Justice KAGAN delivered the opinion of the Court.

Noel J. Francisco, Solicitor General, Joseph H. Hunt, Assistant Attorney General, Edwin S. Kneedler, Deputy Solicitor General, Anthony A. Yang, Assistant to the Solicitor General, Alisa B. Klein, Rachel F. Homer, Attorneys, Department of Justice, Washington, D.C., for Respondent.

Frederick J. Daley, Jr., Meredith Marcus, Daley Disability Law, PC, Chicago, IL, Ishan K. Bhabha, Natacha Y. Lam, Lauren J. Hartz, Jenner & Block LLP, Washington, DC, for Petitioner.

Justice KAGAN delivered the opinion of the Court.

The Social Security Administration (SSA) provides benefits to individuals who cannot obtain work because of a physical or mental disability. To determine whether an applicant is entitled to benefits, the agency may hold an informal hearing examining (among other things) the kind and number of jobs available for someone with the applicant's disability and other characteristics. The agency's factual findings on that score are "conclusive" in judicial review of the benefits decision so long as they are supported by "substantial evidence." 42 U.S.C. § 405(g).

This case arises from the SSA's reliance on an expert's testimony about the availability of certain jobs in the economy. The expert largely based her opinion on private market-survey data. The question presented is whether her refusal to provide that data upon the applicant's request categorically precludes her testimony from counting as "substantial evidence." We hold it does not.

I

Petitioner Michael Biestek once worked as a carpenter and general laborer on construction sites. But he stopped working after he developed degenerative disc disease, Hepatitis C, and depression. He then applied for social security disability benefits, claiming eligibility as of October 2009.

After some preliminary proceedings, the SSA assigned an Administrative Law Judge (ALJ) to hold a hearing on Biestek's application. Those hearings, as described in the Social Security Act, 49 Stat. 620, as amended, 42 U.S.C. § 301 et seq. , are recognizably adjudicative in nature. The ALJ may "receive evidence" and "examine witnesses" about the contested issues in a case. §§ 405(b)(1), 1383(c) (1)(A). But many of the rules governing such hearings are less rigid than those a court would follow. See Richardson v. Perales , 402 U.S. 389, 400–401, 91 S.Ct. 1420, 28 L.Ed.2d 842 (1971). An ALJ is to conduct a disability hearing in "an informal, non-adversarial manner." 20 CFR § 404.900(b) (2018) ; § 416.1400(b). Most notably, an ALJ may receive evidence in a disability hearing that "would not be admissible in court." §§ 404.950(c), 416.1450(c); see 42 U.S.C. §§ 405(b) (1), 1383(c)(1)(A).

To rule on Biestek's application, the ALJ had to determine whether the former construction laborer could successfully transition to less physically demanding work. That required exploring two issues. The ALJ needed to identify the types of jobs Biestek could perform notwithstanding his disabilities. See 20 CFR §§ 404.1560(c)(1), 416.960(c)(1). And the ALJ needed to ascertain whether those kinds of jobs "exist[ed] in significant numbers in the national economy." §§ 404.1560(c)(1), 416.960(c)(1) ; see §§ 404.1566, 416.966.

For guidance on such questions, ALJs often seek the views of "vocational experts." See §§ 404.1566(e), 416.966(e); SSA, Hearings, Appeals, and Litigation Law Manual I–2–5–50 (Aug. 29, 2014). Those experts are professionals under contract with SSA to provide impartial testimony in agency proceedings. See id. , at I–2–1–31.B.1 (June 16, 2016); id. , at I–2–5–48. They must have "expertise" and "current knowledge" of "[w]orking conditions and physical demands of various" jobs; "[k]nowledge of the existence and numbers of [those jobs] in the national economy"; and "[i]nvolvement in or knowledge of placing adult workers[ ] with disabilities[ ] into jobs." Id. , at I–2–1–31.B.1. Many vocational experts simultaneously work in the private sector locating employment for persons with disabilities. See C. Kubitschek & J. Dubin, Social Security Disability Law & Procedure in Federal Court § 3:89 (2019). When offering testimony, the experts may invoke not only publicly available sources but also "information obtained directly from employers" and data otherwise developed from their own "experience in job placement or career counseling." Social Security Ruling, SSR 00–4p, 65 Fed. Reg. 75760 (2000).

