DUNCAN, J.
Law-enforcement officers conducted a warrantless search of defendant’s car based, in part, on an alert by a drug-detection dog. The officers found methamphetamine and marijuana in the car, and the state subsequently charged defendant with one count of possession of methamphetamine, ORS 475.894, and one count of possession of less than one ounce of marijuana, ORS 475.864(3).1 Defendant filed a motion to suppress the results of the search, arguing, inter alia, that the search violated Article I, section 9, of the Oregon Constitution because it was not supported by probable cause.2 The trial court denied the motion. Thereafter, defendant entered conditional guilty pleas, and the trial court entered a judgment convicting defendant of the two possession counts.
Defendant appeals the trial court’s judgment, assigning error to the court’s denial of his motion to suppress. On appeal, defendant renews his argument that the law-enforcement officers did not have probable cause to search his car. He argues that the state failed to establish that the drug-detection dog’s alert was reliable and, therefore, the alert could not factor into the probable cause calculation; he further argues that, without the alert, the officers did not have probable cause. We agree with defendant and, therefore, reverse and remand.
When reviewing a trial court’s denial of a defendant’s motion to suppress, we are bound by the trial court’s findings of fact, provided that the findings are supported by constitutionally sufficient evidence. State v. Ehly, 317 Or 66, 75, 854 P2d 421 (1993). If the court does not make explicit findings on disputed issues of fact, we presume that the court made findings consistent with its ultimate ruling. Id. [695]*695Stated in accordance with those standards, the relevant facts are as follows.
On February 5, 2010, at approximately 6:45 p.m., Oregon State Police troopers Routt and Jackson were in a marked patrol car parked on an overpass above Interstate Highway 84, watching for drivers violating the speed limit. They noticed a car drive up the exit ramp from the highway, stop at the top of the ramp, and park nearby. The driver got out and walked to the front of the car, where he remained for one to two minutes. The troopers could not tell what the driver was doing; it was dark out and they could see only the driver’s silhouette as he stood in front of the car’s headlights. The driver got back into the car and drove down the entrance ramp to return to the highway. At that point, the troopers saw the headlights of a second car turn on. The second car, which was near where the first car had parked, followed the first car down the entrance ramp to the highway.
The troopers decided to follow the cars, which were headed east toward North Powder. While following the cars, the troopers radioed the state police dispatch center and asked a dispatcher to conduct record checks based on the cars’ license plates. The dispatcher informed the troopers that the first car was registered to Cody Sorey and the second car was registered to defendant. The dispatcher also informed the officers that neither Sorey nor defendant had outstanding warrants for their arrest.
The troopers initially focused their attention on Sorey’s car, but did not observe any traffic-code violations. They then turned their attention to defendant’s car and noticed that defendant’s license plate was not illuminated as required by the traffic code and his windows appeared to be tinted darker than allowed by the traffic code.
Routt, who was driving the patrol car, stopped defendant’s car for the two violations. Both troopers got out of the patrol car and walked toward defendant’s car. As Routt approached defendant’s car, he saw that there were three people inside: the driver, a front-seat passenger, and a backseat passenger. Defendant was the driver, and Routt spoke to him through the front passenger-side window. Routt asked [696]*696for, and received, defendant’s driver’s license. According to Routt, defendant was polite and cooperative. However, defendant’s front-seat passenger appeared nervous.
As he later testified, Routt smelled a “very strong odor coming from the car.” The odor smelled “like several different odors mixed together.” Routt was “able to faintly distinguish * * * what [he] believed to be the odor of marijuana.” He “couldn’t tell *** what the other odors were.” Jackson testified that he smelled a “strong odor” coming from defendant’s car. Jackson could not tell “what it was”; he could not tell “if it was marijuana.” According to Jackson, “It was just a strong odor coming from the vehicle.”
Routt took defendant’s license back to the patrol car to “[run] a driver’s license check and [write] the citation.” He contacted the dispatch center and asked a dispatcher to check defendant’s criminal history for narcotics activity. The dispatcher informed Routt that defendant had a prior drug conviction. The dispatcher did not specify the nature or date of the conviction.
While Routt was in the patrol car, Jackson went to the front passenger-side window of defendant’s car to speak with defendant and his passengers. Jackson asked what they were doing, and defendant explained that they had met with Sorey and were on the way to Sorey’s house in North Powder. Jackson noticed that defendant had sores on his face consistent with methamphetamine use. Jackson also noticed that the front-seat passenger was fidgeting and that, during his questioning, she put her foot on a metal case. According to Jackson, the passenger “was nervous and she seemed to be nervous about [the] case.”
