STEPHEN F. WILLIAMS, Circuit Judge:

A jury found Xavier Brooks guilty of various drug charges in October 1990. About six weeks after that trial, police officer Christie Hoyle, the government's chief witness, was fatally shot in the apartment of a fellow officer, David Rowland, evidently with her own service revolver. Later (and apparently without knowledge of the shooting), the judge granted a motion for a new trial, explaining that he was troubled by discrepancies in the testimony, by behavior of the defendant that seemed not to match that of a competent drug dealer, and by a possibly erroneous admission of evidence. See Order and Memorandum of February 19, 1991 (D.D.C. No. 90-0332). At the second trial, the government offered the transcript of Hoyle's original testimony under Federal Rule of Evidence 804(b)(1), which allows use of an unavailable witness's former testimony under some circumstances, and the court admitted it. A second guilty verdict followed. Because the government failed to check pertinent files for information possibly reflecting on Hoyle's credibility, see Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963), we remand for the trial court to require such a search.

* * * * * *

Before the start of the second trial, defense counsel sought either to secure, or to cause the court to examine, any files of the Metropolitan Police Department relating to Hoyle's death. Understandably, counsel was somewhat vague on what he expected the files to show, but the gist of his argument was that a police officer shot with her own revolver in the presence of a fellow officer might have had some sort of problem that would bear on her credibility. As Officer Rowland had been present, defense counsel focussed on the possibility of an "Internal Affairs Division" file on Rowland and his relation to the death. Vernon Gill, General Counsel of the Police Department, attended a preliminary session of court, bringing with him Hoyle's personnel file. The district judge read the personnel file and declared it free of material undermining Hoyle's credibility. In addition, Gill represented to the court that there was no Internal Affairs Division investigation of Hoyle or her death. At the same time, he noted that he had not been asked whether there was a homicide investigation file, and that he had "no knowledge" of any such file but that he presumed one existed. The Assistant U.S. Attorney, inadvertently we assume, recharacterized Gill's statements as a denial that there was a "file by the Internal Affairs Division or anything concerning officer Christie Hoyle" (emphasis added). It is plain that Gill did not deny the existence of an Internal Affairs file on Rowland, which might bear on Hoyle's conduct or character, or the existence of a homicide file on the shooting itself. The assistant's mischaracterization, however, suggests the presence of considerable confusion in the courtroom, confusion that the assistant's mischaracterization doubtless aggravated.

We need not review the evidence at trial, except to note that Hoyle's testimony was plainly critical. She said that she purchased crack from Brooks for $40 and that she saw him hide some ziplock bags containing rock-like substances, bags that other officers later found in the place she identified. Without this evidence, the government had no case against Brooks, as only Hoyle's testimony linked him to the crack she said she bought from him or to the stash of crack found by the other officers.

Brooks's claim here raises two threshold issues that this circuit has expressly left open, see United States v. Kelly, 790 F.2d 130, 135 n. 2 (1986): first, whether the prosecution's Brady obligations include not only a duty to disclose exculpatory information, but also a duty to search possible sources for such information; second, if the duty exists, whether it extends to files in the possession of agencies other than the prosecutor's office (conceived here as simply the U.S. Attorney's office).

In resolving these questions, the Supreme Court's identification of the essence of the Brady rule is of little help:

The heart of the holding in Brady is the prosecution's suppression of evidence, in the face of a defense production request, where the evidence is favorable to the accused and is material either to guilt or to punishment. Important, then, are (a) suppression by the prosecution after a request by the defense, (b) the evidence's favorable character for the defense, and (c) the materiality of the evidence.