At Biestek's hearing, the ALJ asked a vocational expert named Erin O'Callaghan to identify a sampling of "sedentary" jobs that a person with Biestek's disabilities, education, and job history could perform. Tr. 59 (July 21, 2015); see 20 CFR §§ 404.1567(a), 416.967(a) (defining a "sedentary" job as one that "involves sitting" and requires "lifting no more than 10 pounds"). O'Callaghan had served as a vocational expert in SSA proceedings for five years; she also had more than ten years' experience counseling people with disabilities about employment opportunities. See Stachowiak v. Commissioner of Social Security , 2013 WL 593825, *1 (E.D. Mich., Jan. 11, 2013) ; Record in No. 16–10422 (ED Mich.), Doc. 17–13, p. 1274 (resume). In response to the ALJ's query, O'Callaghan listed sedentary jobs "such as a bench assembler [or] sorter" that did not require many skills. Tr. 58–59. And she further testified that 240,000 bench assembler jobs and 120,000 sorter jobs existed in the national economy. See ibid.

On cross-examination, Biestek's attorney asked O'Callaghan "where [she was] getting those [numbers] from." Id. , at 71. O'Callaghan replied that they came from the Bureau of Labor Statistics and her "own individual labor market surveys." Ibid. The lawyer then requested that O'Callaghan turn over the private surveys so he could review them. Ibid. O'Callaghan responded that she wished to keep the surveys confidential because they were "part of [her] client files." Id ., at 72. The lawyer suggested that O'Callaghan could "take the clients' names out." Ibid. But at that point the ALJ interjected that he "would not require" O'Callaghan to produce the files in any form. Ibid. Biestek's counsel asked no further questions about the basis for O'Callaghan's assembler and sorter numbers.

After the hearing concluded, the ALJ issued a decision granting Biestek's application in part and denying it in part. According to the ALJ, Biestek was entitled to benefits beginning in May 2013, when his advancing age (he turned fifty that month) adversely affected his ability to find employment. See App. to Pet. for Cert. 19a, 112a–113a. But before that time, the ALJ held, Biestek's disabilities should not have prevented a "successful adjustment to other work." Id. , at 110a–112a. The ALJ based that conclusion on O'Callaghan's testimony about the availability in the economy of "sedentary unskilled occupations such as bench assembler [or] sorter."Id. , at 111a (emphasis deleted).

Biestek sought review in federal court of the ALJ's denial of benefits for the period between October 2009 and May 2013. On judicial review, an ALJ's factual findings—such as the determination that Biestek could have found sedentary work—"shall be conclusive" if supported by "substantial evidence." 42 U.S.C. § 405(g) ; see supra , at 1151. Biestek contended that O'Callaghan's testimony could not possibly constitute such evidence because she had declined, upon request, to produce her supporting data. See Plaintiff's Motion for Summary Judgment in No. 16–10422 (ED Mich.), Doc. 22, p. 23. But the District Court rejected that argument. See 2017 WL 1173775, *2 (Mar. 30, 2017). And the Court of Appeals for the Sixth Circuit affirmed. See Biestek v. Commissioner of Social Security , 880 F.3d 778 (2017). That court recognized that the Seventh Circuit had adopted the categorical rule Biestek proposed, precluding a vocational expert's testimony from qualifying as substantial if the expert had declined an applicant's request to provide supporting data. See id. , at 790 (citing McKinnie v. Barnhart , 368 F.3d 907, 910–911 (2004) ). But that rule, the Sixth Circuit observed in joining the ranks of unconvinced courts, "ha[d] not been a popular export." 880 F.3d at 790 (internal quotation marks omitted).

And no more is it so today.

II

The phrase "substantial evidence" is a "term of art" used throughout administrative law to describe how courts are to review agency factfinding. T-Mobile South , LLC v. Roswell , 574 U.S. ––––, ––––, 135 S.Ct. 808, 815, 190 L.Ed.2d 679 (2015). Under the substantial-evidence standard, a court looks to an existing administrative record and asks whether it contains "sufficien[t] evidence" to support the agency's factual determinations. Consolidated Edison Co. v. NLRB , 305 U.S. 197, 229, 59 S.Ct. 206, 83 L.Ed. 126 (1938) (emphasis deleted). And whatever the meaning of "substantial" in other contexts, the threshold for such evidentiary sufficiency is not high. Substantial evidence, this Court has said, is "more than a mere scintilla." Ibid. ; see, e.g. , Perales , 402 U.S. at 401, 91 S.Ct. 1420 (internal quotation marks omitted). It means—and means only—"such relevant evidence as a reasonable mind might accept as adequate to support a conclusion." Consolidated Edison , 305 U.S. at 229, 59 S.Ct. 206. See Dickinson v. Zurko , 527 U.S. 150, 153, 119 S.Ct. 1816, 144 L.Ed.2d 143 (1999) (comparing the substantial-evidence standard to the deferential clearly-erroneous standard).

