SELYA, Circuit Judge.
In criminal cases, proffer sessions are commonly used as a means of facilitating plea negotiations. A defendant’s agreement to participate in a proffer session carries with it both potential risks and potential rewards.
To protect the proffering defendant against unfairness, prosecutors customarily offer assurances that they will not use what the defendant reveals at the proffer session against him should plea negotiations fail. In this case of first impression, the government promised the defendant that it would not use against him any “statements made or other information” disclosed at the proffer session.
The proffer session came to naught. When trial became imminent, the government attempted to pull a rabbit out of a hat: it advised the defendant that, notwithstanding its earlier assurances, it planned to introduce at trial voice identification testimony from a police officer, based on what the officer had heard at the proffer session, linking the defendant to a highly incriminating recorded telephone conversation. Although the court found the government’s plan “troubling,” it nonetheless denied the defendant’s motion in limine and allowed the government to present the testimony. The jury returned a guilty verdict.
On appeal, the defendant advances three claims of error, including a renewed challenge to the introduction of the voice identification testimony. With respect to this pivotal issue, we conclude that the admission of the testimony transgressed the proffer agreement and the defendant’s due process rights. On the assumption that such a transgression is subject to harmless error review, we conclude that the proper barometer for that review requires us to ask whether the government has shown that the error was harmless beyond a reasonable doubt. Because the government has not carried this heavy burden, we vacate the defendant’s conviction.
The moral of this story is that, especially when dealing with criminal defendants at proffer sessions, the government must turn square corners. The government did not do so here.
I. BACKGROUND
On February 19, 2010, an agent of the Federal Bureau of Investigation (FBI), Jeffrey Lawrence, working with other law enforcement personnel, set in motion a controlled purchase of crack cocaine. The leading man in the production was a cooperating witness, Robert Williams. The agents equipped Williams with video and audio recorders, and a Massachusetts state trooper, Thomas Fitzgerald, searched him to ensure that he had no drugs or other contraband on his person.
Williams then made a recorded telephone call to the target of the sting, Anthony Hook. During this call, Hook agreed to sell “three and a teenth” of crack cocaine
for $500, and handed the telephone to another individual to arrange the exchange. That individual, a stranger to Williams, set up a meeting at the parking lot of a package store (a retail liquor emporium) and stated that he was wearing a black jacket. Williams replied that he would be driving a blue Chevy Blazer.
After the call, Agent Lawrence gave Williams $500 to make the buy. Williams then drove to the package store under the watchful eye of agents in a surveillance van. Along the way, he stopped for gas and spent a brief interlude unattended inside the gas station’s store.
After Williams arrived at the designated parking lot, defendant-appellant James Melvin, wearing black clothing, entered his vehicle. This encounter was surreptitiously videotaped.
Williams asked the defendant, “What we got right here?” The defendant answered, “Three balls and a teenth.” The videotape shows the defendant removing an unidentifiable object from his waistband. At trial, Williams testified that the object was crack cocaine, which the defendant weighed on a scale and sold to him for $500. The defendant then disembarked from the vehicle.
As Williams began to leave, he was approached by a family friend who “slapped him five” through the open window of his vehicle. Williams’s next stop was a meeting with government agents, where he surrendered a bag of crack cocaine, a scale, and the recording equipment.
The defendant was not immediately arrested. For reasons not relevant here, he later wound up in state custody. Trooper Fitzgerald happened to be present during the defendant’s booking and recognized him as the person who had entered Williams’s vehicle on February 19 (almost two months earlier). The trooper reported this bit of serendipity to Agent Lawrence. On April 22, 2010, the latter showed Williams an array of nine photographs (one of which depicted the defendant) and asked him to identify the seller of the cocaine. Williams chose the defendant’s photograph.
A federal grand jury sitting in the District of Massachusetts subsequently indicted the defendant on a single count of possessing cocaine with intent to distribute.
