. He could have made guideline calculations as well, but only on a variety of farther assumptions about the details of how the crime would occur {e.g., how much money might be taken, injuries done), what the government might charge among alternative offenses, events after the crime (acceptance of responsibility, new criminal history), and the possibility of reasonable departures based on an array of factors — some unpredictable.

To some, the government’s position— .that Lata could reasonably rely only on the 20-year maximum — might also seem unrealistic from the standpoint of fair notice. Absent extremely aggravating chafacteris-tics not present here — e.g., a felony murder in the course of a robbery — it is not easy to imagine a 20-year sentence in this case being imposed under the guidelines (or post-Booker for that matter). In all events, it is unnecessary for the disposition of this case to adopt a flat and final position on due process objections to post- Booker sentences for pr e-Booker crimes.

Here, we think it is enough to resolve this case that even viewed as of the time Lata committed the crime-post-Apprendi but pr e-Blakely — someone in Lata’s position could not reasonably be surprised by the sentence he eventually received. Whether or not exactly the same sentence would necessarily have been imposed by departures under the guidelines is necessarily uncertain. But the sentence imposed is not wildly different than a sentence that might well have been imposed under the guidelines for someone with Lata’s criminal record and offense-related conduct.

Even under mandatory guidelines, a defendant with a criminal record not fully reflected by criminal history points was always on notice that the top of his guideline range might be exceeded. U.S.S.G. § 4A1.3; United States v. Black, 78 F.3d 1, 8 (1st Cir.1996). Lata’s pre-sentence report indicated that an upward departure might be warranted because of the character of his criminal record. Nothing in the guidelines flatly forbad the judge from departing based on the fear induced by the threat of the bomb and gun. The result does not violate the due process clause.

We reserve for the future the case, if one ever arises, in which a sentence is imposed for a pr e-Booker crime that is higher than any that might realistically have been imagined at the time of the crime or based on factors previously discouraged, prohibited, or not recognized under the guidelines. As we have seen, any prospective guideline range estimated before the crime has been committed is far more contingent and uncertain than may be true on the day of sentencing. And, since post-Booker sentences are open to review for reasonableness, Booker, 125 S.Ct. at 765-66, extreme sentences at the very least will rarely survive ordinary review so as to present the naked due process question.

• This brings us to Lata’s remaining argument, which is that the case ought to be remanded for resentencing under the post- Booker regime. Of course, the district judge has already sentenced Lata under a[*113] non-mandatory regime, which is what he would again get on remand. But the district judge did not have the benefit of Booker’s own determination as to what parts of the existing statute remained after severance and just how the , Supreme Court conceived of the relationship of statute, guidelines, and discretion in post- Booker sentencing. Booker, 125 S.Ct. at 764-65.

In his post-Booker brief, Lata’s counsel notes that Lata is over 6Ó and that he is suffering from cancer, so that the prospects of recidivism are arguably no longer very great. Under the guidelines, both age and infirmity were discouraged as bases for departure, § 5H1.1.4; under the statute alone, conceivably they might be afforded more weight in a particular case, especially if they diminished the risk of recurrence. See 18 U.S.C. § 3553(a)(2)(C) (2000) (sentence to take account, inter alia, of need to protect public “from further crimes of the defendant”). Cf. United States v. Heldeman, 402 F.3d 220, 224 (1st Cir.2005).

Because the district judge knew most of the pertinent facts and did not regard himself as constrained by the guidelines, it is far from clear that the result would be different on remand. Admittedly, based on defense counsel’s proffer, Lata’s affliction is possibly much worse than was known at the time of sentencing. But for assessing plain or harmless error — and the failure fully to anticipate Booker itself is treated as “error” under existing doctrine, Antonakopoulos, 399 F.3d at 76 — our focus is primarily upon what was known at the time of sentencing. ^[3]

Nevertheless, under our somewhat mechanical test of what is preserved Booker error, Antonakopoulos, 399 F.3d at 76, the “error” here was manifestly preserved in the district court, as the government freely conceded at oral argument. Under the harmless error test, the government must show beyond, a reasonable doubt that a lower sentence would not be imposed under the post-Booker regime — a test where doubts are resolved in favor of a remand. United States v. Vazquez-Rivera, 407 F.3d 476, 489 (1st Cir.2005). On balance, we think remand is the safest course in this case.

As we have stressed before, a remand even for plain error does not carry any implication that this court expects or endorses a lower sentence. Heldeman, 402 F.3d at 224. This is so a fortiori where the remand arises out of a preserved error and where the harmless error test makes even a modest possibility of change enough' to warrant remand. This general qualification should not require repeating in every case; it should be assumed unless the opinion expressly says otherwise.

The sentence is vacated and the case remanded for resentencing consistent with this decision.

It is so ordered.