Paul Lambert appeals the denial of Disability Insurance Benefits for chronic back pain. The Social Security Administration denied his application initially and on reconsideration, and an administrative law judge ("ALJ") concluded that Lambert suffers from degenerative disc disease that is severely impairing but not disabling. Lambert challenges the ALJ's decision to give little weight to the most recent opinions of his treating neurosurgeon and to discredit his own testimony about the severity of his pain and extent of his limitations. We reverse and remand for further agency proceedings.

I. Background

Lambert applied for benefits in 2012 at age 41 alleging disabling lower back pain since 2011, an onset date that he later amended to 2013. Yet his back problems started long before then. In 2004 discs in his lumbar spine were surgically fused with a rod. In 2008 surgeons repaired the rod. Still, Lambert held several jobs over the years.

In 2010 Lambert began experiencing back pain "most of the time" and thereafter also had "intermittent" pain down his left leg that often caused him to fall. By late 2012 Lambert had tried various treatments, including steroid injections in his spine and pelvis, chiropractic care, medication, and physical therapy. Nothing produced lasting relief, though hydrocodone helped ease the pain. Medical imaging revealed no postsurgical complications or other explanation for his persistent pain. Several neurosurgeons found the cause unclear; three said further surgery was not an option.

In September 2012 a pain specialist attributed Lambert's pain to degenerative disc disease or joint disease of the lower lumbar spine. Months later he diagnosed *772 Lambert with failed back syndrome (meaning he experienced continuous pain despite surgeries) and recommended that he accept his chronic pain and proceed with a pain-management program instead of seeking a surgical cure. The pain specialist also recommended that Lambert consider behavioral therapy to learn coping skills. Lambert followed this advice, but in early 2013 the pain specialist referred him to a neurosurgeon to find the source of the left-leg pain that by this time was causing daily falls.

Neurosurgeon Kamajit Paul began treating Lambert in June 2013 and initially recommended a conservative course of steroid injections to determine if his pain was caused by dysfunction in the left sacroiliac joint. (The sacroiliac joints connect the pelvis to the lower spine and support the weight of the upper body when a person stands. Sacroiliac Joints , MAYO CLINIC , https://www.mayoclinic.org/diseases-conditions/sacroiliiitis/multimedia/sacroiliac-joints/img-20005962 (last visited June 28, 2018).) The pain specialist administered three injections but continued advising Lambert to accept his chronic pain and moderate his activities. Because the injections provided only several hours of relief, Dr. Paul believed Lambert had dysfunction in his left sacroiliac joint and recommended surgery to fuse it. But he cautioned that the surgery offered no guarantee of improvement and that Lambert would still experience some back and leg pain and would "never be 100%."

Dr. Paul performed the left-joint fusion in October 2013, and Lambert's condition initially started to improve. In November Lambert reported minimal pain. After a month of physical therapy, he underwent a functional assessment in January 2014. He was able to walk without an assistive device and reported "improved function at home and in the community." His physical therapist recommended allowing him to return to work with some lifting restrictions, and Dr. Paul released Lambert to light-duty work.

But Lambert's relief was short-lived. In February 2014 he returned to Dr. Paul, now reporting pain on his right side. Dr. Paul was "not sure what [was] happening" and sent Lambert back to his pain specialist. Testing revealed potential dysfunction in the right sacroiliac joint.

In late March 2014, Dr. Paul completed an assessment of Lambert's functional abilities. He diagnosed bilateral joint dysfunction and stated that Lambert experienced continued pain in his right sacroiliac joint and lower back. He opined that in a competitive work situation, Lambert could sit for at least six hours out of eight, stand for 30 minutes at a time (up to two hours total), and walk one block at a time (if allowed to shift between these positions at will). Dr. Paul also noted, however, that Lambert's prognosis was guarded and that he "may develop problems in the upper lumbar spine."

