KAYATTA, Circuit Judge.
Dan Carlos Marchena-Silvestre (“Marchena-Silvestre”) appeals his seventy-two month sentence following his guilty plea to a charge of unlawfully possessing automatic weapons. After careful review of the record, we conclude that the district court’s sentencing determination was infected by plain error.
I. Background
Since Marchena-Silvestre’s sentence followed a guilty plea, we draw the facts from the plea agreement, the change-of-plea colloquy, the presentence investigation report (PSR), and the sentencing hearing transcript.
See United States v. Almonte-Nuñez,
771 F.3d 84, 86 (1st Cir.2014). We rehearse only the facts necessary to form a basis for our analysis.
A. The Offense and Indictment
On October 24, 2013, Puerto Rico law enforcement agents searched Marchena-Silvestre’s apartment pursuant to a search warrant. The agents discovered and seized the following arsenal of firearms and ammunition: (1) an AR-15 assault rifle, unlawfully modified to fire in full automatic mode, equipped with an unlawful short barrel, and loaded with one round in the chamber and thirty-seven rounds in the magazine; (2) a Glock pistol, unlawfully modified to fire in full automatic mode, loaded with one round in the chamber and twelve rounds in the magazine; and (3) an additional 127 rounds of ammunition for the two firearms.
After waiving his
Miranda
rights, Marchena-Silvestre admitted that the firearms and ammunition belonged to him, that he purchased both firearms, and that he also purchased and installed a metal chip that enabled the Glock pistol to fire in full automatic mode. The investigating agents also discovered that the Glock pistol had been stolen from its registered owner. Less than a week after the seizure, a federal grand jury returned an indictment charging Marchena-Silvestre with possessing a machine gun in violation of 18 U.S.C. § 922(o) and possessing a stolen firearm in violation of 18 U.S.C. § 922©.
B. The Plea Agreement
Pursuant to a written plea agreement (the Agreement) with the government, Marchena-Silvestre agreed to plead guilty to possessing the machine gun. In turn, the government agreed to dismiss the charge that he possessed a stolen firearm, so long as Marchena-Silvestre complied with the Agreement’s terms.
Paragraph 7 of the Agreement, entitled “Applicability of United States Sentencing Guidelines,” contained a chart of “Sentencing Guidelines Calculations” for 18 U.S.C. § 922(o) that Marchena-Silvestre and the government agreed to “submit” to the court. The chart included a base offense level of 18,
see
U.S.S.G. § 2K2.1(a)(5), a two-point upward enhancement for a stolen firearm,
see
U.S.S.G. § 2K2.1(b)(4)(A), and a three-point reduction for acceptance of responsibility,
see
U.S.S.G. § 3E1.1, to arrive at a total adjusted offense level of 17. Since the parties did not agree to a criminal history category, the chart then set out the applicable guideline sentencing ranges for criminal history categories I (24-30 months) through VI (51-63 months). Paragraph 9, entitled “Sentence Recommendation,” provided that “the government reserves the right to request a term of imprisonment equal to the
higher end
of the applicable guidelines range and the defendant will request a term .of imprisonment equal to the
lower end
of the applicable guidelines range,” and that “any recommendation by either party for a term of imprisonment above or below the stipulated sentence recommendation constitutes a material breach of the ... Agreement.” The stipulated sentencing recommendations did not bind the district court, and Marchena-Silvestre only retained the right to appeal in the event that the district court did not sentence him within the stipulated guideline sentencing range.
C. The Presentence Investigation Report
The district court accepted Marchena-Silvestre’s guilty plea at the plea colloquy, and instructed the probation department to submit a PSR. The PSR departed from the Agreement by recommending a base offense level of 20 rather than 18, due to the added consideration that the defendant’s unlawful use of controlled substances made him a “prohibited person” under the guidelines.
See
U.S.S.G. § 2K2.1(a)(4)(B). The PSR applied the same two base level adjustments as recommended by the Agreement, resulting in a total offense level of 19 (rather than 17 as calculated in the Agreement).