After speaking with defendant and his passengers, Jackson spoke to two other law-enforcement officers by phone and learned that defendant had a pending charge for distribution of marijuana. One of the officers told Jackson that defendant “dealt in marijuana and methamphetamine.” At that point, which was about 10 to 12 minutes into the traffic stop, Jackson believed that he had reasonable suspicion that defendant was engaged in drug activity. Jackson asked defendant to consent to a search of his car, and [697]*697defendant refused. Jackson then called a K-9 officer, Trooper McDowell, to come to the scene with his drug-detection dog, Mauri. According to McDowell, it is state police policy not to use a drug-detection dog unless there is reasonable suspicion that drugs are present.
McDowell arrived at the scene about 15 to 20 minutes later. He circled defendant’s car with Mauri, who “alerted” to “the driver’s side * * * seam between the door and the body of the car.” Based on Mauri’s alert, McDowell believed there was probable cause to search defendant’s car. McDowell relayed that information to Jackson, and the troopers searched the car and found methamphetamine and marijuana.3
Defendant filed a motion to suppress “all evidence obtained * * * following [the] unlawful extension of [the traffic] stop * * * [and the] unlawful warrantless search.” In the motion, defendant asserted that the troopers unlawfully seized him by expanding the scope of the stop to include a drug investigation that was not supported by reasonable suspicion. Defendant also asserted that McDowell’s deployment of Mauri was a warrantless search that was not supported by probable cause. Defendant later filed a written addendum to his motion, asserting that “the drug dog sniff * * * was unreliable and therefore did not establish probable cause to search the * * * vehicle.”
At the hearing on defendant’s motion to suppress, the state presented evidence about the reliability of Mauri’s alert to defendant’s car, including testimony from McDowell about the training he and Mauri had received. According to McDowell, Mauri was trained by a master trainer, Christina Bunn, who owns Puget Sound Security. Bunn trained Mauri by saturating a towel with an odor, hiding the towel, having Mauri find the towel, and then playing with Mauri as a reward. Mauri was trained to detect marijuana, methamphetamine, cocaine, and heroin.
McDowell began working with Mauri in October 2009, after Mauri had been trained by Bunn. McDowell [698]*698attended a three-week training course to learn, as he testified, “what to look for, what changes of behavior in [Mauri] * * * tell me that she is detecting the presence of a controlled substance.” According to McDowell, an “alert” has two components: an “area alert” and a “specific alert.” An area alert is a change in behavior, such as “[a] head snap, change of direction, tail wagging, intense sniffing, and closed mouth breathing.” A specific alert is a particular behavior— scratching — which Mauri does “on the controlled substance or [as] close to it as she can get.” An alert requires both an area alert and a specific alert. When Mauri alerts, McDowell rewards her by giving her a clean towel as a toy and playing with her. McDowell explained that
“the ultimate goal is when she alerts to a vehicle — when she scratches on an object, whether it’s a vehicle, a locker, whatever the case may be, it’s her belief that she’s scratching at the odor and the towel is coming from the scratch, not from me circling and throwing the towel.”
McDowell and Mauri were initially certified as a dog-handler team in November 2009, less than four months before the February 2010 traffic stop of defendant. McDowell testified that, in order to maintain the certification, he is required to train with Mauri for four hours each week. In addition, McDowell testified that there are “quarterly trainings where the whole K-9 team across the state gathers *** [to] train as a team[,]” but McDowell did not specify whether he and Mauri had been to any such trainings before the traffic stop of defendant. McDowell also testified, “Every time [Mauri has] alerted to the presence of a controlled substance in the field, during the search we have found a controlled substance. So her accuracy record is 100 percent.” But, McDowell did not specify how many times Mauri had alerted in the field.
Regarding Mauri’s use in this case, McDowell explained that he and Mauri circled defendant’s car while Mauri was on a leash. McDowell’s practice is to use the leash to “detail [] [Mauri] to certain parts of the vehicle.” He “[doesn’t] just run around holding the leash.” Instead, he uses the leash to “focus [Mauri] on different areas of the vehicle. Detail her to a door seam. Detail her to a trunk [699]*699seam. Door handles, whatever the case may be.” When asked whether he had detailed Mauri to the area where she alerted to defendant’s car, McDowell answered that it was his job to do so and he hoped that he had “detailed her right to that door seam.”
