Defendant was convicted of murder. On appeal, he contends that the trial court erred in permitting two eyewitnesses to make in-court identifications. Because of “serious questions concerning the reliability of the identification evidence admitted at defendant’s trial,” and “[d]ue to the novelty and complexity” of newly-articulated rules and guidelines governing the admissibility of eyewitness identifications, State v. Lawson/James, 352 Or 724, 765, 291 P3d 673 (2012), we reverse and remand for a new trial.
On New Year’s Eve 2007, two young women—D, 19, and N, 18, friends since their freshman year in high school— took public transportation from their suburban West firm homes to Portland, where they met N’s boyfriend, who lived there. D felt “out of [her] element”; N, evidently, had been in the neighborhood before. After visiting one of N’s boyfriend’s relatives, the young women joined three other people (two men and a woman) and drove to another house, where they planned to attend a New Year’s Eve party. D and N were in the back seat. When they arrived at the house at around 11:30 p.m., they noticed a group of some 25 to 50 men, all African-American, gathered outside. D and N (who are white) then noticed that some of the men were engaged in a fight. The two men in the car with D and N got out and, according to D, “vanished into the crowd.” Remaining in the car with a third woman, D and N “didn’t feel comfortable at the place” and “want[ed] to leave.” At that point, they heard a loud noise that sounded like fireworks. They saw that one of the men who had been in the fight was holding a gun in the air; he was somewhere between 12 and 25 feet from the car. He fired the gun three more times. The young women decided to drive away just as two men, one of whom was one of the original passengers and neither of whom was involved in the fight, jumped in. Before they could leave the scene, however, a man who D and N believed to be the shooter attempted to get in the car as well, but he was repelled by one of the male occupants. After driving approximately two blocks, the car was stopped by police officers, and the occupants were detained for questioning.
*691According to one of the officers, Beniga, D told him that “she didn’t see the shooting and couldn’t really describe much,” and that she “could not give specific descriptions of who was involved.” N could tell Beniga only that the shooter was an African-American man wearing a “do-rag,” had a stocky build, and was in his mid to late twenties.
The man with whom the gun-brandisher was fighting, Monette, died from four gunshot wounds in the chest. Witnesses who had been at the party provided conflicting information regarding the identity of the shooter. Several noted that he put on a ski mask before firing the fatal shots. A ski mask was subsequently found at the crime scene; it contained DNA from defendant and from another man, Porter, who, along with defendant, had been involved in an argument with the victim before it escalated into armed conflict. Porter, who had nine prior felony convictions, was for a time a suspect in the murder of Monette. Ultimately, however, he testified against defendant at trial, hoping that, by doing so, he would receive leniency in a pending federal weapons charge.
Between the night of the crime and the beginning of defendant’s trial, 23 months passed. During that time, the state did not conduct a line-up, photo array, or any other procedure in which the two young women had the opportunity to view defendant or identify him or anybody else as the shooter. Shortly before the trial, D met with the defense attorney and an investigator. According to the attorney, D, in response to questions about the shooter, was able to describe his attire and hair style (“Afro-ish or tight braids”), but was unable to provide further details because “[a] 11 black men look the same” to her. (At trial, D denied making that statement.) Also shortly before trial, the young women met with the prosecutor in his office, where they provided more detailed descriptions of the shooter. Also at that meeting, the prosecutor and D decided that, at trial, he would ask her during her direct examination to identify defendant as the shooter only if she signaled to him from the witness stand that, having observed defendant in court, she could do so.