See
21 U.S.C. § 841(a)(1). The indictment was returned on May 20, 2010. That same day, law enforcement personnel, including Officer James Mazza, a Springfield police officer (who had been part of the February 19 surveillance team), interviewed the defendant at a proffer session. The proffer session was held pursuant to a written proffer agreement drafted by the government.
The proffer session proved fruitless and the defendant proceeded to trial. The government posited that the defendant sold crack cocaine to Williams in the parking lot. As part of its case in chief, the government called Officer Mazza to testify that he had familiarized himself with the defendant’s voice during the proffer session and, through that acquired knowledge, could identify the defendant as the unknown speaker who had participated in the February 19 telephone call. The defense attempted to point the finger at Williams, implying that he had acquired the crack cocaine elsewhere and had
staged matters so as to shift the blame to the defendant.
The jury found the defendant guilty. After the imposition of sentence, this timely appeal followed.
II. ANALYSIS
In this venue, the defendant advances three claims of error. This litany includes a claim that the district court should have suppressed the photo array identification as impermissibly suggestive and unreliable; a claim that Williams’s testimony should have been excluded because a government agent had told him that payment for his services depended on future convictions; and a claim that allowing Officer Mazza’s voice identification testimony violated his due process rights.
We address these arguments in inverse order of difficulty.
A.
The Photo Array.
We need not linger long over the defendant’s challenge to the photo array. The defendant filed a pretrial motion to suppress this identification, arguing that only two of the nine photographs were “close-ups” (one depicting the defendant, the other a man with a beard and a facial scar); and that due to this peculiarity, the array was both unreliable and impermissibly suggestive. The district court deemed this argument unconvincing and so do we.
To withhold photo array identification evidence from a jury, a court must be “persuaded that there was a very substantial likelihood of irreparable misidentification.”
United States v. Rivera-Rivera,
555 F.3d 277, 282 (1st Cir.2009) (internal quotation marks omitted). This analysis proceeds in two steps. First, the court must determine whether the identification procedure was impermissibly suggestive. If it was not, the inquiry ends. If, however, the defendant can successfully negotiate this first step, the court must then examine the totality of the circumstances to ascertain whether the identification was nevertheless reliable.
See United States v. DeCologero,
530 F.3d 36, 62 (1st Cir.2008).
Our review of a district court’s bottom-line conclusions about reliability and suggestiveness is plenary.
See United States v. Espinal-Almeida,
699 F.3d 588, 602 (1st Cir.2012),
cert. denied,
— U.S. -, 133 S.Ct. 1837, 185 L.Ed.2d 845 (2013). We review any underlying findings of fact for clear error.
See id.
The defendant’s assignment of error falters at the first step of the two-step pavane. As the district court noted, the defendant’s photograph shows him only “somewhat closer to the camera” than others. Minor differences within the components of a photo array are to be expected and, without more, do not render the identification procedure impermissibly suggestive.
See, e.g., United States v. Brennick,
405 F.3d 96, 99-100 (1st Cir.2005). Here, there is no “more.”
To dwell on this aspect of the case would be pointless. Jurors should not be treated as gullible dupes,
see Manson v. Brath-waite,
432 U.S. 98, 116, 97 S.Ct. 2243, 53 L.Ed.2d 140 (1977), and photo array identification evidence should be withheld from them “only in extraordinary cases,”
Rivera-Rivera,
555 F.3d at 282 (internal quotation marks omitted). The photo array here was unexceptionable, and the district court’s denial of the motion to suppress was proper.
B.
The Bounty.
In an argument that reminds us that the love of money is the root of all evil,
see
I Timothy 6:10, the defendant asserts that the government’s quondam financial arrangement with Williams — dollars for convictions — so tainted his testimony as to work a violation of the Due Process Clause, U.S. Const, amend. V. This is a closer question, but we ultimately conclude that the prosecutor’s timely intervention and the deployment of effective safeguards sufficed to mitigate the taint of which the defendant complains.
