Defendant Daniel Leon Abercrombie was arrested for driving under the influence of alcohol. A breathalyzer said his blood contained 0.08 percent alcohol - the lowest amount that is illegal per se. But at trial, the government conceded that the breathalyzer has a margin of error. And according to the government's witness, Mr. Abercrombie's blood might have contained just 0.077 percent alcohol - an amount that would have required the jury to acquit him.
Knowing that Mr. Abercrombie was relying on a margin-of-error defense, the government introduced evidence that supported the breath test results. The arresting officer testified that, during three field sobriety tests, Mr. Abercrombie exhibited various "indicia of intoxication." In its rebuttal argument, the government asked the jury to use those "indicia" to find that the breath test results did not exaggerate Mr. Abercrombie's true blood-alcohol content.
Treating the margin of error as irrelevant, the magistrate judge barred an expert from testifying that Mr. Abercrombie has disabilities and that physical manifestations of his disabilities can mimic intoxication. That was error (whether reviewed de novo or for abuse of discretion), and the court is not convinced the error was harmless beyond a reasonable doubt. As a result, Mr. Abercrombie's conviction for driving under the influence is vacated.
I. JURISDICTION
This case was tried by a magistrate judge. 18 U.S.C. § 3401 ; Fed. R. Crim. P. 58. Mr. Abercrombie filed a timely notice of appeal (Doc. # 45) from the judgment of conviction and sentence entered by the magistrate judge (Doc. # 43). This court has appellate jurisdiction. 18 U.S.C. § 3402 ; Fed. R. Crim. P. 58(g)(2).
II. BACKGROUND
A. Arrest on Fort Rucker
One afternoon in January 2017, Daniel Abercrombie drove to the gate of Fort Rucker, a military base in Alabama, with his father in the passenger seat.1 He stopped at the security checkpoint and handed the guard a visitor's pass and a non-driver's identification card. (Doc. # 46, at 33-36.) According to the guard, Mr. Abercrombie spoke slowly, softly, and with a strange "rhythm," and "something was off [about him]." But he did not slur his words, his eyes were not bloodshot, and there were no problems with his driving. In short, the guard saw no signs that he was drunk. (Doc. # 46, at 42-44, 46.) Still, Mr. Abercrombie did not present a valid driver's license, so he was told to pull into an inspection lane. (Doc. # 46, at 37, 44.)
*1206Sergeant Carlos Whitehead, a civilian police officer for the Department of the Army, went to the inspection lane. He stood on the passenger side of the car, next to Mr. Abercrombie's "highly inebriated" father, and smelled alcohol. (Doc. # 46, at 67-68.) Sergeant Whitehead ordered Mr. Abercrombie to step out of the car. An officer searched the car (with consent) and found a beer can on the back seat and a mostly empty whisky bottle on the front floorboard. Sergeant Whitehead also smelled alcohol on Mr. Abercrombie's breath, so he decided to perform three field sobriety tests. (Doc. # 46, at 68-71.)
The first test was the "one-leg stand." This "divided attention test" tested Mr. Abercrombie's "ability to ... physically perform a task and mentally perform a task." (Doc. # 46, at 72.) Sergeant Whitehead first demonstrated the test to Mr. Abercrombie: "I'm going to point the right foot, I'm going to point the toe outward, I'm going to watch the toe, I'm going to count one-thousand one, one-thousand two, one-thousand three, and so on, until I tell you to stop." (Exhibit 1.) When it was his turn, Mr. Abercrombie did not sway or put his foot down for twenty seconds, but Sergeant Whitehead concluded that he displayed two out of four possible "indicia of intoxication." One indicia was that he "used his arms for balance." That is, his left arm was near his side, and his right arm was elevated. (Doc. # 46, at 74, 103-05.) The other indicia was that he "failed to count out loud." But a video of the test reveals that Mr. Abercrombie might have used his fingers to count, and Sergeant Whitehead did not specifically tell him to count out loud. (Exhibit 1; Doc. # 46, at 74, 103-05.)
