Federal agents suspected Nicolas Gomez of involvement in a cocaine-distribution ring operating in Chicago and Milwaukee. A wiretap on the phones of Robert Romero, a known Chicago supplier, revealed a reseller named “Güero” who lived in Milwaukee. The agents believed that Gomez was Güero. When Romero and Güero scheduled a deal for September 3, 2010, the agents followed Romero as he drove from Chicago to Milwaukee and parked his car on a street near Gomez’s house. The two men had a brief conversation next to Romero’s car and then left the scene on foot. Later that day the agents seized Romero’s ear — still parked where he had left it — and found a quarter kilogram of cocaine in the trunk. Gomez was arrested and charged with conspiracy to distribute cocaine and related crimes.
At trial the government introduced more than 50 recorded telephone calls between Romero and Güero detailing their cocaine transactions in the months leading up to September 3. The evidence tying the calls to Gomez’s residence was overwhelming, so his defense was that Güero must have been Victor Reyes, his brother-in-law who lived in the same house. In response the government sought to introduce a small quantity of cocaine found in Gomez’s bedroom at the time of his arrest. Gomez objected, citing Rule 404(b)(1) of the Federal Rules of Evidence, which prohibits the admission of evidence of other crimes, wrongs, or acts for the purpose of showing a person’s character or propensity to behave in a certain way. So-called “other act evidence” is admissible for other purposes, however, see Fed.R.Evid. 404(b)(2), and here the district court admitted the evidence for the purpose of proving Gomez’s identity as Güero.
Gomez was convicted on all counts. On appeal he primarily challenged the district court’s decision to admit the other-act evidence under Rule 404(b)(2). A divided panel affirmed. We reheard the case en banc to clarify the framework for admitting other-act evidence. We now conclude that our circuit’s four-part test should be replaced by an approach that more closely tracks the Federal Rules of Evidence. Applying a rules-based framework here, we hold that the cocaine found in Gomez’s bedroom should not have been admitted, but the error was harmless, so we affirm the judgment.
I. Background
In 2010 federal agents were investigating Romero, a Chicago-based cocaine supplier. A wiretap on his phones revealed that he was regularly conducting business with a reseller in Milwaukee, a man he called “Güero” or “Guerito.” The cell phones Güero used were registered to a residence at 2522 West Mineral Street in Milwaukee where Gomez and his brother-in-law Reyes lived. (Other people lived in the house too, although the record doesn’t identify them.) When Romero and Güero arranged a cocaine sale over the phone, GPS data tracked Romero driving from Chicago to an alley behind Gomez’s house on Mineral Street. When several months of wiretap evidence established this transactional pattern, the only thing left was to catch the conspirators in the act.
The opportunity came on September 2, 2010, when Romero and Güero discussed a *851sale for the following day. On September 3, DEA agents staked out Gomez’s house in Milwaukee, and a separate group of FBI agents followed Romero. He left Chicago in a white Mercedes, drove to Milwaukee, and parked on a street within a block and a half of Gomez’s house. Both groups of agents watched and videorecorded a brief interaction between Gomez and Romero standing next to the parked Mercedes. After a short conversation, the two men shook hands, then parted company and walked away in opposite directions.
Gomez’s route took him past the DEA agents, who stopped him to confirm his identity. Pretending to be part of an anti-gun task force, they patted him down and asked for his name, address, and telephone number. The number Gomez gave was the same number Güero used to arrange the sale that day. The agents let him go and Gomez walked home.
A few minutes later, FBI agents watched as Gomez drove a green minivan down the street and picked up Romero. The two men drove to Mercado El Rey, a nearby restaurant and grocery store, where they met Reyes (Gomez’s brother-in-law and housemate). An FBI agent followed them into the restaurant and photographed the meeting. Surveillance continued as the three men left El Rey and went their separate ways.
Reyes drove away in a tan Suburban and was stopped and identified by a DEA agent. Gomez and Romero must’ve been spooked because neither of them returned to the white Mercedes — Romero took a taxi all the way back to Chicago. Later that day the federal agents seized the abandoned Mercedes. A search of the car at DEA headquarters revealed a quarter kilogram of cocaine hidden in the trunk.
That evening and the following morning, recorded phone calls showed Romero and Güero frantically reviewing the events of September 3. Güero told Romero that the Mercedes had been towed and that he was using a new cell phone. (The number Gomez gave to DEA agents was deactivated that very day.) When Romero asked Güero what had happened after the “three of [them]” left El Rey, Güero responded that his brother-in-law had been stopped by police while driving away from El Rey — exactly what happened to Reyes. Romero then asked Güero when he had been stopped and searched by police, and Güero explained that he was stopped while walking — exactly what happened to Gomez.
On September 29 — almost four weeks later — federal agents arrested Gomez at his home. On the kitchen table were Gomez’s wallet and the cell phone Güero had been using since September 3. When agents searched Gomez’s bedroom, they found a shoe box filled with documents addressed to Gomez, including a phone bill for one of the three cell phones Güero had used up until September 3. Although all three cell phones were registered under Reyes’s name, records from the cellular-service provider confirmed that the billing statements were addressed to Gomez at 2522 West Mineral Street. Agents also found a small quantity of cocaine in the pocket of a pair of pants in Gomez’s bedroom.
Gomez was brought directly to the FBI’s prisoner processing center in Chicago where two agents interviewed him. They played three of the recorded phone calls, including the one describing the events of September 3. Gomez identified his own voice as Güero on all three. He was thereafter charged with conspiracy to possess cocaine with intent to distribute, see 21 U.S.C. § 846, and three counts of using a telephone to facilitate a drug crime, see id. § 843(b).