Moore v. Illinois, 408 U.S. 786, 794-95, 92 S.Ct. 2562, 2567-68, 33 L.Ed.2d 706 (1972). This does not speak directly to information that is available to the prosecution but of which none of the prosecutors was aware. The Court, has, however, made clear that the test is an objective one, not dependent in any way on prosecutorial bad faith, see Brady, 373 U.S. at 87, 83 S.Ct. at 2567-68, so the rule is framed in a way that permits its application to facts of which the prosecutors are ignorant. In one case, the Supreme Court appears to have assumed extension of the rule to unsearched files outside the prosecutor's office. Pennsylvania v. Ritchie, 480 U.S. 39, 57, 107 S.Ct. 989, 1001, 94 L.Ed.2d 40 (1987). We conclude that that assumption is the law and join the three circuits that have addressed the matter, the 3rd, 5th and 7th, in answering both questions affirmatively.

In extending the Brady duty to searches for evidence, the 5th Circuit framed the matter as one of incentives for the government, arguing that without the extension "we would be inviting and placing a premium on conduct unworthy of representatives of the United States Government." United States v. Auten, 632 F.2d 478, 481 (5th Cir. 1980). The 7th Circuit has sounded a similar note, warning that "a prosecutor's office cannot get around Brady by keeping itself in ignorance, or compartmentalizing information about different aspects of a case." Carey v. Duckworth, 738 F.2d 875, 878 (7th Cir. 1984). The 3rd Circuit, while following the 5th Circuit's Auten decision, did not refer to incentives directly, although its observation that the U.S. Attorney's failure to check an obvious database for a key witness's criminal conviction "amounted to conduct unworthy of the United States Attorney's Office", United States v. Perdomo, 929 F.2d 967, 970-71 (3rd Cir. 1991), may rest on the same judgment.

Of course the prosecutor's own interest in avoiding surprise at trial gives him a very considerable incentive to search accessible files for possibly exculpatory evidence, quite independent of Brady. Accordingly there is less need for a judicially constructed incentive than in the classic Brady situation, where prosecutors already possess the information but may have little incentive to divulge it apart from the Brady rule itself. We suspect the courts' willingness to insist on an affirmative duty of inquiry may stem primarily from a sense that an inaccurate conviction based on government failure to turn over an easily turned rock is essentially as offensive as one based on government non-disclosure. See, e.g., Calley v. Callaway, 519 F.2d 184, 223 (5th Cir. 1975) ( en banc) (reflecting concern for "inherent fairness").

The cases finding a duty to search have involved files maintained by branches of government "closely aligned with the prosecution", United States ex rel. Fairman, 769 F.2d 386, 391 (7th Cir. 1985), and in each case the court has found the bureaucratic boundary too weak to limit the duty. Thus in Auten the duty reached the key witness's convictions in the FBI and National Crime Information Center files, see 632 F.2d at 480-81; in United States v. Deutsch, 475 F.2d 55 (5th Cir. 1973) — possible adverse information in the personnel file of a Post Office employee who testified that defendants sought to bribe him; in Fairman (a state prosecution) — the worksheet of a Chicago police officer who had testified for the prosecution at a trial involving an attack on Chicago police officers. But cf. United States v. Zavala, 839 F.2d 523, 528 (9th Cir. 1988) ( Brady does not extend to reports in the hands of the court or probation office). Given the close working relationship between the Washington metropolitan police and the U.S. Attorney for the District of Columbia (who prosecutes both federal and District crimes, in both the federal and Superior courts), a relationship obviously at work in this prosecution, the duty must reach the police department's homicide and Internal Affairs Division files.

The final question, then, is whether the duty to search was triggered here. In some cases, the duty to search flows straight from the nature of the files. In Auten, for example, once the court decided that the Brady duty could ever require prosecutors to search FBI and National Crime Information Center records, it felt no need to explain why that duty applied as to the government's key witness. Where the file's link to the case is less clear, the court must also consider whether there was enough of a prospect of exculpatory materials to warrant a search. Here we think there was.

