dissenting.
At the start of Mr. Keith Schwartz’s trial, the district court inadvertently provided the jury with an erroneous jury instruction. The instruction said that the jury had heard evidence that Keith Schwartz had previously been convicted of a felony. There was no such evidence presented, and Mr. Schwartz moved for a mistrial. This motion was denied.
*758Mr. Schwartz appeals that ruling. In this appeal the parties agree that the instruction was erroneous, but they disagree over the instruction’s harmfulness. In my view, the exroneous instruction was deeply prejudicial, rendering the denial of a mistrial an abuse of discretion. Thus, I would reverse the conviction.
I. The district court erroneously instructed the jury that evidence showed that Mr. Schwartz was a convicted felon.
Mr. Schwartz managed a pain management center, where a doctor prescribed opioids and other drugs for patients. The government prosecuted Mr. Schwartz, alleging that the center served as an unlawful pill mill. The case went to trial.
After swearing in the jurors, the court took a 29-minute recess. During this recess, the court provided each juror with a packet of ;jury instructions. Unbeknownst to the court, however, the packets contained an instruction that had been included by mistake.1 The first sentence of the instruction stated: “You have heard evidence that Keith Schwartz has been convicted of a felony, that is, a crime punishable by imprisonment for a term of years.” See R. vol. I, at 653; R. vol. Ill, at 45.2
After the recess ended, the court read aloud many of the jury instructions. When the court began reading the instruction about evidence of a prior felony conviction, the court realized that this instruction had been mistakenly included. The court stopped reading, asked the courtroom deputy to collect all copies of this instruction, and instructed the jurors to disregard the instruction.
Mr. Schwartz moved for a mistrial, arguing that the instructional error had prejudiced the jury. The court acknowledged the strength of Mr. Schwartz’s argument but denied the motion for a mistrial. Ultimately, no evidence of a prior felony conviction was ever admitted.
II. The instructional error prejudiced the jury.
On appeal, Mr. Schwartz argues that he was entitled to a mistrial, and I agree.3 Little doubt exists regarding the existence of an error, and the majority acknowledges that the instruction was erroneous. The only real issue is whether the error was prejudicial. I believe it was, for the jury instruction created the devastating impression that there was evidence of a prior felony conviction. See Alan D.. Hornstein, Between Rock and a Hard Place: The Right to Testify and Impeachment by Pri- or Conviction, 42 Vill. L. Rev. 1, 1 (1997) (“If the jury learns that a defendant previously has been convicted of a crime, the probability of conviction increases dramatically.”); L. Timothy Perrin, Pricking Boils, Preserving Error: On the Horns of a Dilemma After Ohler v. United States, 34 U.C. Davis L. Rev. 615, 651-52 (2001) (stating that “the admission at trfal of a criminal defendant’s prior convictions often spells doom for a criminal defendant” and *759that “[t]he available empirical data demonstrate that the admission of a prior conviction has an explosive impact on the jury, substantially increasing the likelihood that the jury will convict the defendant of the charged crime”). This impression tainted everything that the jury heard and saw over the course of the 11-day trial.
III. The Standard of Review and the Conflicting Standards for Whether a Mistrial Is Appropriate
The district court denied Mr. Schwartz’s motion for a mistrial. That denial is reviewed for an abuse of discretion. United States v. Morgan, 748 F.3d 1024, 1041 (10th Cir. 2014).
Both parties agree that the court should not have given the jury instruction. The resulting issue is whether the district court abused its discretion by denying Mr. Schwartz’s motion for a mistrial. On that issue, the government and the majority conclude that the denial of a mistrial fell within the district court’s discretion because the error had been harmless.
The majority approaches harmlessness based on our case law involving exposure to extraneous information. I agree with this approach, for the jury instruction exposed the jury to information that was never presented in any of the evidence.
In assessing whether the jury’s exposure to extraneous information was harmless, permitting the denial of a mistrial, we have employed two different standards. United States v. Muessig, 427 F.3d 856, 865 (10th Cir. 2005). Under the first standard, a mistrial is appropriate upon the “slightest possibility” that exposure to the extraneous information affected the verdict. Id. For this standard, the moving party bears the burden of proof. See id. Under the second standard, the jury’s exposure to extraneous information creates a presumption of prejudice. Id. But the non-movant may rebut this presumption by showing that the exposure was harmless. Id. In my view, the instructional error could not be deemed harmless under either standard.