*852At trial the government introduced more than 50 recorded phone calls and testimony from various federal agents to establish the facts we’ve just described. Gomez’s defense was mistaken identity — he claimed that he was simply in the wrong place at the wrong time. He also argued that the government would never be able to show him in possession of cocaine. In response prosecutors sought to introduce the small quantity of cocaine found in Gomez’s bedroom at the time of his arrest. Gomez objected on Rule 404(b)(1) grounds. The trial judge initially denied the government’s request but ultimately admitted the evidence to show Gomez’s identity as Güe-ro.1 In the end Gomez focused his defense on attempting to raise a reasonable doubt about the government’s contention that he was Güero, arguing that it was more likely that Reyes was Romero’s coconspirator. He reminded the jurors that Reyes lived at the same address and had also met with Romero on September 3, and that the cell phones used in the conspiracy were registered in Reyes’s name. The jury convicted Gomez on all counts.
On appeal Gomez primarily challenged the admission of the other-act evidence— the cocaine found in his bedroom on the day of his arrest — and also raised a sentencing issue. A divided panel of this court affirmed, but the disagreement was limited to the evidentiary question. United States v. Gomez, 712 F.3d 1146, 1159 (7th Cir.2013); id. at 1159-63 (Hamilton, J., dissenting). We vacated the panel opinion and granted rehearing en banc on Gomez’s challenge to the admission of the other-act evidence under Rule 404(b). We now reinstate the panel opinion on the sentencing issue and address only the Rule 404(b) question.
II. Discussion
A. The Admissibility of Other-Act Evidence
Rule 404(b) prohibits the admission of evidence of other crimes, wrongs, or acts for the purpose of proving a person’s character or propensity to behave in a certain way, but permits the use of this evidence for other purposes:
(1) Prohibited Uses. Evidence of a crime, wrong, or other act is not admissible to prove a person’s character in order to show that on a particular occasion the person acted in accordance with the character.
(2) Permitted Uses; ... This evidence may be admissible for another purpose, such as proving motive, opportunity, intent, preparation, plan, knowledge, identity, absence of mistake, or lack of accident.
Fed.R.Evid. 404(b).
Our circuit has long used a four-part test to determine when other-act evidence is admissible:
To determine if such evidence is admissible, the district court must engage in a four-pronged analysis and evaluate whether (1) the evidence is directed toward establishing a matter in issue other than the defendant’s propensity to commit the crime charged, (2) the evidence shows that the other act is similar enough and close enough in time to be relevant to the matter in issue, (3) the evidence is sufficient to support a jury finding that the defendant committed the similar act, and (4) the probative value of the evidence is not substantially outweighed by the danger of unfair prejudice.
*853United States v. Zapata, 871 F.2d 616, 620 (7th Cir.1989) (citing United States v. Shackleford, 738 F.2d 776, 779 (7th Cir.1984) for parts (1), (2), and (4) of the test and Huddleston v. United States, 485 U.S. 681, 108 S.Ct. 1496, 99 L.Ed.2d 771 (1988), for part (3)).
Multipart tests are commonplace in our law and can be useful, but sometimes they stray or distract from the legal principles they are designed to implement; over time misapplication of the law can creep in. This is especially regrettable when the law itself provides a clear roadmap for analysis, as the Federal Rules of Evidence generally do. We have noted this problem in the Rule 404(b) context before. Especially in drug cases like this one, other-act evidence is too often admitted almost automatically, without consideration of the “legitimacy of the purpose for which the evidence is to be used and the need for it.” United States v. Miller, 673 F.3d 688, 692 (7th Cir.2012); see also United States v. Jones, 455 F.3d 800, 812 (7th Cir.2006) (Easterbrook, J., concurring) (“Allowing a prosecutor routinely to introduce drug convictions in the case in chief without demonstrating relevance to some concrete dispute between the litigants creates needless risk that a conviction will rest on the forbidden propensity inference.”). Moreover, as we explain here, some aspects of our test lack an adequate basis in the rules.
Our four-part test for evaluating the admissibility of other-act evidence has ceased to be useful. We now abandon it in favor of a more straightforward rules-based approach. This change is less a substantive modification than a shift in paradigm that we hope will produce clarity and better practice in applying the relevant rules of evidence.
1. Rules 401 & 402 (Relevance) and 104 (Relevance Conditioned on a Fact)
All evidentiary questions begin with Rule 402, which contains the general principle that “[r]elevant evidence is admissible” and “[irrelevant evidence is not.” Rule 401 defines relevant evidence as that which is both probative (having “any tendency to make a fact more or less probable than it would be without the evidence”) and material (the fact must be “of consequence in determining the action”).
The second and third factors in our four-part test generally correlate to the basic relevance inquiry under Rules 401 and 402, but the rules do not apply with the rote inflexibility that the test implies. Step three of the test directs the district court to evaluate whether the evidence of the proffered other act is sufficient to support a jury finding that the defendant committed it. Step two asks if the other act is both recent and similar enough to the conduct charged in the case to be relevant (i.e., “of consequence in determining the action”). See Zapata, 871 F.2d at 620.
Step three — the “sufficiency” inquiry — flows from Rule 104(b), which addresses relevance conditioned on a fact: “When the relevance of evidence depends on whether a fact exists, proof must be introduced sufficient to support a finding that the fact does exist.” Fed.R.Evid. 104(b).2 In Huddleston v. United States, the Supreme Court considered whether the admission of other-act evidence requires a preliminary finding by the court that the act has been proved by a prepon*854derance of the evidence. 485 U.S. at 682, 108 S.Ct. 1496. The Court held that it does not. Id. at 689, 108 S.Ct. 1496. Relying on the default principle that relevant evidence is admissible unless a rule specifies otherwise, the Court concluded that nothing in the text or history of Rules 104 or 404(b) requires the judge to find that the proponent has proved the other act before the evidence may be admitted. Id. at 687-89, 108 S.Ct. 1496. Although a preliminary finding by the judge is not required as a condition of admissibility, the Court emphasized that other-act evidence may not be admitted unless the evidence is sufficient for the jury to find by a preponderance of the evidence that the other act was committed. Id. at 689-90, 108 S.Ct. 1496. This requirement remains in full force as a condition of relevance.