The PSR also detailed Marchena-Silves-tre’s criminal history: In 2009, he was convicted of carrying a firearm in violation of Puerto Rico’s Weapons Law (a misdemeanor for which he was fined $300); and in 2013 he was convicted of illegally occupying property owned by the Puerto Rico Housing Department, resulting in a $50 fine. The two convictions resulted in a criminal history category of I. Cross-referencing that category with the total offense level of 19, the PSR recommended a guideline sentencing range of 30 to 37 months.
See
U.S.S.G. ch. 5 pt. A (Sentencing Table). In his sentencing memorandum, Marchena-Silvestre stated that he had “no objections” to the PSR.
D. The Sentencing Hearing
The sentencing hearing began with the government informing the court that it would request a sentence at the “high end range of the guideline sentence.” The court proceeded to summarize the facts of the case based on the PSR, noting the serious and illegal arsenal at the heart of the case. The court then moved to reviewing Marchena-Silvestre’s criminal history, noting that a combination of prior offenses without ’ serious penalties “is what really strikes you when you see this kind of thing.” The court noted what it thought were two prior firearms charges: one a misdemeanor conviction for carrying a firearm without a license, the other an arrest for carrying what the court described as a “nine millimeter nickel plated pistol.” In fact, the second charge as described in the PSR was for carrying a “nickel magazine loaded with three rounds
of .9 caliber ammunition,” a charge dropped for lack of probable cause. No one corrected the court’s misreading.
Given a turn to speak again before the court calculated a guideline sentencing range, the prosecutor claimed that he stood by the terms of the Agreement, yet he recommended a 37-month sentence, equaling the high end of the PSR’s recommended range (rather than the 30-month high end as specified in the Agreement’s chart for a criminal history category of I).
During the ensuing discussion, the district court inexplicably announced that Marchena-Silvestre “has a base offense level of 19,” which was both wrong and contrary to any information that was before the court. The court also neglected to calculate any total offense level. The court made clear that it did not regard the case as a guidelines “heartland” case, and that it felt a lengthier sentence was needed because of the high incidence of criminal violence in the Commonwealth for which there was too little accountability. It recited the “factors to be considered in imposing a sentence” listed in 18 U.S.C. § 3553(a), and made clear that it felt that a “variance is in order under [the] 3553(a) factors.” The court then announced its sentence by beginning with a guideline sentencing range, as follows:
Range is 31 to 41 months. The fine range is 6,000 to 60 thousand, which means nothing in [this] case. Plus supervised release of one to three years. Statutory maximum, ten years. I think that this case, because of the kind of gun, ammunition involved, his prior experiences before the law with guns, requires a sentence of at least 72 months is the sentence I’m imposing.
Unfortunately, the announced guideline sentencing range corresponded to nothing in the PSR nor, for that matter, to any offense level in the sentencing guidelines.
See
U.S.S.G. ch. 5 pt. A (Sentencing Table). Had the court adopted the PSR’s recommendation, the range should have been 30 to 37 months. Even more unfortunately, no one in the courtroom — including even defense counsel — corrected the court.
Piling error on top of errors, when the district court submitted its written statement of reasons, it wrote that the
total
offense level was 19, that the criminal history category was I, and that the guideline sentencing range was 33 to 41 months (not 31 to 41 months as it had stated earlier, or 30 to 37 months as recommended by the PSR). A guideline sentencing range of 33 to 41 months, however, applies to either a total offense level of 20 with a criminal history category of I, or a total offense level of 19 with a criminal history category of II.
See
U.S.S.G. ch. 5 pt. A (Sentencing Table). And, of course, that guideline sentencing range includes a higher floor than the (also incorrect) range announced at the hearing.
II. Analysis
On appeal, Marchena-Silvestre says that he- is entitled to resentencing for three reasons: (1) his sentencing hearing was procedurally flawed, (2) his above-guideline 72-month imprisonment term is substantively unreasonable, and (3) the
government materially breached the plea agreement. Because Marchena-Silvestre did not raise these objections in the district court, we review only for plain error.
See United States v. Dávila-González,
595 F.3d 42, 47 (1st Cir.2010);
see also United States v. Ruiz-Huertas,
792 F.3d 223, 228 (1st Cir.2015). The plain error standard of review places the burden on Marchena-Silvestre to make four showings in order to justify reversal: “(1) that an
error occurred
(2) which was clear or obvious and which not only (3) affected the defendant’s substantial rights, but also (4) seriously impaired the fairness, integrity, or public reputation of judicial proceedings.”