IV. The Government’s Harmlessness Argument
The government argues that
• the jury was exposed to the erroneous instruction for only a short period of time,
• the instructional error was rendered harmless by events unfolding at trial, and
• a poll following the verdict indicated that the error was harmless.
I would reject these arguments.
A. The duration of the erroneous exposure was sufficient to cause prejudice.
The government contends that the- error in this case was harmless because the jury’s exposure to the erroneous instruction “‘lasted mere seconds.’” Appellee’s Ans. Br. at 17 (citation omitted); accord Maj. Op. at 755. For support, the government cites United States v. Morgan, 748 F.3d 1024 (10th Cir. 2014). There we held that allegedly prejudicial testimony was harmless, noting that the “testimony was one short moment in a seven-day trial.” Morgan, 748 F.3d at 1041. I would reject the government’s argument both as a matter of fact and as a matter of law.
First, as a factual matter, the exposure here did not necessarily transpire over just a few seconds. The record does not state exactly when the jury was provided with the instruction packet; the record reveals only that the jury obtained the packet at some point during the recess. That recess lasted 29 minutes, during which the jury *760may have perused the contents of the packets.
Second, even if the jury had seen the instruction for only a few seconds, this would be only one factor to consider. The Morgan court relied on the combination of multiple factors, only one of which was the temporal length of the erroneous exposure. Id. at 1041-42. As discussed below, the unique facts of our case magnified the potential for prejudice. Based on these facts, I would find prejudice even if the duration of the exposure had lasted only a short period of time.
B. The curative instructions and the questioning of Mr. Schwartz’s ex-wife did not render the instructional error harmless.
In the government’s view, the instructional error was rendered harmless by the district court’s curative instructions and the government’s questioning of Mr. Schwartz’s ex-wife, which suggested that Mr, Schwartz had previously faced felony charges. I disagree.
1. The Curative Instructions
The government contends that the dis-. trict court cured the error by telling the jury to disregard the erroneous instruction, adding that at other points during the trial, the court instructed the jury to ignore anything that it had been told to disregard.
We ordinarily presume that juries follow instructions like these. United States v. Morgan, 748 F.3d 1024, 1042 (10th Cir. 2014). And as a general rule, these kinds of instructions are sufficient to avert prejudice. Id. at 1041-42. But there is a limit to what instructions can cure: “[Wjhere the character of [extrinsic evidence] is such that it will create so strong an impression on the minds of the jurors that they will be unable to disregard it in their consideration of the case, although admonished to do so, a mistrial should be ordered.” See Maestas v. United States, 341 F.2d 493, 496 (10th Cir. 1965).
For instance, in Maestas v. United States, we held that curative instructions were insufficient when the jury was exposed to the fact that the defendant had previously been imprisoned. 341 F.2d at 495-97. Maestas mirrors the present case and indicates that the curative instructions here were insufficient.
In Maestas, a governmental informant gave a nonresponsive answer in cross-examination, revealing that the defendant had previously been imprisoned. The defendant immediately moved for a mistrial. Id. at 495. The district court overruled the motion, struck the nonresponsive answer, and instructed the jury to disregard the answer. Id.
Shortly thereafter, the informant gave another nonresponsive answer, again revealing that the defendant had been imprisoned. Id. at 495-96. Again, the defendant unsuccessfully moved for a mistrial; but the court again struck the nonrespon-sive answer and instructed the jury to disregard the answer. Id. at 496.
On appeal, we held that the district court had abused its discretion in refusing to grant a mistrial. Id. at 496-97. We reasoned that the jury must have been prejudiced by the informant’s unresponsive answers. Id. at 496. We acknowledged the two curative instructions, but concluded that they could not have prevented the prejudice. Id. at 495-97.
The prejudice here is even greater than it was in Maestas. There the prejudicial information came from a governmental informant in the midst of the trial. Here the prejudicial information came at the start of the trial from the court itself. These differences magnify the impact of the prejudice *761in our case. Thus, Maestas would prevent reliance on the curative instructions to avert the prejudice to Mr. Schwartz.4
a. The instructional error occurred at the start of Mr. Schwartz’s trial.
The timing of the prejudicial information here differs from the timing in Maestas. The jurors in Maestas were initially able to impartially evaluate the evidence against the defendant. Problems arose only after the jurors learned that the defendant had served time in prison. From that point on, the prejudicial information tainted everything that the jurors saw and heard.
