dissenting.
To establish a violation of the due process clause, Banks had to show that Browning’s credit-card problem was exculpatory evidence. Even then, a new trial is appropriate only if there is “a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.” United States v. Bagley, 473 U.S. 667, 682, 105 S.Ct. 3375, 87 L.Ed.2d 481 (1985).
The district judge never decided whether the credit-card evidence was exculpatory. His opinion, which covers less than two pages, asserts that the evidence was impeaching but does not say why. The judge must have assumed that the credit-card problem gave Browning an incentive to ingratiate herself with the DEA by identifying the substance as cocaine whether it was or not, and that the defense could have used this motive to undercut her testimony. But that is hardly the only available inference.
Because Browning’s assertions could have been verified or refuted by other evidence, her financial difficulties gave her a reason to be extra careful, not a reason to commit perjury. A chemist found to have lied about a substance’s composition will be fired and may be prosecuted. Careless analysis (short of perjury) will be less acceptable from a chemist who is in hot water than from a chemist with a spotless record. Defense counsel would have stepped on a land mine by introducing the credit-card evidence, only to have Browning reply “yes, because I was in trouble for financial issues I took extra care to analyze this sample”; the prosecutor then could have elicited on re-direct that chemists who misidentify a substance are disciplined and reminded the jury that the defense could have had the substance analyzed by a forensic chemist.
My colleagues say (at p. 511) that when “there are at least two plausible explanations [one impeaching and the other not] the choice between them was one for the district court to make.” I agree with this principle. Unfortunately, the district court did not make the choice. The court simply assumed that the credit-card evidence would have undermined rather than strengthened Browning’s testimony. We can’t apply a deferential standard of appellate review when the district court has not confronted the competing inferences and made a reasoned choice. See, e.g., United States v. Beasley, 809 F.2d 1273 (7th Cir.1987).
If the evidence would have undermined Browning’s analysis, a further question must be explored. What would have happened had the prosecutor known about this evidence and revealed it to the defense? The district court assumed that Browning would have testified and been impeached. But there is another possibility. The prosecutor might have used a different chemist. Then no impeaching evidence would have been before the jury, and the trial would have proceeded exactly as it did. That replacement chemist’s testimony would not have been undercut. This is exactly what a declaration filed by the prosecutor (Domagala Declaration at ¶ 6) says would have happened. On this understanding withholding the credit-card information did not hamper the defense. It changed the identity of the chemist but not the substance of any testimony.
According to the majority, the district court “was not required to accept on faith the Government’s assertion that its replacement chemist would have been unimpeachable.” At p. 512. I agree: Never is a district judge required to accept something on faith. Once again, however, the problem is that the district judge did not even mention this subject. No testimony *515was taken; no facts were found; no discretion was exercised.
Nor could the district judge have rejected Domagala’s submission. There is no contrary evidence. A judge cannot preemptively disbelieve statements in an affidavit. At the post-trial hearing when Browning testified about her financial troubles, the district judge could have demanded that the prosecutor produce Do-magala or other proof about what could have been done to replace Browning. After entertaining evidence from both sides, the judge could have made findings of fact. But that was not done.
Everything I have said so far rests on arguments presented to the district court and reiterated in the prosecutor’s appellate brief. There are deeper problems with the district court’s approach, however. Perhaps the prosecutor thought it unnecessary to discuss them, so clear are the errors I have mentioned. But they are still worth attention, because they put the issues in perspective.
The premise of Bagley, which extended Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963), from primary exculpatory evidence to material that could be used to impeach a witness, is that a witness’s credibility can be vital to the strength of the prosecution’s case. That will be so when the testimony is not verifiable (the witness testifies to something seen or heard, for example). It is not so when the witness’s contentions can be checked by a third party. If, for example, a witness testifies that a particular document contains particular language, the best way to find the truth is to look at the document, not to inquire whether the witness has a felony conviction or is in financial trouble or has told a contradictory story to someone else.
Everything that Browning said could have been checked. The question at hand was whether the seized substance is cocaine, not whether Browning had analyzed that substance correctly (or was telling the truth about her own analysis). Conviction (or acquittal) depends on what the substance actually is. Cf. United States v. Moon, 512 F.3d 359 (7th Cir.2008) (lab results are not “testimonial” under the confrontation clause).
Brady applies only to evidence that is known to the government but not the defense. The seized substance, however, was available to the defense. The jury found that Banks was the person from whom the substance had been seized (that conclusion, which is essential to the verdict, is not undermined by any problem in Browning’s testimony), and Banks himself knew what he was carrying. More: the substance could have been analyzed by the defense. The Treasury would have paid for a chemist to analyze it on defendant’s behalf. 18 U.S.C. § 3006A(e)(l). Because the substance was available to both sides, there cannot have been a Brady problem.
If Banks thought that the substance was not cocaine (or even that there was a serious question about its composition), defense counsel would not have relied exclusively on trying to impeach the DEA’s chemist — and at trial defense counsel did not even try to impeach Browning. (Counsel waived cross-examination.) To challenge Browning’s conclusions, the defense would have needed its own expert and could have had the substance analyzed readily. Yet Banks has never denied that the substance is cocaine and has never had it tested (at least, has not argued that tests undermine Browning’s results; perhaps a defense expert confirmed Browning’s conclusions). It would be absurd to hold a new trial to establish, for a second time, something that is uncontested.
*516Once Browning’s credit-card problem came to light, and the prosecutor said that it would have been easy to use another chemist, the district judge had to decide what would have happened. The eviden-tiary hearing at which Browning testified about her finances should have explored two other issues: First, would the prosecution have used another chemist? Second, what would this other chemist have said? Suppose the prosecutor had testified that he would have used another chemist, and the judge had believed this. Then there would be no basis for another trial, unless there is good reason to think that the second chemist would reach a conclusion different from Browning’s. And the best way to find out is to appoint an independent (non-DEA) expert and have that expert analyze the substance.
This case went off the rails because the parties failed to alert the district judge to the distinction between verifiable and non-verifiable testimony. When impeaching evidence about a non-verifiable subject (such as whether a witness saw the defendant shoot the victim) comes out after trial, the only way to probe that subject’s significance is to hold another trial with all evidence placed before the jury. But for a question such as “is substance X cocaine or sugar?”, there is a way to explore materiality without a new trial. Unless a fresh expert analysis of the substance would undermine Browning’s analysis, a new trial would be pointless. The defense bears the burdens of production and persuasion on Brady claims; the gaps in the record mean that the outcome of the trial stands.