WILLIAMS, Senior Circuit Judge:

The district court dismissed an indictment against the five defendants on the ground that the evidence presented to the grand jury, and indeed the decision to prosecute two of the defendants, were tainted by statements of the defendants that for purposes of this appeal are conceded to have been compelled within the meaning of Garrity v. New Jersey, 385 U.S. 493, 87 S.Ct. 616, 17 L.Ed.2d 562 (1967). We reverse and remand as to four of the defendants; the government itself moved to dismiss the indictment against Nicholas Slatten, without prejudice to possible later re-indictment, and the district court's grant of the motion has taken Slatten out of the case for now. United States v. Slough, 677 F.Supp.2d 112, 115-116 & n. 2 (D.D.C.2009).

* * *

On September 16, 2007 a car bomb exploded near the Izdihar Compound in Baghdad, where a U.S. diplomat was conferring with Iraqi officials. American security officials ordered a team from Blackwater Worldwide to evacuate the diplomat to the Green Zone. See U.S. Department of State, U.S. Embassy Baghdad, (Draft) Use of Deadly Force Incident at Nisur Square—Baghdad: Preliminary Report and Findings, Sept. 23, 2007, at 2. Another Blackwater team, Raven 23, headed out of the Green Zone to block traffic at the Nisur Square traffic circle and thus assure the diplomat's safe passage back. (In fact, because a checkpoint had fortuitously been closed, the escort convoy never passed through Nisur Square.) Id. Raven 23 positioned its four vehicles on the south side of the Square and its members started gesturing to stop traffic. Shots were fired; the dispute over who fired at whom and when is the substantive crux of the criminal case underlying this appeal. When the shooting stopped, 14 Iraqi civilians were dead and 20 wounded. Slough, 677 F.Supp.2d at 116.

Within hours of the incident, the Department of State's Diplomatic Security Service ("DSS") conducted brief interviews with each of the 19 members of Raven 23. Id. at 117. Among the 19 were the five defendants in this case, Paul Slough, Nicholas Slatten, Evan Liberty, Dustin Heard and Donald Ball. [Redacted.]*

On September 18, 2007, two days after the incident, all Raven 23 members submitted sworn written statements to the State Department, using a form that included a guarantee that the statement and the information or evidence derived therefrom would not be used in a criminal proceeding against the signer. Slough, 677 F.Supp.2d at 118-19. The government conceded before the district court that under Garrity the September 18 statements must be treated as having been compelled; as to the September 16 statements, the district court so found and the government does not appeal that finding.1

The incident almost immediately became the focus of media attention in both the United States and Iraq. Some of the early articles, published within a few days of the incident, reported that the Blackwater team was attacked, and purported to quote from and otherwise rely on a State Department "incident report," presumably prepared at least in part on the basis of the interviews and statements. See, e.g., Adam Zagorin & Brian Bennett, Iraq Limits Blackwater's Operations, TIME, Sep. 17, 2007, http://www.time.com/time/world/ article/0,8599,1662586,00.html; Sabrina Tavernise, U.S. Contractor Banned by Iraq Over Shootings, N.Y. TIMES, Sept. 18, 2007, at A1. These very same articles, however, also cite Blackwater representatives as making the same assertion ([Redacted.]). See Tavernise. The articles also cite Iraqi officials' statements that Blackwater guards used excessive force. Joshua Partlow & Walter Pincus, Iraq Bans Security Contractor, WASH. POST, Sept. 18, 2007, at A1; Sinan Salaheddin, Iraq Plans Review of Foreign Security Firm Status, ASSOC. PRESS, Sept. 18, 2007.

The September 18 written statements were also leaked to the media. On September 28, 2007, ABC News reported that it had obtained all 19 of the September 18 sworn statements and quoted from some of them. See The Blotter, First Images of Controversial Blackwater Incident, ABC NEWS, Sept. 28, 2007, http://blogs.abcnews. com/theblotter/2007/09/exclusivefirst.html. Defendant Slough's statement was later posted online in its entirety, [Redacted.], and news reports by ABC News and the New York Times, among others, reproduced parts of defendants' and other team members' September 18 statements. [Redacted.]