By contrast, the instructional error here occurred before the jury had even heard opening statements. The jury’s first impression of Mr. Schwartz involved supposed evidence that he was a convicted felon. From the beginning, this prejudicial misinformation cast a shadow over everything that the jury saw and heard.
Scholarship confirms that this sort of first impression was particularly prejudicial. For example, scholars have recognized that individuals generally assign disproportionate weight to what they learn first. See Lawrence S. Wrightsman, The Place of Primacy in Persuading Jurors: Timing of Judges’ Instructions and Impact of Opening Statements, 8 U. Bridgeport L. Rev. 431, 432 (1987) (“Social psychologists have recognized for a long time the importance of early information in the formation and maintenance of the impressions of others.”). Jurors are no exception. The jurors’ initial information about a defendant frequently bears significantly on the outcome. See id. at 431 (arguing, based on studies, “that early information is very important in forming jurors’ impressions of guilt or responsibility in trials and suits”).5
In addition, scholars have recognized that a juror’s mood, focus, and energy level normally change over the course of a trial. At the outset, jurors are generally “anxious, curious, receptive, and looking for someone to trust.” Michael Frost, Ethos, Pathos & Legal Audience, 99 Dick. L. Rev. 85, 111 (1994); see Thomas A. Mauet, Fundamentals of Trial Techniques 42 (3rd ed. 1992); Ronald L. Carlson & Edward J. Imwinkelried, Dynamics of Trial Practice: Problems and Materials 110 (3rd ed. 2002). At this point, jurors “are also more likely to be interested and attentive.” Michael Frost, Ethos, Pathos & Legal Audience, 99 Dick. L. Rev. 85, 112 (1994); see Ronald L. Carlson & Edward J. Imwinkelried, Dynamics of Trial Practice 110 (3d ed. 2002).
But as the trial continues, the jurors’ focus and energy will generally decrease. By the end of the trial, the jurors “will be tired, perhaps bored, and occasionally confused.” Michael Frost, Ethos, Pathos & Legal Audience, 99 Dick. L. Rev. 85, 112 (1994); see Lawrence S. Wrightsman, The *762Place of -Primacy in Persuading Jurors: Timing of Judges’ Instructions and Impact of Opening Statements, 8 U. Bridgeport L. Rev. 431, 432 (1989) (“When observing a time-consuming event like a trial, attention tends to decrease over time because of such factors as boredom, distraction, fatigue, or simply an overload of information.”). In this case, the trial lasted eleven days; the jurors were presumably the most focused, receptive, and energetic at the start, precisely when the jury was exposed to information suggesting that Mr. Schwartz was a convicted felon notwithstanding the absence of any such evidence during the trial.
The timing of the instructional error magnified its prejudicial effect, leaving Mr. Schwartz in an even deeper hole than the defendant in Maestas.
b. The district court itself revealed the prejudicial information.
Maestas also differs from our case with respect to the source of the prejudicial information. In Maestas, the prejudicial information came from a governmental informant. 341 F.2d at 494-96. As a result, the jurors could reasonably question the source of the information. See On Lee v. United States, 343 U.S. 747, 757, 72 S.Ct. 967, 96 L.Ed. 1270 (1952) (“The use of informers, accessories, accomplices, false friends, or any of the other betrayals which are ‘dirty business’ may raise serious questions of credibility.”). Here the prejudicial information about Mr. Schwartz could not reasonably be questioned, for it came from the court itself.
‡ * *
In Maestas, a mistrial was necessary when the jury was exposed to the fact that the defendant had previously been imprisoned. Curative instructions were not enough to avert the prejudice. And the prejudice here was even greater than it had been in Maestas: The erroneous information here was revealed at the start of the trial by the district court. As a result, Maestas forecloses reliance on the curative instructions.
2. The instructional error w^s not rendered harmless by the government’s questioning of Mr. Schwartz’s ex-wife.
The government also argues that the instructional error was harmless because the jury eventually heard about felony charges during the government’s questioning of Mr. Schwartz’s ex-wife. In my view, this questioning did not render the instructional error harmless.
In questioning Mr. Schwartz’s ex-wife, the government’s questions suggested that Mr. Schwartz had been charged with various felonies. But, the government elsewhere downplays the impact of these questions, pointing out that the government “presented no independent evidence of the charges, and [Mr. Schwartz’s ex-wife] did not confirm or refute them.” Appellee’s Ans. Br. at 51. Thus, the jurors could reasonably question whether Mr. Schwartz had even been charged with these felonies.