The second test looked for horizontal gaze nystagmus (involuntary jerking) in Mr. Abercrombie's eyes. Sergeant Whitehead observed nystagmus during the test, so he counted six out of six possible indicia of intoxication. (Doc. # 46, at 79-82.) But Sergeant Whitehead hurried this "HGN test": It was supposed to take 83 to 124 seconds, and he took just 43 seconds to administer it. (Doc. # 46, at 107-16.)
For the third test, Sergeant Whitehead directed Mr. Abercrombie to walk nine steps heel-to-toe in a straight line, turn around, and walk back. Out of eight possible indicia of intoxication this test could reveal, Sergeant Whitehead determined that Mr. Abercrombie exhibited six: He could not keep his balance, "stepped off the line," walked twenty-two steps, made an "incorrect turn," "missed [a] heel to toe," and raised his arms. (Doc. # 46, at 86.) Mr. Abercrombie also started the test before he was told to do so. But as Sergeant Whitehead later admitted, he did not tell Mr. Abercrombie to keep his arms by his side. (Doc. # 46, at 123.)
After concluding that Mr. Abercrombie displayed fourteen out of eighteen indicia of intoxication (plus "glazed eyes," "slurred speech," and "somewhat of a lethargic reaction"), Sergeant Whitehead arrested Mr. Abercrombie and took him to the police station. (Doc. # 46, at 88-89.) There, Sergeant Whitehead used a Draeger Alcotest 7100MKIII-C machine to test Mr. Abercrombie's breath for alcohol. That machine automatically performs two "calibration checks" whenever an officer administers a test. The first check uses a lab sample known to be 0.020 percent alcohol; the machine accepts any reading between 0.015 and 0.025 percent. The second calibration check uses a lab sample known to be 0.080 percent alcohol; the machine accepts any reading between 0.076 and 0.084 percent. See Ala. Admin. Code r. 370-1-1-.01(5)(a)(2) (2017). So according to Sergeant Whitehead, a person whose blood-alcohol content is 0.077 percent might erroneously *1207test above the legal limit. (Doc. # 46, at 131-32, 135-138.)
Mr. Abercrombie's test result was "0.08 g/210L," or 0.08 percent. (Doc. # 58-1, at 2.) Ala. Code § 32-5A-194(a)(5) ("Percent by weight of alcohol in the blood shall be based upon ... grams of alcohol per 210 liters of breath."). He was charged with driving a vehicle while there was "0.08 percent or more by weight of alcohol in his ... blood" in violation of Alabama Code § 32-5A-191(a)(1).2 That statute applies on Fort Rucker under the Assimilative Crimes Act. 18 U.S.C. § 13 ; see United States v. Tyson , 829 F. Supp. 368, 369 (M.D. Ala. 1993). It is an "illegal per se law" that applies "without reference to the effect that alcohol may have on the driver." Curren v. State , 620 So. 2d 739, 740 (Ala. 1993) (cleaned up).
B. Jury Trial
Mr. Abercrombie went to trial in March 2018. A week before trial, he gave notice that he intended to call Dr. David Ghostley, a clinical psychologist, as an expert witness. (Doc. # 20.) Dr. Ghostley had evaluated Mr. Abercrombie, and he reported that Mr. Abercrombie has several psychological conditions: Unspecified Pervasive Developmental Disorder, Generalized Anxiety Disorder, Unspecified Learning Disabilities, Adaptive Behavior Deficits, and Borderline Intellectual Functioning. Dr. Ghostley intended to testify about "the physical manifestations of" those diagnoses. (Doc. # 20, at 1.) To be more specific, he would have testified that those disabilities can be confused with intoxication. (Doc. # 46, at 7-8.)
