1. See, e.g., Larry Laudan, Truth, Error, and Criminal Law: An Essay in Legal Epistemology (2006); H.L. Ho, A Philosophy of Evidence Law (2008); Redmayne, Mike, Exploring the Proof Paradoxes, 14 Legal Theory281 (2008)Google Scholar; Pardo, Michael S. & Allen, Ronald J., Juridical Proof and the Best Explanation, 27 Law & Phil.223 (2008)Google Scholar; Pardo, Michael S., The Field of Evidence and the Field of Knowledge, 24 Law & Phil.321 (2005)Google Scholar; Pritchard, Duncan, Testimony, in 1 The Trial on Trial (Anthony Duff et al. eds., 2004)Google Scholar; Allen, Ronald J. & Leiter, Brian, Naturalized Epistemology and the Law of Evidence, 87 Va. L. Rev.1491 (2001)Google Scholar; Brewer, Scott, Scientific Expert Testimony and Intellectual Due Process, 107 Yale L.J.1535 (1998)Google ScholarPubMed.

2. See Gettier, Edmund L., Is Justified True Belief Knowledge?, 23 Analysis121 (1963)Google Scholar.

3. Id.

4. Id.

5. See Beltrán, Jordi Ferrer, Legal Proof and Fact Finders' Beliefs, 12 Legal Theory293 (2006)Google Scholar; Cohen, L. Jonathan, Should a Jury Say What It Believes or What It Accepts?, 13 Cardozo L. Rev. (1991)Google Scholar.

6. See Ferrer, supra note 5.

7. Id. at 294. In the United States, this sufficiency requirement depends on what reasonable fact finders could conclude based on the evidence and the burden of proof. See Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 250–256 (1986) (explicating the standard for civil cases); Jackson v. Virginia, 443 U.S. 307 (1979) (explicating the standard for criminal cases).

8. See Redmayne, supra note 1, at 299.

9. See id. at 302.

10. Alex Stein, Foundations of Evidence Law 59 (2005).

11. See Ferrer, supra note 5; Ho, supra note 1, at 89–99. For a discussion of the justification of legal verdicts, see Amaya, Amalia, Justification, Coherence, and Epistemic Responsibility in Legal Fact-Finding, 5 Episteme306 (2008)Google Scholar.

12. Ferrer, supra note 5, at 307.

13. Ho, supra note 1, at 93.

14. Id. at 95–96.

15. Id. at 87–89.

16. This criterion could also be added to the belief and belief-plus accounts to create a true-belief account and a true-belief-plus account (without a justification requirement). I discuss these possibilities in the next section.

17. Cf. Glanville Williams, A Short Rejoinder, Crim. L. Rev. 103, 104 (1980) (“There is a miscarriage of justice whenever an innocent man is convicted.”).

18. Eugenio Bulygin, Cognition and Interpretation of Law, in Cognition and Interpretation of Law 20 (L. Gianformaggio & S. Paulson eds., 1995) (“for punishing Tom the law requires not only that the sentence ‘Tom killed Peter’ be true, but also that there be sufficient evidence for the truth of this sentence.”)

19. Ho, supra note 1, at 99. Ho attempts to derive this requirement from the nature of beliefs (i.e., that beliefs aim at truth); however, he acknowledges that it is unclear whether such an argument can succeed. Id.

20. See 3 Anthony Duff et al., The Trial on Trial: Towards a Normative Theory of the Criminal Trial 87–91 (2007); see also Pardo, Field of Evidence, supra note 1, at 322–323 (endorsing a knowledge account for verdicts). A number of scholars have explicitly rejected a knowledge account. See Redmayne, supra note 1; Stein, supra note 10; Ferrer, supra note 5; Ho, supra note 1; Douglas Walton, Witness Testimony Evidence: Argumentation, Artificial Intelligence, and Law (2007); Sorensen, Roy, Future Law: Prepunishment and the Causal Theory of Verdicts, 40 Noûs166 (2006)Google Scholar.

21. Duff et al., supra note 20, at 91.

22. Id.

23. Id.

24. Id. I previously also endorsed a knowledge account on similar grounds: that a justified true verdict may be true by mere coincidence without any connection to the justifying evidence. See Pardo, Field of Evidence, supra note 1, at 322–323.