Today, Biestek argues that the testimony of a vocational expert who (like O'Callaghan) refuses a request for supporting data about job availability can never clear the substantial-evidence bar. See Brief for Petitioner 21–34. As that formulation makes clear, Biestek's proposed rule is categorical, rendering expert testimony insufficient to sustain an ALJ's factfinding whenever such a refusal has occurred. But Biestek hastens to add two caveats. The first is to clarify what the rule is not, the second to stress where its limits lie.

Biestek initially takes pains—and understandably so—to distinguish his argument from a procedural claim. Reply Brief 12–14. At no stage in this litigation, Biestek says, has he ever espoused "a free-standing procedural rule under which a vocational expert would always have to produce [her underlying data] upon request." Id. , at 2. That kind of rule exists in federal court: There, an expert witness must produce all data she has considered in reaching her conclusions. See Fed. Rule Civ. Proc. 26(a)(2)(B). But as Biestek appreciates, no similar requirement applies in SSA hearings. As explained above, Congress intended those proceedings to be "informal" and provided that the "strict rules of evidence, applicable in the courtroom, are not to" apply. Perales , 402 U.S. at 400, 91 S.Ct. 1420 ; see 42 U.S.C. § 405(b)(1) ; supra , at 1152. So Biestek does not press for a "procedural rule" governing "the means through which an evidentiary record [must be] created." Tr. of Oral Arg. 6; Reply Brief 13. Instead, he urges a "substantive rule" for "assess[ing] the quality and quantity of [record] evidence"—which would find testimony like O'Callaghan's inadequate, when taken alone, to support an ALJ's factfinding. Id. , at 12.

And Biestek also emphasizes a limitation within that proposed rule. For the rule to kick in, the applicant must make a demand for the expert's supporting data. See Brief for Petitioner i, 5, 18, 40, 55; Tr. of Oral Arg. 25–26. Consider two cases in which vocational experts rely on, but do not produce, nonpublic information. In the first, the applicant asks for the data; in the second, not. According to Biestek, the expert's testimony in the first case cannot possibly clear the substantial-evidence bar; but in the second case, it may well do so, even though the administrative record is otherwise the same. And Biestek underscores that this difference in outcome has nothing to do with waiver or forfeiture: As he acknowledges, an applicant "cannot waive the substantial evidence standard." Id. , at 27. It is just that the evidentiary problem arises from the expert's refusal of a demand, not from the data's absence alone. In his words, the testimony "can constitute substantial evidence if unchallenged, but not if challenged." Reply Brief 18.

To assess Biestek's proposal, we begin with the parties' common ground: Assuming no demand, a vocational expert's testimony may count as substantial evidence even when unaccompanied by supporting data. Take an example. Suppose an expert has top-of-the-line credentials, including professional qualifications and many years' experience; suppose, too, she has a history of giving sound testimony about job availability in similar cases (perhaps before the same ALJ). Now say that she testifies about the approximate number of various sedentary jobs an applicant for benefits could perform. She explains that she arrived at her figures by surveying a range of representative employers; amassing specific information about their labor needs and employment of people with disabilities; and extrapolating those findings to the national economy by means of a well-accepted methodology. She answers cogently and thoroughly all questions put to her by the ALJ and the applicant's lawyer. And nothing in the rest of the record conflicts with anything she says. But she never produces her survey data. Still, her testimony would be the kind of evidence—far "more than a mere scintilla"—that "a reasonable mind might accept as adequate to support" a finding about job availability. Consolidated Edison , 305 U.S. at 229, 59 S.Ct. 206. Of course, the testimony would be even better—more reliable and probative—if she had produced supporting data; that would be a best practice for the SSA and its experts. And of course, a different (maybe less qualified) expert failing to produce such data might offer testimony that is so feeble, or contradicted, that it would fail to clear the substantial-evidence bar. The point is only—as, again, Biestek accepts—that expert testimony can sometimes surmount that bar absent underlying data.