But let’s suppose the jurors believed that Mr. Schwartz had previously been charged with other felonies. Even then, the instructional error would not be harmless. A charge is merely an accusation. There is a vast, qualitative difference between an accusation of criminality and a finding of guilt. Cf. Michelson v. United States, 335 U.S. 469, 482, 69 S.Ct. 213, 93 L.Ed. 168 (1948) (“Arrest without more does not, in law any more than in reason, impeach the integrity or impair the credibility of a witness. It happens to the innocent as well as the guilty.”). Accordingly, the impression created by the questioning paled in comparison to the impression ere-*763ated by the instructional error. At most, the questioning suggested that Mr. Schwartz had been accused of felonies; the instruction gave the jury the impression that there was evidence of a prior felony conviction, ,
C. The district court’s poll could not reliably indicate whether the jury was prejudiced.
After Mr. Schwartz was found guilty, the district court conducted a poll. In the government’s view, this poll showed that the jurors had put the erroneous instruction aside. I disagree, for the poll was too vague and generic to reliably expose prejudice from the instruction.
The court’s poll consisted of five stock questions:
1. Were these and are these your verdicts?
2. Was each such verdict based on the law I gave you and the instructions and none other than the instructions?
3. Was each such verdict based only on the facts that were admitted into evidence and none other?
4. In your verdicts, did you disregard everything I told you to disregard and not include any such matters in your deliberations?
5. In your verdicts where evidence was limited to a particular purpose and not generally for all purposes, did you limit your consideration of that evidence to that particular purpose?
See R. vol. Ill, at 2238-48. None of the questions directly addressed the instruction.6
For purposes of harmlessness, we have two possible questions. The first possible question is whether even the “slightest possibility” of prejudice existed notwithstanding the vague, generic poll given at the end of the trial. See Part III, above. The second possible question is whether the government proved that the instructional error was harmless based on the jurors’ answers to the poll. See id. In my view, the answer to both questions is “no.”
The answer to both questions remains “no” even when we consider the curative instructions and the questioning of Mr. Schwartz’s ex-wife. Under Maestas, the curative instructions did not avert the prejudice. As for the questioning, there was no independent evidence that Mr. Schwartz had previously been charged with the relevant crimes. Thus, the jury could justifiably doubt whether Mr. Schwartz had previously been charged with these crimes. In addition, even if the jury believed that Mr. Schwartz had previously been charged with these crimes, the prejudice from that belief would pale in comparison to the prejudice from the instructional error.
Y. The government’s evidence did not render the instructional error harmless.
The majority characterizes the government’s evidence as ample. If the majority means that the evidence was enough for a reasonable jury to find guilt,. I agree. But if the majority instead means that the government’s evidence was so overwhelming that the instructional error was harmless,7 I would respectfully disagree. The *764evidence against Mr. Schwartz was not overwhelming, and the jury could reasonably have voted to acquit.
For instance, a jury could reasonably have concluded that Mr. Schwartz had strived to stay within the bounds of the law. For this conclusion, the jury could have drawn on trial testimony indicating that
• Mr. Schwartz had consulted with an attorney to determine whether the pain management center was complying with the applicable regulations,
• Mr. Schwartz had researched the applicable regulations,
• Mr. Schwartz had consistently emailed articles to his staff about legal problems confronting other pain management centers, hoping that these articles would prevent the sort of mistakes that these centers had made,
• Mr. Schwartz had pushed for various safeguards, including but not limited to random urinalysis and the use of Colorado’s Prescription Drug Monitoring Program,8
• Mr. Schwartz had consistently emphasized the importance of following the law, and he would confront the doctor and the staff if he learned that they had bent or broken the rules,
• Mr. Schwartz had emphasized the importance of maintaining pristine, detailed records; had encouraged the use of electronic records; and had promoted the appropriate documentation of treatment,
• Mr, Schwartz had encouraged a 20% reduction in medications to help ensure that the center stayed within the bounds of the law,9 and
• Mr. Schwartz had urged the offering of discounts to encourage use of alternative therapies, such as physical therapy, that could decrease or even eliminate the need for opioids.
In addition, former employees testified that
• while working for the center, they had not believed that they were doing anything illegal, and
• the doctor, not Mr. Schwartz, had the final say on all medical-related questions.