United States v. Duarte,
246 F.3d 56, 60 (1st Cir.2001).
A. Failure to Calculate the Guideline Sentencing Range
“ ‘[Flailing to calculate (or improperly calculating) the Guidelines range’ is a ‘significant procedural error.’ ”
United States v. Tavares,
705 F.3d 4, 25 (1st Cir.2013) (alteration in original) (quoting
Gall v. United States,
552 U.S. 38, 51, 128 S.Ct. 586, 169 L.Ed.2d 445 (2007)). The court’s calculation failures in this case were obvious and several. It neglected to calculate a total offense level, misstated the base offense level, and settled on a non-existent guideline sentencing range extending four months longer than the upper end of the range recommended by the PSR.
These
errors
and their obviousness easily satisfy the first two requirements for a successful plain error challenge. The closer question is whether Marchena-Silvestre also satisfies the requirement that he show that the obvious errors “affected [his] substantial rights.”
Duarte,
246 F.3d at 60. In the sentencing context, we construe this requirement as imposing a “burden of showing a reasonable likelihood ‘that, but for the error, the district court would have imposed a different, more favorable sentence.’ ”
United States v. Ortiz,
741 F.3d 288, 293-94 (1st Cir.2014) (quoting
United States v. Turbides-Leonardo,
468 F.3d 34, 39 (1st Cir.2006)). For the following reasons, we think that Marchena-Silvestre has shown such a reasonable likelihood.
We begin with the role of the guidelines calculation itself. We need tread no new ground in pointing out what precedent already makes clear about the required nature of that calculation:
Although the Sentencing Guidelines are now advisory rather than.mandatory, district courts are still required to ‘begin all sentencing proceedings by correctly calculating the applicable Guidelines range.’
Gall,
552 U.S. at 49, 128 S.Ct. 586. Only after a court has correctly calculated the applicable [guideline sentencing range] and evaluated the factors set out in 18 U.S.C. § 3553(a) can it properly exercise its discretion to sentence a defendant within or outside the applicable Guidelines range. Far from a meaningless exercise, the requirement that the district court begin by correctly calculating the [guideline sentencing range] serves an important function; it provides ‘a framework or starting point’ to guide the exercise of the court’s discretion.
Freeman v. United States,
— U.S. -, 131 S.Ct. 2685, 2692, 180 L.Ed.2d 519 (2011). Starting with such a framework gives the sentencing judge ‘an idea of the sentences imposed on equivalent offenders elsewhere,’ which in turn ‘promote[s] uniformity and fairness’ in sentencing.
United States v. Rodríguez,
630 F.3d 39, 41 (1st Cir.2010). Thus, even though sentencing judges are free to impose non-Guidelines sentences in appropriate cases, ‘district courts must still give respectful consideration to the now-advisory Guidelines (and their accompanying policy statements).’
Pepper v. United
States,
562 U.S. 476, 131 S.Ct. 1229, 1247, 179 L.Ed.2d 196 (2011).
United States v. Millán-Isaac,
749 F.3d 57, 66-67 (1st Cir.2014).
It is therefore fair to presume in the ordinary case that there is a reasonable likelihood that any variance added onto that starting point moves the end point beyond where it would have been but for the error in the starting point.
See Ortiz,
741 F.3d at 294 (“[TJhere is every reason to believe that the court used the [guideline sentencing range] as an anchoring point from which to vary.”);
Rodríguez,
630 F.3d at 41 (explaining that the sentencing judge must “start out by calculating the proper Guidelines range — a step so critical that a calculation error will usually require resentencing”).
This is not to say that every error in calculating the guideline sentencing range calls for reversal under plain error analysis, or even under harmless error analysis.