But if that is true, why should one additional fact—a refusal to a request for that data—make a vocational expert's testimony categorically inadequate? Assume that an applicant challenges our hypothetical expert to turn over her supporting data; and assume the expert declines because the data reveals private information about her clients and making careful redactions will take a fair bit of time. Nothing in the expert's refusal changes her testimony (as described above) about job availability. Nor does it alter any other material in the record. So if our expert's opinion was sufficient—i.e. , qualified as substantial evidence—before the refusal, it is hard to see why the opinion has to be insufficient afterward.

Biestek suggests two reasons for that non-obvious result. First, he contends that the expert's rejection of a request for backup data necessarily "cast[s her testimony] into doubt." Reply Brief 16. And second, he avers that the refusal inevitably "deprives an applicant of the material necessary for an effective cross-examination." Id. , at 2. But Biestek states his arguments too broadly—and the nuggets of truth they contain cannot justify his proposed flat rule.

Consider Biestek's claim about how an expert's refusal undercuts her credibility. Biestek here invokes the established idea of an "adverse inference": If an expert declines to back up her testimony with information in her control, then the factfinder has a reason to think she is hiding something. See id. , at 16 (citing cases). We do not dispute that possibility—but the inference is far from always required. If an ALJ has no other reason to trust the expert, or finds her testimony iffy on its face, her refusal of the applicant's demand for supporting data may properly tip the scales against her opinion. (Indeed, more can be said: Even if the applicant makes no demand, such an expert's withholding of data may count against her.) But if (as in our prior hypothetical example, see supra , at 1154 – 1156) the ALJ views the expert and her testimony as otherwise trustworthy, and thinks she has good reason to keep her data private, her rejection of an applicant's demand need not make a difference. So too when a court reviews the ALJ's decision under the deferential substantial-evidence standard. In some cases, the refusal to disclose data, considered along with other shortcomings, will prevent a court from finding that "a reasonable mind" could accept the expert's testimony. Consolidated Edison , 305 U.S. at 229, 59 S.Ct. 206. But in other cases, that refusal will have no such consequence. Even taking it into account, the expert's opinion will qualify as "more than a mere scintilla" of evidence supporting the ALJ's conclusion. Which is to say it will count, contra Biestek, as substantial.

And much the same is true of Biestek's claim that an expert's refusal precludes meaningful cross-examination. We agree with Biestek that an ALJ and reviewing court may properly consider obstacles to such questioning when deciding how much to credit an expert's opinion. See Perales , 402 U.S. at 402–406, 91 S.Ct. 1420. But Biestek goes too far in suggesting that the refusal to provide supporting data always interferes with effective cross-examination, or that the absence of such testing always requires treating an opinion as unreliable. Even without specific data, an applicant may probe the strength of testimony by asking an expert about (for example) her sources and methods—where she got the information at issue and how she analyzed it and derived her conclusions.See, e.g. , Chavez v. Berryhill , 895 F.3d 962, 969–970 (CA7 2018). And even without significant testing, a factfinder may conclude that testimony has sufficient indicia of reliability to support a conclusion about whether an applicant could find work. Indeed, Biestek effectively concedes both those points in cases where supporting data is missing, so long as an expert has not refused an applicant's demand. See supra , at 1154 – 1155. But once that much is acknowledged, Biestek's argument cannot hold. For with or without an express refusal, the absence of data places the selfsame limits on cross-examination.

Where Biestek goes wrong, at bottom, is in pressing for a categorical rule, applying to every case in which a vocational expert refuses a request for underlying data. Sometimes an expert's withholding of such data, when combined with other aspects of the record, will prevent her testimony from qualifying as substantial evidence. That would be so, for example, if the expert has no good reason to keep the data private and her testimony lacks other markers of reliability. But sometimes the reservation of data will have no such effect. Even though the applicant might wish for the data, the expert's testimony still will clear (even handily so) the more-than-a-mere-scintilla threshold. The inquiry, as is usually true in determining the substantiality of evidence, is case-by-case. See, e.g. , Perales , 402 U.S. at 399, 410, 91 S.Ct. 1420 (rejecting a categorical rule pertaining to the substantiality of medical reports in a disability hearing). It takes into account all features of the vocational expert's testimony, as well as the rest of the administrative record. And in so doing, it defers to the presiding ALJ, who has seen the hearing up close.

That much is sufficient to decide this case. Biestek petitioned us only to adopt the categorical rule we have now rejected. He did not ask us to decide whether, in the absence of that rule, substantial evidence supported the ALJ in denying him benefits. Accordingly, we affirm the Court of Appeals' judgment.

It is so ordered.

Justice SOTOMAYOR, dissenting.