Step two of the test, which requires an inquiry into the similarity and timing of the other act, is loosely connected to the basic principles of relevance found in Rules 401 and 402. See United States v. Foster, 652 F.3d 776, 785-86 (7th Cir.2011) (explaining that “the comparison of [the defendánt’s] prior acts to the charged crimes” is “directed at establishing the relevancy of the 404(b) evidence”) (citing United States v. Lloyd, 71 F.3d 1256, 1264-65 (7th Cir.1995)). But the strength of this inquiry varies depending on the particular theory of admissibility. For example, one permissible purpose for the introduction of other-act evidence is to prove a defendant’s identity through a “distinctive manner of operation, or modus operandi.” United States v. Simpson, 479 F.3d 492, 497-98 (7th Cir.2007), abrogated in part on other grounds by United States v. Boone, 628 F.3d 927, 933 (7th Cir.2010). A prior act will be relevant to this purpose when it “ ‘bears a singular strong resemblance to the pattern of the offense charged’ with the similarities between the two crimes ‘sufficiently idiosyncratic to permit an inference of pattern.’ ” Id. at 498 (quoting United States v. Thomas, 321 F.3d 627, 634-35 (7th Cir.2003)). Sometimes the prior bad act may be too dissimilar to be relevant to show a distinctive pattern, leaving only the forbidden propensity inference. Id.
On the other hand, the need to check for similarity and recency may be substantially diminished or nonexistent depending on the particular purpose for which the evidence is offered. See United States v. Torres, 977 F.2d 321, 326 (7th Cir.1992); United States v. Beasley, 809 F.2d 1273, 1277 (7th Cir.1987). In some cases the relative similarity of the other act to the charged offense may be unimportant as a test of relevance. See, e.g., Foster, 652 F.3d at 785-86 (holding that the similarity of the other-act evidence to the charged offense was “of exceedingly minimal significance” when evidence of a prior check-cashing scheme was introduced to show a criminal relationship between the defendant and his accomplice in an armed bank robbery); United States v. Shriver, 842 F.2d 968, 974 (7th Cir.1988) (same with respect to motive). Recognizing this, we have repeatedly said that the “similarity” requirement for admitting other-act evidence is not “unduly rigid,” Foster, 652 F.3d at 785, but instead is “loosely interpreted and applied,” United States v. Vargas, 552 F.3d 550, 555 (7th Cir.2008).
Our discussion thus far should illustrate the problem of treating the “similarity” and “timing” factors as formal boxes to check in the admissibility analysis. It’s far too tempting to stop at superficial comparisons without meaningfully analyzing how the similarity and recency of the prior bad act affect its relevance in the unique circumstances of the case. And the similarity and timing of the other act may not bear on the relevance question at all. We *855think it best to return to a framework that weighs the relevance of other-act evidence directly.
To restate the principle in positive terms: The extent to which a proffered “other crime, wrong, or act” is close in time and similar to the conduct at issue in the case may have a bearing on its relevance, which is the starting point for all evidence questions, but the importance of testing for similarity and recency will depend on the specific purpose for which the other-act evidence is offered. The proponent of the other-act evidence should address its relevance directly, without the straightjacket of an artificial checklist.
2. Rule 404(b)
Rule 404(b) excludes relevant evidence of other crimes, wrongs, or acts if the purpose is to show a person’s propensity to behave in a certain way, but other-act evidence may be admitted for “another purpose” including, but not limited to, “proving motive, opportunity, intent, preparation, plan, knowledge, identity, absence of mistake, or lack of accident.” Fed. R.Evid. 404(b). The rule is straightforward enough, but confusion arises because admissibility is keyed to the purpose for which the evidence is offered, and other-act evidence is usually capable of being used for multiple purposes, one of which is propensity.3 See Beasley, 809 F.2d at 1279-80 (“Almost any bad act evidence simultaneously condemns by besmirching character and by showing one or more of ‘motive, opportunity, intent, preparation, plan, knowledge, identity, or absence of mistake or accident’.... ”). In the criminal context, and especially in drug cases, few defendants are new to criminal activity and the range of possible defenses is fairly limited, so at least three of the permitted purposes listed in the rule- — knowledge, intent, and identity — are routinely in play.
Because other-act evidence can serve several purposes at once, evidentiary disputes under Rule 404(b) often raise the following question: Does a permissible ultimate purpose (say, proof of the defendant’s knowledge or intent) cleanse an impermissible subsidiary purpose (propensity)? On the surface the rule seems to permit this. But if subsection (b)(2) of the rule allows the admission of other bad acts whenever they can be connected to the defendant’s knowledge, intent, or identity (or some other plausible non-propensity purpose), then the bar against propensity evidence would be virtually meaningless. We have made this point before. See, e.g., United States v. McMillan, 744 F.3d 1033, 1038 (7th Cir.2014); Miller, 673 F.3d at 696 (explaining that “if applied mechanically,” the permitted purposes listed in the rule “would overwhelm the central principle” of the rule against propensity evidence (quoting Beasley, 809 F.2d at 1279)).