The witnesses that the government relied on most heavily before the grand jury—Raven 23 members Adam Frost, Matthew Murphy and Mark Mealy—admitted to having read these news reports, and it soon became apparent that parts of their testimony may have been tainted by their exposure. In an effort to safeguard its case, the government decided to present a redacted case to a second grand jury, which returned an indictment against the defendants, finding that there was probable cause to believe that defendants committed (and attempted to commit) voluntary manslaughter and weapons violations. Slough, 677 F.Supp.2d at 127-28.

The defendants moved to dismiss the indictment as tainted. As required by Kastigar v. United States, 406 U.S. 441, 92 S.Ct. 1653, 32 L.Ed.2d 212 (1972), the district court held a hearing to determine the existence and extent of any taint. It found that exposure to defendants' statements had tainted much of the evidence presented to the second grand jury—the testimony of security guards Frost and Murphy and Iraqi witnesses and victims, Frost's written journal, the factual proffer and debriefing of Jeremy Ridgeway (a Raven 23 member who had been indicted and had pleaded guilty), and physical evidence recovered by DSS from the scene of the crime—and had also tainted the prosecutors' decision to indict defendants Heard and Ball. The district court thus dismissed the indictment as to all five defendants. The government now appeals. We review the district court's findings that the government used a defendant's immunized statement for clear error, United States v. North, 910 F.2d 843, 855 (D.C.Cir.1990) ("North I"), a standard that is met for any finding that was "induced by an erroneous view of the law," United States v. Kilroy, 27 F.3d 679, 687 (D.C.Cir.1994) (internal quotations omitted).

* * *

The Fifth Amendment bars the government from compelling self-incriminating testimony from individuals. If the government nevertheless decides to require an individual to testify, it must offer him immunity that puts him in "substantially the same position as if [he] had claimed his privilege." See Kastigar, 406 U.S. at 458-59, 92 S.Ct. 1653. In a later prosecution of the individual, the government cannot use his immunized testimony itself or any evidence that was tainted— substantively derived, "shaped, altered, or affected," North I, 910 F.2d at 863—by exposure to the immunized testimony. Nor can the government use it to develop investigatory leads, to focus an investigation on a witness, Kastigar, 406 U.S. at 460, 92 S.Ct. 1653, or to motivate another witness to give incriminating testimony. United States v. Rinaldi, 808 F.2d 1579, 1584 n. 7 (D.C.Cir.1987). In North itself, for example, after North gave his immunized testimony former National Security Advisor Robert C. McFarlane had requested a second hearing before special investigating committees "in order to respond" thereto, and we found error in the district court's having admitted McFarlane's trial testimony without having determined "what use—if any" he had made of North's. North I, 910 F.2d at 864. More generally, evidentiary content (potentially including a witness's whole testimony, as where his very availability was derived from or caused by immunized statements) will share the constitutional ban on use of the immunized statements. Kilroy, 27 F.3d at 687. Below we deal explicitly with situations where evidence's content or availability is derived from both immunized statements and independent factors.

In building a case against a defendant who received use immunity for his statements, the government must prove, by a preponderance of the evidence, that "all of the evidence it proposes to use was derived from legitimate independent sources." North I, 910 F.2d at 854 (quoting Kastigar, 406 U.S. at 461-62, 92 S.Ct. 1653, internal quotations omitted). As the district court observed, proof that a witness was "never exposed to immunized testimony" or that the investigators memorialized (or "canned") a witness's testimony before exposure, Slough, 677 F.Supp.2d at 132 (citing North I, 910 F.2d at 872), would obviously satisfy the requirement. But a failure by the government to make either showing does not end the district court's inquiry. North I requires the court to parse the evidence "witness-by-witness" and "if necessary, . . . line-by-line and item-by-item," 910 F.2d at 872, and to "separate the wheat of the witnesses' unspoiled memory from the chaff of [the] immunized testimony," id. at 862. This sifting is particularly important in cases where, as here, a witness was exposed to a defendant's immunized statement but testifies to facts not included in that statement.

In sifting the record as to taint of the evidence before the indicting grand jury, the district court made a number of systemic errors based on an erroneous legal analysis.

First, the district court erred by treating evidence, including the testimony of Frost, Murphy, Ridgeway and the Iraqi witnesses, and the Frost journal, as single lumps and excluding them in their entirety when at the most only some portion of the content was tainted—it made no effort to decide what parts of the testimony or the journal were free of taint. Prima facie, this error applies (for example) to all elements of testimony that do not overlap with the content of the immunized statements. North I requires the court to segregate tainted parts of the evidence from those parts that either could not have been tainted (because there is no overlap) or were shown to be untainted by a preponderance of the evidence. 910 F.2d at 872. Even in instances where there could be no possible claim that the immunized statements caused the witness to speak up (as in some variant of the McFarlane instance), the district court found that the government had failed to fulfill its burden; yet the court never identified what the government could have done besides pointing to the complete absence of overlap, or why it should have been required to show more.