The Court focuses on the propriety of a categorical rule that precludes private data that a vocational expert refuses to provide upon request from qualifying as " ‘substantial evidence.’ " See ante , at 1152. I agree with Justice GORSUCH that the question presented by this case encompasses an inquiry not just into the propriety of a categorical rule in such circumstances but also into whether the substantial-evidence standard was met in the narrower circumstances of Michael Biestek's case. See post , at 1161 – 1162 (dissenting opinion). For the reasons that Justice GORSUCH sets out, the vocational expert's conclusory testimony in this case, offered without even a hint of support, did not constitute substantial evidence.

Once Biestek established that he had impairments, the agency bore the burden of proving that work opportunities were available to someone with his disabilities and individual characteristics. 20 CFR § 416.912(b)(3) (2018). To meet that burden, the agency relied on a vocational expert's testimony that Biestek could qualify for one of 240,000 "bench assembler" jobs or 120,000 "sorter" jobs nationwide. Tr. 59 (July 21, 2015). The expert said that those numbers were based in part on her "professional experience." Id. , at 61. When Biestek's counsel understandably asked for more details, the expert said only that she got the numbers from a publicly available source as well as from her "own individual labor market surveys" that were part of confidential client files. Id. , at 71; see id. , at 67, 71–72. Biestek's counsel asked if the names in the files could be redacted, but the administrative law judge (ALJ) interrupted and ruled that she would not require the surveys to be produced in redacted form. Id. , at 72; see also id ., at 67.

Perhaps the ALJ would have allowed Biestek's counsel to ask followup questions about the basis for the testimony at that point, and perhaps Biestek's counsel should have tried to do so. But a Social Security proceeding is "inquisitorial rather than adversarial." Sims v. Apfel , 530 U.S. 103, 110–111, 120 S.Ct. 2080, 147 L.Ed.2d 80 (2000) ; see 20 CFR §§ 404.900(b), 416.1400(b). The ALJ acts as "an examiner charged with developing the facts," Richardson v. Perales , 402 U.S. 389, 410, 91 S.Ct. 1420, 28 L.Ed.2d 842 (1971), and has a duty to "develop the arguments both for and against granting benefits," Sims , 530 U.S. at 111, 120 S.Ct. 2080 ; see also Social Security Ruling, SSR 00–4P, 65 Fed. Reg. 75760 (2000) (noting "the adjudicator's duty to fully develop the record"). Here, instead of taking steps to ensure that the claimant had a basis from which effective cross-examination could be made and thus the record could be developed, the ALJ cut off that process by intervening when Biestek's counsel asked about the possibility of redaction.

The result was that the expert offered no detail whatsoever on the basis for her testimony. She did not say whom she had surveyed, how many surveys she had conducted, or what information she had gathered, nor did she offer any other explanation of the data on which she relied. In conjunction with the failure to proffer the surveys themselves, the expert's conclusory testimony alone could not constitute substantial evidence to support the ALJ's factfinding.

I agree with much of Justice GORSUCH's reasoning. I emphasize that I do not foreclose the possibility that a more developed record could justify an ALJ's reliance on vocational-expert testimony in some circumstances even if the expert does not produce records underlying that testimony on request. An expert may have legitimate reasons for not turning over data, such as the burden of gathering records or confidentiality concerns that redaction cannot address. In those circumstances, as the majority suggests, the agency may be able to support an expert's testimony in ways other than by providing underlying data, such as by offering a fulsome description of the data and methodology on which the expert relies. See ante , at 1162. The agency simply did not do so here.

Justice GORSUCH, with whom Justice GINSBURG joins, dissenting.

Walk for a moment in Michael Biestek's shoes. As part of your application for disability benefits, you’ve proven that you suffer from serious health problems and can't return to your old construction job. Like many cases, yours turns on whether a significant number of other jobs remain that someone of your age, education, and experience, and with your physical limitations, could perform. When it comes to that question, the Social Security Administration bears the burden of proof. To meet its burden in your case, the agency chooses to rest on the testimony of a vocational expert the agency hired as an independent contractor. The expert asserts there are 120,000 "sorter" and 240,000 "bench assembler" jobs nationwide that you could perform even with your disabilities.

