dissenting.
It is our obligation in habeas corpus to defer to the state courts, not to clean up after them. The trial judge in this case left a gaping hole in the record when he “found” that Maria Balderrama lied and *594the prosecutor she accused of suppressing evidence told the truth, without bothering to hear either one of them testify. Rather than remanding for an evidentiary hearing, the Illinois appellate court compounded the problem when, “drawing] inferences of fact from the record,” it dismissed Bal-derrama’s post-trial statement as “highly incredible.” Today, my colleagues conclude not only that no harm was done, but that it is an entirely “sensible” approach for courts to disbelieve recantations as a matter of principle, whatever the circumstances. See ante at 593. Their reasoning would summarily foreclose relief not only to Mendiola, but to any defendant convicted on the testimony of an eyewitness who later recants, and I cannot join it.
1.
Maria Balderrama was a key prosecution witness whose impartiality the prosecutors trumpeted in closing arguments. She did not identify Mendiola as the man who shot Manuel Gutierrez, but she corroborated the testimony of the one and only witness who did — Francisco Carabez. The prosecutor’s own words reveal how important she was to the State’s case:
What else do you have, ladies and gentlemen? Maria Balderamma [sic]. Let’s talk about Maria Balderamma [sic] for a second. Thirteen-year-old girl that came in and told you what she saw. She saw and she heard the victim begging for his life. She tells you he’s laying there on the ground saying, I’m not nothing, I’m not nothing, don’t shoot me, please don’t shoot. She says the man took out the gun, fired the gun numerous times. And what’s her description of the shooter? Not too tall, not too short. Was he fat? No, not really fat. Was he thin? No, not really thin. Not what the Latin King members, the defense witnesses, tell you a concocted defense of some short, fat guy. Maria Balderamma [sic] was right across the street. She says she couldn’t see his face. She was trembling when she viewed the lineup.
What she tells you, ladies and gentlemen, supports everything that Francisco Carabez said, everything he said about what happened.
F141-42 (emphasis added). See also id. at F149 (“And you know Francisco Carabez is telling the truth because it’s supported by all of the evidence in this case.”); F197 (“Two credible witnesses, Angelo Torres, is enough. Maria Balderamma [sic] is enough.”).
Balderrama’s sworn post-trial statement was as material as her testimony at trial. What she alleges is that while she was on the witness stand, she realized that Mendi-ola was not the person who shot Gutierrez. She recognized Mendiola (whom she did not know by name) from the neighborhood, where she had seen him eating out with his parents on one occasion and in church on two others. C117-18. And she was “positive” that Mendiola was not the shooter, because Mendiola did not have a ponytail, did not have dark skin, and was taller than the person she had seen kill Gutierrez. C118-19. Balderrama goes on to allege that immediately after she testified (and while the trial was still underway), she approached one of the prosecutors and repeatedly told him that “[t]hat is not the guy” who shot Gutierrez. C123. He instructed her to say nothing to either Men-diola’s attorneys or his parents, however, and until the trial was over, she heeded that instruction. C123-24.
If Balderrama is telling the truth, the prosecutor suppressed exculpatory evidence in violation of Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963). By Balderrama’s account, she got enough of a look at the shooter and had enough of a visual acquaintance with Men-diola to know that he was not the killer. Her testimony to that effect would have left the State without the corroboration of Carabez’s testimony that it so emphasized in arguments to the jury. That corroboration cannot be dismissed as cumulative or *595immaterial. By all accounts, the events that culminated in the murder of Gutierrez unfolded very quickly in a chaotic environment. Because there was no physical evidence implicating Mendiola in the murder, the State’s case rested almost entirely on eyewitness testimony — and in great measure upon the testimony of a single witness, Carabez. Balderrama was the only prosecution witness who could confirm Carabez’s account of what occurred and his description of the person who shot Gutierrez. (Torres could only testify as to what Carabez told him.) She also was, as my colleagues acknowledge, one of very few witnesses whose credibility was unim-peached. Ante at 592.