The district court excluded, for example, all of the testimony of Frost and Murphy, finding that the two guards "had been thoroughly immersed" in defendants' immunized statements by virtue of having read news reports about the Nisur Square incident. Slough, 677 F.Supp.2d at 144. But Frost's and Murphy's grand jury testimony included specific recollections with no referent either in defendants' immunized statements or news reports derived therefrom. [Redacted.] As these elements of Frost's testimony have no antecedent in the immunized statements, they cannot be tainted (unless somehow the statements caused Frost's testimony in some subtler way). Similarly, [Redacted.]. The list goes on, as the government points out in its briefs. Appellant Br. at 62-64; Appellant Reply. Br. at 13 (describing parts of Frost's and Murphy's testimony not overlapping with the statements); Appellant Br. at 89-93 (describing parts of Jeremy Ridgeway's statements that could not have been tainted); Appellant Reply Br. at 26-27 (describing parts of Iraqi witness testimony that did not overlap with any immunized statements that appeared in news reports). We will not catalogue every instance of non-overlap here, but North's mandate that the district court parse the record "line-by-line" clearly requires such review, not only for the rest of Frost's and Murphy's testimony, but also for the Iraqi witnesses' testimony and Ridgeway's proffer and statement.

In United States v. North, 920 F.2d 940 (D.C.Cir.1990) ("North II"), we noted that the defendant bears the burden of laying "a firm `foundation' resting on more than `suspicion'" that proffered evidence was tainted by exposure to immunized testimony. Id. at 949 & n. 9 (quoting from Lawn v. United States, 355 U.S. 339, 348-49, 78 S.Ct. 311, 2 L.Ed.2d 321 (1958)). A witness's prior exposure to immunized statements can hardly be said to meet that burden as to completely non-overlapping points—defense counsel conceded as much at oral argument. See Oral Arg. Transcript 54, 56, 58. Of course, defendants may fill that gap by submitting additional evidence; again, the McFarlane episode in the North case may be a model, though it is worth noting that our disposition there left unresolved whether McFarlane's responding to North's testimony in fact constituted a forbidden "use" within the meaning of Kastigar. North I, 910 F.2d at 864.

Second (and closely related), the district court erred by failing to conduct a proper independent-source analysis as required by Kastigar, 406 U.S. at 460, 92 S.Ct. 1653, and Rinaldi, 808 F.2d at 1582. In particular, the district court erred by finding that any evidence responding to allegations that Raven 23 was attacked was tainted, even where no information specific to a particular defendant was included, and the supposedly tainting sources in fact encompassed multiple, equivalent assertions by non-defendants. Many of the news reports were based on the State Department spot and incident reports, which, in turn, were in part based on statements by all 19 guards ([Redacted.]), not simply the five defendants' immunized statements. See, e.g., Zagorin & Bennett (citing from the incident report that "the motorcade was engaged with small arms fire from several locations" and "returned fire").

Moreover, the State Department reports were not the only sources offered in the news stories to support the claim of [Redacted.]—the very same articles also cite Blackwater representatives as making the claim. There is no suggestion in the district court's opinion that Blackwater management learned the specifics of [Redacted.] from the State Department reports; [Redacted.] Where two independent sources of evidence, one tainted and one not, are possible antecedents of particular testimony, the tainted source's presence doesn't ipso facto establish taint. (Moreover, a witness's testimony need not have any exterior antecedent, i.e., any precursor other than the witness's perceptions of what happened.) Speaking of a government decision to pursue a line of investigation, for instance, the Second Circuit said, "[I]f it appears that that pursuit could have been motivated by both tainted and independent factors, the court must determine whether the government would have taken the same steps `entirely apart from the motivating effect of the immunized testimony.'" United States v. Nanni, 59 F.3d 1425, 1432 (2d Cir.1995) (citing United States v. Biaggi, 909 F.2d 662, 689 (2d Cir.1990); Biaggi in turn drew on broader legal sources, such as those governing claims of dismissal for exercise of First Amendment rights, Mt. Healthy City School District Board of Education v. Doyle, 429 U.S. 274, 97 S.Ct. 568, 50 L.Ed.2d 471 (1977)). The same principle—a goal of removing any net effect on either side—must apply to any circumstance (e.g., a segment of testimony, a witness's decision to speak up) claimed to be an effect of immunized testimony. Immunity, properly construed, "leaves the witness and the Federal Government in substantially the same position as if the witness had claimed his privilege." Kastigar, 406 U.S. at 458-59, 92 S.Ct. 1653 (emphasis added, internal quotations omitted) (quoting Murphy v. Waterfront Comm'n, 378 U.S. 52, 79, 84 S.Ct. 1594, 12 L.Ed.2d 678 (1964)). To preserve that symmetry, obviously courts cannot bar the government from use of evidence that it would have obtained in the absence of the immunized statement.