In April 2014 Dr. Paul surgically fused Lambert's right sacroiliac joint. Lambert returned to physical therapy and in early June reported that his preoperative pain had resolved. But later that month Lambert told his physical therapist that the pain on the left side of his lower back had returned; he said it was minimal but interfered with sleep. A week later Lambert said the pain had worsened and now prevented him from walking as far as he could just weeks before. In July 2014 Lambert told Dr. Paul that he had been experiencing pain-exacerbated by activity-for as long as four weeks. X-rays revealed intact surgical hardware without abnormality, so Dr. Paul thought the pain did not stem from the recent fusion. He directed Lambert to proceed with a previously scheduled functional assessment.

*773 In late July 2014, Lambert's physical therapist performed the functional assessment and observed a "significant decrease" in his capabilities since the January assessment. The July assessment revealed that Lambert had "significant limitations" in sitting, standing, and walking; he required position changes every 15 minutes; and he had a limp that grew more severe as he walked. The physical therapist opined that Lambert's work tolerance "would be low to sedentary" and recommended that he follow up with a physician.

Dr. Paul examined Lambert once more at the end of July 2014. Lambert had pain and restricted motion in his back but normal coordination with no muscle atrophy or weakness. Based on his examination and the therapist's functional assessment, Dr. Paul said Lambert was limited to 15 minutes of sitting or standing at a time and needed to change positions frequently. He concluded that Lambert's "functional capacity [was] markedly reduced to the extent that ... he cannot do even sedentary work." Dr. Paul again referred Lambert to the pain specialist because his back pain could not be controlled by surgery. In August Dr. Paul opined that Lambert "would not be able to tolerate a work situation" because his "persist[ent] low back pain" had worsened, is severe, and is not expected to improve.

At an August 2014 hearing before an ALJ, Lambert testified that in 2011 he fell down the stairs in his home and his employer laid him off so he could obtain unemployment-insurance benefits without having to search for work while he "figured out what was going on." The ALJ voiced concern over Lambert's receipt of unemployment benefits after his alleged onset date. Lambert then orally amended his onset date to January 1, 2013.

Lambert testified that he could not sit, stand, or walk for more than 15 minutes without a lot of pain in his lower back. He said that nothing entirely relieved the pain, including hydrocodone, Vicodin, and participation in the pain-management program. He had stopped doing housework, used a mounted seat while showering, and needed his wife's help putting on and removing his socks. Lambert testified that around the time of his first sacroiliac joint surgery in the fall of 2013, he had started taking college classes online in the hope of possibly working as a music teacher-but his pain had progressed so much since March 2014 that he could not focus and had to withdraw from classes scheduled for fall 2014.

The ALJ applied the standard multistep analysis, see 20 C.F.R. §§ 404.1520 (a), 416.920(a), and concluded that Lambert was not disabled after the amended onset date of January 1, 2013. As relevant here the ALJ determined that Lambert was severely impaired by degenerative disc disease ; that he nonetheless had the residual functional capacity ("RFC") to perform sedentary work with specified limitations; and that based on the testimony of a vocational expert, he was capable of working as a sorter, assembler, order clerk, or office helper.

In delineating Lambert's RFC, the ALJ concluded that Lambert's alleged symptoms were caused by medically determinable impairments, but that the severity of his pain and his claimed functional limitations were "not substantiated by the medical and other evidence of record." The ALJ also said that Lambert's receipt of unemployment benefits after the initial alleged onset date and his subsequent attempt to "moot" the issue by amending his onset date "reflect[ed] adversely on his credibility."

The ALJ gave "little weight" to Dr. Paul's opinions from July and August 2014. They were, in the ALJ's view, inconsistent with his March 2014 opinion and his objective *774 findings. The ALJ also characterized Dr. Paul's opinion that Lambert would not be able to tolerate a work situation as a legal conclusion reserved to the Commissioner. On the other hand, the ALJ gave "considerable weight" to two opinions of state agency consultants who reviewed Lambert's medical records in August 2012 (before the amended onset date) and April 2013 (four months after the amended onset date) and concluded that Lambert could perform sedentary work with restrictions.

The Appeals Council denied review. A district judge affirmed the Commissioner's decision and Lambert appeals.

II. Analysis

Lambert challenges the ALJ's decision to give little weight to the most recent opinions of his treating neurosurgeon, Dr. Paul. He also argues that the ALJ was wrong to discredit his testimony about the severity of his symptoms and limitations.