One need only look to United States v. Agurs, 427 U.S. 97, 96 S.Ct. 2392, 49 L.Ed.2d 342 (1976), for confirmation that the assertions in Balderrama’s post-trial statement are material. In Agurs, the Supreme Court held that the prosecutor’s suppression of exculpatory evidence in violation of Brady will demand a new trial “if the omitted evidence creates a reasonable doubt that did not otherwise exist[.]” Id. at 112, 96 S.Ct. at 2402. By way of illustration, the Court remarked:
If, for example, one of only two eyewitnesses to a crime had told the prosecutor that the defendant was definitely not its perpetrator and if this statement was not disclosed to the defense, no court would hesitate to reverse a conviction resting on the testimony of the other eyewitnesses....
Id. at 112 n. 21, 96 S.Ct. at 2402 n. 21, quoting Comment, Brady v. Maryland and The Prosecutor’s Duty to Disclose, 40 U. Chi. L.Rev. 112, 125 (1972). This is almost exactly the situation we have here. One of the two eyewitnesses who were central to the State’s case purportedly told the prosecutor that Mendiola was not the assailant, and this exculpatory information was kept from the defense until after he was convicted. In view of the fact that the State relied upon Balderrama to bolster the testimony of the only witness who could identify Mendiola as the killer, her statement (if credited) creates “a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different.” Strickler v. Greene, 527 U.S. 263, 119 S.Ct. 1936, 1948, 144 L.Ed.2d 286 (1999) (quoting United States v. Bagley, 473 U.S. 667, 682, 105 S.Ct. 3375, 3383, 87 L.Ed.2d 481 (1985)); Kyles v. Whitley, 514 U.S. 419, 433-34, 115 S.Ct. 1555, 1565, 131 L.Ed.2d 490 (1995).1
2.
The key question, of course, is whether Balderrama’s post-trial assertions are credible. But for what has transpired in this case, there would be no need to point *596out that credibility assessments require an evidentiary hearing. We honor that rule probably more than a hundred times a year in reviewing summary judgment rulings alone. See Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 255, 106 S.Ct. 2505, 2513, 91 L.Ed.2d 202 (1986). The reasons are as familiar as the rule. Ascertaining whether a witness is telling the truth — as yet an entirely unscientific task — demands an opportunity for the factfinder to look her in the eye, observe her demeanor, note the dryness of her brow, hear the inflections in her voice, and in general to observe how she holds up on cross-examination. E.g., Anderson v. City of Bessemer City, N.C., 470 U.S. 564, 575, 105 S.Ct. 1504, 1512, 84 L.Ed.2d 518 (1985); United States v. Mancillas, 183 F.3d 682, 701 n. 22 (7th Cir.1999) (quoting United States v. Garcia, 66 F.3d 851, 856 (7th Cir.1995)), cert. denied, — U.S.-, 120 S.Ct. 1271, 146 L.Ed.2d 221 (2000). Cold paper records supply none of this information.
What Mendiola appropriately asked for, and what he was entitled to, was a brief evidentiary hearing so that the court could determine whether Balderrama or the prosecutor was telling the truth. Balder-rama’s statement, which Mendiola submitted in support of his request, laid more than an adequate foundation for the hearing. Her statement was given under oath, and it set forth in detail when and why she realized that Mendiola was not the person she had seen shoot Gutierrez, as well as the circumstances and content of the conversation in which she disclosed this information to the prosecutor. Nothing more was required to demonstrate that an evi-dentiary hearing was necessary.
But rather than devote twenty minutes or so to such a hearing, or to articulate sound reasons why a hearing was unnecessary, the trial judge simply “found” that Balderrama was lying. The sole piece of evidence that the court actually had before it at that point was Balderrama’s sworn statement. Beyond that, it had only an out-of-court verbal assurance from the prosecutor identified in Balderrama’s statement denying that there had ever been a conversation in which she told him that Mendiola was not the shooter. G6, G12. That assurance was relayed to the court by the prosecutor’s colleague. Nonetheless, the court proceeded to find the hearsay repetition of the prosecutor’s denial to be more credible than Balderra-ma’s sworn statement:
I find there was no violations [sic] of the document Brady versus Maryland because I accept [Assistant State’s Attorney] Mr. Berlin’s word through the representations of [his colleague] Mr. Rogers here in court. Mr. Berlin being an officer of the court that he never had a conversation with Maria Balderamma [sic] or she stated that Mr. Mendiola was not the shooter. I find that conversation did not exist; that Maria Balder-amma [sic] a young lady from the community for whatever motivated her, made in fact a false statement to the defense attorneys in the presence of the court reporter post-trial. That assertion due to the fact I find Mr. Berlin to be a credible outstanding state’s attorney. He’s not in any way encroached nondisclosure under Brady versus Maryland.
