OPINION OF THE COURT
Sergeant Timothy J. Michael was convicted of wrongfully using amphetamines because three airmen said he did. They claimed that on 20 May, Michael participated when they used “crank” (amphetamine) at an apartment shared by two of these airmen, again later in a parked car, and yet again back at the apartment. Appellant complains that this evidence is legally and factually insufficient to establish that he used an illegal substance.
The Court of Military Appeals has clearly set out the standard for reviewing the sufficiency of the evidence:
The test for [legal sufficiency] is whether, considering the evidence in the light most favorable to the prosecution, a reasonable factfinder could have found all the essential elements beyond a reasonable doubt. Jackson v. Virginia, 443 U.S. 307, 319, 99 S.Ct. 2781, 2789, 61 L.Ed.2d 560 (1979). For factual sufficiency, the test is whether, after weighing the evidence in the record of trial and making allowances for not having personally observed the witnesses, the members of the Court of Military Review are themselves convinced of the accused’s guilt beyond a reasonable doubt.
United States v. Turner, 25 M.J. 324, 325 (C.M.A.1987).
I
The evidence is legally sufficient. There was ample testimony from these witnesses about the substance used and its effect on them. United States v. Day, 20 M.J. 213 (C.M.A.1985); United States v. Tyler, 17 M.J. 381 (C.M.A.1984). There is more than enough evidence to support a finding by the military judge of each essential element beyond a reasonable doubt.
II
Appellant’s claim of factual insufficiency is more troubling, particularly in light of his complaint of collusion among prosecution witnesses. With the evidence in substantial conflict, the prosecution added a significant unknown into the equation. On the morning of the trial, the prosecutor sent the second of his three witnesses to see the third who was then in jail as a result of his own court-martial, to refresh his recollection. We are called upon to decide whether this tactic tipped the scales in the appellant’s favor.
It is always disturbing when a non-commissioned officer with an otherwise excellent record is marked with a conviction, based solely on the testimony of admitted habitual drug users who have everything to gain by “helping” the investigating agents and the prosecution.1 This is particularly so when the accused presents strong evidence in rebuttal. With the evidence so evenly balanced, we cannot tolerate any suggestion that the prosecution had its finger on the scale.
In the case at bar, the principal witness against Sergeant Michael, an Airman Heber, was just such an admitted habitual drug user. Heber’s drug involvement became so severe that his wife turned him in to the Office of Special Investigations (OSI). The other two witnesses were Airman Giovanni, who had been Heber’s roommate, and Airman Wise, both of whom, along with appellant, were implicated by Heber. Giovanni received a bad-conduct discharge, 3 months confinement, reduction and forfeitures for his drug use. No charges had been brought against either *902Heber or Wise by the time they testified here. All clearly had something to gain by helping the prosecution.
The tales told by these three differ among themselves in minor detail and are directly contradicted in some important particulars by four defense witnesses. Two of these witnesses were Sergeant Michael and his wife. The other two were apparently totally unrelated to the appellant, and had no apparent favor to curry.
The reason the testimony of Wise and Giovanni differed at trial only in minor detail, appellant argues, was because the prosecutor sent Wise to Giovanni’s jail cell the morning of the trial to help him “refresh his recollection.” He points out that there were significant differences between Wise’s testimony at the pretrial investigation conducted pursuant to Article 32, UCMJ and his testimony at trial.2 Giovanni did not testify at the pretrial investigation, so any similar comparison with him is not possible.3
Giovanni explained his collaboration with Wise as follows:
RECROSS EXAMINATION
Questions by the Civilian Defense Counsel:
Q: You testified earlier that you had talked to Wise today—
TC: Objection. Beyond the scope, your honor.
MJ: Overruled.
—and you indicated you wanted to get some things straight, is that correct?
A: (Amn. Giovanni) Yes, ma’am.
Q: So you all were comparing notes?
A: Basically, ma’am.
Q: To make sure that your testimony was in line for court today?
A: I just wanted to see if what I remembered was correct.
It was uncontroverted that it was the prosecutor who arranged for Amn Wise to visit Giovanni. He testified: “Captain [K] told me this morning on the phone, to go over to confinement and talk to him.”