The district court also found that these early news reports tainted Frost's journal and his testimony when he addressed the claim of [Redacted.] (Indeed, when armed guards shoot a number of people in a crowd, it doesn't take Hercule Poirot to start wondering what the crowd was doing.) The district court erred by failing to consider whether Frost's testimony and journal, as well as other evidence challenging the story that [Redacted.], were more probably than not derived from sources other than defendants' immunized statements.

Third, the district court applied the wrong legal standard when it excluded Frost's journal and his testimony simply because the news reports based on some of the immunized statements were "a cause" for his writing it. Slough, 677 F.Supp.2d at 151. Defendants cite our language in Hylton to the effect that if Hylton's immunized statements "were a cause of [a key witness's] decision to plead and testify against Hylton, [the witness's] testimony was impermissible even if the government had prior knowledge of [the witness's] role." 294 F.3d at 134. But Hylton did not decide that any causal role was necessarily fatal. To do so would have been to reverse North I's (and Kastigar's) references to independent sources, as well as Kastigar's own explicit view that immunity, properly applied, "leaves the witness and the Federal Government in substantially the same position as if the witness had claimed his privilege." Kastigar, 406 U.S. at 458-59, 92 S.Ct. 1653 (emphasis added, internal quotations omitted). Finally, Hylton's entire focus was on explaining why a defendant was correctly asserting ineffective assistance of counsel, because of the latter's failure to make a Kastigar objection, not on resolving the nuances of multiple sources or causes.

Thus, only if the government on remand fails to establish by a preponderance that Frost would have written the journal or testified in the absence of exposure to defendants' immunized statements would use of the journal and testimony be barred under Kastigar. Of course, the defendants' communications transmitted to Frost via the media are relevant against the government in this analysis only to the extent that they actually added to the information flowing through from non-defendant sources.

This takes us to a fourth systemic error. To the extent that evidence tainted by the impact of one defendant's immunized statements may be found to have accounted for the indictment of that defendant, it does not follow that the indictment of any other defendant was tainted. The district court assumed the contrary. Slough, 677 F.Supp.2d at 166 & n. 66. Although the prosecution presented a single indictment against all five defendants, each defendant was charged individually and therefore the presence, extent and possible harmfulness of the taint must be assessed individually.

Defendants argue that the government proceeded on a joint liability theory that would render defendant-by-defendant taint assessment unsuitable. They point to a prosecutor's statement to the grand jury that it was "charging [the defendants] jointly, with each of these shootings because they're working together." Grand Jury Tr., Dec. 2, 2008, PM, at 10-11. But in context the reference does not suggest government adoption of the broad theory espied by defendants. The prosecutor had explained to the grand jury that for aider and abettor liability a defendant need not have fired a fatal or wounding shot. Even shots that hit no one could aid and abet directly harmful shots by making it "difficult for victims to run that direction to safety." Grand Jury Tr., Nov. 20, 2008, AM, at 16. But guilt was individual: a vote to indict required jurors to be satisfied "that there's probable cause that each of the people we've identified . . . did, in fact, shoot their weapons that day . . . [a]nd joined in this, in what happened."2 Id. In context it is plain that the snippet identified by defendants is no more than a reference back to the government's aider and abettor theory.