See United States v. Marsh,
561 F.3d 81, 86 (1st Cir.2009) (declining to resolve a dispute over whether the court erroneously applied an upward departure under the guidelines because “the district court stated that it would have reached the same result in a non-Guideline setting”);
United States v. Gerhard,
615 F.3d 7, 35 (1st Cir.2010) (similar);
cf. Williams v. United States,
503 U.S. 193, 202-03, 112 S.Ct. 1112, 117 L.Ed.2d 341 (1992) (explaining that remand is required under harmless error analysis “only if the sentence was imposed as a
result of
an incorrect application of the Guidelines” (internal quotation marks omitted)). A sentencing court might, for example, make it clear that it was aware of a possible flaw in its calculation of a guideline sentencing range, and explain that its sentence would nevertheless be the same under an alternative analysis pressed by the party that ultimately appealed.
See, e.g, Tavares,
705 F.3d at 24-28 (deeming district court’s error in not conclusively choosing between the parties’ proposed sentencing ranges harmless because the court indicated its understanding of the competing calculations and then stated it would elect a sentence above either range). Here, though, there is no such explanation by the court. Instead the government asks us to infer such a view, relying on the small number of months by which the court erred, the relatively much greater size of the variance, and the round year nature of the sentence, all on top of a fair description of the section 3553(a) factors and a clear intent to issue a variant sentence.
While the case for such an inference is certainly plausible, we think it falls short of serving as an adequate substitute for a “clear statement by the court” that would be sufficient to “diminish the potential of the [guideline sentencing range] to influence the sentence actually imposed.”
Ortiz,
741 F.3d at 294 (citing
United States v. McGhee,
651 F.3d 153, 159 (1st Cir.2011)). Here, the district court attempted to calculate a guideline sentencing range, described this case by reference to its differences from the “heartland of cases” within that range, and, throughout the hearing, recited a litany of justifications under section 3553(a) for varying upward from that range given the specific characteristics of the defendant and the crime. There is nothing wrong with this approach — unless one starts at the wrong yard marker.
It is, of course, true that the district court declared before hearing any argument that this was “not a guideline case.” Seizing on this declaration, the government argues that the incorrect calculation could not have materially affected the sentence. This argument directly conflicts with the government’s own assertions— with which we agree — that the court
“did
consider the Guidelines,” and then exercised its discretion to vary upward. And
we have already explained that the district court’s repeated references to the guidelines and the “heartland of cases” within those guidelines indicate to us that the guidelines served as a starting point from which the court imposed an upward variance. It follows that if the district court had correctly calculated a lower starting point, then there is at least a reasonable likelihood that it would have landed on a sentence shorter than 72 months (even if just a few months shorter). Nothing in this record provides any indication clear enough to overbear the probative force of this logical presumption.
See Ortiz,
741 F.3d at 294 (finding that since “the record contain[ed] no suggestion that the court considered the dimensions of the [guideline sentencing range] to be irrelevant,” an error in calculating defendant’s criminal history score caused plain error);
United States v. Fagans,
406 F.3d 138, 141 (2d Cir.2005) (explaining that “an incorrect calculation of the applicable Guidelines range will taint not only a Guidelines sentence, ... but also a non-Guidelines sentence, which may have been explicitly selected with what was thought to be the applicable Guidelines range as a frame of reference”).
Turning to the last prong of plain error review, we need not tarry. . The district court’s repeated failures to calculate the guideline sentencing range correctly, or to explain its calculation, all in a fashion that created a higher range than was recommended by either the PSR or the Agreement, compromised the fairness and integrity of the proceeding.
See Olano,
507 U.S. at 736, 113 S.Ct. 1770. We therefore conclude that, under plain error review, we should exercise our discretion to vacate Marchena-Silvestre’s sentence and afford him a new sentencing hearing.
See United States v. González-Castillo,
562 F.3d 80, 84 (1st Cir.2009).
B. Alleged Breach of the Plea Agreement
Our decision to remand for resen-tencing due to procedural error does not end our analysis, because the parties disagree concerning the meaning of the plea agreement that will still apply on resen-tencing, and we typically grant specific performance as a remedy where the government’s breach of a plea agreement leads to reversible error.