G18-19.
This was an extraordinary turn of events. Without having heard a single witness testify, and based solely on the unsworn assurances of a prosecutor who was not even before the court,2 the court simply took the prosecution’s word and *597labeled Balderrama a liar. It did not say that her post-trial statement was incredible as a matter of law, it did not say that her statement was immaterial. It simply chose not to believe her, without any of the process that normally attends such credibility determinations.
3.
Faced with the blatant impropriety of the trial court’s credibility determination, the Illinois Appellate Court felt the need in the first instance to become a factfinder itself. “[A]n appellate court may draw inferences of fact from the record before it,” the court proclaimed. People v. Mendiola, No. 1-95-2874, Order, at 20 (Ill.App. July 21, 1997) (hereinafter, “Order”), citing ill. Sup.Ct. Rule 366(a)(4).3 And without further ado — indeed, without any explanation at all — the court simply announced: “We find the content of Maria Balderra-ma’s post-trial statement regarding the conduct of the assistant State’s Attorneys in this case to be highly incredible.” Order at 20.
The appellate court’s finding is even less sound than the trial court’s. The trial judge, at least, had heard Balderrama testify at trial and, within the confines of the courtroom, had seen the prosecutors at work. See ante at 593. The appellate court, by contrast, had only a cold record before it. It had no business making credibility determinations. See Cabana v. Bullock, 474 U.S. 376, 388 n. 5, 106 S.Ct. 689, 698 n. 5, 88 L.Ed.2d 704 (1986).4 What the court meant to say, perhaps, when it “found” Balderrama’s post-trial statement to be “highly incredible,” was that no reasonable finder of fact could believe it — that her statement was incredible as a matter of law. See, e.g., Anderson v. Bessemer City, N.C., 470 U.S. at 575, 105 S.Ct. at 1512; Kidd v. Illinois State Police, 167 F.3d 1084, 1095-96 (7th Cir.1999). That won’t fly either. To accuse a prosecutor of misconduct, as Balderrama did, is a grave matter. None of us wants to believe that an officer of the court would instruct a witness to keep exculpatory information to herself. But is it beyond the realm of possibility? Regrettably, it is not. See Lockett v. Blackburn, 571 F.2d 309 (5th Cir.1978) (State encouraged and helped confidential informants who wit*598nessed defendant’s sale of heroin to undercover agent to leave state before trial, rendering defendant unable to subpoena them); see also, e.g., United States v. Boyd, 833 F.Supp. 1277 (N.D.Ill.1993) (Aspen, J.), aff'd, 55 F.3d 239 (7th Cir.1995); Walker v. City of New York, 974 F.2d 293, 294-95 (2d Cir.1992), cert. denied, 507 U.S. 961, 113 S.Ct. 1387, 122 L.Ed.2d 762 (1993), and cert. denied, 507 U.S. 972, 113 S.Ct. 1412, 122 L.Ed.2d 784 (1993); Ex Parte Davis, 957 S.W.2d 9, 10-11 (Tex. Crim.App.1997), cert. denied, 523 U.S. 1023, 118 S.Ct. 1307, 140 L.Ed.2d 472 (1998); Commonwealth v. Smith, 532 Pa. 177, 615 A.2d 321, 322-23 (1992). Balder-rama’s allegations are straightforward and plausible, and we are given no other reason to believe that her testimony — if she were ever allowed to give it — could not be credited. So far as the record reveals, she is not delusional, she suffers no impairments in her perception, nor is she an incompetent witness — she was, after all, a jewel in the State’s case against Mendiola. A factfinder would of course be free not to believe her, but before she can be deemed incredible, her testimony must be heard.