According to McDowell, when he and Mauri circled defendant’s car, Mauri’s “area alert” was “intense sniffing, closed mouth breathing after she came out of the driver’s side door.” After she sniffed the driver’s-side door seam, she “backed out and gave a specific alert.” McDowell testified that, after Mauri’s specific alert, “[s]he was rewarded. She had her towel reward. We played a little bit and she went back in the car.”
After the hearing on defendant’s motion to suppress, the trial court issued a letter opinion denying the motion. In its opinion, the court ruled that the troopers had reasonable suspicion to extend the traffic stop into a drug investigation based on the “drug crime ‘indicators’” that they had observed.4 The court also ruled that, after Mauri alerted to defendant’s car, the troopers had probable cause to search defendant’s car. The court explained:
“The drug search by the dog did not involve any entry into the vehicle until there was ‘probabl[e] cause’ of illegal drugs following the ‘hit’ by the drug dog. Everything prior to the drug dog search was proper. The actual entry of the vehicle was based on exigent circumstances, thus a warrant was not required. The drugs were properly seized.”
As mentioned, after the trial court denied his motion to suppress, defendant entered conditional pleas of guilty, reserving the right to appeal the court’s denial of his motion. See ORS 135.335(3).5 On appeal, defendant argues [700]*700only that the trial court erred in concluding that the troopers had probable cause to search his car.6 Specifically, defendant argues that the trial court erred because the “state failed to establish a sufficient record of [Mauri’s] reliability.” In response, the state argues that defendant did not preserve his appellate argument, but even if he did, the state established that Mauri’s alert was reliable.
We first turn to the state’s preservation argument. As mentioned, defendant filed a motion to suppress, as well as an addendum to that motion in which he specifically asserted that “the drug dog sniff * * * was unreliable and therefore did not establish probable cause to search the * * * vehicle.” At the hearing on defendant’s motion, the prosecutor objected to the addendum on the ground that it was untimely. The trial court overruled the objection. The state then presented its evidence, including McDowell’s testimony about the training he and Mauri had received. When the prosecutor asked McDowell about Mauri’s field performance, defense counsel objected, and the prosecutor responded that the question was relevant in light of “[t]he addendum to the defendant’s motion to suppress.” The trial court agreed with the prosecutor, telling defense counsel, “[y]ou’ve raised the reliability of the dog[.]”
We conclude that defendant preserved his appellate argument by filing the addendum to his motion to suppress. As both the prosecutor and the trial court recognized, the addendum raised the issue of Mauri’s reliability.
Having concluded that defendant preserved his appellate argument, we turn to the merits. As noted, defendant asserts that the troopers’ search of his car violated Article I, section 9, of the Oregon Constitution, which prohibits unreasonable searches and seizures. The troopers did not have a warrant to search defendant’s car, and a warrantless search is per se unreasonable unless it falls within “one of the few specifically established and well-delineated exceptions to the warrant requirement.” State v. Davis, 295 Or 227, 237, 666 P2d 802 (1983) (internal quotation marks [701]*701omitted). The state has the burden of proving that a warrantless search falls within such an exception. Id. Whether the state has carried that burden is a question of law, which we review for errors of law. Id. at 238.
One exception to the warrant requirement is the “exigent circumstances exception,” under which law-enforcement officers may conduct a warrantless search when both probable cause and exigent circumstances exist. State v. Meharry, 342 Or 173, 177, 149 P3d 1155 (2006). “Exigent circumstances include, among other things, situations in which immediate action is necessary to prevent the disappearance, dissipation, or destruction of evidence.” Id. The “automobile exception” is “a subset of the exigent circumstances exception” under which the “mobility of a vehicle, by itself, creates an exigency.” Id. Under the automobile exception, a law-enforcement officer may search a lawfully stopped vehicle if (1) the officer has probable cause to believe that the vehicle contains contraband or evidence of a crime and (2) the vehicle was mobile at the time “the police encounter it in connection with a crime.” State v. Brown, 301 Or 268, 277, 721 P2d 1357 (1986); State v. Kurokawa-Lasciak, 351 Or 179, 192, 263 P3d 336 (2011).
In this case, the state argues that the troopers’ warrantless search of defendant’s car was justified by the automobile exception. Defendant does not dispute that his car was mobile when the troopers stopped him for the traffic violations. Thus, the general issue on appeal is whether the troopers had probable cause to believe that defendant’s car contained contraband or evidence of a crime.
