ORDER

Tina Kellems applied for disability benefits, but an administrative law judge denied her claim, reasoning that her testimony about her pain-induced functional limitations was not credible. The ALJ supported his credibility determination with a number of factual findings, including a finding that Kellems had engaged in drug-seeking behavior. A magistrate judge, presiding with the consent of the parties, affirmed the ALJ's decision, and Kellems appeals. We conclude that the ALJ's finding that Kellems engaged in drug-seeking behavior is not supported in the record. Because it is impossible to determine whether the ALJ would have made the same credibility determination had he not erroneously accused Kellems of engaging in drug-seeking behavior, we reverse the district court's judgment and remand this case to the Social Security Administration for further proceedings.

I.

In her application for disability benefits, Kellems claimed that she became disabled in November 2001, when Randall Megeff, her primary-care doctor, diagnosed her with chronic pain syndrome, depression, and hypertension. Judith Lee-Sigler, a specialist in physical medicine who worked in the same clinic, began treating Kellems' pain in February 2002 and diagnosed her with fibromyalgia, cervical degenerative disc disease, and lower back pain. Although Kellems was attending nursing school at the time, she later put her studies on hold because of her pain. She was also being paid for the somewhat strenuous work of caring for her husband's elderly relative; she had to lift the woman out of a wheelchair as often as 25 times each day, which aggravated her pain. But she could not afford to quit, and the record suggests that she was still performing this work as late as May 2003.

Dr. Lee-Sigler told Kellems in July 2002 that she was taking too many narcotic pain medications. Kellems was "frustrated" but "willing to work with" her doctor to discontinue some of them. In November 2002 Dr. Lee-Sigler concluded that Kellems' pain required expert treatment and referred her to a chronic-pain center in another city around 80 miles away. Kellems saw Dr. Lee-Sigler a few weeks later to ask for prescription refills. She admitted that she had been taking some of her husband's pain medication because she had run out, so the doctor agreed to refill Kellems' prescription one final time to tide her over until she could make it to the pain center. Eventually a doctor at the pain center devised a treatment plan for Kellems, but the 90-minute drive was too long for her to make on a regular basis. So Dr. Lee-Sigler agreed to administer the treatment plan by prescribing the recommended medications, including methadone. Over the next few months Dr. Lee-Sigler suggested that Kellems should also enroll in physical therapy and exercise more, but there is no evidence in the record that Kellems pursued the doctor's suggestions.

In January 2004 Dr. Lee-Sigler refused to continue administering Kellems' treatment plan. She did not approve of the plan's use of methadone and suggested that "an exercise and behavioral modification program" would do more good. Dr. Megeff agreed to prescribe some of the treatment plan's medications on a temporary basis, but not methadone. So Kellems began an unfruitful search for a pain center that was convenient and that would accept her as a patient. At one point in March 2004 she ran out of methadone and requested some from an emergency-room doctor. But she promptly told Dr. Megeff about the incident; in response he referred her to Dr. Lee-Sigler, who wrote one final methadone prescription and told her how to discontinue the medication gradually.

In May 2004 Dr. Megeff agreed to manage Kellems' pain medications on a permanent basis. He explained that he believed she was in real pain and did not suspect she was an addict. But nevertheless he insisted that "if I am going to manage her medications there are going to be rules." First he required Kellems to stop taking methadone. He also required her to agree to a "pain contract" in which she promised to accept "ongoing psychological or psychiatric support for stress management" and not to seek medication from other doctors. Dr. Megeff said he would "likely" stop treating her if she broke her promises. But there is no evidence in the record that Kellems ever ran afoul of the agreement or that Dr. Megeff ever discussed discontinuing her treatment.

While he was treating Kellems' pain, Dr. Megeff referred her to several specialists, including a genetics clinic, a neurologist, and a spine doctor; she agreed to all these appointments without incident. However, she told Dr. Megeff that she was "not interested" in seeing a psychiatrist. Dr. Megeff also urged her to discontinue all nonessential medications. When he told her this in October 2005 she was "leery," but in May 2006 she agreed to try. The record does not show whether she succeeded, although Dr. Megeff's treatment notes suggest she did not. In August 2006 Kellems briefly sought to replace Dr. Megeff. She visited another primary-care doctor and told him that she was dissatisfied with Dr. Megeff's treatment plan. She asked whether he could prescribe her new pain medications, but he told her he could not think of any she had not yet tried, so she never went back. In April 2007 Dr. Megeff supported Kellems' disability claim by submitting a letter that corroborated her reports of pain and concluded that she was not able to perform any work.

At a hearing before an ALJ in April 2007 Kellems rated her pain most days at ten on a ten-point scale. She testified that she could not sit for more than thirty minutes or lift anything more than two pounds. She said she sat in a recliner all day and could not do any household chores. Even though her medications were "doing the best they can," she explained, they were still doing a "bad" job relieving her pain.

The ALJ found that Kellems was not disabled because she could still perform "light work with occasional postural limitations." Although the ALJ believed that she was experiencing real pain, he did not find her testimony about the intensity of her pain or her pain-induced limitations completely credible. He made several factual findings to support his credibility determination:

The undersigned notes that the record as a whole, reveals inconsistencies in the claimant's description of her pain, and exaggerations of her pain complaints (for example alleging pain at a level 10 constantly-despite not seeking emergency room treatment), her continued work activity after the alleged onset date, the evidence of drug-seeking behavior and refusal to follow recommendations such as seeking psychiatric care or weaning from drugs as recommended.