The government never challenged whether Dr. Ghostley was an expert under Federal Rule of Evidence 702. But minutes before trial, the government questioned the relevance and materiality of his testimony. (Doc. # 46, at 7.) Defense counsel explained that Mr. Abercrombie was making a margin-of-error defense and that he anticipated the government would "elicit testimony as to the indicia of intoxication to corroborate the breath test results." (Doc. # 46, at 7.) Dr. Ghostley's testimony would have cut against that corroboration, supporting the margin-of-error defense. The judge then opined that performance on a field test is relevant only to establish that an officer had "reasonable suspicion" to administer a breath test. (Doc. # 46, at 8.) The judge stated:
No. The .08 - as far as the breathalyzer, if they do it right, it either gives a result or it doesn't. If it's [a] .08 and that's what the limit is, that's the limit. It doesn't matter at that point. It's what it is. But the indicia - the other indicia, they're offering that for the purposes of whether or not they can get to the point of giving him the breathalyzer test. So I don't - Ghostley's testimony on that I don't find - I find that's going to be more confusing, and it's not going to be - it's not going to help as far as the issue at all of whether or not he was intoxicated. It's not like it's close and it's right on the line. It doesn't matter whether it's on the line or not. If that's what it is and you hit that on the breathalyzer, that's what it is.
(Doc. # 46, at 9.) The judge thus implicitly based his ruling on Rules 402 and 403 of the Federal Rules of Evidence. Defense *1208counsel's follow-up request to keep the government from introducing evidence of the field tests was denied. (Doc. # 46, at 10.)
Defense counsel's opening statement previewed its margin-of-error defense and alluded to Mr. Abercrombie's disabilities. (Doc. # 46, at 30.) But the judge soon told counsel: "Once that test is given, you cannot argue that the test ... is not proof one way or another. If they lay the foundation, you can't argue, well, we don't think he's drunk because of other things." (Doc. # 46, at 54.) During trial, defense counsel advanced the margin-of-error defense by challenging whether Mr. Abercrombie failed the field tests. (Doc. # 46, at 95-140.) But the judge once again asserted that field tests simply give a reason for more tests and are irrelevant to proving intoxication. (Doc. # 46, at 132-33.) When defense counsel again tried to explain his margin-of-error defense, the judge stated: "Whether there's a margin of error or not, you knew the thing was calibrated, did you not?" (Doc. # 46, at 133.)
Mr. Abercrombie testified in his own defense. He admitted that he drank beer and wine earlier that day. (Doc. # 46, at 144.) He also testified that he took "special ed" classes, was "nervous" during the field tests, and had some trouble keeping up with Sergeant Whitehead's instructions during the field tests (though it "wasn't that bad"). (Doc. # 46, at 141, 145-46.)
Finally, in its closing argument, the government conceded that Mr. Abercrombie's blood-alcohol content could have been below the legal limit. (Doc. # 46, at 154 ("Is it possible that it was 0.076? Yes.").) Yet it stood by the test results since "the machine was working properly" and since Mr. Abercrombie drank alcohol that day. (Doc. # 46, at 154-55.) On rebuttal, the government drove the point home, asking the jury to decide between two scenarios. In the first scenario:
Mr. Abercrombie was one of the single most unlucky people on earth that day in that everything went against him. He just happened to have drank some that day, and these tests, the field sobriety tests that the officer has done at least a hundred of, he just misread them all. Didn't do it right. The machine was acting up. Everything just went wrong and just a series of unfortunate coincidences that led Mr. Abercrombie into this courtroom today.
(Doc. # 46, at 161.) In the second, "more logical, more reasonable situation":
He told us he had been drinking that day. When he pulls up to the gate, there's a bottle of Jack Daniel's and an open Ice House beer that he says wasn't drinking. You can take him at his word if you want. That he, in fact, was intoxicated enough that of 18 total potential clues the officer could look for in a field sobriety test, he nailed 14 of them. 14 of the 18. Every possible clue on the HGN. To be honest with you, there's some that I thought the officer missed. He didn't check the box that said he started too soon on the walk and turn. But if you watch the video, he did. He lined up and started walking, and the officer told him, I didn't tell you to go yet. So I would argue that he hit seven of the eight clues on the walk and turn. Now, they may say that he wasn't stumbling. I watched that video. He was stumbling. He nearly fell down. So the more reasonable thing is that he was so intoxicated that he nailed 14, maybe 15 of the clues on the HGN and that he had consumed enough alcohol that when he went and gave a breath sample, his lowest sample was .08.
(Doc. # 46, at 161-62.) Thus, from the beginning to the end of the trial, the government *1209relied on Mr. Abercrombie's physical actions to bolster the breath test results.