25. See Robert B. Brandom, Making It Explicit 157–163 (1998).

26. Moreover, beliefs and acceptances often converge. They may diverge, however, in cases with mandatory evidentiary presumptions.

27. For a discussion of material and probatory adjudicative errors, see Laudan, supra note 1, at 12–13.

28. The three accounts differ on what that level is—belief, belief-plus, or justification.

29. I am putting aside issues of jury nullification, in which an acquittal may not be an error when the evidence establishes proof beyond a reasonable doubt.

30. See In re Davis, 130 S. Ct. 1, 2–3 (2009) (Scalia, J., dissenting) (“This Court has never held that the Constitution forbids the execution of a convicted defendant who has had a full and fair trial but is later able to convince a habeas court that he is ‘actually’ innocent.”). The dissent came from an order directing a federal district court to hold an evidentiary hearing on whether a state prisoner, currently sentenced to death, could prove his innocence. Whether state prisoners can challenge their convictions through federal habeas review solely on grounds of “actual innocence” is technically an open question for the Supreme Court. See District Attorney's Office v. Osborne, 129 S. Ct. 2308 (2009); House v. Bell, 547 U.S. 518, 555 (2006). Under current law, federal courts may grant habeas relief to state prisoners only when their convictions are “contrary to, or an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” 28 U.S.C. §2254(d)(1). Thus, Justice Scalia argued, Davis cannot get relief even if the district court concludes that Davis has clearly proven his innocence. Justice Thomas joined the dissent. Justice Stevens (joined by Justices Ginsburg and Breyer) concurred in the order and argued that Davis may still get relief. The case is likely to return to the Supreme Court if Davis's proof is persuasive. The analysis above regarding the probatory and material perspectives clarifies the disagreement among the justices.

31. Cf. John Hawthorne & Stanley, Jason, Knowledge and Action, 105 J. Phil. 571 (2008)Google Scholar (arguing that knowledge provides the norm for action but that some failures to comply are excusable or less blameworthy than others).

32. Even when legal decision rules are conceptualized probabilistically (e.g., proof beyond a 0.5 likelihood in civil cases or beyond a 0.9 likelihood in criminal cases), a finding is materially accurate when it is 100 percent accurate in the sense that there is a finding when the disputed proposition is true and no finding when it is false.

33. See In re Davis, supra note 30, at 1–2; see supra note 30.

34. By contrast, possible disagreements with truth can be reconciled by holding truth constant and focusing on the legal proof side. My analysis assumes a minimally realist conception of truth in the sense that the truth transcends the evidence and what occurs at trial.

35. Much of this disagreement may have to do with the fact that there is no generally accepted account of epistemic justification. See William P. Alston, Beyond “Justification”: Dimensions of Epistemic Evaluation (2005).

36. See David Lewis, Elusive Knowledge, in Papers in Metaphysics and Epistemology 418 (1999).

37. See Jason Stanley, Knowledge and Practical Interests (2005).

38. For a discussion of rules in legal proof and epistemology, see Schauer, Frederick, In Defense of Rule-Based Evidence Law—and Epistemology Too, 5 Episteme295 (2008)Google Scholar.

39. See Hawthorne & Stanley, supra note 31.

40. See Stanley, supra note 37.

41. Material errors may be false positives (i.e., finding a false proposition to be proven) or false negatives (i.e., finding a true proposition to be not proven). From a systemic perspective, the relevant policy considerations concern not only minimizing total errors but also whether to prefer one type of error over another.

42. From this perspective, it does not matter whether any individual verdict is justified. A system that produced more accurate verdicts would be superior to a system that produced fewer, even if the latter produced more verdicts that were justified. Epistemic justification is no guarantee of truth—this is one lesson of the Gettier problem—and so it is, of course, possible that a highly accurate proof system could have few justified verdicts and that a highly inaccurate system could have many justified verdicts. These possibilities, however, run counter to the basic presuppositions of evidence law and legal proof that (1) verdicts better supported by the available evidence (and thus more justified than their alternatives) are more likely to be true; and (2) verdicts supported by better evidence (and thus more likely to be justified than verdicts based on worse evidence) are more likely to be true.