The state argues that Mauri’s alert, when combined with other information that the troopers had about defendant, gave rise to probable cause to search defendant’s car. The state does not argue that the other information was sufficient to give rise to probable cause in and of itself. Thus, the specific issue on appeal is whether the state established that Mauri was sufficiently reliable for her alert to contribute to the conclusion that there was probable cause to search defendant’s car.
[702]*702As the Supreme Court held in State v. Foster, 350 Or 161, 177, 252 P3d 292 (2011), and its companion case, State v. Helzer, 350 Or 153, 156, 252 P3d 288 (2011), an alert by a properly trained and reliable drug-detection dog can provide probable cause to search. “Whether a particular alert by a particular dog provides probable cause is an issue that requires an individualized inquiry, based on the totality of the circumstances known to police, which typically will include such considerations as the dog’s and its handler’s training, certification, and performance.” Foster, 350 Or at 163.
Foster and Helzer illustrate how a court is to assess the reliability of an alert by a drug-detection dog. In each case, the defendant moved to suppress evidence that law-enforcement officers obtained as a result of a warrantless search of his car, arguing that the officers, who searched based on an alert by a drug-detection dog, did not have probable cause to search because the alert was not sufficiently reliable. As a result of the different evidentiary records in the cases, the Supreme Court held that the alert in Foster was sufficiently reliable to provide probable cause, but the alert in Helzer was not. In the course of doing so, the court described the potential for erroneous drug-detection-dog alerts, the type of training and testing necessary to protect against erroneous alerts, and the evidence that the state must present to establish that a dog-handler team has successfully completed that type of training and testing. Because Foster and Helzer provide guidance for assessing the reliability of an alert by a drug-detection dog, we describe them in some detail.
In Foster, the challenged alert was by a drug-detection dog named Benny. Benny was a certified drug-detection dog, but, as the Supreme Court noted, the state did not rely “on the bare fact that Benny was ‘certified’ as establishing, even as a prima facie matter, Benny’s reliability.” 350 Or at 171 n 6. Given the “the lack of any statutory standards for or official state oversight of the certification process,” the court held that “ [t] o meet its burden to establish probable cause, the state thus had to produce evidence of Benny’s training and [the Oregon Police Canine Association’s (OPCA’s)] [703]*703certification program.” Id. To do so, the state presented testimony from Benny’s handler, Officer Hulke, and a master dog trainer, Officer Fyfe, who was employed by the OPCA, an organization that tests and certifies drug-detection dogs. Id. at 164. The witnesses described how Benny and Hulke had been trained initially, how they had been tested and certified, and how they continued to train after being certified. Id. at 165-66.
Fyfe selected Benny from an animal shelter for training. Id. at 167. Benny was then paired with Hulke, and Benny and Hulke completed 70 hours of initial training, using the “play-reward” method. Id. According to Hulke and Fyfe, such training
“involves pairing a dog with a specific handler. The dogs are trained to detect heroin, cocaine, methamphetamine, and marijuana. Initially, the handler exposes the dog to a training aid such as a tennis ball that has been submerged in the drug and familiarizes the dog with the odor by playing with it. Then the trainer hides the ball, and the dog learns to sniff it out. Next, the trainer hides the drug rather than a tennis ball. When the dog finds the drug, it is rewarded by being allowed to play with a favorite toy.”
Id. at 165 (internal quotation marks omitted).
Hulke trained Benny in “a variety of settings, such as vehicles, schools, and wineries, where he introduced distractions, such as other animals or animal feces.” Id. at 167. Sometimes, Hulke had someone else hide the training aid “to avoid cuing Benny to the aid with his own scent.” Id. Hulke received training assistance from a master trainer certified by the OPCA. Id.
After their initial training, Hulke and Benny passed the certification test on their second attempt. Id. As described by Fyfe and summarized by the Court of Appeals,
“[T]he test involves testing the dog’s ability to detect the presence of drug-scented items in two rooms, three vehicles, seven packages, and an open area. To prevent handler cuing and to force the teams to perform under stress, the handler does not know how many items, if any, the dog should alert to on each deployment. Each environment may have up to three drug packages hidden in it, or may have [704]*704none at all. The environments also have distractors, such as dirty clothing and urine. To pass the portion of the test in rooms and vehicles, a dog must have at least a 90 percent accuracy rate. For the packages and open area, the dog must be 100 percent accurate. Approximately 25 percent of the dogs fail the test. If a dog passes the test, it is certified for one year and must complete the test again to be recertified.”