C. Conviction and Appeal
The jury convicted Mr. Abercrombie, and he was sentenced to two years' probation. (Doc. # 43, at 1-2; Doc. # 46, at 174.) Mr. Abercrombie filed a timely notice of appeal. (Doc. # 45.) He challenges his conviction under the Fifth and Sixth Amendments, both of which guarantee the right to present a complete defense. (Doc. # 53, at 5.)3 The government did not file a timely response brief. (See Docs. # 54, 55, 56.) When it did file a brief, it argued that Mr. Abercrombie's demeanor was "irrelevant." (Doc. # 58, at 11.) This appeal is now ripe.
III. STANDARDS OF REVIEW
Evidentiary rulings are reviewed "for clear abuse of discretion." United States v. Machado , 886 F.3d 1070, 1085 n.14 (11th Cir. 2018). Even if an abuse occurred, the ruling will stand unless it caused "a substantial prejudicial effect." Id. But if a ruling "implicates a constitutional question," that question is reviewed de novo. Id.
IV. DISCUSSION
"Implicit in a criminal defendant's constitutional rights under the Fifth and Sixth Amendments is the right to present evidence in his or her favor." Machado , 886 F.3d at 1085.4 And, generally speaking, a court "should not prohibit a defendant from presenting a theory of defense to the jury." United States v. Thompson , 25 F.3d 1558, 1564 (11th Cir. 1994). At the same time, however, a defendant "does not have an unfettered right to offer testimony that is incompetent, privileged, or otherwise inadmissible under standard rules of evidence." Taylor v. Illinois , 484 U.S. 400, 410, 108 S.Ct. 646, 98 L.Ed.2d 798 (1988) ; see also Holmes v. South Carolina , 547 U.S. 319, 326, 126 S.Ct. 1727, 164 L.Ed.2d 503 (2006).
The Eleventh Circuit noted this tension in United States v. Hurn , 368 F.3d 1359 (11th Cir. 2004), and summarized four categories of evidence that courts must let defendants introduce:
First , a defendant must generally be permitted to introduce evidence directly pertaining to any of the actual elements of the charged offense or an affirmative defense. Second , a defendant must generally be permitted to introduce evidence pertaining to collateral matters that, through a reasonable chain of inferences, could make the existence of one or more of the elements of the charged offense or an affirmative defense more or less certain. Third , a defendant generally has the right to introduce evidence that is not itself tied to any of the elements of a crime or affirmative defense, but that could have a substantial impact on the credibility of an important government witness. Finally , a defendant must generally be permitted to introduce evidence that, while not directly or indirectly relevant to any of the elements of the charged events, nevertheless tends to place the story presented by the prosecution in a significantly different light, such that a *1210reasonable jury might receive it differently.
Id. at 1363 (footnote omitted; emphasis added). Hurn used the adverb "generally" to note that "otherwise relevant evidence may sometimes validly be excluded under the Rules of Evidence." Id. at 1363 n.2. "Nevertheless," Hurn continued, "the fact that a particular rule of evidence requires the exclusion of certain evidence is not dispositive." Id. In some situations, "particular applications of a generally valid rule may unconstitutionally deny a defendant his rights." Id.
Dr. Ghostley's testimony should have been admitted under both the second and the fourth Hurn categories. At trial, the government conceded that the margin-of-error defense is valid, and it relied on Mr. Abercrombie's demeanor to bolster the breath test results. But if physical manifestations of Mr. Abercrombie's disabilities can be confused with intoxication, that would tend to make it less likely that he was drunk. That would, in turn, help the margin-of-error defense by making it harder to corroborate the breath test results. It would have also put the government's evidence about the field tests in a different light. Thus, it was error to exclude Dr. Ghostley's testimony. See United States v. Pemberton , 479 F. App'x 264, 268 (11th Cir. 2012) (per curiam) (holding testimony should have been admitted under both the second and the fourth Hurn categories). Based on the evidence as a whole, the court is not convinced that the error was harmless.