43. By analogy, suppose the proof process were instead a mechanical test from a “guilt machine” that purported to tell us whether a particular defendant was guilty. If the machine reported a result of “guilty,” the defendant could object on the ground that the test itself does a poor job of sorting guilty from innocent defendants (systemic epistemic concerns); or alternatively, the defendant could object on the ground that even if the test is generally good at identifying guilt, there was a problem in administering his test that undermined the particular result (case-specific epistemic concerns). Both appear to be legitimate challenges that a defendant could raise, either of which would undermine the epistemic success of a legal verdict.

44. Sorensen, supra note 20, is an exception. I discuss Sorensen's examples in the next section. For a discussion of the problem in the context of substantive criminal law, see Douglas N. Husak & Craig A. Callender, Wilful Ignorance, Knowledge, and the “Equal Culpability” Thesis: A Study of the Deeper Significance of the Principle of Equality, 1994 Wis. L. Rev. 29, 44–53; and for a discussion of the problem in the context of witness testimony and hearsay evidence, see Pardo, Michael S., Testimony, 82 Tul. L. Rev.119 (2007)Google Scholar.

45. See William G. Lycan, On the Gettier Problem Problem, in Epistemology Futures 148 (Stephen Hetherington ed., 2006) (“the Gettier problem became a leading focus, if not the focus, of disenchantment with the definition-and-counterexample method of analytic philosophy. In some cases the disenchantment spilled over into scorn; there were slighting references to the ‘S knows that P’ crowd.”). But see Timothy Williamson, The Philosophy of Philosophy 179 (2007) (referring to Gettier's article as the “canonical example in the literature on philosophical thought experiments”).

46. Weinberg, Jonathan M., Nichols, Shaun & Stich, Stephen, Normativity and Epistemic Intuitions, 29 Phil. Topics429–460 (2001)Google Scholar, reprinted in Experimental Philosophy 17 (J. Knobe & S. Nichols eds., 2008).

47. The argument is an abductive one from specific cases to general conclusions; in other words, the examples are given epistemic priority over any general definitions or principles. Moreover, as Williamson explains, the hypothetical cases themselves (not psychological facts about intuitions) provide evidence for evaluating the counterfactual conditionals they raise. See Williamson, supra note 45, at 235. Hypothetical Gettier cases provide counterexamples to the modal claim that knowledge coincides with JTBs in all possible cases; actual Gettier cases provide counterexamples to the nonmodal claim that they coincide in all actual cases. Id. at 193.

48. Gettier, supra note 2, at 1–2.

49. Id. at 2–3.

50. More specifically, Gettier cases often involve instances of bad luck (undermining justification) followed by good luck (making the belief true). See Zagzebski, Linda, The Inescapability of Gettier Problems, 44 Phil. Q.65 (1994)Google Scholar. For a more developed and nuanced discussion of the varieties of luck in epistemology, see Duncan Pritchard, Epistemic Luck (2005).

51. Lycan, supra note 45, at 154; I. Scheffler, Conditions of Knowledge (1965).

52. Lycan, supra note 45, at 154; R.M. Chisolm, Theory of Knowledge (1966).

53. Goldman, Alvin I., Discrimination and Perceptual Knowledge, 73 J. Phil.771 (1976)Google Scholar.

54. For a good overview of the literature on the Gettier problem, see Lycan, supra note 45.

55. See Zagzebski, supra note 50.

56. See Lycan, supra note 45.

57. See Pardo, Field of Evidence, supra note 1, at 322–323. As in Gettier's examples, the legal examples assume we know the truth. One objection raised during a presentation of an early version of this paper was that we typically never know the truth in legal settings—we only have better or worse evidence—and hence I am not entitled to this assumption. Assuming truth, however, simply allows us to reflect more clearly on the conceptual point regarding a possible disconnect between truth and justification manifested in Gettier cases; it is not an empirical claim about whether we know the truth in any actual cases nor a metaphysical claim about whether such knowledge is possible. In any event, those persuaded by this objection could reframe the examples by describing them as involving two perspectives: our own external perspective, in which we have better evidence, and the fact finders' internal one, in which they are presented with the trial evidence. In these reframed examples, my analysis would still go through in the relevant respects. The jury would arrive at the same conclusion as we do, although coincidentally; the Gettier-ized gap would now be between our justification and their justification (rather than between the truth and their justification).