*856To resolve this inherent tension in the rule, we have cautioned that it’s not enough for the proponent of the other-act evidence simply to point to a purpose in the “permitted” 'list and assert that the other-act evidence, is relevant to it. Rule 404(b) is not just concerned with the ultimate conclusion, but also with the chain of reasoning that supports the non-propensity purpose for admitting the evidence. United States v. Reed, 744 F.3d 519, 524-25 (7th Cir.2014); United States v. Lee, 724 F.3d 968, 976-77 (7th Cir.2013); Miller, 673 F.3d at 697-98. In other words, the rule allows the use of other-act evidence only when its admission is supported by some propensity-free chain of reasoning. Lee, 724 F.3d at 978 (“When one looks beyond the purposes for which the evidence is being offered and considers what inferences the jury is being asked to draw from that evidence, and by what chain of logic, it will sometimes become clear ... that despite the label, the jury is essentially being asked to rely on the evidence as proof of the defendant’s propensity to commit the charged offense.”); Miller, 673 F.3d at 697-99; United States v. Jones, 389 F.3d 753, 757 (7th Cir.2004), vacated on other grounds by Jones v. United States, 545 U.S. 1125, 125 S.Ct. 2948, 162 L.Ed.2d 864 (2005). This is not to say that other-act evidence must be excluded whenever a propensity inference can be drawn; rather, Rule 404(b) excludes the evidence if its relevance to “another purpose” is established only through the forbidden propensity inference.
Spotting a hidden propensity inference is not always easy. See Jones, 389 F.3d at 757. For this reason, although we have long required the record to reflect a “principled exercise of discretion” by the district court, Beasley, 809 F.2d at 1279, we have more recently emphasized the importance of identifying the non-propensity theory that makes the other-act evidence relevant and specifically asking how the evidence tends to make a particular fact of consequence more or less probable. For example, in United States v. Ciesiolka, 614 F.3d 347, 355 (7th Cir.2010), we noted that it was critical to “delineate precisely the legitimate ends to which the evidence could be applied.” In Miller we explained that the court should ask “more specifically how” the other-act evidence is relevant to a permitted purpose in order to help expose impermissible uses of other-act evidence for pure propensity purposes. 673 F.3d at 699. In United States v. Richards, 719 F.3d 746 (7th Cir.2013), we said that the “district court[ ] must consider specifically how the prior conviction tends to serve the non-propensity exception.” Id. at 759 (internal quotation marks and alterations omitted). And in Lee we explained that the court must “consider the chain of logic by which the jury is being asked to glean the defendant’s knowledge, intent, etc., from proof of his prior misdeeds.” 724 F.3d at 976-77.
The principle that emerges from these recent cases is that the district court should not just ask whether the proposed other-act evidence is relevant to a non-propensity purpose but how exactly the evidence is relevant to that purpose — or more specifically, how the evidence is relevant without relying on a propensity inference. Careful attention to these questions will help identify evidence that serves no permissible purpose.
3. Rule 403
Finally, even if other-act evidence is relevant without relying on a propensity inference, it may be excluded under Rule 403, which applies “with full force” in this context, Miller, 673 F.3d at 696, and gives the district court discretion to exclude relevant evidence if its proba*857tive value is “substantially outweighed by a danger of ... unfair prejudice,” Fed. R.Evid. 403. Other-act evidence raises special concerns about unfair prejudice because it almost always carries some risk that the jury will draw the forbidden propensity inference.4 See Lee, 724 F.3d at 976 (describing the “inherent risk of prejudice that such evidence poses to the defendant”). Rule 403 does much of the heavy lifting in the admissibility analysis by excluding other-act evidence that may be slightly probative through a non-propensity theory but has a high likelihood of creating unfair prejudice by leading a jury to draw conclusions based on propensity. See Miller, 673 F.3d at 697 (explaining that under Rule 403 “all bad acts evidence must be balanced for probative value and unfair prejudice”); United States v. Chapman, 692 F.3d 822, 827 (7th Cir.2012) (“The admission of [other-act] evidence always carries with it some risk of unfair prejudice to the defendant, but the critical issue is whether that risk is sufficiently outweighed by other factors.”).
i. The degree to which a fact is contested
One important issue in Rule 403 balancing in this context is the extent to which the non-propensity factual proposition actually is contested in the case. For example, if a defendant offers to concede or stipulate to the fact for which the evidence is offered, additional evidence may have little probative value. See, e.g., Old Chief v. United States, 519 U.S. 172, 191-92, 117 S.Ct. 644, 136 L.Ed.2d 574 (1997) (holding that a defendant’s stipulation to a prior felony conviction removes its probative value in a prosecution for unlawful possession of a firearm by a felon). Of course, there are various degrees of factual disagreement in a trial, and stipulations are at one end of that spectrum. See generally 2 EdwáRD J. Imwinrelried, Uncharged Misoonduct Evidenoe §§ 8:10— 15 (2004) (reviewing various forms of disagreement and their effect on the admissibility of other-act evidence). Because each case is unique, Rule 403 balancing is a highly context-specific inquiry; there are few categorical rules. See Miller, 673 F.3d at 696-97. The general guiding principle is that the degree to which the non-propensity issue actually is disputed in the case will affect the probative value of the other-act evidence. See United States v. Causey, 748 F.3d 310, 318 (7th Cir.2014); Lee, 724 F.3d at 976; Miller, 673 F.3d at 696-97.
On the other hand, there are a few discrete circumstances in which we can say as a categorical matter that other-act evidence is substantially more prejudicial than probative. The Supreme Court’s decision in Old Chief is one example. There, the defendant was charged with unlawful possession of a firearm by a felon and offered to stipulate to his status as a felon to prevent the jury from learning the details of his prior conviction. The government rejected his offer to stipulate, and the district court permitted the prosecutor to introduce the prior conviction over the defendant’s Rule 403 objection. 519 U.S. at 177-78, 117 S.Ct. 644.