A. Dr. Ghostley's testimony was relevant to a collateral matter that affected an element of the offense.
On appeal, the government argues that any evidence about Mr. Abercrombie's demeanor is irrelevant: "The issue that the jury was there to decide was whether Abercrombie had 0.08% or more by weight of alcohol in his blood. Dr. Ghostley's testimony would give the jury no insight whatsoever into this question." (Doc. # 58, at 11.) The government's premise - that physical intoxication is not an element of the offense - is correct. But its conclusion - that evidence of physical intoxication is therefore irrelevant - does not follow. The government recognized this at trial.
Under Alabama Code § 32-5A-191(a)(1), no one may "drive or be in actual physical control of any vehicle while ... [t]here is 0.08 percent or more by weight of alcohol in his or her blood." Section 191(a)(1) is an "illegal per se law," meaning that it applies "without reference to the effect that alcohol may have on the driver." Curren , 620 So. 2d at 740 (cleaned up). The only two elements of the crime are (1) driving, or actual physical control of a vehicle; and (2) a blood-alcohol concentration of 0.080 percent or greater. See id. The government need not prove that the driver was "drunk" or physically "under the influence" of alcohol. See id. at 742.
Yet physical indications of intoxication are still relevant in § 191(a)(1) cases. If, for example, a driver fails a field sobriety test, an officer may be justified in giving that driver a breath test. See Meininger v. State , 704 So. 2d 1034, 1039 (Ala. 1997) ; Babers v. City of Tallassee , 152 F. Supp. 2d 1298, 1306 (M.D. Ala. 2001) ("The results of the field sobriety tests support a finding of probable cause."). And whether the driver is physically "under the influence" is a collateral matter that can reinforce or undermine breath test results.
As the Alabama Supreme Court explained in Curren v. State , a defendant charged with violating § 191(a)(1) "can offer to rebut the ... evidence that his blood alcohol content was [0.08 percent] when he *1211was found driving." 620 So. 2d at 742.5 The defendant may, for example, argue that a breath test is "incorrect since [he] did not exhibit physical signs of intoxication consistent with having [0.08 percent] blood alcohol concentration or higher." Id. at 743 (cleaned up). Or he may show that the testing device "is inaccurate." Id. (citation omitted). Or he may show "that he had not consumed enough alcohol in the relevant time to reach the level indicated by the chemical test results." Id. (citation omitted). Simply put, the defendant is not always stuck with his breath test results - those results are evidence of "the weight of alcohol in his ... blood. " Ala. Code § 32-5A-191(a)(1) (emphasis added). If the results are inaccurate, the defendant might be not guilty.
At trial, the government conceded that Mr. Abercrombie's true blood-alcohol concentration could have been below the legal limit. (Doc. # 46, at 154.) Sergeant Whitehead testified that Mr. Abercrombie's blood-alcohol content could have been 0.077 percent. (Doc. # 46, at 138.) The government then used Mr. Abercrombie's demeanor (his performance on the field tests) to argue that the breath test results were accurate. (Doc. # 46, at 160-62.) See State v. Gai , 366 Mont. 408, 288 P.3d 164, 169 (2012) (affirming a per se DUI conviction despite a margin-of-error defense because the driver had physical signs of intoxication); State v. Weeks , 634 A.2d 1275, 1277 (Me. 1993) (same).
But what was good for the goose should have been good for the gander. Given the margin of error, it was proper for Mr. Abercrombie to argue that the breath test results were inflated. He could support that argument by showing he "did not exhibit physical signs of intoxication consistent with having" a 0.08 percent blood-alcohol content. Curren , 620 So. 2d at 743 (cleaned up). Just as evidence about the field tests could support the breath test results, evidence about the physical manifestations of his disabilities could have undermined the breath test results. That is, if purported "indicia of intoxication" were actually indicia of a disability, it would have been easier to argue that Mr. Abercrombie should benefit from the margin of error.
B. Because it closely pertained to a relevant collateral matter, Dr. Ghostley's testimony should have been admitted.
A defendant has "the right to introduce evidence ... that makes the existence or non-existence of some collateral matter somewhat more or less likely, where that collateral matter bears a sufficiently close relationship to an element of the offense." Hurn , 368 F.3d at 1364. Although the chain of inferences cannot be "too long, dubious, or attenuated," id. at 1366, the right applies with special force "where the government attempts to use [the] collateral matter as the basis for securing a conviction," id. at 1365.