We recognize that the probability is low that any of the police department files will yield information drawing Ms. Hoyle's credibility into question. She was a victim of a violent death, quite possibly either suicide or homicide, and we cannot embrace the idea that mendacity is materially more common among such victims than among the population at large. Taking this above the range of utter speculation, however, are the presence of a fellow officer, plus the possible use of her service revolver. The combination suggests some link between the death and her work, and thus (by extension) to some trouble connected with her work.

We think it highly relevant that defense counsel pinpointed files that can be searched without difficulty — police investigations into the death of Christie Hoyle. The defendant was not asking the U.S. Attorney's office to examine some sprawling mass of records; whoever is in charge of the investigation(s) into the Hoyle death should have been able very quickly to alert the assistant U.S. Attorney to any findings or clues as to character flaws or behavior that might tend to undermine Hoyle's testimony.

Studying the cases explicitly inquiring into the necessary degree of notice, we find this closer to the ones finding the duty triggered than to the ones finding the possibility of exculpatory materials purely speculative. The closest case is United States v. Deutsch, supra, in which the defendant argued that the government should have checked the personnel file of the key prosecution witness, a Post Office employee. There was a faint suggestion that he was in trouble at work; the employee had given what the court regarded as "evasive" answers on the subject, at first denying any trouble and then acknowledging his supervisor's objection to his personal appearance. Deutsch, 475 F.2d at 57-58.

Other cases sensibly warn against reliance on utter speculation. Thus in United States v. Navarro, 737 F.2d 625, 630-31 (7th Cir. 1984), the court considered a claim that the government should have checked an Immigration and Naturalization Service file on a government witness who testified against the defendant, theorizing that the file would contain information about a deal between the witness and the INS that might call the witness's testimony into doubt. Calling the defendant's theory a mere shot in the dark, id. at 632, the court held that "[m]ere speculation that a government file may contain Brady material" was not enough to require an in camera examination, id. at 631. See also United States v. Andrus, 775 F.2d 825, 843 (7th Cir. 1985) (rejecting claim to see the personal files of all the law enforcement witnesses "without even a hint" of impeaching material); United States v. Romo, 914 F.2d 889, 899 (7th Cir. 1990) (rejecting similar omnibus claim); cf. United States v. Balliviero, 708 F.2d 934, 943 (5th Cir. 1983) (similar). The suggestions here carry the case enough beyond pure speculation (though just barely).

As has proved true of the other aspects of Brady jurisprudence, no formula defining the scope of the duty to search can be expected to yield easily predicted results. Where, as here, however, there is an explicit request for an apparently very easy examination, and a non-trivial prospect that the examination, might yield material exculpatory information, we think the prosecution should make the inquiry. As the burden of the proposed examination rises, clearly the likelihood of a pay-off must also rise before the government can be put to the effort. Cf. Navarro, 737 F.2d at 631-32 (noting that in prior case rejecting attempted Brady application the problem was "compounded by the extremely large number of government files involved").

On remand, the district court should require the U.S. Attorney's office to do what it should have done earlier: to review (or have a suitably responsible person in the Metropolitan Police Department review) any homicide and any Internal Affairs Division files of the Department that may contain material exculpatory information (undermining the credibility of Hoyle, either as a general matter or with special reference to this case). This treatment — a directive to the prosecution to examine the files — accords with the Court's explanation in Pennsylvania v. Ritchie, 480 U.S. 39, 59, 107 S.Ct. 989, 1002, 94 L.Ed.2d 40 (1987), in which the Court found that a state statute preventing disclosure of files concerning child abuse maintained by the Pennsylvania Children and Youth Services could not trump the prosecution's Brady duties. The Court said: "In the typical case where a defendant makes only a general request for exculpatory material under Brady . . ., it is the State that decides which information must be disclosed. Unless defense counsel becomes aware that other exculpatory evidence was withheld and brings it to the court's attention, the prosecutor's decision on disclosure is final." Id. at 59, 107 S.Ct. at 1002 (footnote omitted). We require the prosecution to fulfill this Brady duty.