The ALJ also thought that Kellems' behavior at the hearing was inconsistent with her testimony: although she said she could not sit for more than thirty minutes, he noted, she sat through the hour-long hearing without incident. In addition the ALJ based his credibility determination on the absence of objective medical evidence explaining the source of Kellems' pain. Finally the ALJ refused to give Dr. Megeff's opinion controlling weight, reasoning that it was inconsistent with the record and "based mostly on the claimant's subjective complaints of ongoing pain and her description of inability to engage in activities."

II.

Kellems argues on appeal that the ALJ's credibility determination is tainted by an erroneous finding that she engaged in drug-seeking behavior. She insists that, rather than engaging in drug-seeking behavior, she merely took the medications that her doctors prescribed to relieve her crippling pain. The ALJ's contrary finding, she argues, "grossly distorts the record" by ignoring the "longitudinal history" of her pain medications, her doctors' confidence that her pain was real, and her compliance with her doctors' orders.

Several cases approve discounting the testimony of a claimant who has engaged in drug-seeking behavior, see Simila v. Astrue, 573 F.3d 503, 519 (7th Cir. 2009); Poppa v. Astrue, 569 F.3d 1167, 1171 (10th Cir. 2009); Berger v. Astrue, 516 F.3d 539, 546 (7th Cir. 2008); Anderson v. Barnhart, 344 F.3d 809, 815 (8th Cir. 2003); Edlund v. Massanari, 253 F.3d 1152, 1157 (9th Cir. 2001), but none has defined what constitutes drug-seeking behavior. Claimants in these cases do have a common thread, however; each obtained, or attempted to obtain, pain medication by deceiving or manipulating a medical professional. In Simila, for example, the claimant was refused a refill on pain medication after his doctor discovered that he was abusing cocaine; undeterred, he obtained a refill from another doctor by claiming that the first doctor had turned him away for financial reasons. 573 F.3d at 519. Likewise in Poppa the claimant disobeyed her doctor's orders not to increase the dosage of her pain medication and at least twice asked the doctor to postdate her prescriptions; she also wanted extra medication for an upcoming vacation but was not forthcoming about how long she would be away. 569 F.3d at 1171-72. And in Anderson the claimant's doctor observed that his patient was "`somewhat manipulative' in his efforts to convince him of his pain." 344 F.3d at 815.

In this case, however, we do not see any evidence in the record that Kellems obtained, or attempted to obtain, pain medication by deceiving or manipulating a medical professional. In particular we do not think that the supposedly incriminating incidents that the ALJ pointed to in Dr. Megeff's treatment notes are convincing evidence of drug-seeking behavior. Although Kellems never followed through on her promise to pursue mental-health treatment, she did seek treatment with a genetics center, a neurologist, and a spine specialist; her willingness to undergo these referrals, presumably designed to locate an objective source of her pain, undercuts any inference that her failure to see a psychiatrist was a drug-seeking tactic. The ALJ also found that Kellems rebuffed Dr. Megeff's recommendation to discontinue all nonessential medications, but the record shows the opposite. She stopped taking methadone in May 2004 and agreed to try eliminating all extraneous medications in May 2006. In fact there is no evidence in the record that Dr. Megeff ever observed Kellems exploiting deceptive or manipulative drug-seeking tactics. To the contrary, he said, "I do not feel that she has exhibited typical addictive behaviors." He explained that she had never "lost" prescriptions and, other than seeking methadone at the emergency room in March 2004, had "not sought care with multiple physicians." There is no evidence that she ever broke the terms of her "pain contract"; indeed Dr. Megeff was still managing her medications when the ALJ made his finding.

The commissioner cites additional examples of drug-seeking behavior that the ALJ mentioned in his recitation of Kellems' medical history but did not rely on in making his credibility determination. Of course SEC v. Chenery Corp., 318 U.S. 80, 87-88, 63 S.Ct. 454, 87 L.Ed. 626 (1943), forbids the commissioner to defend the ALJ's decision on grounds not "embraced" by ALJ. Parker v. Astrue, 597 F.3d 920, 922 (7th Cir. 2010). But in any event none of these incidents is persuasive. Kellems' methadone-seeking trip to the emergency room in March 2004 was an isolated event that occurred while she was in withdrawal; she immediately told her doctors, who instructed her how to discontinue the drug properly. And she attempted to replace Dr. Megeff in August 2006 because she was frustrated with her lack of progress, not because she was seeking a secret second source of medications.

Yet even a credibility determination that is partially flawed and based on factual findings that "are a bit harsh" will be disturbed only if the ultimate conclusion is "patently wrong." Simila, 573 F.3d at 517; Berger, 516 F.3d at 546. And indeed the ALJ made other factual findings to support his credibility determination. But, given his mistaken impression that Kellems was engaged in drug-seeking behavior, we will not rely on these additional findings. A credibility determination is a "judgment call" that encompasses all relevant factors. See Allord v. Barnhart, 455 F.3d 818, 821 (7th Cir. 2006). The record shows that the ALJ's mistaken finding was a significant component of his credibility determination. And so it is not possible to say, for example, whether he would have seen Kellems' behavior at the hearing in the same light had he not wrongly suspected that she was a drug-seeker. See Powers v. Apfel, 207 F.3d 431, 436 (7th Cir. 2000) (approving ALJ's direct observation of claimant if "one of several factors" supporting credibility determination). Nor do we know what he would have made of her seemingly inconsistent work history; if he did not already distrust her, he might well have found that her ability to perform heavy lifting in May 2003 said little about whether she was telling the truth when she testified to her pain-induced limitations in April 2007. We think that a reasonable factfinder would be entitled to come out the other way, and we decline to speculate what the ALJ might have done had he had not erred in finding that Kellems engaged in drug-seeking behavior. See Allord, 455 F.3d at 821-22. Under these circumstances the ALJ's credibility determination was patently wrong.

We REVERSE the district court's judgment and REMAND the case to the Social Security Administration for further proceedings.