Id. at 165-66 (brackets and internal quotation marks omitted). Thus, the test required that the dog-handler team demonstrate an ability to locate drugs in a variety of settings with a high rate of accuracy, and the test was designed to protect against the possibility that dogs would locate the hidden drug based on cues from their handlers.
After passing the test, Hulke trained Benny for an additional 230 hours before using Benny in the field. Id. at 168. Hulke continued to train Benny, with the assistance of a master trainer from the OPCA, and he and Benny were recertified approximately two weeks before the alert at issue in Foster. Id. at 168. According to Hulke’s records, 66 percent of Benny’s alerts in the field had resulted in searches that led to the discovery of seizable items. Id.
For his part, the defendant in Foster presented evidence challenging both (1) whether an alert by a drug-detection dog can give rise to probable cause to search, given that dogs can detect residual odors of drugs and, if it can, (2) whether Benny’s alert was sufficiently reliable, given how he was trained.
In support of his first challenge, the defendant presented expert evidence that “a dog is capable of detecting trace amounts and residual odors of a drug that may remain after the odor-emanating drug is no longer present, or that may be carried by an object or a person who had contact with drugs in another location,” and, as a result, “when a dog alerts to a location and a subsequent search of that location does not result in the seizure of drugs or drug paraphernalia, there is no way to determine whether the dog alerted to a residual odor or whether the alert was a result of dog or handler error.” Id. at 167. Based on that evidence, [705]*705the defendant argued that an alert by an odor-trained dog “does not signify the presence of drugs; it signifies only the presence of the odor of drugs” and, therefore, an “officer cannot have probable cause to believe drugs will be found in a search of the area to which the dog alerted.” Id. at 172.
The Supreme Court disagreed, rejecting the defendant’s argument that “a dog trained to alert to drugs based on their odor, rather than the actual presence of drugs, is inherently too unreliable.” Id. at 171, 173. The court concluded that the fact that a dog is capable of detecting residual odors “does not ipso facto render it unreasonable to believe that drugs or other seizable things are probably present.” Id. at 173. (Emphasis in original.) Observing that probable cause “does not require certainty,” id. at 172, the court explained,
“First, if the dog is properly trained and handled, the likelihood that the dog’s alert indicates the presence of an illegal drug remains a substantial one. Second, and significantly, even if the odor is a residual one only, there is no substantial — let alone equal or greater — likelihood of a purely innocent explanation for the presence of the odor of drugs. Either illegal drugs are present, or something that was in contact with illegal drugs and carries the odor is present.”
Id. at 173. Therefore, the court concluded that “probable cause to search may arise from the alert of a trained drug-detection dog despite the possibility that the alert is to a residual odor of an illegal drug rather than an odor emanating from the actual drug.” Id. at 174 (emphasis in original).
In support of his second challenge — viz., that Benny’s alert was not sufficiently reliable, given Benny’s training — the defendant disputed the effectiveness of the play-reward method used to train Benny. The defendant presented an expert witness, Woodford, who testified that the play-reward method was inferior to another method, the imprint method,7 and that it did not adequately protect against the possibility that a dog would alert based on [706]*706human cues, as opposed to drug odors. Id. at 166. Woodford suggested that
“a major flaw with the play-reward system is that the dog may simply be detecting the scent of the handler who handled the drugs when they were hidden, or may be picking up on subtle cues from the handler, such as changes in heart rate or breathing patterns when the dog is near the place where the handler has hidden the drugs.”
Id. (internal quotation marks omitted).
In response to Woodford’s criticism, the state presented additional testimony from Fyfe, the OPCA master trainer, who explained that, although the play-reward training and testing involves the hiding of training aids by humans,
“during play-reward testing, a dog’s handlers do not place the drugs in the environments. Moreover, the people who do place the drugs in the environments handle the drug packages with gloves and also go to various points in the environments and disturb the environments, so that ‘our dogs are able to go out there and not alert to skin, crushed grass, disturbed earth where we bury the drug training aid[.]’ Fyfe also testified that the play-reward system is the only system in general use in the United States for training drug-detection dogs.”
Id. at 167 (some internal quotation marks omitted; brackets in original).