Where did these numbers come from? The expert says she relied on data from the Bureau of Labor Statistics and her own private surveys. But it turns out the Bureau can't be the source; its numbers aren't that specific. The source—if there is a source—must be the expert's private surveys. So you ask to see them. The expert refuses—she says they’re part of confidential client files. You reply by pointing out that any confidential client information can be redacted. But rather than ordering the data produced, the hearing examiner, herself a Social Security Administration employee, jumps in to say that won't be necessary. Even without the data, the examiner states in her decision on your disability claim, the expert's say-so warrants "great weight" and is more than enough evidence to deny your application. Case closed. App. to Pet. for Cert. 111a–112a, 118a–119a.

Would you say this decision was based on "substantial evidence"? Count me with Judge Easterbrook and the Seventh Circuit in thinking that an agency expert's bottom-line conclusion, supported only by a claim of readily available evidence that she refuses to produce on request, fails to satisfy the government's statutory burden of producing substantial evidence of available other work. See Donahue v. Barnhart , 279 F.3d 441, 446 (CA7 2002) ; McKinnie v. Barnhart , 368 F.3d 907, 910–911 (CA7 2004) (per curiam ).

Start with the legal standard. The Social Security Act of 1935 requires the agency to support its conclusions about the number of available jobs with "substantial evidence." 42 U.S.C. § 405(g). Congress borrowed that standard from civil litigation practice, where reviewing courts may overturn a jury verdict when the record lacks "substantial evidence"—that is, evidence sufficient to permit a reasonable jury to reach the verdict it did. Much the same standard governs summary judgment and directed verdict practice today. See 2 K. Hickman & R. Pierce, Administrative Law § 10.2.1, pp. 1082–1085 (6th ed. 2019); Anderson v. Liberty Lobby, Inc. , 477 U.S. 242, 252, 106 S.Ct. 2505, 91 L.Ed.2d 202 (1986) ; NLRB v. Columbian Enameling & Stamping Co. , 306 U.S. 292, 300, 59 S.Ct. 501, 83 L.Ed. 660 (1939).

Next, consider what we know about this standard. Witness testimony that's clearly wrong as a matter of fact cannot be substantial evidence. See Scott v. Harris , 550 U.S. 372, 380, 127 S.Ct. 1769, 167 L.Ed.2d 686 (2007). Falsified evidence isn't substantial evidence. See, e.g ., Firemen's and Policemen's Civil Serv. Comm'n v. Brinkmeyer , 662 S.W.2d 953, 956 (Tex. 1984). Speculation isn't substantial evidence. See, e.g. , Cao He Lin v. Department of Justice , 428 F.3d 391, 400 (CA2 2005) ; Alpo Petfoods, Inc. v. NLRB , 126 F.3d 246, 250 (CA4 1997). And, maybe most pointedly for our purposes, courts have held that a party or expert who supplies only conclusory assertions fails this standard too. See, e.g. , Lujan v. National Wildlife Federation , 497 U.S. 871, 888, 110 S.Ct. 3177, 111 L.Ed.2d 695 (1990) ("The object of [summary-judgment practice] is not to replace conclusory allegations of the complaint or answer with conclusory allegations of an affidavit"); Regents of Univ. of Minn. v. AGA Medical Corp. , 717 F.3d 929, 941 (CA Fed. 2013) ("conclusory expert assertions cannot raise triable issues of material fact") (collecting cases); Mid-State Fertilizer Co. v. Exchange Nat. Bank of Chicago , 877 F.2d 1333, 1339 (CA7 1989) ("An expert who supplies nothing but a bottom line supplies nothing of value to the judicial process"); Sea Robin Pipeline Co. v. FERC , 795 F.2d 182, 188 (CADC 1986) ("[I]nordinate faith in the conclusory assertions of an expert ... cannot satisfy the requirement [of] substantial evidence").

If clearly mistaken evidence, fake evidence, speculative evidence, and conclusory evidence aren't substantial evidence, the evidence here shouldn't be either. The case hinges on an expert who (a) claims to possess evidence on the dispositive legal question that can be found nowhere else in the record, but (b) offers only a conclusion about its contents, and (c) refuses to supply the evidence when requested without showing that it can't readily be made available. What reasonable factfinder would rely on evidence like that? It seems just the sort of conclusory evidence courts have long held insufficient to meet the substantial evidence standard. And thanks to its conclusory nature, for all anyone can tell it may have come out of a hat—and, thus, may wind up being clearly mistaken, fake, or speculative evidence too. Unsurprisingly given all this, the government fails to cite even a single authority blessing the sort of evidence here as substantial evidence, despite the standard's long history and widespread use.