*692The signal arrangement never occurred.1 A few moments into the state’s direct examination of D, the court experienced a power failure and had to take a 19-minute recess. During that time, D happened to see defendant in the hallway outside of the court room. She began to hyperventilate and exclaimed, “Oh, my God that’s him, that’s him, that’s him.”2 Shortly after court reconvened, the prosecutor asked her to describe the shooter and then asked, “The person that you saw with a gun in the street that you’ve described, do you see anybody—well, is that person in this courtroom?” Defense counsel objected, and a lengthy discussion among the court and the parties’ attorneys ensued out of the presence of the jury. Defense counsel explained that his objection was based on the Due Process Clause of the United States Constitution, as well as OEC 403, under which relevant evidence is not admissible “if its probative value is substantially outweighed by the danger of unfair prejudice”—although the court and counsel agreed that the OEC 403 question “is now ultimately a constitutional analysis in the circumstance.”The court concluded that the in-court identification was suggestive, but that it was not unduly prejudicial and that exclusion was not “required by the constitution.” D subsequently identified defendant as the shooter, and (over another objection by defendant) testified that her “degree of certainness” was “[l]ike a 95.” When N took the stand and was asked if she could identify anybody in the courtroom as the shooter, the defense again objected, the objection was again overruled, and she identified defendant. The jury ultimately returned a guilty verdict. This appeal ensued.
In his opening brief, defendant argued that the trial court misapplied the then-governing analysis of eyewitness identification as established by the Supreme Court in State *693v. Classen, 285 Or 221, 232, 590 P2d 1198 (1979). In that case, the court announced a two-part test:
“First, the court must determine whether the process leading to the offered identification was suggestive or needlessly departed from procedures prescribed to avoid such suggestiveness. If so, then the prosecution must satisfy the court that ‘the proffered identification has a source independent of the suggestive confrontation’ or photographic display, see Commonwealth v. Botelho, [369 Mass 860, 343 NE 2d 876, 881 (1976)] (citing cases), or that other aspects of the identification at the time it was made substantially exclude the risk that it resulted from the suggestive procedure.”
Id. (footnote omitted). “An identification is unduly suggestive if it unfairly singles out or points to a defendant as the suspect to be identified for a known crime * * *.” State v. Rector/Tremaine, 82 Or App 466, 477, 729 P2d 1 (1986), rev den, 302 Or 614 (1987). Examples include improperly constituted lineups, single or dual item photographic “arrays” or “throw-downs,” coached witnesses, on-scene identifications of a suspect handcuffed and in a police car, and so forth. See, e.g., State v. James, 240 Or App 324, 327, 245 P3d 705 (2011), aff’d, 352 Or 724, 291 P3d 673 (2012) (giving examples). According to defendant, the in-court identification was egregiously suggestive: Defendant was seated at the defense table, hence the only suspect in the room, and it was obvious that the state was singling him out and pointing to him as the guilty party. Further, defendant maintained that the in-court identification “needlessly departed from procedures,” Classen, 285 Or at 232, because the state had no legitimate reason not to have conducted a proper lineup or photographic array in the two years since the crime. Thus, defendant argued, the court should have moved to the second step, that is, a consideration of factors that would determine “whether [the] identification had been made independent of suggestive procedures ” Lawson/James, 352 Or at 737 (characterizing Classen). Such factors include
“the opportunity that the witness had at the time to get a clear view of the persons involved in the crime and the attention he or she gave to their identifying features, the *694timing and completeness of the description given by the witness after the event, the certainty expressed by the witness in that description and in making the subsequent identification, and, of course, the lapse of time between the original observation and the subsequent identification.”
Id. at 737-38 (quoting Classen, 285 Or at 232-33). Had. the trial court considered those factors, defendant argued, it would necessarily have concluded that the identifications were unacceptably unreliable. The witnesses’ observations occurred at night, neither witness could give a detailed description at the time of the event, and nearly two years had elapsed between the original observation and the subsequent identification. Only one Classen factor pointed to reliability: both witnesses expressed certainty.
