PER CURIAM. When a person applies for disability benefits, the Social Security Administration evaluates whether signifi- cant numbers of jobs exist in the national economy for someone with that person’s limitations. Administrative law judges often rely on vocational experts to estimate these job numbers. But ALJs cannot afford complete discretion to vocational experts. Instead, when a claimant challenges a vocational expert’s job-number estimate, the ALJ must inquire whether the methodology used by the expert is reliable. In this case, the vocational expert enlisted by the agency to estimate the number of jobs suitable for Randall Ruenger omitted crucial de- tails about her methodology, such as the source of her job numbers and the reason she used the equal distribution method. But the ALJ nevertheless relied on the expert’s testimony. Because substantial evidence does not support the ALJ’s decision, we vacate and remand for further proceedings.

I Randall Ruenger applied for benefits in 2015, alleging that he had limited use of his left arm and mental impairments including anxiety and depression. He eventually received a hearing before an ALJ in 2018. Applying the five-step inquiry found in 20 C.F.R. § 416.920, the ALJ determined that Ruenger had not worked within the claim period (step one); that his mental and physical impairments were severe (step two) but did not presumptively establish a disability (step three); and that he had the capacity to perform light work with certain physical and social limitations (step four). At the fifth and fi- nal step of the inquiry, the ALJ determined—based on a vocational expert’s testimony—that Ruenger could still perform jobs that exist nationwide in significant numbers. Accordingly, the ALJ denied Ruenger’s application.

Some context about step five is necessary. At this step, the agency bears the burden of demonstrating that there are significant numbers of jobs in the national economy for someone with the claimant’s abilities and limitations. See 20 C.F.R. § 416.960(c)(2). Because estimating job numbers is no easy feat, ALJs commonly rely on the testimony of vocational experts—professionals with experience in job placement and knowledge of working conditions. See Biestek v. Berryhill, 139 S. Ct. 1148, 1152 (2019); 20 C.F.R. § 416.966(e).

To provide tailored job-number estimates, vocational experts use various sources, and the expert here consulted three. The first is the Dictionary of Occupational Titles (DOT), a publication produced by the Department of Labor that lists job titles and their requirements. The DOT was last revised thirty years ago, leaving many of its job descriptions outdated. See Browning v. Colvin, 766 F.3d 702, 709 (7th Cir. 2014) (“No doubt many of the jobs [in the DOT] have changed and some have disappeared.”). Since 2008, the Social Security Administration has been working on a project to replace the DOT with an updated publication—a development this court continues to invite. See Chavez v. Berryhill, 895 F.3d 962, 966 (7th Cir. 2018). In any event, the DOT does not estimate how many positions exist in the national economy for each job title.

Because of this, vocational experts commonly use another source produced by the Department of Labor that does provide job-number estimates: the Occupational Employment Survey. Unfortunately for vocational experts, this publication organizes its estimates not by DOT job titles but by another classification system, the “standard occupational classification” (SOC) system. SOC codes sort jobs into broad occupational categories, such as “mathematicians” (SOC 15-2021) or “electrical engineers” (SOC 17-2071), that each encompass multiple DOT job titles. See May 2020 National Occupational Employment and Wage Estimates, U.S. BUREAU OF LABOR STATISTICS, https://www.bls.gov/oes/current/oes_nat.htm (last visited January 6, 2022). This creates a matching problem: vocational experts can identify the number of jobs in the larger SOC grouping but cannot identify how those jobs are distributed among individual DOT job titles within that grouping. See Chavez, 895 F.3d at 965–66.

To bridge this gap, vocational experts sometimes turn to a third source, the Occupational Employment Quarterly, which estimates the number of jobs available in the national economy for each DOT job title. It does so by using the “equal distribution method,” a calculation that simply divides the number of jobs estimated for an SOC code by the number of DOT titles contained within that SOC code. We have repeatedly questioned the accuracy of the equal distribution method, see, e.g., Alaura v. Colvin, 797 F.3d 503, 507–08 (7th Cir. 2015); Voigt v. Colvin, 781 F.3d 871, 879 (7th Cir. 2015); Browning, 766 F.3d at 709, because it illogically assumes that each DOT job title within an SOC code exists in equal numbers in the national economy. See Chavez, 895 F.3d at 966.

Here, the vocational expert testified that jobs in three categories existed in significant numbers for someone with Ruenger’s limitations: cafeteria attendant (106,000 jobs), office helper (214,000 jobs), and packager (316,000 jobs). She also provided three DOT job titles—“cafeteria attendant (hotel & restaurant),” “office helper (clerical),” and “packager opera- tor, automatic (tobacco)”—as examples of particular occupations included within her estimates. When the ALJ asked her to explain the methodology behind these estimates, she described a two-part process. First, she compiled job numbers from the Department of Labor’s Occupational Employment Survey. She testified that instead of using SOC codes, she looked through the industries listed in the Occupational Employment Survey and added up the estimates for “names of jobs” that were suitable for Ruenger. Second, she testified that she checked her job-number estimates against the Occupational Employment Quarterly, keeping her own estimates only when they came within 100 jobs of the estimate set forth in the Occupational Employment Quarterly’s comparable occupational grouping.