Grudgingly indulging the assumption that Balderrama’s statement might be true, the Illinois appellate court alternatively posited that it was not material in the Brady sense — that it was unlikely to have affected the outcome of the trial, in other words. Order at 20-21; see Strickler, 119 S.Ct. at 1948, 1952; Kyles, 514 U.S. at 434, 115 S.Ct. at 1565-66. For three reasons, each of which again revolves around her credibility, the court was confident that a jury would have discounted Balderrama’s allegations: (1) “young Maria’s post-trial statement lacks credibility because it seems to result from defense counsel’s leading questions”; (2) her statement “was contradicted by nearly every other witness”; and (3) cross-examination by Mendiola’s attorney at trial “significantly undercut Maria’s ability to inculpate or exculpate any suspect.” Order at 21-21 (emphasis in original). The district court thought that this was a reasonable materiality assessment, and sustained the state court judgment on that basis. Mendiola v. Carter, No. 98 C 3183, 1998 WL 748276, at *8 (N.D.Ill. Oct. 22) (Kocoras, J.). On closer inspection, however, the appellate court’s materiality evaluation turns out to be as flawed as the rest of its analysis.
That Mendiola’s lawyers may have asked some leading questions when they elicited Balderrama’s post-trial statement is entirely beside the point. Her statement was submitted solely as prima facie proof of the need for a hearing. See G8. In that sense, it was no different from an affidavit, which typically is drafted by an attorney in terms favorable to his client. Had the trial court allowed Mendiola to put Balderrama on the witness stand, his attorneys no doubt would have questioned her in an appropriate manner; and the State, I might add, would have enjoyed the opportunity to cross-examine her. But one party cannot conduct a hearing on his own, and so we are left with the record as it is. To discount the evidentiary value of Balderrama’s statement because of its form ignores the fact that it is the trial court that is to blame for refusing to hold a hearing, not Mendiola.
To say that Balderrama’s statement “was contradicted by nearly every ether witness” (Order at 21) is flat-out wrong. What contradicted Balderrama, the court believed, was the testimony of several eyewitnesses that the man who shot Gutierrez “wore a hood.” Order at 19. If he was wearing a hood, the appellate court apparently reasoned, Balderrama could not possibly have known that he wore his hair in a ponytail — and the ponytail is one of the reasons she gives for her late realization that Mendiola was not the shooter. C118. Yet, one can wear a hooded sweatshirt or jacket without wearing the hood up, and if the shooter’s hood was down during any portion of the encounter, it would have been possible for Balderrama to see a ponytail. In fact, of the five eyewitnesses who testified about the shooting, only *599one — Carabez—indicated that the assailant raised his hood at the beginning of the encounter. B68, B91. Three others testified that the killer left his hood down until after he shot Gutierrez. D150, E38, E65. The fifth — Balderrama—could not recall whether he had a hood or not. B124.
Finally, to suggest that Balderrama’s ability to inculpate or exculpate anyone as Gutierrez’s assailant had already been cast into doubt at trial ignores the basis for Balderrama’s assertion that Mendiola was not the shooter. Balderrama had indeed effectively conceded on cross-examination that she did not get a good look at the killer and did not see his face. B123-24. The two characteristics that in her mind ruled out Mendiola as the assailant, however — his complexion and lack of a ponytail (C118-19) — are not features that require a good look at someone’s visage to discern. Nothing in this record gives us reason to doubt that Balderrama in fact saw a ponytail and dark complexion on the assailant. (Carabez, for example, testified that the shooter’s complexion was “[bjrownish.” B73.) The State itself was content to rely on her description of the shooter to bolster Carabez’s identification of Mendiola. F142.
4.