The state, for its part, argued primarily that Classen does not apply to an in-court identification unless that identification has been tainted by a prior improper out-of-court identification procedure. That is so, the state argued, because Classen derives from federal Due Process Clause jurisprudence, and an in-court identification is per se free from due process concerns: It takes place in front of the defendant’s counsel and the jury, which can gauge the procedure for itself, and the witness who makes the identification is subject to cross-examination. Here, the state maintained, the in-court identifications were free from prior contamination; “[u]nder those circumstances, this case afforded defendant his due process.” Further, the state argued, each in-court identification had sufficiently reliable sources. D “had a clear and close-up view of defendant at the time of the murder, she paid attention to his physical features, [and] she expressed certainty in her identification.” According to the state, N testified that
“before the shooting, her attention was focused on defendant. * * * [He] was 20 to 25 feet away from her. But she said that defendant got closer to her as her car started to drive away from the area. She indicated that defendant came up to her car window and that she got a good look at him.”
Neither of the young women had been drinking or was otherwise impaired.
*695After this case was briefed and argued, the Oregon Supreme Court “revise [d]” the test set out in Classen. Lawson/James, 352 Or at 727. Notably, the court relegated the federally based due process aspect of Classen to backup status; like all federal constitutional principles, it would apply only if state law, fully and correctly applied, failed adequately to vindicate a person’s federally guaranteed right. Instead, questions of the admissibility of eyewitness testimony are to be decided in the first instance by reference to “applicable provisions of the Oregon Evidence Code (OEC).” Id. Initially, the proponent of the evidence has the burden of establishing that the identification is relevant, OEC 401; OEC 402, and, although eyewitness identifications are almost always relevant, the proponent nonetheless has to establish by a preponderance of the evidence that the identification derives from what the witness actually saw, OEC 602, and not from unduly suggestive procedures, OEC 701. Lawson/James, 352 Or at 761-62. Once the proponent establishes relevance, the burden shifts to the opponent, who can succeed in having the evidence excluded only by establishing, under OEC 403, that its probative value is substantially outweighed by the danger of unfair prejudice, confusion of the issues, misleading the jury, or undue delay. Id. at 762. Thus, questions involving the effect, if any, of suggestive (and therefore unreliable) identification procedures can apparently be addressed at both the first stage of the inquiry in order to determine relevance, because an identification that derives from suggestive techniques instead of witness perception is so unreliable that it is not relevant to any disputed issue, and in the second stage because an identification that derives from suggestive techniques is unreliable and hence of little probative value. At both stages, the questions present legal issues for the court.
Much of Lawson / James is devoted to a discussion, based on review of voluminous scientific research, of the variables that affect whether a particular eyewitness identification is reliable. “System variables,” those that “refer to the circumstances surrounding the identification procedure itself that are generally within the control of those administering the procedure,” id. at 740, include (as relevant to this *696case): (1) whether the procedures are conducted by an administrator who does not know the identity of the suspect (the “blinder” the administrator, the more reliable the identification), id. at 741-42; (2) whether the procedures are such that the witness is presented with a single subject for identification (“inherently suggestive,” especially when not conducted “immediately after the witness has observed a criminal perpetrator in action”), id. at 742-43; and (3) whether there has been post-event “contamination” such as suggestive questioning, id. at 743.3 “Estimator variables,” that is, “characteristics of the witness, the alleged perpetrator, and the environmental conditions of the event,” id. at 740, include (again as relevant here): (1) stress (high levels “can have a negative effect” on accuracy), id. at 744; (2) witness attention, id.; (3) duration of exposure (longer exposure results in more accuracy), id.; (4) environmental viewing conditions such as distance and lighting, id. at 744-45; (5) the physical and mental condition of the witness, id. at 745; (6) the distinctive characteristics of the suspect (witnesses “are significantly better at identifying members of their own race”), id.; (7) speed of identification (“[a]ccurate identifications generally tend to be made faster”), id.; (8) level of certainty (“witness confidence or certainty is not a good indicator of identification accuracy”), id. (emphasis added); and (9) “memory decay” (“Memory generally decays over time. Decay rates are exponential rather than linear[.]”) id. at 746.