The relevant facts are not disputed. Williams did not agree to serve as a government cooperator for love of country but, rather, because the government promised to compensate him. Between December of 2010 and March of 2011, Agent Lawrence told him on approximately six occasions that he would be paid only when the government had obtained three convictions through his cooperation.
When the prosecutor learned of this tawdry arrangement, he put an end to it. On March 16, 2011, he revealed the arrangement to defense counsel and explained that he had directed Agent Lawrence to disabuse Williams of the notion that future payments would be contingent upon convictions. At a meeting held soon thereafter, Agent Lawrence, flanked by two prosecutors, told Williams that he would be paid for his services, not for convictions.
Prior to trial, the defendant moved in limine to exclude Williams’s testimony as irrevocably corrupted by the original bounty arrangement. The district court denied the motion, citing the government’s later disavowal and disclaimer of that arrangement.
In appealing this ruling, the defendant places heavy reliance on prior decisions of this court that contain language condemning witness compensation contingent upon convictions.
United States v. Cresta,
825 F.2d 538 (1st Cir.1987);
United States v. Dailey,
759 F.2d 192 (1st Cir.1985). This language, he insists, should have led the district court to exclude Williams’s testimony.
In
Dailey,
we examined plea agreements in which the government’s promise to recommend leniency depended on the “value” or “benefit” of cooperation to the government. 759 F.2d at 194-97. We held that “established safeguards” — informing the jury of the precise nature of the arrangement, cross-examination, and specific instructions to weigh the cooperating witness’s testimony with care — would adequately protect the defendant’s due process rights against any potential prejudice arising from the government’s promise.
Id.
at 196-97, 200.
Dailey
is not directly applicable to this case because we took pains to distinguish the agreements at issue there from those in which a “promised benefit is made contingent upon a resulting indictment or conviction.”
Id.
at 197. However, our decision acknowledged in dictum that we could “think of no instance in which the government would be justified in making a promised benefit contingent upon the return of an indictment or a guilty verdict.”
Id.
at 201. We added “that benefits made contingent upon subsequent indictments or convictions skate very close to, if indeed they do not cross, the limits imposed by the due process clause.”
Id.
at 201 n. 9.
Dictum in
Cresta
reinforced these observations. Though not faced with an actual bounty arrangement there, we reiterated that “we would not condone” an agreement promising such contingent benefits. 825 F.2d at 547.
We do not retreat from what we said in
Dailey
and
Cresta:
we continue to view
the practice of conditioning a witness’s benefits upon the incidence of future indictments or convictions as pernicious. A cooperator whose eyes are fixed on such a prize might well be tempted to wander from the path of candor and, thus, unduly prejudice the criminal defendant.
We add, moreover, that such bounty arrangements are likely to boomerang against prosecutors. After all, fee arrangements that are contingent upon the securing of indictments or convictions detract mightily from a cooperator’s credibility at trial. We are, therefore, puzzled as to why the government would employ such a tactic more than a quarter century after we had twice disapproved of it.
Here, however, the story does not end with a witness laboring under the belief that conviction — and conviction alone — will earn him his keep. Before Williams was called to testify, the prosecutor pulled him back from the abyss and removed the unsavory lure of dollars for convictions. None of our prior cases address such a situation, even in dictum.
In our view, the prosecutor’s intervention, coupled with the district court’s prophylactic instructions, makes all the difference. With the bounty taken unequivocally off the table, we think that the question of how to proceed was within the sound discretion of the trial court.
Here, the court decided that sunlight was a sufficient disinfectant. The nature of the bounty arrangement and the prosecutor’s efforts to rectify the problem were fully disclosed to the jury. Williams was subject to .cross-examination on these points as well as on his overall testimony. As a further safeguard, the district court gave pellucid instructions that the jury should consider the testimony of the cooperating witness “with particular caution” and an eye toward “whether his testimony has been affected by his interest in maintaining his relationship with the government or by any of the benefits he has received from the government.” In the circumstances of this case, we think that this set of safeguards was adequate to protect the defendant’s due process rights. When all is said and done, we cannot set aside a defendant’s conviction simply because we disapprove of some aspect of the government’s handling of the case.