The Supreme Court reversed. Id. at 192, 117 S.Ct. 644. The Court began by acknowledging the “familiar, standard rule that the prosecution is entitled to prove its *858case by evidence of its own choice, or, more exactly, that a criminal defendant may not stipulate or admit his way out of the full evidentiary force of the case as the Government chooses to present it.” Id. at 186-87, 117 S.Ct. 644. But when a defendant’s status as a felon is an element of the offense and he offers to stipulate to that fact, the evidence of his prior conviction loses its probative value, leaving only a substantial risk of unfair prejudice. Id. at 191-92, 117 S.Ct. 644. In this specific situation, a district court abuses its discretion in admitting the details of the prior conviction. Id.
Our circuit also requires special caution when other-act evidence is offered to prove intent, which though a permissible non-propensity purpose is nonetheless “most likely to blend with improper propensity uses.” Miller, 673 F.3d at 698. In cases involving general-intent crimes — e.g., drug-distribution offenses (as distinct from drug conspiracies or possession of drugs with intent to distribute) — we have adopted a rule that other-act evidence is not admissible to show intent unless the defendant puts intent “at issue” beyond a general denial of guilt. See, e.g., United States v. Hicks, 635 F.3d 1063, 1070-71 (7th Cir.2011); Shackleford, 738 F.2d at 781, overruled in part on other grounds by Huddleston v. United States, 485 U.S. 681, 108 S.Ct. 1496, 99 L.Ed.2d 771. Our most complete explanation of this rule comes from Shackleford:
We have previously distinguished between situations in which intent is in issue because the government must show specific intent as an essential element of the crime and when intent is only a formal issue that can be inferred from the act. When the crime charged requires proof of specific intent, we have held that, because it is a material element to be proved by the government, it is necessarily in issue and the government may submit evidence of other acts in an attempt to establish the matter in its case-in-chief, assuming the other requirements of Rules 404(b) and 403 are satisfied.... On the other hand, we have stated that when intent is only a formal issue, so that proof of the proscribed act gives rise to an inference of intent, then unless the government has reason to believe that the defense will raise intent as an issue, evidence of other acts directed toward this issue should not be used in the government’s case-in-chief and should not be admitted until the defendant raises the issue.
738 F.2d at 781 (emphasis added).
The specific-intent/general-intent distinction in the Rule 404(b) context is sometimes misunderstood. The critical point is that for general-intent crimes, the defendant’s intent can be inferred from the act itself, so intent is not “automatically” at issue. The paradigm case involves a charge of distribution of drugs, see Hicks, 635 F.3d at 1070-71, a general-intent crime for which the government need only show that the defendant physically transferred the drugs; the jury can infer from that act that the defendant’s intent was to distribute them. Hence our rule that “[b]eeause unlawful distribution [of drugs] is a general intent crime, in order for the government to introduce prior bad acts to show intent, the defendant must put his intent at issue first.” Id.; see also United States v. Manganellis, 864 F.2d 528, 539 (7th Cir.1988).
In contrast, we have repeatedly rejected a similar rule for specific-intent crimes because in this class of cases “intent is automatically at issue.” United States v. Conner, 583 F.3d 1011, 1022 (7th Cir.2009) (collecting cases). Unfortunately, this line of precedent too frequently has been seen as a rule of automatic admission *859for other-act evidence in eases of specific-intent crimes. See Lee, 724 F.3d.at 981; Miller, 673 F.3d at 698-99. We firmly rejected that notion in Miller, emphasizing that other-act evidence is always subject to Rule 403 balancing. 673 F.3d at 696-98. We explained that although “[ijntent can be ‘automatically at issue’ because it is an element of a specific intent crime,” other-act evidence offered to prove intent “can still be completely irrelevant to that issue, or relevant only in an impermissible way.” Id. at 697-98. We have reiterated these themes in other recent cases. See, e.g., Lee, 724 F.3d at 976 (“Simply because a subject like intent is formally at issue when the defendant has claimed innocence and the government is obliged to prove his intent as an element of his guilt does not automatically open the door to proof of the defendant’s other wrongful acts for purposes of establishing his intent.”); United States v. Earls, 704 F.3d 466, 471 (7th Cir.2012) (“Rule 404(b) does not provide a rule of automatic admission whenever bad acts evidence can be plausibly linked to another purpose.... The Rule 402 requirement of relevance and the unfair prejudice balancing inquiries of Rule 403 still apply with full force.” (internal quotation marks omitted)).
To summarize then, when intent is not “at issue” — when the defendant is charged with a general-intent crime and does not meaningfully dispute intent — other-act evidence is not admissible to prove intent because its probative value will always be substantially outweighed by the risk of unfair prejudice. In contrast, when intent is “at issue”- — in cases involving specific-intent crimes or because the defendant makes it an issue in a case involving a general-intent crime — other-act evidence may be admissible to prove intent, but it must be relevant without relying on a propensity inference, and its probative value must not be substantially outweighed by the risk of unfair prejudice. And again, the degree to which the non-propensity issue actually is contested may have a bearing on the probative value of the other-act evidence.
Before moving on, we pause to note a point raised by the government that one of our recent cases could be read to suggest a generally applicable rule that other-act evidence may not be admitted unless the defendant “meaningfully dispute[s]” the non-propensity issue for which the evidence is offered. Richards, 719 F.3d at 759 (citing Miller, 673 F.3d at 697). Richards never actually held that, but for clarity’s sake, we reiterate that there is no such categorical rule or prerequisite, and we decline to adopt one now. Nothing in the Rules of Evidence supports imposing such a universal prerequisite to the admission of other-act evidence. Indeed, the advisory committee explicitly disapproves of any general requirement of this sort. See Fed.R.Evid. 401 advisory committee’s notes (1972) (“The fact to which the evidence is directed need not be in dispute .... [T]he ruling should be made on the basis of such considerations as waste of time and undue prejudice (see Rule 403), rather than under any general requirement that evidence is admissible only if directed to matters in dispute.”); Fed. R.Evid. 404 advisory committee’s notes (1972) (“No mechanical solution is offered.”). Moreover, as we have noted, the Supreme Court has specifically endorsed “the accepted rule that the prosecution is entitled to prove its case free from any defendant’s option to stipulate the evidence away.” Old Chief, 519 U.S. at 189, 117 S.Ct. 644. The Court held in Old Chief that “if ... there [is] a justification for receiving evidence of the nature of prior acts on some issue other than status 0i.e., to prove motive, opportunity, intent, preparation, plan, knowledge, identity, or *860absence of mistake or accident ... ), Rule 404(b) guarantees the opportunity to seek its admission.” Id. at 190, 117 S.Ct. 644 (internal quotation marks omitted).