At this point in the trial, appellant moved for a mistrial. The military judge’s denial of that motion is also asserted as error. It is hornbook law that a mistrial is one of the most extreme remedies. As the Court of Military Appeals recently held:
Declaration of a mistrial is a drastic remedy, and such relief will be granted only to prevent a manifest injustice against the accused. See United States v. Pastor, 8 M.J. 280, 281 (C.M.A.1980). The decision to grant a mistrial rests within the military judge’s discretion, and we will not reverse his determination absent clear evidence of abuse of discretion.
United States v. Rushatz, 31 M.J. 450, 456 (C.M.A.1990). We are convinced that the judge did not abuse his discretion.
We must still decide whether trial counsel nonetheless committed error by sending one critical witness to “compare notes” with another critical witness the morning of the trial. We think not. While we certainly do not endorse the practice and believe it to have been an error in judgement, we conclude that it did not constitute legal error.
MiLR.Evid. 615 empowers the judge to exclude witnesses from the courtroom to keep them from hearing the testimony of other witnesses. We do not believe this rule should be extended to prohibit communications among witnesses outside the courtroom, before the trial even starts. The United States District Court for the Eastern District of Kentucky, while construing the parallel provision in the Federal Rules of Evidence,4 was asked to find that *903the trial judge had erred in not granting a request to prohibit out of the courtroom conversations among witnesses. The court found such an order to be generally “unrealistic.” (They did acknowledge, however, that in some cases “where the result turns on fine nuances in the testimony, it may be necessary ... to enforce the more stringent restrictions.”) United States v. Scharstein, 531 F.Supp. 460, 464 (E.D.Ky. 1982).
Were it not for our Article 66 (10 U.S.C. § 866) mandate to find the verdict to be also “correct in fact,” we would be finished with this inquiry. We cannot, however, overlook the impact of this “comparing [of] notes” upon our evaluation of the factual sufficiency of this evidence.
We have here the testimony of a principal prosecution witness, corroborated by the two who “cooperated.” As noted above, all had much to gain by their cooperation. On the other hand, we have the appellant’s sworn denial, buttressed in some details by his wife. These two also have much to gain. Two witnesses however, who would seem to have no stake in the outcome, dispute key points of the story as told by the three prosecution witnesses. Would the verdict have been different had there been serious discrepancies in the testimony of Wise and Giovanni? — if Wise’s testimony had only gone as far as his testimony in the Article 32 Investigation, and he hadn’t remembered the additional details he recounted on the stand (which conformed his story with that of Heber)?
Normally, we rely on cross-examination to ferret out and highlight such differences and discrepancies in trial testimony. Allowing these two witnesses to confer just hours before their testimony to “compare notes,” in our opinion robbed the defense of their ability to effectively test their recollections with cross-examination. As the Supreme Court of New Mexico observed while construing a state rule which prohibits witnesses from discussing their testimony outside the courtroom:
The purpose of the rule is to maintain the effectiveness of cross-examination by preventing witnesses from using knowledge of prior testimony to recast their own testimony, perhaps even with honorable motives, to conform to what other witnesses have said or to otherwise anticipate possible cross-examination.
United States v. Reynolds, 111 N.M. 263, 804 P.2d 1082, 1088 (App.1990). See United States v. Croom, 24 M.J. 373, 375 (C.M.A.1987).
The trial judge handled this situation well, and we recognize that he heard and observed the demeanor of the witnesses as well as the manner in which they testified. Article 66(c), UCMJ. The cooperation of Wise and Giovanni was thoroughly explored and taken into consideration. In the exercise of our independent responsibilities, however, we are just not able to come to the same conclusion.
Although it was Giovanni’s memory that the prosecution sought to refresh, it was Wise’s testimony that varied substantially at trial from his sworn pretrial statement which is part of the Article 32 investigation report.5 We do not know the reason for this variance, but this whole cooperative enterprise involving the prosecutor and these two critical prosecution witnesses creates in our mind a substantial doubt as to their credibility. With the scales already very close to even, any such doubt persuades us to resolve this case in favor of the appellant, so we reverse. Article 66(c), UCMJ. Because we do not believe that this doubt can now ever be resolved, the charge and specification are
DISMISSED.
Senior Judge PRATT and Judge McLAUTHLIN concur.