Given the margin of error in the breath test results, whether Mr. Abercrombie was physically intoxicated was relevant to the weight of alcohol in his blood. It tended to make it more or less likely that the breath test results were accurate. And since the government used this collateral matter to secure a conviction, Mr. Abercrombie had the right to introduce his own evidence. To see why, consider United States v. Word , 129 F.3d 1209 (11th Cir. 1997), and United States v. Lankford , 955 F.2d 1545 (11th Cir. 1992).
*1212In Word , a defendant named Dawn Dailey was charged with securities fraud. 129 F.3d at 1211. "At trial, the government argued that the jury could infer that Dailey willfully and knowingly committed fraud because she was romantically involved with the man responsible for the fraudulent scheme's creation - Calvin Word." Id. The government specifically relied on its romantic-involvement theory in its closing argument. But Dailey was not allowed to present evidence to counter the government's argument. Id. at 1212. Instead, the district court barred her from testifying "that Word physically and emotionally abused her and ... did not share much information with her." Id. The Eleventh Circuit held that excluding testimony about Dailey's relationship with Word was reversible error - particularly given the government's trial strategy. Id. at 1213 ("Because the government's strategy made the romantic relationship significant, that exclusion amounted to an abuse of discretion and a reversible error.").
In Lankford , the district court erred when it excluded the defendant's qualified expert. 955 F.2d at 1553. The defendant, a former political candidate, did not report a check he received during his campaign, and he was charged with willfully filing a false federal income tax return. Id. at 1547. The defendant testified that he thought the check was a gift that he did not need to report (that he did not "willfully" violate the law). Id. at 1551. The government elicited expert testimony that the defendant should have reported the check. Id. Yet the trial court excluded the defendant's expert, who would have testified that the defendant made a reasonable mistake. Id. The Eleventh Circuit held it was wrong to exclude the defendant's expert, an error "compounded" by the fact that the government was allowed to call its expert. Id. "It is an abuse of discretion," the appellate court explained, "to exclude the otherwise admissible opinion of a party's expert on a critical issue, while allowing the opinion of his adversary's expert on the same issue." Id. (citation omitted).
Just like it was error to exclude evidence in Word and Lankford , it was error to exclude Dr. Ghostley's testimony. The government introduced evidence that Mr. Abercrombie was physically intoxicated. (Doc. # 46, at 71-89.) It introduced that evidence through Sergeant Whitehead, and it took pains to establish that Sergeant Whitehead is trained to detect intoxication. (Doc. # 46, at 57-64.) That the evidence supported a finding of probable cause to test is a given. But the government went further, using the field tests to argue that Mr. Abercrombie was intoxicated. (Doc. # 46, at 160-62.) Dr. Ghostley would have testified that Mr. Abercrombie's disabilities can be confused with intoxication, which would have responded to the government's evidence and helped undermine the breath test results. That bears a close, logical relationship to an element of the offense, so it should have been admitted.
C. Dr. Ghostley's testimony also would have put Mr. Abercrombie's actions in a different light.
A defendant also has the right to "complete the picture" if the selective presentation of evidence "might color a jury's assessment of the material facts of the case." Hurn , 368 F.3d at 1367. If, for example, the government's evidence "tends" to make "entirely legitimate, normal, or accepted acts appear unusual or suspicious," the defendant "has the right to introduce additional evidence to dispel this unjustified taint." Id. This rule applies to Dr. Ghostley's testimony. Once again, consider two other cases.
In United States v. Todd , 108 F.3d 1329 (11th Cir. 1997), a businessman was *1213charged with embezzling funds from his employees' pension plan. Id. at 1330. His defense was that his employees were "compensated and treated so well" that they authorized him to use pension plan funds to keep the business from going under. Id. at 1331. The district court excluded evidence that employees wanted the defendant to save the business, but the government still presented evidence that employees did not authorize the use of pension plan funds. During rebuttal, the government argued that "it simply defies logic to think that the employees would authorize this." Id. (emphasis omitted). The Eleventh Circuit reversed on appeal. Especially given the government's argument on rebuttal, that court reasoned, the defendant should have been allowed to present evidence about his employees. See id. at 1334.