Though not a fixed requirement, we reiterate that the district court should consider the degree to which the non-propensity issue actually is contested when evaluating the probative value of the proposed other-act evidence. Because other-act evidence almost always carries a risk of unfair prejudice, sensitivity to' the real factual disputes in the case is critical to meaningful Rule 403 balancing.
In sum, to overcome an opponent’s objection to the introduction of other-act evidence, the proponent of the evidence must first establish that the other act is relevant to a specific purpose other than the person’s character or propensity to behave in a certain way. See Fed.R.Evid. 401, 402, 404(b). Other-act evidence need not be excluded whenever a propensity inference can be drawn. But its relevance to “another purpose” must be established through a chain of reasoning that does not rely on the forbidden inference that the person has a certain character and acted in accordance with that character, on the occasion charged in the case. If the proponent can make this initial showing, the district court must in every case assess whether the -probative value of the other-act evidence is substantially outweighed by the risk of unfair prejudice and may exclude the evidence under Rule 403 if the risk is too great. The court’s Rule 403 balancing should take account of the extent to which the non-propensity fact for which the evidence is offered actually is at issue in the case.
ii. Jury instructions
Appropriate jury instructions may help to reduce the risk of unfair prejudice inherent in other-act evidence. See, e.g., United States v. Carter, 695 F.3d 690, 702 (7th. Cir.2012); Fed.R.Evid. 403 advisory committee’s notes (1972) (explaining that the effectiveness of limiting instructions are a factor in weighing the danger of unfair prejudice). A limiting instruction must be given upon request. See Fed. R.Evid. 105. But a defendant may choose to go without one to avoid highlighting the evidence. See Jones, 455 F.3d at 811 (Easterbrook, J., concurring). We caution against judicial freelancing in this area; sua sponte limiting instructions in the middle of trial, when the evidence is admitted, may preempt a defense preference to let the evidence come in without the added emphasis of a limiting instruction. The court should consult counsel about whether and when to give a limiting instruction.
Rule 404(b) already requires that the prosecutor notify the defendant before trial of an intent to use other-act evidence, provided the defense has requested this notice (though the judge may excuse a lack of pretrial notice for good cause). See Fed.R.Evid. 404(b)(2)(A)-(B). Motions in limine are common even in the absence of a defense request for notice. As soon as it becomes clear that other-act evidence may be part of the prosecution’s case, the district judge should raise with the defendant whether the defense wants a limiting instruction (since Rule 105 requires it to be given on request) and, if so, what kind.
When given, the limiting instruction should be customized to the case rather than boilerplate. Id. at 811-12 (“A good limiting instruction needs to be concrete so that the jury understands what it legitimately may do with the evidence.”). In order to “effectively distinguish appropriate from inappropriate inferences,” id. at 812, jurors should be told in plain language the specific purpose for which the *861evidence is offered and that they should not draw any conclusions about the defendant’s character or infer that on a particular occasion the defendant acted in accordance with a character trait. Our circuit’s Pattern Jury Instruction 3.11 is a good starting point, but it needs customization for the particular case.
Moreover, we see no reason to keep the jury in the dark about the rationale for the rule against propensity inferences. Lay people are capable of understanding the foundational principle in our system of justice that “we try cases, rather than persons.” People v. Allen, 429 Mich. 558, 420 N.W.2d 499, 504 (1988); see also United States v. Linares, 367 F.3d 941, 945 (D.C.Cir.2004) (explaining that a component of the presumption of innocence is that “a defendant must be tried for what he did, not for who he is”); 1 Imwinkel-RIED, supra, § 1:3 (2004) (explaining that the rule against propensity evidence guards against the illicit temptation to “penalize the defendant for his or her past misdeeds”). The court’s limiting instruction would be more effective if it told the jurors that they must not use the other-act evidence to infer that the defendant has a certain character and acted “in character” in the present case because it does not follow from the defendant’s past acts that he committed the particular crime charged in the case.
Finally, the instruction would be improved by tying the limiting principle to the prosecution’s burden of proof. The jurors should be reminded that the government’s duty is to prove beyond a reasonable doubt every element of the specific crime charged, and it cannot discharge its burden by inviting an inference that the defendant is a person whose past acts suggest a willingness or propensity to commit crimes.
B. Application of a Rules-Based Framework
Here, the district court allowed the government to use the evidence of the user quantity of cocaine found in Gomez’s bedroom for the purpose of proving his identity as Güero. That was error, but for reasons we will explain, the error was harmless.
1. The Evidence Was Relevant Only Through a Propensity Inference
Because the proponent of the other-act evidence must explain how it is relevant to a non-propensity purpose, the government needed a rationale for connecting the cocaine found in Gomez’s bedroom to his identity as Güero without relying on the forbidden propensity inference. As we’ve discussed, one accepted way to use other-act evidence to prove identity is to argue that the perpetrator had a distinctive modus operandi. Gomez suggests that we should limit the “identity” uses of other-act evidence to this theory. We see no reason to do so. Indeed, we have previously said that modus operandi is not the exclusive theory for admitting other acts to prove identity. Simpson, 479 F.3d at 498. Accordingly, we reject Gomez’s invitation to artificially limit the ways in which other-act evidence can be admitted to prove identity. At the same time, however, a defense of mistaken identity does not by itself give the government a green light to use other-act evidence.