58. This idea appears to be at the core of Justice Stevens's concurrence in In re Davis. See supra note 30.

59. For example, suppose a jury is instructed to find the mens rea element of a crime proven if the defendant had one of two mutually exclusive mental states.

60. For a more realistic example, consider an eyewitness identification that turns out to be true, but given the conditions, the witness would have made a (false) positive identification if several other people had been presented to the witness instead.

61. The verdict is problematic primarily because it is unsafe—that is, in a number of similar possible worlds the jury would have reached the same result and been in error.

62. Alternatively, one might propose strengthening the concept of justification and concluding that these conclusions are not really justified. However, this runs the risk of collapsing justification into truth. For a discussion, see Zagzebski, supra note 50.

63. Sorensen, supra note 20.

64. Id. at 170.

65. Gilbert Harman, Thought 143–144 (1973).

66. Sorensen, supra note 20, at 171.

67. Id.

68. Id. (“Although the jurors do not know that the assassin killed the civil rights leader . . . their verdict is still just.”).

69. See Lycan, supra note 45.

70. Moreover, the idea that false or misleading evidence not presented could undermine the epistemic success of agents will seem strange to most evidence scholars, I think, given that excluding misleading evidence is one of the main strategies for preserving the epistemic integrity of verdicts.

71. Sorensen, supra note 20, at 171.

72. Id.

73. Id.

74. Id.

75. Id. at 172.

76. His theory would exclude any “forward-looking” evidence. See Thomson, Judith Jarvis, Liability and Individualized Evidence, 49 Law & Contemp. Probs.199 (1986)Google Scholar; see also Redmayne, supra note 1, at 296 (noting that Sorensen's theory is problematic because it would exclude motive evidence).

77. An analogy with science may be helpful. Well-established scientific facts or theories may fail to qualify as “knowledge” under some epistemological theories or conceptions of knowledge, but this does not render them epistemically equivalent to baseless conclusions or mere conjecture. Nor does it render a concept of “scientific knowledge” useless for the philosophy of science or other theoretical endeavors. Likewise, even if legal proof fails to generate genuine knowledge as that concept is understood by some (many) epistemologists, this neither reduces legal judgments to the level of guesses or conjecture, nor does it render a concept of “legal knowledge” theoretically useless for the philosophy of law.

78. See Weinberg, Nichols & Stich, supra note 46. The simple fact that there is disagreement about the examples does not by itself undermine the examples as sources of evidence. For a discussion of this point, see Williamson, supra note 45, at 192. For example, there is often disagreement among eyewitnesses, but eyewitness testimony remains an important source of knowledge.

79. See Lycan, supra note 45; Frank Jackson, Book Review: Experimental Philosophy (Knobe & Nichols eds., 2008), Notre Dame Phil. Rev. (2008). For an alternative explanation of disagreement about Gettier examples, see Williamson, supra note 45, at 183–190, who suggests that it may have to do with having a better or worse classificatory ability to apply the relevant concepts. He draws an analogy to law students, who learn an ability to apply legal concepts more successfully than those without legal training.

80. See Jackson, supra note 79.

81. They are better because they are safer. See supra note 61.

82. See Jackson v. Virginia, 443 U.S. 307 (1979).

83. Id.

84. Cf. Duff et al., supra note 20, at 91 (concluding that a convicted defendant in such a situation is entitled to have the conviction reversed).

85. A similar conclusion applies to the other legal examples. In Disjunctive Verdict, if we remove the evidence from the false side of the disjunct, there is now insufficient evidence to support a conviction. Likewise, in Fake Cabs, if we admit evidence about the fake cabs, then the evidence is no longer sufficient. Even if the system that produced these verdicts is generally reliable, the examples present case-specific epistemic defects that undermine the verdicts.

86. See Jonathan L. Kvanvig, The Value of Knowledge and the Pursuit of Understanding (2003); Kaplan, Mark, It's Not What You Know That Counts, 82 J. Phil.350 (1985)Google Scholar.