Veteran Social Security practitioners must be feeling a sense of déjà vu . Half a century ago, Judge Henry Friendly encountered Kerner v. Flemming , 283 F.2d 916 (CA2 1960). There, the agency's hearing examiner offered "nothing save [his own] speculation" to support his holding that the claimant "could in fact obtain substantial gainful employment." Id. , at 921. The Second Circuit firmly explained that this kind of conclusory claim is insufficient to meet the substantial evidence standard. In response, the Social Security Administration began hiring vocational experts, like the one in this case, to document the number of jobs available to a given claimant. But if the government can do what it did in this case, it's hard to see what all the trouble was for. The agency might still rest decisions on a hunch—just so long as the hunch comes from an agency contractor rather than an agency examiner.

Instead of addressing the realities of this case, the government asks us to imagine a hypothetical one. Assume, it says, that no one had requested the underlying data. In those circumstances, the government points out, even Mr. Biestek appears to accept that the agency's decision could have stood. And if that's true, the government asks, why should it make a difference if we add only one additional fact—the expert's refusal to produce the data? See ante , at 1154 – 1156 (presenting the same argument).

The answer is an old and familiar one. The refusal to supply readily available evidentiary support for a conclusion strongly suggests that the conclusion is, well, unsupported. See, e.g. , Interstate Circuit, Inc. v. United States , 306 U.S. 208, 226, 59 S.Ct. 467, 83 L.Ed. 610 (1939) ("The production of weak evidence when strong is available can lead only to the conclusion that the strong would have been adverse"); Clifton v. United States , 4 How. 242, 248, 11 L.Ed. 957 (1846) (the withholding of "more direct" proof suggests that "if the more perfect exposition had been given it would have laid open deficiencies and objections which the more obscure and uncertain testimony was intended to conceal"); 31A C.J. S., Evidence § 156(2), p. 402 (1964) ("The unfavorable inference ... is especially applicable where the party withholding the evidence has had notice or has been ordered to produce it"). Meanwhile, a similar inference may not arise if no one's bothered to ask for the evidence, or if the evidence is shown to be unavailable for a good reason. In cases like those, there may be just too many other plausible and innocent excuses for the evidence's absence. Maybe, for example, nobody bothered to seek the underlying data because everyone knew what it would show.

Fine, the Court responds, all that's true enough. But even if we accept that an expert's failure to produce the evidence underlying her conclusion may support an inference that her conclusion is unsupported, that doesn't mean such an inference must follow. Whether an inference is appropriate depends on the facts of the particular case. See ante , at 1156 – 1157.

But what more do we need to know about the facts of this case? All of the relevant facts are undisputed, and it remains only to decide the legal question whether they meet the substantial evidence standard. We know that the expert offered a firm and exact conclusion about the number of available jobs. We know that the expert claimed to have private information to support her conclusion. We know Mr. Biestek requested that information and we have no reason to think any confidentiality concerns could not have been addressed. We know, too, that the hearing examiner had "no other reason to trust the expert['s]" numbers beyond her say-so. Ibid. Finally and looking to the law, we know that a witness's bare conclusion is regularly held insufficient to meet the substantial evidence threshold—and we know that the government hasn't cited a single case finding substantial evidence on so little. This is exactly the sort of case where an adverse inference should "tip the scales." Ibid.

With so much now weighing against the government, everything seems to turn on a final hypothetical. Now we are asked to imagine that the expert had offered detailed oral testimony about the withheld data. Her testimony was so detailed, we are asked to suppose, that Mr. Biestek could have thoroughly tested the data's reliability through cross-examination. (You might wonder just how effective this cross-examination could be if Mr. Biestek didn't have access to the data. But overlook that.) Surely in those circumstances it wouldn't matter whether the expert failed to produce the data even in bad faith. Any failure to produce would be harmless as a matter of law because the expert's testimony, all by itself, would amount to substantial evidence on which a rational factfinder might rely. Ante , at 1156 – 1157.

The problem is that this imaginary case has nothing to teach us about our real one. In Mr. Biestek's case, it is undisputed that the expert offered only a bare conclusion about the number of available jobs. No other relevant testimony was offered or received: no testimony about the underlying data, no testimony about its specific sources, no testimony about its reliability. In our real case, there is simply no way to shrug off the failure to produce the data as harmless error. To the contrary, and as we have seen, cases like this routinely fail to satisfy the substantial evidence standard. And if the government has a "duty to fully develop the record," ante , at 1158 (SOTOMAYOR, J., dissenting), that conclusion should follow all the more strongly.