In so invoking Ritchie, we note that the decision's provision of a remedy presents a puzzle. The Ritchie Court explained that the decision whether information shall be disclosed normally rests in the first instance with the prosecutor, but then without explanation remanded for in camera review of the disputed files by the trial judge. We have found no case interpreting Ritchie in the circumstances present here — a need to choose who will search files as yet unsearched by the prosecutors themselves. But we note that post- Ritchie cases considering demands for disclosure of files to the defense appear to have followed the Court's words rather than its action, regarding prosecutorial review of possible Brady materials as normally sufficient, see, e.g., United States v. Presser, 844 F.2d 1275, 1281-82 (6th Cir. 1988); United States v. Clark, 957 F.2d 248, 251-52 (6th Cir. 1992); United States v. Pottorf, 769 F.Supp. 1176, 1179 (D.Kan. 1991), and reserving in camera inspection by the court for cases where the defense had become "aware that . . . exculpatory evidence was withheld", Ritchie 480 U.S. at 59, 107 S.Ct. at 1002. See, e.g., Miller v. Dugger, 820 F.2d 1135, 1136-37 (11th Cir. 1987) (involving witness's prior grand jury testimony, which appeared sure to conflict with either trial or deposition testimony as they conflicted); United States v. Kiszewski, 877 F.2d 210, 215-16 (2d Cir. 1989) ( in camera review ordered when, after prosecutor's denial that any Brady material existed, prosecutor revealed that FBI agent witness's file contained complaints that he was "on the take"); see also United States v. Pou, 953 F.2d 363, 367 (8th Cir. 1992) (attributing result in Kiszewski to fact that Brady request identified "[v]ery specific documents . . . with possible impeachment value"). While the Ritchie Court's actual decision to order in camera judicial review rather than prosecutorial review remains unexplained, we speculate that it may have arisen from a circumstance not present here — a need to balance a special interest in secrecy (apart from normal prosecutorial interests) against the interest in defense access to marginal exculpatory material. In any event, we think it right to take Ritchie at its word. Although the defendant has pinpointed specific files, he has not identified "exculpatory evidence [that the prosecution] withheld", Ritchie, 480 U.S. at 59, 107 S.Ct. at 1002, so the case calls for the usual prosecutorial rather than judicial examination.

* * * * * *

In addition to his Brady claim, the defendant raises three others, which we can reject briefly. First, he argues that the district court erred in ruling it could not depart downward from the range specified by the sentencing guidelines because of weaknesses in the government's case. See Transcript of Sentencing, August 16, 1991 at 7-8. We agree with the district judge. If the case has such vulnerabilities that a reasonable jury could not find guilt beyond a reasonable doubt, obviously the court must grant a motion for acquittal. But if the evidence withstands that test, we think that to permit departures based only on an individual assessment of the evidence would invite the sort of discrepancies the guidelines were intended to minimize.

Second, defendant says it was error to admit the original testimony of officer Hoyle. But the evidence satisfies the terms of Rule 804(b)(1), and we can not say the trial court abused the discretion that rule or Rule 403 permits. Ohio v. Roberts, 448 U.S. 56, 66, 73, 100 S.Ct. 2531, 2542-43, 65 L.Ed.2d 597 (1980); United States v. Miller, 904 F.2d 65, 67-68 (D.C. Cir. 1990).

Finally, Brooks argues that the district court erred in quashing a subpoena duces tecum under Fed.R.Crim.P. 17(c), seeking materials related to the investigation of the death of Officer Hoyle. This was not an abuse of discretion, as Brooks was improperly trying to use the subpoena as a discovery tool. United States v. Nixon, 418 U.S. 683, 698-700, 94 S.Ct. 3090, 3102-3104, 41 L.Ed.2d 1039 (1974).

* * * * * *

In order to permit the district judge to require full compliance with Brady, the case is

Remanded.