A. Treating Neurosurgeon's Opinions

Lambert contests the ALJ's decision to give little weight to Dr. Paul's July and August 2014 opinions that his pain had worsened to the point that he could not tolerate even sedentary work. Because Dr. Paul is a treating physician, his opinion on the nature and severity of Lambert's medical condition is entitled to controlling weight if it is well supported by medical findings and consistent with other record evidence. See 20 C.F.R. §§ 404 .1520c(a) (2017) ; Gerstner v. Berryhill , 879 F.3d 257 , 261 (7th Cir. 2018) (noting that this treating-physician rule applies only to claims filed before March 27, 2017).

Lambert identifies multiple flaws in the ALJ's decision. He argues that the ALJ overlooked medical evidence substantiating Dr. Paul's most recent opinions, wrongly found the opinions inconsistent with Dr. Paul's earlier opinion from March 2014, and failed to consider the relevant factors for evaluating medical source opinions set forth in 20 C.F.R. § 404.1527 (c). He also challenges the ALJ's decision to discount Dr. Paul's opinion as a legal conclusion outside a doctor's role. Finally, Lambert asserts that the ALJ failed to explain why Dr. Paul's opinions were entitled to less weight than those of the agency physicians rendered before some of the key medical evidence was compiled.

We agree that the ALJ's reasons for giving little weight to Dr. Paul's most recent opinions are inadequate to "build an accurate and logical bridge between the evidence and the result." Beardsley v. Colvin , 758 F.3d 834 , 837 (7th Cir. 2014). First, the ALJ said there was no objective basis for Dr. Paul's opinion about Lambert's symptoms as of July 2014 because "x-rays revealed good fusion and good position of the [sacroiliac] joint." But no medical source opined that the imaging results were inconsistent with Lambert's complaints of disabling pain. Indeed, throughout Lambert's treatment history, medical imaging ruled out specific, objective causes of his ongoing pain-yet his doctors performed surgeries, prescribed powerful pain medications, and recommended long-term pain-management techniques for his suite of chronic back problems. ALJs must rely on expert opinions instead of determining the significance of particular medical findings themselves. Meuser v. Colvin , 838 F.3d 905 , 911 (7th Cir. 2016) ; Stage v. Colvin , 812 F.3d 1121 , 1125 (7th Cir. 2016) ; see also Goins v. Colvin , 764 F.3d 677 , 680 (7th Cir. 2014) (remanding where the ALJ "play[ed] doctor" by summarizing the MRI results without subjecting them to professional medical scrutiny). The ALJ's finding of a mismatch between the objective evidence and the treating neurosurgeon's opinion failed to heed that principle.

Second, the ALJ discounted Dr. Paul's mid-2014 opinions because Lambert's *775 symptoms after the April 2014 surgery had an "unclear" cause. But Dr. Paul's conclusion, shared by the pain specialist, was that Lambert has a chronic back condition that may never be cured. Dr. Paul attempted to treat Lambert's pain through sacroiliac joint surgeries; after surgical intervention failed, he returned Lambert to long-term pain management. Degenerative conditions often get worse over time, see Hill v. Colvin , 807 F.3d 862 , 868-69 (7th Cir. 2015), and Lambert has an extensive history of surgeries that yielded only temporary or partial relief. Given his chronic back condition, it was improper to reject Dr. Paul's mid-2014 assessment merely because he could not isolate the source of the pain. See Cole v. Colvin , 831 F.3d 411 , 416 (7th Cir. 2016) ; Parker v. Astrue , 597 F.3d 920 , 922 (7th Cir. 2010).

Third, it was wrong for the ALJ to say that "no evidence" showed that Lambert's pain after the April 2014 surgery would not "respond to conservative treatment." Lambert's back pain was consistently classified as incurable or chronic by Dr. Paul, by the pain specialist, and by Lambert's primary-care physician. And a surgical consultant who reviewed Lambert's records before both sacroiliac joint surgeries rated his back and joint pain as "severe" and unresponsive to conservative treatment.