The Supreme Court rejected the defendant’s arguments that were based on Woodford’s testimony. First, with respect to Woodford’s testimony that the play-reward method was inferior to the imprint method, the court pointed out that “[probable cause] does not require an officer to rely on the best or most reliable source of information if some other, but still reliable, source provides the officer with the requisite reasonable belief.” Id. at 174. Second, with respect to Woodford’s testimony that the play-reward method did not adequately protect against alerts based on human cues, the court pointed out that Woodford’s criticism was directed at the play-reward method in general and, as a result, “was not dog specific or otherwise directed specifically to Benny’s reliability or the particular training that Benny had received.” Id. [707]*707at 175 n 9. According to the court, “It did not account for the fact that the training * * * [had] been designed to avoid, or at least minimize, a major flaw that Woodford identified in play-reward training {i.e., the scent and physical cues that a handler may give to a dog when the handler is involved in hiding the drugs used in training).” Id. at 175.
The Supreme Court also discussed Benny’s “approximate 66 percent ‘find’ rate — that is, his field performance records show[ing] that actual drugs were found in 66 percent of the areas searched after he alerted.” Id. at 176. The court concluded:
“The fact that no actual drugs were found in 34 percent or so of Benny’s deployments is, in the absence of other information, a neutral consideration. It is impossible to know in those circumstances whether Benny detected the residual odor of an illegal drug (a correct alert, but not one that led to the successful recovery of evidence of drugs) or whether, as defendant suggests, Benny did not alert to an odor at all, but alerted instead in response to a handler cue or a ‘Pavlovian’ desire for a reward (a ‘false’ alert).”
Id. at 176-77. Accordingly, the court observed, “field reports should not be the full measure — or perhaps even the most meaningful measure — of a dog’s reliability; more telling is the dog’s performance in controlled circumstances, such as those involved in testing for certification, where the dog’s ability to find and signal the presence of drugs can be accurately gauged.” Id. at 177 (citing State v. Nguyen, 726 NW 2d 871, 878 (SD 2007)); see also Florida v. Harris,___US___, ___, 133 S Ct 1050, 1056, 185 L Ed 2d 61 (2013) (“[I]n most cases [,]” records of a drug-detection dog’s field performance “have relatively limited import [,]” and “[e]rrors may abound in such records.”).8
[708]*708In sum, in Foster, the record established that Hulke and Benny (1) “went through a thorough training regimen,” 350 Or at 176; (2) successfully completed a certification test that required Benny to locate hidden drugs in a variety of settings, with a high rate of accuracy, and under circumstances that were designed to reduce the possibility that Benny would locate the drugs based on handler cues or scent or on physical trails left by the humans who hid the drugs; (3) continued to train after they were certified; and (4) were recertified shortly before the alert at issue. It also established that Benny had a “‘find’ rate” that “confirm[ed] that [he could] accurately detect the odor of drugs present in an environment, as he was trained to do.” Id. Accordingly, the Supreme Court concluded that, “based on the totality of the circumstances bearing on Benny’s particular reliability” in the case, “Benny’s alert to defendant’s car gave the officers probable cause to search it for illegal drugs or other seizable evidence of illegal drug activity.” Id. at 178.
The Supreme Court reached the opposite conclusion in Helzer, which involved an alert by a drug-detection dog named Babe. As the court described it, the evidentiary record regarding Babe’s reliability was “sparse.” Helzer, 350 Or at 156. “[T]he state relied exclusively on the testimony” of Babe’s handler, Stokoe, “to describe how Babe had been certified and trained in drug detection, and how Stokoe had learned to work with Babe as her handler.” Id.
At the hearing on the defendant’s motion to suppress, Stokoe testified that Babe had been trained by “a private business or organization, ‘Code Three Canine’” before he acquired her. Id. Stokoe did not participate in that training and did not testify about the nature or details of that training. After Stokoe acquired Babe, he completed a two-week training program with an instructor from Code Three [709]*709Canine in order to “learn how to work with Babe and how to train her to maintain and improve her skills.” Id. at 157. But, Stokoe did not provide any details about that training or the requirements that had to be met to complete it successfully:
“Stokoe was not asked to describe the details of that training, either in terms of what he was taught about handling Babe to maximize her accuracy and reliability, or what he learned through that training about Babe’s reliability. He testified only that he and Babe, to complete the training successfully, had to ‘meet the standards of the trainer at the time.’”
Id.
After the two-week training program, Stokoe trained Babe for at least 300 hours over a period of approximately six months. Id. During that time, Stokoe did not work with any formal trainers, and, “[w]ith the exception of about 10 days on which other dog-handler officers or members of his family assisted him, Stokoe worked with Babe entirely on his own.” Id.