What leads the Court to a different conclusion? It says that it views Mr. Biestek's petition as raising only the "categorical" question whether an expert's failure to produce underlying data always and in "every case" precludes her testimony from qualifying as substantial evidence. Ante , at 1152, 1156 – 1157. And once the question is ratcheted up to that level of abstraction, of course it is easy enough to shoot it down: just point to a series of hypothetical cases where the record contains additional justification for the expert's failure to produce or additional evidence to support her opinion. In such counterfactual cases, the failure to produce either would not be enough to give rise to an adverse inference under traditional legal principles or could be held harmless as a matter of law. See ante , at 1154 – 1157.

But as I understand Mr. Biestek's submission, it does not require an all-or-nothing approach that would cover "every case." As the Court acknowledges, Mr. Biestek has focused us "on the Seventh Circuit's categorical rule." Ante , at 1154, n. 1. And that "rule" targets the narrower "category" of circumstances we have here—where an expert " ‘give[s] a bottom line,’ " fails to provide evidence "underlying that bottom line" when challenged, and fails to show the evidence is unavailable. McKinnie , 368 F.3d at 911 (quoting Donahue , 279 F.3d at 446 ). What to do about that category falls well within the question presented: "[w]hether a vocational expert's testimony can constitute substantial evidence of ‘other work’ ... when the expert fails upon the applicant's request to provide the underlying data on which that testimony is premised." Pet. for Cert. i. The answer to that question may be "always," "never," or—as the Court itself seems to acknowledge—"[s]ometimes." Ante , at 1157. And if the answer is "sometimes," the critical question becomes "in what circumstances"?

I suppose we could stop short and leave everyone guessing. But another option is to follow the Seventh Circuit's lead, resolve the smaller yet still significant "category" of cases like the one before us, and in that way begin to offer lower courts meaningful guidance in this important area. While I would not hesitate to take this course and make plain that cases like Mr. Biestek's fail the substantial evidence standard, I understand the Court today to choose the first option and leave these matters for another day.

There is good news and bad news in this. If my understanding of the Court's opinion is correct, the good news is that the Court remains open to the possibility that in real-world cases like Mr. Biestek's, lower courts may—and even should—find the substantial evidence test unmet. The bad news is that we must wait to find out, leaving many people and courts in limbo in the meantime. Cases with facts like Mr. Biestek's appear to be all too common. See, e.g. , Dubin, Overcoming Gridlock: Campbell After a Quarter-Century and Bureaucratically Rational Gap-Filling in Mass Justice Adjudication in the Social Security Administration's Disability Programs, 62 Admin. L. Rev. 937, 966 (2010). And many courts have erred in them by finding the substantial evidence test met, as the Sixth Circuit did in the case now before us. Some courts have even conflated the substantial evidence standard—a substantive standard governing what's needed to sustain a judgment as a matter of law—with procedural rules governing the admission of evidence. These courts have mistakenly suggested that, because the Federal Rules of Evidence don't apply in Social Security proceedings, anything an expert says will suffice to meet the agency's burden of proof. See, e.g ., Welsh v. Commissioner of Social Security , 662 Fed. Appx. 105, 109–110 (CA3 2016) ; Bayliss v. Barnhart , 427 F.3d 1211, 1218, and n. 4 (CA9 2005). Definitively resolving this case would have provided more useful guidance for practitioners and lower courts that have struggled with a significant category of cases like Mr. Biestek's, all while affording him the relief the law promises in disputes like his.

The principle that the government must support its allegations with substantial evidence, not conclusions and secret evidence, guards against arbitrary executive decisionmaking. See Friendly, " Some Kind of Hearing," 123 U. Pa. L. Rev. 1267, 1313–1314 (1975). Without it, people like Mr. Biestek are left to the mercy of a bureaucrat's caprice. Over 100 years ago, in ICC v. Louisville & Nashville R. Co. , 227 U.S. 88, 33 S.Ct. 185, 57 L.Ed. 431 (1913), the government sought to justify an agency order binding private parties without producing the information on which the agency had relied. The government argued that its findings should be "presumed to have been supported." Id. , at 93, 33 S.Ct. 185. In essence, the government sought the right to "act upon any sort of secret evidence." Gellhorn, Official Notice in Administrative Adjudication, 20 Texas L. Rev. 131, 145 (1941). This Court did not approve of that practice then, and I would not have hesitated to make clear that we do not approve of it today.

I respectfully dissent.