Fourth, the ALJ overlooked the extent to which findings in the July 2014 functional assessment supported Dr. Paul's most recent opinions. Dr. Paul opined, consistent with the physical therapist's July assessment, that Lambert could sit or stand for only 15 minutes and needed frequent position changes, and he also concluded that Lambert's "functional capacity [was] markedly reduced to the extent that ... he [could] not do even sedentary work." The physical therapist had said that Lambert's tolerance for sedentary work would be low and (unlike in January) did not specifically opine that Lambert could return to work. So the functional assessment supports Dr. Paul's opinion. An ALJ's failure to consider findings that support a treating physician's opinion is error. Minnick v. Colvin , 775 F.3d 929 , 938 (7th Cir. 2015).

Fifth, the ALJ decided, without explanation or record support, that Dr. Paul's most recent opinion was inconsistent with his earlier assessment. Dr. Paul's treatment notes reveal that his later views were an update -the product of repeated failed attempts to treat Lambert's pain through surgeries. From his first examination of Lambert in June 2013, Dr. Paul's notes refer to his uncertainty about the source of Lambert's pain. His March 2014 opinion discusses the likelihood of continued back problems and includes a "guarded" prognosis. Between that opinion and the one he issued four months later, Lambert underwent another surgery, had trouble with physical therapy, and completed a functional assessment-all of which supported Dr. Paul's July and August opinions that Lambert's condition had worsened. Physicians may update their views without being inconsistent if their later opinions are based on a patient's changed condition. See Scrogham v. Colvin , 765 F.3d 685 , 696-97 (7th Cir. 2014). There is no unexplained inconsistency here.

Relatedly, the ALJ weighed Dr. Paul's opinions without considering the regulatory factors listed in 20 C.F.R. § 404.1527 (c). ALJs must evaluate a treating physician's noncontrolling opinion by considering the treatment relationship's length, nature, and extent; the opinion's supporting explanation and consistency with other evidence; and any specialty of the physician. Moss v. Astrue , 555 F.3d 556 , 561 (7th Cir. 2009) ; Bauer v. Astrue , 532 F.3d 606 , 608 (7th Cir. 2008). Dr. Paul, a neurosurgeon specializing in spinal disorders *776 , treated Lambert's back problems for over a year, examined him at least 15 times, and performed two sacroiliac joint surgeries. Dr. Paul based his most recent opinions on the physical therapist's July functional assessment and on his own contemporaneous examinations. Yet the ALJ did not explain his view of these factors in assigning little weight to Dr. Paul's opinions.

Finally, the ALJ wrongly discounted Dr. Paul's opinion as an improper legal conclusion by a medical professional. Whether a claimant qualifies for benefits is a question of law, Garcia v. Colvin , 741 F.3d 758 , 760 (7th Cir. 2013), but a medical opinion that a claimant is unable to work is not an improper legal conclusion, Bjornson v. Astrue , 671 F.3d 640 , 647-48 (7th Cir. 2012) (remanding for the ALJ to consider the opinion that the claimant "remained unable to work") (citing 20 C.F.R. § 404.1527 (e)(1) ). Indeed, ALJs must consider medical opinions about a patient's ability to work full time because they are relevant to the RFC determination. Garcia , 741 F.3d at 760 . Here Dr. Paul's most recent opinion is that Lambert's chronic back pain is so limiting that he no longer can tolerate even sedentary work. That's not an improper legal conclusion.

The ALJ's flawed analysis of Dr. Paul's opinions is compounded by his failure to explain why he gave considerable weight to the opinions of the agency physicians that Lambert could perform sedentary work. It is puzzling why the ALJ would credit these opinions while discounting Dr. Paul's as an improper legal conclusion. See Bjornson , 671 F.3d at 648 . That inconsistency aside, an ALJ must weigh medical opinions by applying the regulatory factors in 20 C.F.R. § 404.1527 (c)(2). The reviewing consultants rendered their opinions before Lambert was treated by the pain specialist, before Dr. Paul fused Lambert's sacroiliac joints in failed attempts to alleviate his pain, before Dr. Paul opined that Lambert's pain had worsened and his limitations had degraded since 2011, and before the physical therapist found Lambert's functional abilities had diminished. ALJs may not rely on outdated opinions of agency consultants "if later evidence containing new, significant medical diagnoses reasonably could have changed the reviewing physician's opinion." Moreno v. Berryhill , 882 F.3d 722 , 728 (7th Cir. 2018).