Based on this testimony, the jury could reasonably have concluded that
• there was no unlawful conspiracy and
• any illegal activity at the center was attributable to the doctor, not to Mr. Schwartz.
*765The jury could also justifiably question the strength of the government’s evidence. That evidence largely consisted of the experiences of three patients. But the jury could reasonably have downplayed the experiences of these three patients because the center had supplied treatment for hundreds of other patients. ■
The government also relied heavily on the experiences of an undercover agent who had posed as a patient. The jury could reasonably downplay this evidence, too, because the undercover agent had attended the center shortly after it had taken over the patient base from the prior owner. At this point, the jury could reason, the center had to deal with a flood of new patients, many of whom were frantic because their former doctor was gone and they were running out of medication. The record also indicates that the center was forced to treat some of these patients without referring to their medical records because they had been seized by law enforcement.
Based on the evidence as a whole, a jury could reasonably have voted to acquit. Thus, I would draw two conclusions based on the two possible standards:
1. There was at least the “slightest possibility” that the outcome would have been different without the instructional error.
2. The government failed to prove that the instructional error was harmless.
These two conclusions lead me to believe that the district court abused its discretion in denying Mr. Schwartz’s motion for a mistrial.
* * *
The district court unquestionably erred by telling the jury that evidence existed even though none was ever presented. The subject-matter was devastating, a felony conviction, tainting the jury’s perception of Mr. Schwartz from the outset. Without this devastating error, there was at least the “slightest possibility” of a different outcome. And none of the government’s arguments would have satisfied its potential burden to prove that the instructional error was harmless. Thus, whichever harmlessness standard applies, the district court’s error was not harmless and I would reverse the conviction. Because the majority affirms, I respectfully dissent.
.A second jury instruction was also inadvertently included in the packet. The first sentence of this jury instruction referred to "evidence of other crimes, wrongs or acts engaged in by Mr. Schwartz.” R. vol. I, at 655; see R. vol. Ill, at 46. Before the jury deliberated, this instruction was read without objection. Thus, for the sake of argument, I assume that any error related to this jury instruction would have been harmless.
. Like the majority, I presume that the jurors read this instruction. Maj. Op. at 754.
. Because I would reverse on this ground, I do not analyze Mr. Schwartz’s other appeal points.
. In Maestas, the informant provided the jury with the extraneous information at two separate times. See pp. 7-8, But that fact does not distinguish Maestas from our case. As noted above, the jury here was provided with the erroneous instruction during the recess and could have read the instruction at least once. Then, the jurors could have read the instruction a second time when following along as the judge read the instructions aloud. Finally, after the judge announced that the instruction was to be collected, any juror whose curiosity had been aroused could have looked down and read the instruction for a third time. Thus, the jury could have been exposed to the improper information multiple times.
. The district court acknowledged this tendency, telling the jurors that "[i]t’s a natural tendency to develop an impression and then stay with it as you go along.” R. vol. Ill, at 11.
, The government appears to contend that "individually polling the jurors on whether they read or understood the erroneously included instruction!] ] may have compounded the problem” of jury prejudice. Appellee’s Ans. Br. at 22 n.6. But the jury was polled after it had already issued its verdict. Thus, a more specific poll could not have affected the verdict.
. The government does not argue that the instructional error was harmless based on the *764strength of the evidence against Mr. Schwartz.
. As the Centers for Disease Control and Prevention explains on its website,
Prescription Drug Monitoring Programs (PDMPs) are state-run electronic databases used to track the prescribing and dispensing of controlled prescription drugs to patients. They are designed to monitor this information for suspected abuse or diversion (i.e., channeling drugs into illegal use), and can give a prescriber or pharmacist critical information regarding a patient’s controlled substance prescription history.
Prescription Drug Monitoring Programs (PDMPs), Centers for Disease Control and Prevention, https://www.cdc.gov/ drugoverdose/pdmp/index.html (last updated March 21, 2017). This webpage does not appear in the appellate record, but is subject to judicial notice. See New Mexico ex rel. Richardson v. Bureau of Land Mgmt., 565 F.3d 683, 702 n.22 (10th Cir. 2009) (taking judicial notice of materials on the websites of two federal agencies); see also Winzler v. Toyota Motor Sales U.S.A., Inc., 681 F.3d 1208, 1213 (10th Cir. 2012) ("The contents of an administrative agency's publicly available files .., traditionally qualify for judicial notice.,,."),
. The center achieved the 20% reduction encouraged by Mr. Schwartz.