To support its identity theory of relevance here, the government relies on two eases in which other-act evidence was admitted in response to a defense of mistaken identity: United States v. Brown, 471 F.3d 802 (7th Cir.2006), and United States v. Gibson, 170 F.3d 673 (7th Cir.1999). Brown is inapplicable. There, the defendant was alleged to have participated in a drug transaction that was interrupted by *862the police. He fled the scene, but another man involved in the deal was apprehended and agreed to cooperate with law enforcement. 471 F.3d at 804. The cooperator identified Brown as the drug buyer who fled the scene when the police arrived. Brown’s defense was mistaken identity— he claimed that he had nothing to do with the cooperator — not on this specific occasion or at any other time. Id. at 806. In response the government introduced evidence that Brown had purchased drugs from the cooperator on many prior occasions. We approved this use of other-act evidence because it undercut Brown’s claim that the cooperator was lying. Id.
The basic theory of admissibility in Brown — that evidence of other transactions between the defendant and a witness is admissible to bolster the witness’s identification of the defendant as a participant in the charged transaction — is a distinct and widely acknowledged theory of admissibility under Rule 404(b). See 1 ImwinK-elRied, supra, § 3:7 (2006). But it has no application in this case. The other-act evidence here — a user quantity of cocaine found in Gomez’s bedroom 26 days after the conspiracy ended — did not serve that purpose.
Gibson is more on point. Gibson was charged with distributing cocaine in 1996. His defense was that it was not him, but his brother, who sold the cocaine. In response the government introduced Gibson’s postarrest statement to the FBI that he regularly sold cocaine from 1994 to 1997 (he denied involvement in the particular sale at issue in the case). We upheld the prosecution’s use of Gibson’s statement about his history of drug dealing “because his primary defense at trial was that of mistaken identity.” Gibson, 170 F.3d at 679.
Gibson is hard to square with United States v. Simpson, 479 F.3d 492, a nearly identical case that reached the opposite conclusion. Simpson was charged with selling cocaine, and as in Gibson his defense was mistaken identity. The government introduced Simpson’s statement to the FBI admitting that he had been dealing cocaine for three or four years leading up to the charged crime. We held that this evidence was improperly admitted to prove identity because it could only be relevant to that issue by way of an impermissible propensity inference. Id. at 497-98. We distinguished Gibson by highlighting that Gibson had specifically pointed the finger at an alternative culprit — his brother — and that Gibson’s history of drug dealing made it more likely that he — not his brother — was the drug dealer. Id. at 499 n. 1. In contrast, Simpson argued more generally that the police had the wrong guy. Id.
Here, as in Gibson, Gomez’s mistaken-identity defense singled out another person — his brother-in-law and housemate Victor Reyes — as the “real” Güero. The government introduced the user quantity of cocaine found in Gomez’s bedroom for the purpose of showing that as between the two, it was more likely that Gomez was Güero. If the distinction drawn in Simpson is valid, then Gibson controls as the more closely analogous precedent. But the distinction does not hold up. Simpson relied on a supposed difference in the probative value of propensity evidence when a mistaken-identity defense focuses on a particular alternative suspect rather than arguing mistaken identity more generally. But Rule 404(b) does not allow propensity evidence when it is probative enough; it bars propensity evidence as a categorical matter. In Gibson, just as in Simpson, the evidence of the defendant’s history of drug dealing tended to prove his identity as a participant in the charged drug deal only by way of a forbidden propensity inference: Once a drug dealer, al*863ways a drug dealer. Gibson and Simpson cannot be reconciled. We now conclude that Gibson did not survive our recent decisions in Miller and Lee.
This case is yet another example of the importance of asking how exactly the proposed other-act evidence is relevant without relying on propensity. The government maintains that the cocaine found in Gomez’s bedroom was admissible because it made it more probable that Gomez — not Reyes — was Güero. But that is just to say that the evidence is relevant; it doesn’t tell us how the evidence is relevant in a propensity-free way. On that pivotal question, the government has little to offer.
If there had been some basis to argue that the cocaine in Gomez’s bedroom was the product of the conspiracy, then it might have qualified as direct evidence of Gomez’s participation in Romero’s drug ring and Rule 404(b) would not apply. See United States v. Phillips, 745 F.3d 829, 833 (7th Cir.2014); United States v. Adams, 628 F.3d 407, 414 (7th Cir.2010); United States v. Alviar, 573 F.3d 526, 538 (7th Cir.2009). But the government doesn’t make that argument, for good reason. The cocaine in Gomez’s room was a small user quantity; it was half as pure as the quarter kilogram of cocaine found in Romero’s car; and it was recovered almost four weeks after Romero and Gomez ceased doing business together.
In the end, the government offers no theory other than propensity to connect the cocaine found in Gomez’s bedroom to his identity as Güero, Romero’s coconspir-ator. The government’s sole theory is that Gomez’s possession of a user quantity of cocaine 26 days after the conspiracy ended shows that he, rather than Reyes, was Güero. That argument is extraordinarily weak, but the more important point is that it rests on pure propensity: Because Gomez possessed a small quantity of cocaine at the time of his arrest, he must have been involved in the cocaine-distribution conspiracy. The district court should not have admitted this evidence.
2. The Error Was Harmless
Evidentiary errors are subject to review for harmlessness. See Fed. R.CRIM.P. 52(a); McMillan, 744 F.3d at 1034; United States v. Vargas, 689 F.3d 867, 875 (7th Cir.2012). In this context, “[t]he test for harmless error is whether, in the mind of the average juror, the prosecution’s case would have been significantly less persuasive had the improper evidence been excluded.” Vargas, 689 F.3d at 875 (citing United States v. Loughry, 660 F.3d 965, 975 (7th Cir.2011)). The evidence identifying Gomez as Güero was quite compelling and would not have been less so had the other-act evidence been excluded.