The government responds that any error in weighing Dr. Paul's opinions is harmless. An error is harmless only if we are convinced that the ALJ would reach the same result on remand. See McKinzey v. Astrue , 641 F.3d 884 , 892 (7th Cir. 2011). Here the outcome is not foreordained; at the very least, the ALJ formulated an RFC without including Dr. Paul's most recent opinions.

The government also argues that under the Social Security Act, Lambert needed to prove that he was unable to work for an identifiable, continuous 12-month period. This argument misreads the statute. The Act does not specify how long a claimant must be unable to engage in substantial gainful activity. Instead it is the claimant's "medically determinable physical or mental impairment" that must have "lasted or can be expected to last for a continuous period of not less than 12 months." 42 U.S.C. § 423 (d)(1)(A). Lambert's impairment from degenerative disc disease continued from at least January 2013 through August 2014, well over a year. The government's reading of the Act would preclude benefits for anyone with an impairment that causes 12 months of bad days with good days interspersed. See Bauer v. Astrue , 532 F.3d 606 , 609 (7th Cir. 2008) ("A person who has a chronic disease, whether physical or psychiatric, and is under continuous treatment for it *777 with heavy drugs, is likely to have better days and worse days.... Suppose that half the time she is well enough that she could work[ ] and half the time she is not. Then she could not hold down a full-time job.").

B. Lambert's Testimony

Lambert also argues that the ALJ wrongly discredited his testimony about the severity of his back pain and related functional limitations. We will overturn an ALJ's adverse credibility determination only if it is unsupported by substantial evidence or rests on legally improper analysis. Ghiselli v. Colvin , 837 F.3d 771 , 778-79 (7th Cir. 2016).

Lambert contends that the adverse credibility determination rests on a misinterpretation of the medical records. The ALJ stressed that "imaging studies have consistently show[n] good fusions and intact hardware," and Lambert's exams revealed "normal coordination" and "no atrophy or deficits in motor, strength, or sensory" abilities. But none of Lambert's physicians interpreted these medical findings as inconsistent with his reports of recurrent and worsening pain and functional limitations. Even when tests showed no hardware malfunction, coordination issues, or strength deficits, Lambert's physicians continued to treat his pain. The ALJ also failed to acknowledge that Lambert's physical exams showed only brief periods of coordination and strength, which is consistent with his testimony that sitting, standing, or walking for more than 15 minutes exacerbated his pain. See Gerstner , 879 F.3d at 264 (noting that the claimant's performance in exams did not undermine her allegation that pain was triggered by prolonged activity).

Similarly, the ALJ emphasized that Lambert had "good responses" to surgeries, physical therapy, and medication when the medical records actually show that these treatments were ineffective at either consistently or decisively improving his chronic pain or resolving his functional limitations. See Stark v. Colvin , 813 F.3d 684 , 687 (7th Cir. 2016) (remanding where the ALJ mentioned ongoing treatments in passing but did not consider whether it relieved the claimant's pain). Indeed, two sacroiliac joint surgeries failed to eliminate Lambert's lower back and leg pain, and physical therapy in 2013 improved his functioning for only a brief period.

The ALJ also noted that Lambert said he had good control of pain with medication in January 2013, that he had pain relief with hydrocodone in July 2013, and that he was no longer taking medications on March 4, 2014. But one physician opined that hydrocodone interferes with Lambert's ability to work. And Lambert resumed the hydrocodone within a month of stopping it in 2014 and continued to experience pain even when taking the medication. The ALJ's evaluation of Lambert's need for narcotic pain medication omitted these important details.

The ALJ further concluded, contrary to the evidentiary record, that Lambert's symptoms were "only intermittent." The ALJ relied on an April 2013 treatment note by the pain specialist, but that note describes Lambert's falls and leg pain as intermittent while characterizing his back pain as chronic. Indeed, the ALJ's description conflicts with many other treatment notes in which Lambert's pain specialist, his primary-care physician, and Dr. Paul assessed his back condition in chronic terms.