See, e.g., United States v. Santana,
6 F.3d 1, 9-11 (1st Cir.1993) (discussing limits of supervisory power). On this record, we conclude that the prosecutor’s timely and unequivocal correction of Agent Lawrence’s blunder combined with the effective safeguards that were employed at trial sufficed to prevent a due process violation.
C.
The Proffer Agreement.
The defendant’s final claim of error is the most vexing. The defendant agreed to appear at a proffer session. That session was governed by a written agreement, drafted by the government, which stated in pertinent part:
1. No statements made or other information provided by [the defendant] will be used by the United States Attorney directly against him, except for purposes of cross-examination and/or impeachment. ...
2. The government may make derivative use of, or may pursue any investigative leads suggested by, any statements made or other information provided by [the defendant] in the course of the proffer.
Officer Mazza attended this proffer session on behalf of the government and became familiar with the defendant’s voice through his attendance. When plea negotiations cratered and trial became necessary, the government sought to have him identify the defendant’s voice on the incul-
patory recording of the February 19 telephone call. The defendant moved in li-mine to exclude this testimony, arguing that it was a direct use of “other information” provided by the defendant — his voice tone and characteristics — at the proffer session. Even though the district court acknowledged that the government’s tactic left “a bad taste,” it denied the motion. The court reasoned that the Criminal Rules protect a defendant’s statements at a proffer session but not the sound of his voice.
The district court’s reasoning fails to take into account the actual language of the proffer agreement. To this extent, its approach was incorrect. Informal immunity agreements, such as proffer agreements, “are shaped ... by the language of the contract conferring the immunity.”
United States v. Hogan,
862 F.2d 386, 388 (1st Cir.1988). The district court should have looked, therefore, to the text of the proffer agreement. The meaning of the proffer agreement presents a question of law, which engenders de novo review.
See, e.g., United States v. Atwood,
963 F.2d 476, 478 (1st Cir.1992). In performing this interpretative task, we are guided chiefly by contract-law principles.
See United States v. McLaughlin,
957 F.2d 12, 16 (1st Cir.1992);
Hogan,
862 F.2d at 388.
In the case at hand, the relevant contract-law principle is that “whenever possible, contracts should be construed to give effect to every word, clause, and phrase.”
Crowe v. Bolduc,
365 F.3d 86, 97 (1st Cir.2004);
see United States v. Alegria,
192 F.3d 179, 185 (1st Cir.1999). Thus, the phrase “other information,” which appears in the proffer agreement, must mean something different from the antecedent term “statements” (which refers to statements that the defendant might make at the proffer session). Any other interpretation would render the phrase “other information” redundant (and, thus, meaningless).
Viewed in context, we think that the natural, common-sense interpretation of the phrase “other information provided by [the defendant]” in the proffer agreement encompasses any knowledge the government gained at the proffer session because of the defendant’s presence there.
See, e.g.,
Webster’s Third New International Dictionary 1160 (2002) (defining “information” as “knowledge communicated by others or obtained from investigation, study, or instruction”). As such, knowledge of the defendant’s voice tone, inflections, and speaking characteristics gleaned from the defendant’s speech at the proffer session is “other information” that cannot be used against the defendant in the government’s case in chief. But that is exactly what the government did: it elicited testimony from Officer Mazza on direct examination about the knowledge of the defendant’s voice that he acquired at the proffer session and then, building on that foundation, had Officer Mazza identify the defendant as the third person who spoke during the recorded conversation between Williams and Hook.
If more were needed — and we doubt that it is — another settled principle of contract interpretation counsels in favor of reading the phrase “other information provided” by the defendant broadly. Under the doctrine of contra proferentem an ambiguous term in a contract normally should be construed against the interest of the drafter, and this praxis applies to proffer agreements.