* * *

As we noted, the district court found the indictments of Heard and Ball independently and fatally tainted on the theory that their immunized statements motivated the prosecutor's decision to seek their indictment. Neither Kastigar nor North states that non-evidentiary uses of immunized statements are barred. Kastigar prohibits the use of immunized evidence as an "investigatory lead" to other derivative evidence that would then be used against the defendant. 406 U.S. at 460, 92 S.Ct. 1653. In North I, after a substantial review of the various circuits' decisions on the matter, we concluded: "Thus, even assuming without deciding that a prosecutor cannot make non-evidentiary use of immunized testimony, in the case before us the [Independent Counsel] did not do so." 910 F.2d at 860. We then went on (though under the heading "`Nonevidentiary' Use," id. at 856) to rule out the use of immunized testimony to refresh the memories of witnesses, id. at 860-63. That is, of course, an indirect evidentiary use.

Here, as the government does not challenge the factual finding on the decision to indict, we must assume its correctness and are thus forced to resolve the issue left unsettled in North I. In the absence of clear Supreme Court or D.C. Circuit precedent, North I turned to relevant decisions in other circuits for guidance and noted a circuit split: the Third and Eight Circuits suggested that Kastigar banned all non-evidentiary uses; the First, Second, Ninth and Eleventh Circuits found otherwise. North I, 910 F.2d at 857. Since North was decided, the Seventh Circuit has joined the latter group, holding that Kastigar is not concerned with "the exercise of prosecutorial discretion." United States v. Cozzi, 613 F.3d 725, 729 (7th Cir.2010) (citing a number of post-North I decisions; internal quotations omitted). In the end, at least as to decisions to indict, we join those circuits refusing to find such decisions vulnerable on the ground of links to immunized statements. The Eleventh Circuit observed in United States v. Byrd, 765 F.2d 1524, 1531 (11th Cir.1985), that such a rule would turn use immunity into transactional immunity. That is a bit of an overstatement; after all, prosecutors could, by construction of firewalls (along with the associated incremental personnel costs), assure that such decisions were made without risk of taint. But defendants' proposed rule clearly would entangle the court in what has hitherto normally been internal prosecutorial decision-making. And it would open a new field for courts' having to make complex causal judgments of the sort already required to assure clean evidence.

Continuing its discussion of non-evidentiary uses, Slough, 677 F.Supp.2d at 158-65, the district court also asserted that defendants' September 16 statements must have been useful to the prosecution and must have guided the government's investigation, id. at 163, but it never detailed what statements, independent of innocent sources, played exactly what role. We cannot uphold the judgment of dismissal to the extent that it rests on such vague propositions.

We further note that the district court lumped physical evidence collected by the DSS under the non-evidentiary-use rubric and found it to be tainted. Slough, 677 F.Supp.2d at 164-65. Insofar as physical evidence was presented to the grand jury, that classification is surely wrong—if the immunized statements led to discovery of physical evidence that was before the grand jury, it should be analyzed as an evidentiary use. As with uses purportedly leading to testimony or to the Frost journal, the district court's judgment was subject to the same errors reviewed above as to nonphysical evidence. (We note that especially as to physical evidence, but in principle as to all evidence alleged to be tainted, the independent sources that might undercut any taint would include actual facts on the ground, such as the locations of vehicles that were shot and bullet strikes, which would lead investigators to look for shell casings from any incoming fire, not to mention make inquiries of potential witnesses.)

If the excluded physical evidence was not presented to the grand jury, as the government's briefs suggest, Appellant Br. at 117 n. 43; Appellant Reply Br. at 38, then the district court's consideration and exclusion of that evidence appears premature. The district court acknowledged as much, noting that the "search [for physical evidence that it excluded] may have been highly relevant to the criminal case eventually brought against the defendants." Slough, 677 F.Supp.2d at 165. To the extent that the court ordered the Kastigar hearing simply to determine the status of the indictment, it would not properly reach the issue of possible use at trial. North II, 910 F.2d at 947-48.

Finally, although the district court disapproved of the prosecutor Kohl's explanation to the grand jury that some of defendants' statements were immunized, it did not find that mentioning the existence of immunized statements constituted a prohibited use under Kastigar. See Slough, 677 F.Supp.2d at 128. As the district court appears not to have relied on Kohl's explanation, we will not rule on the matter.

* * *

We find that the district court's findings depend on "an erroneous view of the law." Kilroy, 27 F.3d at 687. We thus vacate and remand the case for the court to determine, as to each defendant, what evidence —if any—the government presented against him that was tainted as to him, and, in the case of any such presentation, whether in light of the entire record the government had shown it to have been harmless beyond a reasonable doubt. North I, 910 F.2d at 873.

So ordered.