The ALJ adopted the vocational expert’s testimony over Ruenger’s objection. The expert’s testimony was reliable, the ALJ determined, because she articulated a specific method that was based on her experience and consistent with the DOT.

Ruenger appealed to the district court, contending that the vocational expert’s estimates were unreliable. The district court upheld the Commissioner’s decision, concluding that the ALJ sufficiently established the reliability of the job numbers by confirming the expert’s qualifications, ensuring that her testimony was consistent with the DOT, and asking her about her methodology. The court also approved of the expert’s use of the equal distribution method because she used it merely to corroborate the estimates she compiled based on her knowledge and experience. Ruenger then sought our review, again challenging the reliability of the vocational expert’s job-number estimates.

II On appeal, we ask whether substantial evidence supports the ALJ’s conclusion that there are significant numbers of jobs in the national economy for Ruenger to perform. See 42 U.S.C. § 405(g) (requiring Commissioner’s findings to be sustained if supported by substantial evidence). In the context of job-number estimates, substantial evidence requires the ALJ to ensure that the vocational expert’s estimate is the product of a reliable methodology. See Brace v. Saul, 970 F.3d 818, 821–22 (7th Cir. 2020). A methodology is reliable when it is based on “well-accepted” sources and the vocational expert explains her methodology “cogently and thoroughly.” Biestek, 139 S. Ct. at 1155. And when, as here, the claimant challenges the job-number estimate, the ALJ must compel the vocational expert to offer a “reasoned and principled explanation” of the methodology she used to produce the estimate. Chavez, 895 F.3d at 970. The expert’s explanation must be sufficient to instill some confidence that the estimate was not “conjured out of whole cloth.” Donahue v. Barnhart, 279 F.3d 441, 446 (7th Cir. 2002).

Ruenger first argues that the way in which the vocational expert compiled her job numbers is unclear. We agree. Because the expert failed to set forth an understandable methodology, we cannot review her methodology, let alone con- firm that it was reliable. The expert testified that she did not use SOC codes but instead added estimates for “names of jobs” within industries listed in the Occupational Employment Survey. Yet the Occupational Employment Survey fur- nishes job estimates only by SOC codes. Those codes are accompanied by job names (such as “marketing managers” or “computer programmers”), and the job-number estimates for each code can be adjusted for certain industries, but the job names do not exist independent of the codes. See May 2020 National Industry-Specific Occupational Employment and Wage Estimates, U.S. BUREAU OF LABOR STATISTICS, https://www.bls.gov/oes/current/oessrci.htm (last visited January 6, 2022). In other words, the vocational expert obscured the origin of her job estimates and even denied the most likely source—SOC codes. Without this fundamental information, the vocational expert’s testimony could not have provided the ALJ with sufficient confidence that her methodology was reliable.

The Supreme Court’s decision in Biestek supports this conclusion. Biestek, 139 S. Ct. at 1157. There, the Court declined to impose a categorical rule making a vocational expert’s testimony unreliable whenever she refuses to provide data. But it also held that an expert’s testimony will not qualify as substantial evidence when she keeps data private without good reason and her testimony lacks other markers of reliability. Id. Here, the expert had no interest in confidentiality because her data came from the Occupational Employment Survey, a publicly available source. And the issue in this case is not that the expert failed to provide specific numbers, but that her testimony contained inconsistencies and lacked the clarity needed for the ALJ to have confidence in her estimates. Id. at 1155 (testimony meets the substantial evidence threshold when the vocational expert “cogently and thoroughly” describes a well-accepted methodology).

Second, Ruenger argues that the vocational expert failed to justify her use of the equal distribution method. We again agree because her testimony lacked any indication why she trusted the method in this circumstance. True, as the district court found, she did not calculate her own estimates using the equal distribution method. Still, she relied on the method because she kept her own estimates only when they were consistent with the Occupational Employment Quarterly, a publication that uses the method to calculate its estimates. We have never enjoined the use of the equal distribution method, but we have required that a vocational expert justify her use of it. See Chavez, 895 F.3d at 969. Like the expert in Chavez, the expert here failed to justify her choice by, for example, draw- ing on her past experiences with the method or knowledge of job markets. Id. And like the expert in Brace, the expert here “never claimed that [her] method for estimating job numbers is a well-accepted one, much less explained why that is so.” Brace, 970 F.3d at 822. Nor did she testify about why her estimates should come within 100 jobs of the estimates in the Occupational Employment Quarterly. After all, her estimates ac- counted for Ruenger’s limitations, many of which are not ac- counted for by the Occupational Employment Quarterly.