According to the Supreme Court, Stokoe testified about his training of Babe “in general terms only.” Id. As summarized by the court, he testified that,
“[i]n his training, he rewarded Babe for an alert by playing with her using a rolled-up cloth. He said that he used ‘training aids’ to teach Babe to find drugs. Those training aids involved use of ‘a sample of the drug’ that ranged in amounts from ‘residual odors up to a pound.’ Stokoe also used ‘blanks,’ which consisted of empty plastic containers, to ensure that Babe alerted to drugs and not the odor of the plastic containers that he used to hide training aides. The blanks that Stokoe used included ‘blank’ vehicles — that is, vehicles that Stokoe believed would have no odor — so that he could be sure that Babe was not alerting to vehicles ‘just because she’s used to alerting to vehicles.’ And Stokoe described using various foods to make sure that Babe did not alert to the scent of food rather than to that of drugs. Although Stokoe kept records of his training using drug samples, he kept no records of Babe’s responses on blank containers and vehicles. He trained with Babe ‘wherever we [could] find a place to train at.’”
[710]*710Id. “After that self-directed training period, Stokoe and Babe attended a four-day (32 hour) training session with the [OPCA].” Id. at 157-58. Stokoe “gave no details or even a generalized description” of the OPCA training. Id. at 158. After the training, Stokoe and Babe were certified by the OPCA, but Stokoe did not describe the standards that the OPCA applied for the certification. Id.
In addition to Stokoe’s testimony, the state presented Stokoe’s records of Babe’s performance in the field over a period of several months. Stokoe kept records of when Babe was deployed and alerted, but not of when she was deployed but did not alert. Id. at 158.
At the outset of its analysis of the sufficiency of the state’s evidence, the Supreme Court stated, as it had in Foster, that,
“[t]o assess the dog’s and handler’s abilities based on their training and certification, *** more is needed than the fact that the two have received certification by a private organization. The record must provide information about the training that the dog and handler underwent, and the standards they had to meet to achieve certification.”
Id. Applying that requirement, the court concluded that the state’s evidence fell short.
In reaching its conclusion, the Supreme Court pointed out that the state had failed to present specific information about the training provided by Code Three Canine, either before or after Stokoe acquired Babe. Id. at 158-59. The court also pointed out that, although Stokoe provided some information about his self-directed training, in that he “testified vaguely to his use of blanks and food distraction,” “he provided no information beyond that to explain how his training builds accuracy and reliability in both Babe’s abilities and his handling of Babe.” I'd. at 159. Regarding Stokoe’s records of Babe’s performance in the field, the Supreme Court reiterated its holding in Foster-. A dog’s performance during controlled tests is more telling than a dog’s field performance because, in the field, it is difficult to determine if a dog has failed to alert to an area when it should have (because the area is unlikely to be searched) or has alerted [711]*711to an area when it should not have (because it is impossible to determine whether an alert that does not result in the discovery of drugs was based on no odor or a residual odor). Helzer, 350 Or at 159. The court then concluded that, given the limited information in the record about Babe’s training and certification, “the field performance records * * * [were] insufficient to establish that Babe was sufficiently reliable to provide probable cause to search.”9 Id. at 159-60.
Together, Foster and Helzer establish several principles relevant to this case: (1) whether a particular alert by aparticular drug-detection dog is reliable must be determined on a case-by-case basis; (2) the factors relevant to that determination include the dog-handler team’s training, testing, and certification; but (3) the simple fact that a team has been trained, tested, and certified is not enough to establish that an alert is reliable; rather, the type of training, testing, and certification matters. That is because, as Foster and Helzer illustrate, a dog-handler team must be trained in a manner that ensures that the dog alerts in response to drug odors, as opposed to, for example, a desire for a reward, nondrug odors, or handler cues or physical or scent trails left by the person who hid the drugs. Similarly, a dog-handler team must be tested in a controlled environment, where precautions against human cuing have been taken and the dog’s accuracy can be assessed because the persons conducting the test know where the dog should alert and where it should not.
Foster and Helzer also establish that the value of a dog’s field records “may depend on whether their significance is sufficiently developed through testimony at the hearing or is self-evident[,]” Helzer, 350 Or at 160 n 4, and that, in all events, the value of field records is limited because it is unlikely that either false positives or false negatives will be detected in the field, Foster, 350 Or at 177; Helzer, 350 Or at 159.
[712]*712With those principles in mind, we return to the record in this case. As described above, the state’s evidence about the reliability of the alert in this case was limited to McDowell’s testimony.