The ALJ also overread or overlooked important evidence in discrediting Lambert's testimony about his limited activities of daily living. The ALJ relied on a March 2014 note from a physical therapist who reported that Lambert assessed his recovery from left sacroiliac joint surgery *778 as "excellent." But the ALJ glossed over another part of that same note in which Lambert complained about increasing right -side lower back pain that woke him every four to five hours, giving him a sleep pattern that was "very disruptive" to his life. Nor did the ALJ acknowledge that Lambert underwent another surgery and experienced worsened symptoms in the five months or so between his March 2014 comment and the August 2014 hearing.

Lambert next contends that the ALJ wrongly discounted his severe back pain after the most recent sacroiliac joint surgery. The ALJ found that Lambert's complaints were unsupported by objective evidence. To be sure, a lack of objective support from physical examinations and test results is relevant, but an ALJ may not discredit pain complaints solely because they lack objective corroboration. Pierce v. Colvin , 739 F.3d 1046 , 1050 (7th Cir. 2014) ; Parker , 597 F.3d at 921-22 . Indeed, the recurrence of Lambert's symptoms is consistent with Dr. Paul's warning that surgery was not guaranteed to alleviate his lower back pain and the pain specialist's recommendation that Lambert pursue long-term pain-management strategies for his incurable back condition. Lambert "has undergone painful and risky procedures in attempts to alleviate his pain, actions that would seem to support the credibility of his claims regarding the severity of his pain." Israel v. Colvin , 840 F.3d 432 , 441 (7th Cir. 2016).

Last, Lambert argues that the ALJ's adverse credibility determination rests on an improper analysis of his receipt of unemployment benefits before his amended onset date. The ALJ said that changing the alleged onset date to a period after the lapse of his unemployment benefits "evidenced either an intent to return to work as required by state unemployment law or misuse of the law." But a claimant's desire to work is not evidence that the claimant has embellished his limitations, see Gerstner , 879 F.3d at 265 -especially here, where Lambert's hopeful view of his prognosis was not shared by his doctors, who repeatedly urged him to shift from seeking a cure to focusing on pain management. And although unemployment benefits may be relevant if a claimant has represented to the State that he is able to work during the period for which he has applied for federal disability benefits, Schmidt v. Barnhart , 395 F.3d 737 , 746 (7th Cir. 2005), any work-ready representation that Lambert made, personally or by presumption under state law, occurred before the amended onset date.

If the ALJ meant to rely on Lambert's work-ready representation before the amended onset date, the ALJ failed to explain why this affected his evaluation of limitations documented more than a year later-particularly in light of the degenerative nature of Lambert's back condition. See Scrogham , 765 F.3d at 699 (remanding because the ALJ relied on receipt of unemployment benefits to discount claimant's symptoms without considering the progressive nature of disease). The record reveals that when Lambert simultaneously received unemployment benefits and claimed disability, he hoped that medical intervention might reduce his back pain to the point that he could return to work. No one knew at that time whether he would be able to work again.

Finally, we note a more fundamental problem with the ALJ's reliance on Lambert's application for unemployment compensation to discount his credibility in seeking disability benefits. Under the Social Security Act, the line between disabled and nondisabled can be very difficult to chart accurately. And the Social Security system is designed to encourage everyone who can work to do so. Consider, for example, the first step in the five-step analysis *779 of disability. No matter how severe the physical or mental challenges might be, if the person is actually working ("substantial gainful activity"), the person is not disabled. 20 C.F.R. §§ 404.1520 (a), 416.920(a). And a person who is not certain whether he will qualify for Social Security disability surely has, and should have , a strong incentive to keep looking for work and to pursue unemployment compensation as an interim source of income. An ALJ should not discount a claimant's credibility based on an application for unemployment compensation without taking these incentives and pressures into account. The ALJ's opinion here does not indicate that he did.

III. Conclusion

In sum, the ALJ failed to properly assess the treating neurosurgeon's most recent opinions about Lambert's impairments and limitations and Lambert's testimony about his symptoms. Accordingly, we REVERSE the judgment in favor of the Commissioner and REMAND for further proceedings.