concurring:
The opinion, which I join in full, aptly describes why the government failed to meet its burden of proving that Mark breached his informal immunity agreement with federal prosecutors. I write separately to emphasize that this case is a textbook lesson in the importance of documentation with regard to immunity deals. When it comes to proving breach of an immunity agreement, the government should do better than “he said, she said.”
*1107The government routinely enters into agreements in which it promises leniency in exchange for cooperation with an investigation or prosecution. The threshold question of whether a deal was made in the first place is often the subject of dispute. See, e.g., United States v. Aleman, 286 F.3d 86, 90 (2d Cir.2002) (noting the parties’ “failure to agree on the existence” of an agreement); United States v. Thompson, 25 F.3d 1558, 1562 (11th Cir.1994) (evaluating an “alleged oral grant of immunity”). Here, the government candidly acknowledges it made such an agreement directly with Mark, though it inexplicably failed to disclose the existence of the agreement to his counsel, who learned about it by happenstance while questioning another witness at Mark’s trial.
Granting immunity is a big deal. Claiming that a defendant breached the agreed-upon terms is an equally big deal. Prosecuting someone who was previously granted immunity implicates “more ... than just the liberty of [a] defendant. At stake is the honor of the government[,] public confidence in the fair administration of justice, and the efficient administration of justice in a federal scheme of government.” United States v. Carter, 454 F.2d 426, 428 (4th Cir.1972) (en banc). Failure to document the breach, even with something as simple as a file note, a memo, or a reference in the target letter is dumbfounding. See United States v. Harvey, 869 F.2d 1439, 1443 (11th Cir.1989) (en banc) (“[Tjhis appeal would not be necessary had the United States Attorneys ... reduced their agreement ... to writing.”).
When the government alleges breach stemming from the failure to cooperate sufficiently, a typical response is that the defendant complied with the literal terms of the deal. See, e.g., Ricketts v. Adamson, 483 U.S. 1, 11, 107 S.Ct. 2680, 97 L.Ed.2d 1 (1987) (noting that the defendant’s argulnent that his plea agreement did not require him to testify was “an interpretation of the agreement that proved erroneous”); United States v. Floyd, 1 F.3d 867, 868-70 (9th Cir.1993) (highlighting the difference between an agreement that requires a defendant to “testify fully and truthfully” and an agreement that requires her to “cooperate”); United States v. Irvine, 756 F.2d 708, 710-11 (9th Cir.1985) (per curiam) (holding that defendant’s malfeasance constituted a breach notwithstanding his accurate “testimony” because the agreement proscribed any and all “deception”). This case tracks that pattern in the broadest sense: Mark claims that he complied with the terms of his immunity agreement by cooperating with the government’s investigation, while the prosecutors maintain that he breached by giving evasive and contradictory answers to their questions.
In its details, however, this case is far from typical. Specifically, the government claims that Mark breached the immunity agreement during a July 2011 phone call with two Assistant United States Attorneys and an FBI agent. Unlike a defendant who attempts to characterize his less-than-complete cooperation as compliance with the literal terms of an agreement, Mark disputes the very event — the phone call — that prosecutors assert is the genesis of the breach. Mark says that this phone call never took place at all.
The government bears the burden of proving the breach. Op. 1105. When confronted with contradictory evidence about whether a breach occurred, the district court must make factual findings about the defendant’s compliance. See Floyd, 1 F.3d at 871. As a threshold matter, the government needed to prove that a call between prosecutors and Mark took place in July 2011. Absent a phone call, the claim of breach collapses. Two prosecutors testified that they and an FBI agent called *1108Mark. Mark denied that the call took place, and the FBI agent testified that he had no recollection of the call. While the district court is generally entitled to base its rulings on the credibility of witnesses, this case takes on a different patina due to the absence of any documentary evidence about the call itself, let alone the content of the supposed breach.
It is undisputed that the prosecutors spoke with Mark in February 2011 as part of their preparation for the Mazzarella-Grimm trial. Fast forward only five months and the prosecutors were in the throes of preparing for the trial that had been continued. In contrast to the well-documented February call, which was memorialized in phone records and a followup memo by an FBI agent, the purported July call yielded no records. It is difficult to conceive that a trial preparation call involving two attorneys, one FBI agent, and a key witness in a multimillion dollar fraud case would not trigger a substantial paper trail or at least a scrap of documentary evidence. But here, neither the prosecutors nor the FBI could find any notes, correspondence, or other documents verifying that a call had taken place, let alone the results of the call. See Harvey, 869 F.2d at 1443 (describing the government’s failure to take notes during witness interviews as “astonishing”). The only near-contemporaneous document the government produced — the target letter sent to Mark in August 2011 — contains no reference to a July call. The absence of any notes or documentation coupled with complete silence in the target letter is nothing short of remarkable.
In light of this abysmal record, the district court was forced to base its ruling on little more than a swearing contest. Interestingly, the swearing contest pitted the prosecutors against not just Mark but also an FBI agent who had no recollection of the call. Significantly, the court did not make an adverse credibility finding regarding Mark. Following the district court’s initial ruling, Mark requested that the government produce phone records to corroborate that the call took place. In response, the government for the first time attempted to reconstruct evidence of the call. The government’s belated search for phone records, however, demonstrated that the central details of the prosecutors’ testimony were inaccurate: no call was placed from the United States Attorneys’ office to Mark’s cell phone in July 2011. At that point, the district court should have dismissed the indictment or granted a new evidentiary hearing. Notably the government has now eschewed any suggestion of another evidentiary hearing.
Whenever a defendant is prosecuted after having been given immunity, it is eminently foreseeable that he will advance every legitimate argument that his immunity should have remained intact. Cases involving oral grants of immunity and undocumented breaches “create confusion for the government and for the courts.” Harvey, 869 F.2d at 1443. Contemporaneous documentation is thus critical to detail the scope and terms of the agreement and equally critical to establish whether a breach occurred. Setting up a claim of breach without any documentation puts the court and counsel in the unenviable position of reconstructing a breach solely through a swearing contest and a hypothetical reimagining of events. To be sure, the government holds the cards in such a situation. But when the defendant’s liberty rests on those cards, common sense, fairness, and confidence in the system demand more. I therefore concur in the opinion remanding with instructions to dismiss the indictment.