My colleagues’ own effort to rescue the state courts’ finding fares no better and in one important respect, I believe, makes matters worse. Declaring first that there is “[p]lenty of support” in the record for the state courts’ assessment of Balderra-ma’s credibility, ante at 593, they tick off a list of circumstances which, in their view, suggest that her post-trial statement is false, ante at 593. None of these factors, however, supports the trial court’s decision to dismiss Balderrama as a liar without first holding a hearing. Second, my colleagues make the sweeping declaration that “[disbelief of recantations is sensible” because it “promotes truthful testimony during trial.” Ante at 593. Maybe, maybe not. But not all recantations are false, and a generalized pronouncement that trial courts are free to disregard any and all recantations as a matter of policy rules out any relief to those convicted on the basis of perjury or otherwise inaccurate testimony.
The fact that the trial court heard Bal-derrama testify as a witness at trial (see ante at 593) lends little or no support to the determination that her post-trial statement is incredible. That point would be relevant if Balderrama were attempting to withdraw or change her prior testimony. Had she identified Mendiola as the shooter at trial, for example, the trial court would indeed have had some basis for assessing the veracity of her post-trial assertion that Mendiola was not the shooter — having already heard and seen the witness say that X is true under oath, a court has a frame of reference for assessing the credibility of her subsequent statement that X is false. See, e.g., United States v. Provost, 969 F.2d 617, 619-20 (8th Cir.1992), cert. denied, 506 U.S. 1056, 113 S.Ct. 986, 122 L.Ed.2d 139 (1993), cited ante at 593. Put another way, when a witness wishes to take back what she has already asserted under oath, she has some explaining to do; and if a reasonable explanation is not forthcoming, the court need not let her testify a second time. See United States v. Stewart, 198 F.3d 984, 986 (7th Cir.1999), cited ante at 593; see also Higgins v. Mississippi 217 F.3d 951, 954, 955 (7th Cir.2000). Balderrama’s statement is not a true recantation in this sense, however. At trial, Balderrama never implicated Mendiola as the person who shot Gutierrez. Instead, she recounted the sequence of events that culminated in the shooting and gave a vague description of the shooter. In no respect does her post-trial statement conflict with the substance of her testimony. The statement certainly does convey information that Balderrama did not disclose at trial, but in virtually every instance, this was information that she was never asked about at trial. She was never asked, for example, whether the shooter *600wore his hair in a ponytail. She was never asked if Mendiola appeared taller, shorter, darker, or lighter than the shooter. She was never asked, in fact, whether she recognized Mendiola at all. Her posttrial statement thus stands in contrast to a true recantation, which often amounts to a confession of perjury. To the extent it requires explanation, she gives it.
My colleagues also suggest that Balder-rama’s statement “made little sense.” Ante at 593. “Why would she testify as she did,” they ask, “and then sing a different song immediately after leaving the stand?” Id. (emphasis in original). As I have just explained, however, Balderra-ma’s post-trial statement does not amount to a different song so much as an additional verse. So far as the record discloses, when Balderrama took the witness stand, no one thought that she could or would identify Mendiola as Gutierrez’s assailant. She had not picked Mendiola out of the line-up (see B116), she had not seen the shooter’s face (B115-16), and could only describe the shooter in general terms (B116, B124). Her own realization that she did recognize Mendiola (as someone other than the shooter) did not occur until she was on the witness stand. C119. Because no one else was the wiser at that point, she was not asked while on the stand whether she recognized him. The fact that she (allegedly) spoke up immediately after she finished testifying if anything lends credibility to her account of events. Would my colleagues think her allegations more credible if she had waited a month or two (or a year or two) before approaching the prosecutor?
The notion that the trial judge was entitled to credit the prosecutor’s unsworn, hearsay denial because prosecutors have “reputational interests in telling the truth” (ante at 593) makes insufficient room for the real world. Yes, prosecutors, like other attorneys, have an interest in preserving their credibility; confessing small sins serves that interest. But I hardly think it likely that a prosecutor who actually instructs a witness to suppress exculpatory information is going to be forthcoming about it, when it is only her word against his. If indeed a prosecutor has engaged in the kind of serious misconduct that Bal-derrama alleges, arguably it would not be in his “reputational interest” to acknowledge the impropriety. Confessing to conduct that amounts to the obstruction of justice will not do much to advance a prosecutor’s career. Whatever we might think in the abstract, the important point is that we know that some prosecutors do engage in this sort of misconduct (see, e.g., cases cited supra at 597-98), and that they don’t always tell the truth about it. Balderra-ma’s allegations are within the realm of possibility. Under the circumstances, the trial judge was obligated to hear both Bal-derrama and the prosecutor testify before deciding who was telling the truth.