First, Romero addressed his coconspirator as “Güero” in numerous recorded phone calls, confirming that it was always the same person on the other end of the line in the months leading up to the September 3 transaction. GPS data showed Romero driving to the alley behind Gomez’s house after arranging sales with Güero. On September 2 Romero again called Güero to arrange a cocaine sale for the next day. Romero drove to Milwaukee on September 3 as planned and parked his Mercedes near Gomez’s house, consistent with past practice. Federal agents observed — and a videorecording captured — a brief conversation between Gomez and Romero as they stood next to the Mercedes. Importantly, when the two men parted company and left on foot in opposite directions, the agents stopped Gomez, asked for his identification, and he gave them the same phone number that Güero had used to set up the sale that day. Later that day the agents seized and searched the abandoned Mercedes and found a quarter kilogram of cocaine in the trunk.
*864Next, the jury heard other evidence corroborating Gomez’s identity as Güero. The three cell phones Güero used were registered to the house on Mineral Street that Gomez shared with Reyes, but the bills were sent to Gomez at that address, and he had one of the bills in a box in his bedroom. True, the phones could be linked to Reyes too, and Reyes joined Romero and Gomez at the meeting at El Rey after Gomez was stopped and searched by DEA agents. So it’s no wonder Gomez focused his defense on pointing the finger at Reyes.
But other evidence convincingly refuted the theory that Reyes was Güero. During Gomez’s postarrest interview, federal agents played several of the recorded phone calls for him, including one that took place after the aborted cocaine sale on September 3. Gomez admitted that Guero’s voice on the recordings was his. This admission is quite powerful on its own, but because Gomez challenged the agents’ credibility, we press on with our harmless-error analysis. See United States v. Robinson, 724 F.3d 878, 888 (7th Cir.2013).
In a recorded phone call on September 4 retracing the events of the previous day, Güero describes in the first -person how the police stopped him as he walked away from his conversation with Romero; we know that happened to Gomez, not Reyes. Earlier in that same phone call Romero asked Güero what happened to Reyes, and Güero responded with an account in the third person about what happened to his brother-in-law after the three men left El Rey.5 Finally, when Gomez was arrested at his home nearly four weeks after the conspiracy ended, his wallet was found on the kitchen table right next to the phone Güe-ro had used since September 3.
In short, the prosecution’s case was strong and would not have been any less persuasive had the other-act evidence been excluded — or at least it would not have been significantly less persuasive in the mind of the average juror. The link between the small quantity of cocaine found in Gomez’s bedroom and his identity as Güero was extremely weak; the government has never really articulated a coherent theory for why this evidence helped its case. Excluding it would not have seriously diminished the strength of the prosecution’s case against Gomez. We’re satisfied that the error was harmless.
Affirmed.
. The court also allowed the other-act evidence for the purpose of showing Gomez's knowledge and the absence of mistake. The government no longer defends the admission of the evidence for these purposes, so we do not address them here.
. The text of Rule 104(b) was slightly different at the time of Huddleston v. United States, 485 U.S. 681, 108 S.Ct. 1496, 99 L.Ed.2d 771 (1988), but the change to the current language was intended to be purely stylistic, so we quote the current text. Fed.R.Evid. 104 advisory committee's notes (2011 amendments).
. A common misconception about Rule 404(b) is that it establishes a rule of exclusion subject to certain exceptions. That's not quite right. The text of the rule does not say that propensity evidence is inadmissible except when it is used to prove motive, opportunity, intent, etc. Rather, it says that propensity evidence — other-act evidence offered to prove a person’s character and inviting an inference that he acted in conformity therewith — is categorically inadmissible. Fed.R.Evid. 404(b)(1). But the rule also acknowledges that there may be "another” use for other-act evidence — i.e., a different, non-propensity use. Fed.R.Evid. 404(b)(2). So it's technically incorrect to characterize the purposes listed in subsection (2) as "exceptions” to the rule of subsection (1). The Rules of Evidence do contain some true exceptions to the rule against propensity evidence, but they’re found elsewhere — notably in Rules 412 through 415, which are limited to sexual-assault cases. In contrast, the purposes enumerated in subsection (2) of Rule 404(b) simply identify situations in which the rule of subsection (1) by its terms does not apply.
. Rule 403 also gives the court discretion to exclude relevant evidence if its probative value is substantially outweighed by the danger of “confusing the issues, misleading the jury, undue delay, wasting time, or needlessly presenting cumulative evidence.” Fed.R.Evid 403. Other-act evidence may implicate these concerns as well, but the more common problem is the risk of unfair prejudice.
. Our dissenting colleagues object that the government did not press this interpretation of the September 4 phone call before the jury. We disagree. It's true that the prosecutor did not specifically contrast the agents' stop of Gomez (on foot) and Reyes (in his car). But the prosecutor discussed the September 4 phone call in some detail in closing argument, explaining that:
[I]n another call, Call 64171, Gomez says- — • Gomez tells Romero that Gomez had been stopped when he was walking and had turned the corner. Now it’s no coincidence that Güero, the man on this phone with Roberto Romero, is telling Romero about a stop that occurred — that occurred to him that day exactly the same way it happened to Nicolas Gomez that day. It's no coincidence because Nicolas Gomez is Güero. The context of these calls, combined with the other evidence, is just one more kind of proof that Nicolas Gomez was Roberto Romero’s partner in all of these calls.
After touching on some of the other evidence, the prosecutor returned to this point, telling the jurors that they could be confident beyond a reasonable doubt that Gomez was Güero "because of the context of the post-stop phone calls where [Güero] talks about things that only happened to Nicolas Gomez.”