In United States v. Sheffield , 992 F.2d 1164, 1166 (11th Cir. 1993), a civilian who worked on a military base was charged with embezzling government property by ordering his subordinates to use government time and materials to make fishing gear. But the defendant was excluded from introducing evidence that the base had a custom of making fishing gear retirement gifts. Id. at 1167. The Eleventh Circuit reversed, explaining that evidence of that custom would have put the defendant's actions in a different light and would have been relevant to his mental state. See id. at 1170 ("Without this evidence of the gift making custom, production of fishing lure molds on a [military base] must have seemed to the jury like the oddball project of a renegade fisherman.").
Mr. Abercrombie's demeanor was relevant information. A law-enforcement officer testified that Mr. Abercrombie's demeanor was evidence of intoxication. The government relied on that during rebuttal. See Todd , 108 F.3d at 1334 ("Perhaps, the most convincing reason that the evidence is relevant are the comments made by the government in its rebuttal."). But though it went to his margin-of-error defense, Mr. Abercrombie could not put his demeanor in a different light. That was error.
D. It was an abuse of discretion to exclude Dr. Ghostley's testimony under Rule 403.
Dr. Ghostley's testimony was excluded because it would mislead the jury (Doc. # 46, at 9), an implicit reference to Federal Rule of Evidence 403. See Fed. R. Evid. 403 ("The court may exclude relevant evidence if its probative value is substantially outweighed by a danger of one or more of the following: unfair prejudice, confusing the issues, misleading the jury, undue delay, wasting time, or needlessly presenting cumulative evidence."). When it comes to expert testimony, trial judges must be careful to consider Rule 403. See United States v. Frazier , 387 F.3d 1244, 1263 (11th Cir. 2004) (en banc) ("Simply put, expert testimony may be assigned talismanic significance in the eyes of lay jurors, and, therefore, the district courts must take care to weigh the value of such evidence against its potential to mislead or confuse."). Trial judges serve a vital gatekeeping function, and a defendant is bound by evidentiary rules. Id. at 1272. For example, a defendant has no right to introduce unreliable "expert" testimony. Id. at 1271-72.
At the same time, however, "a district court's discretion to exclude evidence under Rule 403 is narrowly circumscribed." Aycock v. R.J. Reynolds Tobacco Co. , 769 F.3d 1063, 1069 (11th Cir. 2014) (cleaned up). A court may exclude relevant evidence under Rule 403 only if its probative value is substantially outweighed by certain dangers. This is a proportionality inquiry, and the balance "should be struck *1214in favor of admissibility." United States v. Elkins , 885 F.2d 775, 784 (11th Cir. 1989) ; see Aycock , 769 F.3d at 1069 (instructing courts to "look at the evidence in a light most favorable to admission") (citation omitted). " Rule 403 is an extraordinary remedy that must be used sparingly because it results in the exclusion of concededly probative evidence." United States v. US Infrastructure, Inc. , 576 F.3d 1195, 1211 (11th Cir. 2009).
A decision to exclude relevant evidence under Rule 403 is reviewed for abuse of discretion, and it is an abuse of discretion to incorrectly or unreasonably apply the law. Aycock , 769 F.3d at 1069. Based on the record, it was an abuse of discretion to exclude Dr. Ghostley's testimony before trial.
Dr. Ghostley's testimony was also excluded because the judge believed that proof of intoxication was irrelevant to proof of a defendant's blood-alcohol content. To the contrary, Mr. Abercrombie's physical demeanor was relevant. Dr. Ghostley was also a key part of Mr. Abercrombie's defense, and that should have affected the balance between the probative value of Dr. Ghostley's testimony and the possibility of misleading the jury. And the chance that Dr. Ghostley's testimony would be "confusing" (Doc. # 46, at 9) did not necessarily mean that possibility substantially outweighed his testimony's probative value.