See, e.g., United States v. Hill,
643 F.3d 807, 875-76 (11th Cir.2011). Here, the phrase “other information provided” by the defendant cannot be said unambiguously to exclude the tone, inflection, and other characteristics of the defendant’s voice.
Struggling to parry these thrusts, the government suggests that the language “statements made and other information provided” in the proffer agreement was meant to mirror the distinction between use immunity for oral testimony and act-of-production immunity for documents and other tangible things.
See
18 U.S.C. §§ 6001-6003. This suggestion is nothing more than an attempt to rewrite the proffer agreement: the government could have framed such a dichotomy when it drafted the proffer agreement, but it did not do so. Whatever may be the contours of formal use immunity granted by courts, it is the language of the proffer agreement, not the language of the immunity statute, that defines the parties’ rights and responsibilities arising out of a proffer session.
See Hogan,
862 F.2d at 388. Consequently, we reject out of hand the government’s strained attempt to forge an analogy.
The government also contends that the phrase “other information” cannot encompass the defendant’s tone, inflections, and other vocal characteristics because it could have successfully moved to compel a voice exemplar from the defendant.
See United States v. Dionisio,
410 U.S. 1, 5-7, 93 S.Ct. 764, 35 L.Ed.2d 67 (1973). To buttress this contention, the government muses that the proffer agreement should not be read to protect information that the government could have procured by compulsion.
This contention is futile. We are not concerned with what the government might have done but, rather, with what the government did.
See United States v. Santana,
175 F.3d 57, 64 n. 6 (1st Cir.1999). The government never moved for a voice exemplar here. Instead, it relied on its (mistaken) interpretation of the proffer agreement. Much as it might like to do so, it cannot reinvent that agreement now.
The short of it is that the existence of a parallel universe in which the government properly obtained its sought-after information does not render its misstep in this one any less violative of the defendant’s due process rights. A constitutional end may not be reached by unconstitutional means.
In a last-ditch effort to save the day, the government notes that the proffer agreement permits derivative uses of “statements made or other information provided” by the defendant during the proffer session. Clinging to this clause, it cursorily suggests that Officer Mazza’s voice identification testimony was a derivative, not direct, use of information from the proffer session. This is simply wrong: the government made a direct, not a derivative, use of the proscribed information.
See, e.g., United States v. Pielago,
135 F.3d 703, 710-11 (11th Cir.1998) (explaining that introducing information obtained from the defendant against him is a direct use of that information).
The short of it is that the government, having agreed not to use information provided by the defendant at the proffer session, reneged on that promise and used such information at trial. We will not countenance that sleight-of-hand. As we have said, “the government must turn square corners when it undertakes a criminal prosecution.”
Ferrara v. United States,
456 F.3d 278, 280 (1st Cir.2006).
Our conclusion that the district court erred in admitting Officer Mazza’s voice identification testimony does not end our odyssey. We must also decide whether the admission of that testimony should be considered harmless.
There are two standards for measuring harmless error in a criminal case. The less grueling standard (viewed from the government’s coign of vantage) allows a conviction to stand as long as a reviewing court can say “with fair assurance, after pondering all that happened without stripping the erroneous action from the whole, that the judgment was not substantially swayed by the error.”
United States v. Sasso,
695 F.3d 25, 29 (1st Cir.2012) (quoting
Kotteakos v. United States,
328 U.S. 750, 765, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)). This less grueling standard applies chiefly to errors of a non-constitutional dimension. The stricter standard, which applies to errors of constitutional dimension, requires reversal unless the government proves “beyond a reasonable doubt that the error did not influence the verdict.”
Id.
(citing
Chapman v. California,
386 U.S. 18, 23-24, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967)).
Laboring to put the genie back in the bottle, the government takes the position, without explanation or elaboration, that the less grueling standard should apply to this error. We do not agree.