Each of these concerns could have been avoided by further testimony from the expert, but the ALJ did not press her to elaborate upon her methodology. Although the ALJ asked her to describe her methodology, substantial evidence requires more. The ALJ must “hold the [vocational expert] to account for the reliability of [her] job-number estimates.” See Chavez, 895 F.3d at 970. But even after cross-examination raised doubts about the expert’s methodology, the ALJ here did not ask the expert to clarify what she meant by “names of jobs,” whether such names are different from SOC codes, or the reason she used the equal distribution method. Thus, nothing in the administrative record allows us to conclude that the vocational expert’s estimates reasonably approximate the number of suitable jobs that exist for Ruenger.

We are mindful of the time constraints and heavy case- loads faced by ALJs. But when a claimant challenges a vocational expert’s job-number estimates, the ALJ has a duty to spend time inquiring into the expert’s methodology. See Chavez, 895 F.3d at 970. This may require that ALJs ask more questions of vocational experts or slow down proceedings to give claimants a greater opportunity to pose their own questions. Otherwise, ALJs risk shifting the agency’s evidentiary burden to the claimant. Id.

As we determined in Brace and Chavez, a new step-five hearing is needed to explore the evidentiary gap in this case. At the hearing, the vocational expert may be able to expand on her testimony or make some other showing that significant jobs exist for Ruenger. See Chavez, 895 F.3d at 970–71; Brace, 970 F.3d at 823. Ruenger in turn will have the opportunity to challenge any such showing.

We VACATE and REMAND for further proceedings consistent with this opinion.

SCUDDER, Circuit Judge, concurring. I write separately to underscore the significance of ALJs failing to ensure a sound record when eliciting assistance from vocational experts for job-number estimates at step five of the disability inquiry.

All three judges on this panel, assisted by very talented law clerks, read the transcript of the VE’s testimony multiple times. The parties’ counsel surely read it many more times still. And yet nobody can explain with coherence or confi- dence what the VE did to arrive at her job-numbers estimate. To my eye, the VE’s testimony seemed rushed and rote, as if she expected certain questions and gave hurried and mechan- ical answers, without taking care—even in response to repeated objection—to explain what she did to arrive at the job-numbers estimate or why that method was reliable. We cannot make sense of the testimony—all of which came from a VE with substantial experience.

Some excerpts help illustrate the point. In response to questioning by Mr. Ruenger’s counsel as to the origin of her job-number estimates, the VE testified that she did not base her estimates on SOC codes. Instead, she said, “I do not use [SOC] code[s]. I use the names of jobs under the different— they’re sorted sort of by industry …. I go to each industry and look at the thing that would refer to that kind of a job in that industry …. So the [SOC] code is meaningless.”

Soon after, and plainly confused, the ALJ intervened to ask the VE to explain her methodology clearly and succinctly. The VE responded with this:

So I would look it up in the Occupational Employment Quarterly, and then I would look up under all of the industries in the Department of Labor wage and earning statistics things that refer to, say, packaging in that industry because I know those jobs in those industries from seeing them [in my professional experience] …. Then I add those numbers. Usually I have maybe nine amounts, like 1,200, 56,000 something, 20-some thousand something. I [add] those up, and if the total I get for those jobs that I know and have seen done is within 100 of the amount given by Occupational Employment Quarterly, then I use that as an example. If it isn’t close or it is hundreds of jobs apart, I throw it away, and I never use that.

No matter how many times we read this testimony, we cannot discern the VE’s methodology. To be sure, we recognize many of the VE’s references and much of the related administrative lingo. But recognizing dots does not tell us how, if at all, they connect. Faced with a transcript like this one, attempting to conduct judicial review is an exercise in futility. That is why we have concluded that substantial evidence did not support the ALJ’s denial of benefits for Randall Ruenger.

The concern underlying this evidentiary shortcoming extends beyond Mr. Ruenger’s case. The issue is more systemic. Since 2008, the Social Security Administration has been prom- ising courts and claimants alike that a new, unified jobs system—designed to simplify the process of compiling job-num- ber estimates—will soon be available. More than a decade later, the Administration has not completed its work. So today’s world is a distinct second best, with VEs made to cross-reference data points from multiple nonconversant data sets live on the witness stand at seemingly breakneck speed. There has to be a better way.

At the very least, the record would benefit from everyone slowing down when VEs take the stand. A disability determination may well mark the difference between income and no income for the claimant. With so much at stake in these proceedings, it is essential that a reviewing court be able to decipher the evidentiary record. Tapping the brake pedal may go a long way toward making everything more transparent.

We also have to imagine that there exist stopgap measures that can improve this process in the short term. At oral argument, we discussed with Mr. Ruenger’s counsel, who has substantial experience in these cases, the possibility of VEs preparing and providing brief written summaries of their meth- odologies to enter into the record. Maybe that would help things some.

In the end, though, it is not our place to prescribe a way forward. Perhaps the Commissioner will read this opinion as an invitation to bring long-awaited and much-needed improvement to this aspect of administrative disability determinations.