See United States v. Clark,
55 F.3d 9, 14 (1st Cir.1995). We therefore address the claimed breach, reviewing once again for plain error. In so doing, we construe the terms and conditions in plea agreements in accordance with traditional principles of contract law,
see United States v. Murphy-Cordero,
715 F.3d 398, 400 (1st Cir.2013) (interpreting a waiver of appeal clause), looking outside the document only as necessary to provide illuminating context or resolve ambiguities in the writing,
see United States v. Alegria,
192 F.3d 179, 183 (1st Cir.1999).
The parties appear to agree — as do we — that the existence of a breach turns on the meaning of the phrase “applicable guidelines range” in Paragraph 9 (“Sentence Recommendation”) of the Agreement. In Marchena-Silvestre’s view, the phrase refers to the range identified by the Agreement itself, in the chart in Paragraph 7 (“Applicability of United States
Sentencing Guidelines”), thereby requiring the government to recommend a sentence no greater than 30 months. The words of the Agreement strongly support this view. One naturally presumes that the undefined term “applicable guidelines range” would refer precisely to the guideline settlement ranges set forth in the immediately prior section of the Agreement called “Applicability of United States Sentencing Guidelines.” If this were not the intended cross-reference, and one must look outside the Agreement to figure out the applicable guideline sentencing range, one would be left to ask: Does it refer to the ranges specified in the PSR, or to those found by the district court?
The government in its brief answers this question by insisting that the “applicable guidelines range” means “the advisory Guidelines’ range found applicable at the sentencing hearing.” But the government itself adopted as its recommendation the range set forth in the PSR before the district court found the applicable range. Nor did the government revise its recommendation when it learned of the court’s different (and higher) calculation. Given this sequence of events, the government’s argument is like the thirteenth chime of a clock: you not only know it’s wrong, but it causes you to wonder about everything you heal’d before.
Returning to the Agreement itself, we observe that if the government were correct, the chart in Paragraph 7, which occupies nearly an entire page of the Agreement, would have no apparent purpose. Conversely, under Marchena-Silvestre’s reading, it serves the obvious purpose of setting out the “applicable sentencing guidelines” that limited the range of sentences the parties could recommend at the hearing. Confronted with this observation at oral argument, the government offered that the chart’s purpose was “transparency” for the defendant. But if this construction of the chart as a gratuitous, unnecessary, and non-binding educational illustration were correct, we think the chart would only be capable of confusing the defendant and setting an expectation that could both go unmet by the government’s recommendation later on and provide possible cause for a withdrawal of the plea.
Marchena-Silvestre’s reading also finds strong support in the repeated reference to a stipulation between the parties in Paragraphs 8 and 9. After Paragraph 7’s chart sets out a guideline sentencing range for each of the six criminal history categories, all based on a total offense level of 17, Paragraph 8 states that “[t]he parties do
not
stipulate as to any Criminal History Category for Defendant.” It would be entirely unnecessary to make such a declaration unless the parties
did
stipulate to the other variable in the chart’s calculations, i.e., the total offense level of 17. Paragraph 9 then goes on to state that “[t]he parties agree that any recommendation ... below or above the
stipulated sentence recommendation
constitutes a material breach” of the Agreement. (Emphasis supplied). From this language one naturally concludes that, if the district court selected a criminal history category of I, the parties would be prohibited from arguing for a sentence outside the range provided in the corresponding section of Paragraph 7’s chart.
We recently put the government on notice that its similar reading of an analogous plea agreement was “anfractuous.”
Almonte-Nuñez,
771 F.3d at 89. In
Almonte-Nuñez,
we were asked whether “the defendant [was] foreclosed from appealing [because] he was ‘sentenced in accordance with the terms and conditions set forth in the Sentence Recommendation provisions’ of the Agreement,” pursuant to the agreement’s waiver of appeal clause.
Id.
at 88. Paragraph 7 (“Sentencing Guidelines Calculations”) of that agreement included a chart that calculated the total offense level to be 25.
Id.
at 88.. Paragraph 8 (“Sentence Recommendation”) then provided that “the defendant may argue for the lower end of the applicable guideline range and the government may argue for the higher end of the guideline range applicable to defendant’s Criminal History Category[.]”
We interpreted the agreement to mean that “for the defendant to have been sentenced in accordance with the terms of the sentence recommendation provisions, he would have had to be sentenced within a [guideline sentencing range] derived from an offense level of 25.”
Almonte-Nuñez,
771 F.3d at 88. Since the district court imposed a sentence thirteen months higher than the high end of the guideline sentencing range corresponding to a total offense level of 25 and the highest possible criminal history category, “[i]t followed], as night follows day, that the sentences ... were not in conformity with the Agreement’s sentence recommendation provisions.”