Finally, the suggestion that there are “sensible” policy reasons to disbelieve recantations categorically (ante at 593) is bothersome. Recantations should be viewed with a healthy dose of skepticism, for all of the reasons my colleagues have cited. But as we know, witnesses don’t always wait until after they leave the courtroom to dissemble. Just as a witness may recant her testimony later to appease the defendant and his allies, she may also lie in the first instance, perhaps to appease the prosecution, to protect someone else, or to exculpate herself. In short, the recantation on occasion represents the truth.5 Deciding when that is so is by no *601means an easy task, and when the trial judge has given the witness’s change of heart due consideration, his judgment is entitled. to our deference. But when a court rejects a plausible recantation out of hand, without any of the process that attends a valid credibility assessment, we owe the court’s finding no respect. To sustain such a summary determination, as we do today, is to unnecessarily exalt the sovereignty of state courts over due process and the pursuit of truth.
5.
Four state judges, my brothers note in closing, have chosen to disbelieve Balder-rama’s post-trial statement exonerating Mendiola. Ante at 593. With all due respect to my colleagues on the Illinois courts, it would not matter if 100 of them had done so, since not one has actually heard what Balderrama has to say. And given that Balderrama and the prosecutor are the only two people who know whether her allegations are true, Mendiola cannot possibly marshal the clear and convincing evidence needed to show that the state courts’ credibility assessment is wrong (see ante at 592, 593) unless and until he is given the chance to put Balderrama on the witness stand.
The gravity of the trial court’s mistake is demonstrated by the lengths to which the Illinois appellate court, and now this court, have gone to compensate for it. Had the trial judge simply done his job and conducted an evidentiary hearing that would have enabled him to decide whether Balderrama’s post-trial statement is truthful, we would not be here today. Instead, we find ourselves struggling to prop up credibility assessments fashioned of smoke and mirrors. It is not our province to second-guess state courts, but neither is it our province to gloss over their mistakes. This case should be remanded to the district court for the evidentiary hearing that the state courts have refused to give Men-diola.
I respectfully dissent.
. See, e.g., Kyles, 514 U.S. at 445, 115 S.Ct. at 1571 ("the effective impeachment of one eyewitness can call for a new trial even though the attack does not extend directly to others ....”) (citing Agurs); Cannon v. Alabama, 558 F.2d 1211, 1215-16 (5th Cir.1977) (new trial ordered where government failed to disclose eyewitness who had identified someone other than defendant as perpetrator), cert. denied, 434 U.S. 1087, 98 S.Ct. 1281, 55 L.Ed.2d 792 (1978); Jackson v. Wainwright, 390 F.2d 288, 298-99 (5th Cir.1968) (new trial ordered where prosecution failed to disclose pre-trial statements of eyewitness who said assailant had lighter complexion than defendant); United States ex rel. Meers v. Wilkins, 326 F.2d 135, 138-40 (2d Cir.1964) (new trial ordered where prosecution failed to disclose two eyewitnesses who told police that defendant did not participate in robbery); Watkins v. Miller, 92 F.Supp.2d 824, 843-47 (S.D.Ind.2000) (Hamilton, J.) (habeas corpus granted where prosecution failed to disclose, inter alia, existence of eyewitness who observed abduction of murder victim at time when defendant had solid alibi and who gave description of abductor that could not have been defendant); United States v. Sheehan, 442 F.Supp. 1003, 1008-09 (D.Mass.1977) (new trial ordered where government failed to disclose existence of only eyewitness who saw unmasked faces of bank robbers and who was unable to identify defendant as one of the robbers); In re Kapatos, 208 F.Supp. 883, 888-89 (S.D.N.Y.1962) (habeas corpus granted and new trial ordered where State failed to disclose pre-trial statement and grand jury testimony of witness who indicated that defendant was not one of two men he saw fleeing scene of murder).