Proffer agreements are sui generis, and the contract-law principles that courts use in construing them are glossed with a concern that the defendant’s consent to appear at a proffer session should not become a lever that can be used to uproot his right to fundamental fairness under the Due Process Clause.
See United States v. $87,118.00 in U.S. Currency,
95 F.3d 511, 517 (7th Cir.1996). “Unlike the normal commercial contract,” it is “due process [that] requires that the government adhere to the terms of any ... immunity agreement it makes.”
United States v. Pelletier,
898 F.2d 297, 302 (2d Cir.1990);
see Hill,
643 F.3d at 874.
Because the government’s adherence to the terms of the proffer agreement is insured by the Due Process Clause, its failure to adhere is perforce of constitutional dimension. It follows inexorably that the stricter harmless-error standard applies to such a failure.
See, e.g., Hill,
643 F.3d at 877-79. For present purposes, this means that the government must show beyond any reasonable doubt that the guilty verdict was not influenced by the trial court’s erroneous admission of Officer Mazza’s voice identification testimony.
The government strives to minimize the force of Officer Mazza’s testimony. It asseverates that the defendant’s identity was not in question because he admitted that he was the man in Williams’s vehicle. Thus, its thesis runs, Officer Mazza’s testimony could not have caused any harm. This asseveration, however, is little more than magical thinking; it confuses and conflates the identity of the person in Williams’s vehicle with the identity of the person who spoke during the antecedent telephone call. We are directed to no portion of the record in which the defendant conceded the latter point.
The government has a fallback position. It boasts that evidence of the defendant’s guilt was so overwhelming as to render the tainted voice identification benign.
This proposition relies heavily on Williams’s tes
timony and the videotaped recording of what transpired in Williams’s vehicle. But probative force, like beauty, often lies in the eye of the beholder. Here, the videotape showed no exchange of drugs, and Williams — especially in light of the aborted bounty arrangement — hardly could be regarded as a paragon of veracity.
These issues are not merely theoretical. The defendant argued that Williams set him up presumably to earn the bounty, and that Williams, not the defendant, supplied the drugs. While the government’s evidence was adequate to sustain the defendant’s conviction against a sufficiency challenge, “more is required to show harmlessness beyond a reasonable doubt.”
United States v. Perez-Ruiz,
353 F.3d 1, 19 (1st Cir.2003).
In the end, Officer Mazza’s voice identification testimony bore on both furcula of the defendant’s two-fold argument for acquittal. For one thing, the defendant claimed that it was Williams who was responsible for the drugs that were later confiscated by the authorities; yet Officer Mazza’s voice identification pegged the vehicle encounter as a pre-arranged drug deal and, thus, undermined the defendant’s argument that the encounter was merely a staged show. For another thing, the defendant vigorously attacked Williams’s credibility; yet Officer Mazza’s testimony bolstered Williams’s account of who had participated in the crucial telephone call.
To cinch matters, we need look no further than the prosecutor’s summation to gauge the importance of Officer Mazza’s voice identification testimony. The prosecutor emphasized that the jurors would “know that it was the defendant who got in the car and who’s on the telephone call beforehand because of the voice identification made by Officer Mazza.” The government should not be allowed to have it both ways.
Cf. United States v. Tierney,
760 F.2d 382, 388 (1st Cir.1985) (“Having one’s cake and eating it, too, is not in fashion in this circuit.”). If Officer Mazza’s voice identification testimony was critical (as the prosecutor told the jury), the government should not now be heard to dismiss that testimony as trivia.
To sum up, we cannot say beyond a reasonable doubt that the guilty verdict in this case was uninfluenced by the erroneous admission of Officer Mazza’s voice identification testimony. Therefore, the defendant’s conviction cannot stand.
III. CONCLUSION
We need go no further. For the reasons elucidated above, we vacate the defendant’s conviction and remand for a new trial in which the district court shall exclude any testimony that is a direct use of “statements made or other information” garnered from the defendant’s proffer session.
Vacated and Remanded.