Id.
at 88.
The government also argued in
Al-monte-Nuñez,
as it does here, that “the defendant was sentenced in conformance with the sentence recommendation provisions because those provisions did not lock in a particular [guideline sentencing range].”
Id:
at 89. We responded by stating that “the Agreement unambiguously set the offense level at 25 and barred arguments in favor of further adjustments.”
Id.
at 89. We similarly rejected the notion that, since the district court retained ultimate sentencing discretion, the government could shift its recommendation based on what occurred at the hearing.
Id.
at 88-89. And we noted that, “[w]ith minimal effort, the government could have drafted a waiver clause having the effect that it unrealistically ascribes to the language actually used in the Agreement.”
Id.
at 89 n. 1 (citing
United States v. Isom,
580 F.3d 43, 51 (1st Cir.2009) (considering an appeal waiver provision that applied “if the sentence imposed by the Court is within the guideline range
determined by the Court
or lower.” (emphasis supplied) (internal quotation mark omitted))).
So, the Agreement itself tilts heavily in favor of the interpretation that Marchena-Silvestre urges we adopt. The problem is that his own counsel — who presumably well understood the Agreement — failed to objeet when the government recommended a sentence outside the stipulated ranges. Was this an unwitting forfeiture? Or was it instead extrinsic evidence that the Agreement should be read as the government applied it in fact (albeit not as the government claims on appeal)? This would seem to be an issue on which many of the points for each side are own-goals.
We are tempted to rely on the standard of review as the deciding factor, given the Supreme Court’s guidance that “the second prong of plain-error review ... will often have some ‘bite’ in plea-agreement cases. Not all breaches will be clear or obvious. Plea agreements are not always models of draftsmanship, so the scope of the Government’s commitments will on occasion be open to doubt.”
Puckett v. United States,
556 U.S. 129, 142, 129 S.Ct. 1423, 173 L.Ed.2d 266 (2009). But such a reliance offers little pragmatic sense in this case. We are remanding for a new sentencing anyhow, at which defense counsel this time will presumably insist on a recommendation consistent with the chart in Paragraph 7 of the Agreement. Of
course, the prosecution has common sense, too. We cannot imagine that it will continue to use this same form to document new plea agreements if it wants to argue for a sentence in a range not reflected in the form. We note, too, that in another case before us the prosecution appears not to have pressed for an interpretation of the agreement like that for which it advocates here.
See United States v. Cirilo,
No. 14-1793, at 2-3 (1st Cir. Sept. 24, 2015). The difference between the ranges for which the parties respectively argue in this case is not great, so prudent counsel may well err on the safe side rather than create a problematic, preserved issue for appeal. In short, there is a good chance that the issue will never arise again in this case, nor in any future case should the government abandon its awkward plea agreement template. We therefore exercise our discretion not to finally adjudicate the issue at this stage of this continuing proceeding.
Finally, to leave room for the government to reassess its position on remand, we direct that a different judge shall preside over Marchena-Silvestre’s sentencing proceedings. The possibility that the government breached the plea agreement,
see United States v. Kurkculer,
918 F.2d 295, 300 (1st Cir.1990) (stating this court’s “repeatedly expressed ...
preference
for ... resentencing before a different judge” when the government breaches a plea agreement), and the fact that the judge appeared to have made up his mind that Marchena-Silvestre deserved a six year imprisonment term without knowing the correct guideline sentencing range,
see Mawson v. United States,
463 F.2d 29, 31 (1st Cir.1972) (per curiam) (explaining that “[i]t is difficult for a judge, having once made up his mind, to resentence a defendant”), counsel in favor of fresh eyes, “both for the judge’s sake, and the appearance of justice,”
id. See also United States v. Hanono-Surujun,
914 F.2d 15, 20 (1st Cir.1990) (ordering that resentenc-ing occur before a different judge due to the district court’s failure to comply with a federal rule and its sharp upward variance from the sentencing guidelines).
III. Conclusion
Marehena-Silvestre’s sentence is
vacated
and this matter is remanded for resen-tencing before a different judge.