The government was allowed to introduce evidence about the field tests. Though Mr. Abercrombie did not challenge whether there was probable cause to administer a breath test, the government was also allowed to question Sergeant Whitehead at-length about the field tests. (Doc. # 46, at 68-89.) The government played a video of the field tests. (Exhibit 1.) It then relied on the field tests (and the video) in its closing arguments, asking the jury to corroborate the breath test results based on Mr. Abercrombie's physical condition. See Word , 129 F.3d at 1211 (stressing that the government's closing argument relied on evidence the defendant was not permitted to rebut); Todd , 108 F.3d at 1331 (same). Thus, it was an abuse of discretion to exclude Dr. Ghostley's testimony.
E. The court is not convinced the error was harmless.
Exclusion of Dr. Ghostley's testimony does not necessarily mean that Mr. Abercrombie is entitled to a new trial. If the error was "harmless beyond a reasonable doubt," his conviction stands. United States v. Willner , 795 F.3d 1297, 1321 (11th Cir. 2015) (quoting Chapman v. California , 386 U.S. 18, 24, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967) ). But if there is a reasonable doubt about whether the result would have been the same, his conviction must be vacated. United States v. Hernandez , 141 F.3d 1042, 1050 (11th Cir. 1998). The government has not made any arguments about harmlessness. And based on an independent review of the evidence, the court cannot say beyond a reasonable doubt that the jury would have convicted Mr. Abercrombie even if Dr. Ghostley had testified.
Dr. Ghostley's testimony would have added new, relevant information, and it could have negated inferences drawn from the field tests. Without evidence of Mr. Abercrombie's disabilities, there were only two ways to interpret Mr. Abercrombie's failures on the field tests: Either he was drunk, or Sergeant Whitehead incorrectly administered the tests. Dr. Ghostley's testimony would have added a third option, one that set a different baseline for evaluating Mr. Abercrombie's behavior. It would have given the jury a reason to believe that Mr. Abercrombie acted the way he did not because of alcohol, but *1215because of his disabilities. Given that additional reason to conclude that the breath test results were exaggerated, the jury might have found that Mr. Abercrombie's blood-alcohol content was one-thousandth of one percent below the legal limit.
The significance of the field tests is underscored by the government's reliance on them in its closing argument, reliance that suggests the government thought it had to shore up its case. See United States v. Edwards , 576 F.2d 1152, 1155 (5th Cir. 1978) ; cf. United States v. Hands , 184 F.3d 1322, 1332 (11th Cir. 1999) ("When assessing the effect of the evidentiary error upon the case as a whole, we also must consider the prosecutor's closing argument."). And though there might have been sufficient evidence to convict Mr. Abercrombie, that does not mean the error was harmless. See United States v. Marshall , 173 F.3d 1312, 1318 & n.15 (11th Cir. 1999). In short, the court cannot say beyond a reasonable doubt that the error was harmless.
V. CONCLUSION
It is ORDERED that Defendant's conviction and sentence for driving under the influence of alcohol is VACATED and that he be given a new trial.
DONE this 28th day of May, 2019.
All references to "Mr. Abercrombie" are references to Defendant Daniel Abercrombie.
This was Count One of the Information. Mr. Abercrombie was also charged with driving with a suspended license (Count Two), open container (Count Three), driving without a current license plate (Count Four), and failing to provide proof of insurance (Count Five). (Doc. # 1, at 1-2.) He was convicted of Counts Two and Three, and the court dismissed Counts Four and Five. (Doc. # 43, at 1; Doc. # 46, at 174-80.) None of these other counts matter for this appeal.
Mr. Abercrombie raised other arguments, too, but there is no need to address them here.
See U.S. Const. amend. V ("No person shall be ... deprived of life, liberty, or property, without due process of law...."); U.S. Const. amend. VI (guaranteeing a defendant's "right ... to have compulsory process for obtaining witnesses in his favor").
When Curren was decided, the legal limit was 0.10 percent. The limit was later reduced to 0.08 percent. See Act of Aug. 9, 1995, No. 784, § 2, 1995 Ala. Laws 1862, 1864.