McDowell testified that he had acquired Mauri after she had been trained by a private entity, Puget Sound Security. McDowell did not participate in that initial training, and he described it in general terms only. McDowell’s description of Mauri’s initial training is similar to the description of the towel-reward training used by the handler in Helzer, and it has the same shortcoming: It provides no information about how the training “builds accuracy and reliability” in the abilities of the dog-handler team. 350 Or at 159.
After Mauri’s initial training, McDowell — like the handler in Helzer — attended his own training with Mauri, but — again like the handler in Helzer — McDowell did not provide any details about that training or the requirements, if any, that he and Mauri had to meet to complete it successfully. McDowell testified only that the purpose of the training was to teach him to recognize the changes in Mauri’s behavior that constituted area and specific alerts. McDowell was not asked to describe, and did not describe, what he learned about Mauri’s reliability or what, if anything, he was taught “about handling [Mauri] to maximize her accuracy and reliability.” Id. at 157. Nor did he describe, for example, what methods, if any, he was taught to distinguish between Mauri’s general behaviors during a deployment and the seemingly similar behaviors that constitute an area alert or how to detail Mauri to particular areas without improperly cuing her to alert.
Similarly, McDowell testified that he and Mauri were certified as a dog-handler team, but he did not testify about the requirements for certification. The record does not contain any information about whether McDowell and Mauri had to complete a test to be certified, and, even assuming that they did, the record does not contain any information about what the test involved, whether controls were used to protect against human cuing, or what accuracy rate, if any, was required to pass the test. See, e.g., Foster, 350 Or [713]*713at 165-66 describing controlled test that required a dog-handler team to search two rooms and three vehicles with a 90 percent accuracy rate and to search seven packages and an open area with a 100 percent accuracy rate).
McDowell testified that, after he and Mauri were certified, he continued to train Mauri. But, again, his description of the training is limited and does not explain how his training could promote reliable alerts. The record does not contain any information about whether McDowell trained Mauri in a variety of settings, used distractors, or had other people hide training aids, or whether such practices would improve Mauri’s reliability. See Helzer, 350 Or at 159 (although handler “testified vaguely to his use of blanks and food distractions * * *, he provided no information beyond that to explain how his training builds accuracy and reliability”).
McDowell did testify about Mauri’s field performance, stating, “Every time [Mauri has] alerted to the presence of a controlled substance in the field, during the search we have found a controlled substance. So her accuracy record is 100 percent.” But, that evidence is not useful for two reasons. First, field records are of limited utility. Second, and more importantly, there is not enough information about Mauri’s performance, in the limited time period between her certification in November 2009 and the traffic stop of defendant in February 2010, to draw any inferences from her reported accuracy rate. The record does not reveal how many times Mauri alerted, and, without that information, there is no basis for concluding that her reported accuracy rate can be attributed to skill as opposed to luck.
We conclude that the record before us is even weaker than that in Helzer and it fails to establish that Mauri’s alert was reliable enough to even contribute to a conclusion that there was probable cause to search defendant’s car. In doing so, we are cognizant of a footnote in Helzer in which the Supreme Court stated, in dicta, that
“[t]he fact that a particular dog alert in a given case is not sufficiently reliable to provide probable cause does not necessarily foreclose any reliance on or consideration of the alert. Assuming that the state can establish that the alert [714]*714has some degree of reliability, the alert may still be considered along with other information in the totality of the circumstances analysis of whether officers searched with probable cause.”
350 Or at 156 n 2 (emphasis in original). Relying on that footnote, the state argues that, “[g]iven the other information known to [the troopers] and the degree to which [Mauri’s] reliability was established on this record, the trial court properly concluded that [the troopers] had probable cause.” The difficulty with the state’s argument is that the record does not establish Mauri’s reliability to any degree. As recounted above, it does not indicate that any steps were taken, at any point in her training and testing, to protect against the possibility that her alerts in training and testing were based on handler cues or human trails, as opposed to the detection of drugs. The only evidence that the state can, and does, point to, reduces to the fact that Mauri was trained and, as McDowell testified, her “accuracy record is 100 percent.” But, as explained above, without additional information about Mauri’s field performance, we cannot draw any inferences about her reliability from her reported accuracy rate.
Because the record does not establish that Mauri’s alert was sufficiently reliable to contribute to a conclusion that there was probable cause to search defendant’s car, and because — as the trial court concluded and the state does not dispute — the troopers did not have probable cause before the alert, the trial court erred in denying defendant’s motion to suppress the results of the search. Therefore, we reverse and remand for further proceedings.
Counts 1 and 2 reversed and remanded; otherwise affirmed.