. "Hearsay testimony is presumptively unreliable under the common law because the opposing party has no opportunity to cross-examine and test the declarant's truthfulness under oath before the factfinder.” United States v. Shukri, 207 F.3d 412, 417 (7th Cir.2000), citing 5 John H. Wigmore, Evidence in Trials at Common Law § 1368, at 37, § 1420, at 251 (rev. ed.1974), and Mccormick On Evidence § 245, at 728 (Edward W. Cleary ed., 3d ed.1984). The prosecutor's out-of-court statement in this case bears none of the indi-cia of reliability that would overcome this presumption. See, e.g., Fed.R.Evid. 804(b).
. The supreme court rule grants a variety of discretionary powers to a reviewing court in Illinois to exercise "on such terms as it deems just,” including the authority to "draw inferences of fact.” Rule 366(a)(4). Illinois appellate courts resort to this authority when they make a finding that the facts of record virtually compel, e.g., In re Marriage of Bennett, 225 Ill.App.3d 828, 167 Ill.Dec. 308, 587 N.E.2d 577, 579-81 (1992), and likewise when they reject a factual assertion that is wholly inconsistent with the record, e.g., In re Marriage of Johnson, 206 Ill.App.3d 992, 151 Ill.Dec. 891, 565 N.E.2d 162, 163-64 (1990). Nothing in the rule, however, authorizes a reviewing court to resolve a genuine swearing contest. See, e.g., Zaderaka v. Illinois Human Rights Com’n, 131 Ill.2d 172, 137 Ill.Dec. 31, 545 N.E.2d 684, 688 (1989).
. Of course I agree that appeUate courts have the ability and authority under appropriate circumstances to make factual determinations, Sumner v. Mata, 449 U.S. 539, 546-47, 101 S.Ct. 764, 769, 66 L.Ed.2d 722 (1981) ("Sumner I”), and that such determinations are presumed to be correct, 28 U.S.C. § 2254(e)(1). See ante at 592. Thus, for example, when a state appellate court examines the record and determines that the trial judge did not rely on an impermissible factor in sentencing the defendant, Wainwright v. Goode, 464 U.S. 78, 85, 104 S.Ct. 378, 382-83, 78 L.Ed.2d 187 (1983), that a witness gave an accurate, detailed description of the defendant, Sumner v. Mata, 455 U.S. 591, 597, 102 S.Ct. 1303, 1307, 71 L.Ed.2d 480 (1982) (per curiam) ("Sumner II”), or that the record reveals or does not reveal racial discrimination in the selection of jurors, Mitchell v. Rees, 114 F.3d 571, 576-77 (6th Cir.1997), cert. denied, 522 U.S. 1120, 118 S.Ct. 1062, 140 L.Ed.2d 122 (1998), we owe that assessment deference. See Sumner II, 455 U.S. at 597-98, 102 S.Ct. at 1307; Sumner I, 449 U.S. at 546-47, 101 S.Ct. at 769. But the presumption of correctness falls away when the appellate court engages in factfinding without the sort of information or process that render factual assessments reliable. Making a credibility assessment based on a paper record alone is a classic example. Bullock, 474 U.S. at 388 n. 5, 106 S.Ct. at 698 n. 5.
. Illinoisans will no doubt recall Cathleen Crowell Webb, who stirred a nationwide controversy in 1985 when she recanted her charge that Gary Dotson had kidnapped and raped her in 1977. Her recantation led the Governor to commute Dotson’s sentence, although it did not initially persuade the Illinois courts to vacate his conviction. See People v. Dotson, 163 Ill.App.3d 419, 114 Ill.Dec. 563, 516 N.E.2d 718 (1987). Notably, however, the trial judge in Dotson at least heard Webb's recantation on the witness stand before rejecting it as incredible. See id. at 719. Years later, when DNA testing ruled out Dotson as *601the source of biological material found on Webb's undergarments, a new trial was ordered and the State dropped the charges against him. See Matt O’Connor, State dismisses Dotson rape case, Chicago Tribune, Aug. 15, 1989, at 1.