In light of the significant differences between the Classen analysis and its “revision” in Lawson/James, and also in light of the rule that we adjudicate cases under the law as it exists at the time of our review, State v. Jury, 185 Or App 132, 136, 57 P3d 970 (2002), rev den, 335 Or 504 (2003), the parties submitted supplemental briefs. Defendant applies the system and estimator variables to the facts and, not surprisingly, concludes once again that the court erred in admitting the disputed testimony. In response, the state does not renew the primary argument that it made in its original brief; it does not argue, in other words, that in-court identifications are per se admissible unless tainted by a prior *697improper or suggestive out-of-court identification. That argument was wisely abandoned. It derived from the fact that Classen was based on due process concerns, and reasoned that, to the extent that untainted in-court identifications are themselves suggestive, they are nevertheless not fundamentally unfair because they take place in front of the fact-finder and defendant’s counsel. Lawson/James, in contrast, focuses on the reliability of the identification under the rules of evidence. Although the underlying presumption is apparently that a reliable identification is fair while an unreliable one is not, that presumption would appear to be more appropriately addressed in a separate and subsequent due process analysis in the event that the identification is found to be reliable under the rules of evidence. Reliability is determined with reference to the system and estimator variables, and there is no reason to presume that those variables necessarily differ as between in-court and out-of-court identifications. Thus, the state, like defendant, correctly focuses its supplemental brief on those variables and how they apply to the facts of this case. We do the same, and we conclude that defendant has the better argument.
Here, the system variables weigh heavily against reliability. The state used only one identification “procedure,” the in-court identification, and it was substantially the same with each of the witnesses. The court found that procedure to be suggestive, albeit not so much so as to require suppression, and the state does not disagree. “Suggestive,” however, is an understatement. The only system variable associated with the in-court identification that would support reliability was the pretrial instruction to the witnesses; the prosecutors told each witness that she did not have to identify defendant if she did not, in fact, recognize him. Yet even with respect to that variable, neither witness was advised of the possibility that she could positively conclude that defendant was not the shooter, or what she should do if that were the case. Meanwhile, on the other side of the scales, the procedure was not “conducted by a ‘blind’ administrator—a person who does not know the identity of the suspect.” Lawson/James, 352 Or at 741. There was “suggestive questioning.” Id. at 743. The prosecutor asked D, ‘You had described an individual that had a gun. And my question [is]: [The] *698individual that had a gun, do you see that person in this courtroom today?”, to which the witness responded, ‘Yes, sir.” He later asked N, “The person that you saw that did the shooting * * * and that then approached the car, * * * is that person—do you recognize anybody meeting that definition * * * in this courtroom today?”, to which N answered, ‘Yes.” In each case, the witness then pointed out the only person in the room who was neither an officer of the court nor a spectator—the man sitting accused at the defense table.
Most significantly, the procedure was similar to, but significantly more suggestive than, a “showup,” which is “inherently suggestive * * * because the witness is always aware of whom police officers have targeted as a suspect.” Id. Defendant was obviously that man; moreover, it was also obvious that the state’s prosecutorial apparatus had confirmed the police officers’ initial identification. Further, “[a] showup is most likely to be reliable when it occurs immediately after the witness has observed a criminal perpetrator in action.” Id. Here, the in-court “showup” occurred two years after the crime.
The estimator variables do little to increase the reliability of the in-court identifications. The variables are not identical with respect to each of the witnesses, so we address each witness in turn. Supporting reliability, D testified that her attention was focused on the shooter, both shortly after he fired the shots and when he approached the car. Her perception was not impaired by alcohol or anything else. And she was able to give a physical description of the shooter, albeit one that was couched in racially stereotypical terms (“close Afro or braids,” “broadish nose and big lips”).
On the other hand, D was highly stressed at the time of the crime, and “[h]igh levels of stress can have a negative effect on a witness’s ability to make accurate identifications.” Id. at 744. She testified that, from the very beginning of the excursion into the Portland neighborhood where the events took place, she felt “out of [her] element” because she was “new to the area.” When she arrived at the location of the party, she was left in the car with only two other women; for that reason, she and N “didn’t feel comfortable at the place and * * * wanted to leave.” She agreed that the incident was *699“traumatic” and “very upsetting.” The entire incident took place quickly; the car from which the witnesses viewed the shooting left the scene quickly, and, although D testified that the man she subsequently identified as defendant put his face against the car window, he was quickly repelled. The incident occurred at night. D is white and defendant is African American; “Witnesses are significantly better at identifying members of their own race than those of other races.” Id. at 745.
Most significantly, the in-court identification occurred two years after the crime, with no intervening opportunity to observe defendant. Lawson/James tells us not only that “[m]emory generally decays over time,” but that “[d]ecay rates are exponential rather than linear.” Id. at 746. Compounding the significance of this time gap is the fact that, in the immediate aftermath of the crime, D told one of the investigating police officers that “she didn’t see the shooting and couldn’t really describe much” and that she “could not give specific descriptions of who was involved.” Thus, to credit D’s in-court testimony, including her in-court identification as well as her testimony regarding what she saw at the scene of the crime, a factfinder would have to not only disregard the scientific precepts adopted in Lawson! James, but to turn them upside down by believing that memory improves over time.4 At the least, the time gap completely undermines D’s level of certainty, which she estimated as “a 95” (presumably out of 100). Even under otherwise favorable circumstances, “witness confidence or certainty is not a good indicator of identification accuracy,” and it “has substantial potential to influence jurors.” Id. at 745.
The reliability of N’s testimony is undermined by most of the same variables that undermine D’s. Like D, N was under stress at the time of the crime; she testified that she “was very upset” by what she had observed. The circumstances of her observation at the crime scene were the *700same as D’s: a rapid succession of startling events, at night. Because her boyfriend was African American, she may not have been as susceptible to “own-race bias,” id., as D. But her testimony and identification of defendant, like D’s, occurred two years after the event, and must be seen in the context of her statement to Beniga in the immediate aftermath of the event that “the only description [she] could provide then was that the shooter was a black male, had a stocky build, and [she] guessed he was in his mid to late 20s.”
In sum, the facts in this case closely resemble those in State v. Wesley, 254 Or App 697, 716, 295 P3d 1147 (2013), where we concluded:
“[T]he identification in this case is problematic under Lawson / James for several reasons, including, most notably, that the eyewitness maintained before the suggestive procedure that he was unable to identify the perpetrators, the eyewitness provided no description at all of the perpetrators before the suggestive procedures, and the eyewitness’s opportunity to view the perpetrators was relatively brief and occurred while he was attempting to avoid being shot. In light of those concerns, we conclude that a new hearing on the admissibility of the challenged eyewitness identification is required, based on the considerations prescribed in Lawson / James.”
We reach the same conclusion here.5
Reversed and remanded.
As the state acknowledges, if the signal arrangement had occurred and “the witness [had been] unable to identify anyone in the courtroom, that procedure could have resulted in exculpatory evidence that would have had to have been disclosed to the defense.” We are unable to discern any reason why, if the prosecutor intended to disclose the nonsignal to the defense, he would have arranged the secret signal process in the first place.
The prosecutor disclosed this encounter to the court and defense counsel during an in camera conversation. The disclosure was not testimony, and the jury never heard it.
Other “system” factors, not relevant to this case, deal with the mechanics of line-ups, photo arrays, and mug shot viewings, none of which occurred here.
D herself acknowledged the corrosive effect that time has on memory. When asked on cross-examination if she recalled parts of the conversation she had with a police officer immediately after the crime, she replied, “It was two-and-a-half years ago, ma’am. I do not remember that I said that to the detective that night, but if I did, then he probably did say that.”
We recognize that in-court identifications are a staple of criminal trials, and we do not hold or even imply that they are per se unreliable. This case deals with one particular trial and the unusual situation in which, among other things, there was a significant time lapse, an inability to provide particular contemporaneous descriptions or details, and no intermediate, nonsuggestive out-of-court identification.