SUTTON, Circuit Judge.

Michael Bigelow's federal habeas corpus petition — challenging his state-court convictions for kidnapping, assault and arson — is before us for a second time. At his criminal trial, Bigelow maintained that he was in another city on the day of the crime. In his first appeal to us, we reversed the district court's denial of the writ, instructing it to hold an evidentiary hearing to determine whether Bigelow's trial counsel failed adequately to investigate this alibi defense, particularly after a corroborating witness stepped forward a few days before the criminal trial. See Bigelow v. Williams, 367 F.3d 562, 565 (6th Cir. 2004). On remand, the district court determined that Bigelow was entitled to relief because his counsel had provided ineffective assistance in violation of the Sixth and Fourteenth Amendments, and it granted the writ. We affirm.

I.

Because we set forth the facts of the crime in some detail in resolving the first appeal, see Bigelow, 367 F.3d at 566-67, we recount them only briefly here. On June 17, 1993, a man accosted Charlotte Schrier while she was sitting in her parked car. Id. at 566. He forced her to drive him to another location, then cut her arm with a razor blade. Id. She kicked her assailant in the groin and ran away, at which point the man set her car on fire. Id.

The police questioned Bigelow based on his resemblance to a composite sketch of the attacker and placed him under arrest after Schrier identified him as the culprit. Id. at 567. After he rejected a plea offer, the state tried Bigelow for kidnapping, felonious assault and arson. Id. At his trial, he presented an alibi witness, Vernon Greenlee, who testified that Bigelow had worked with him in Columbus, Ohio (150 miles away from Toledo) on the day of the crime. Id. The jury apparently rejected this defense, convicting Bigelow of all three crimes, after which the court sentenced him to a 20 to 42 year prison term. Id.

Bigelow unsuccessfully appealed his conviction, then brought a state post-conviction petition, arguing that his trial counsel had provided ineffective assistance of counsel. Id. The state trial court initially dismissed his claim, but the appeals court reversed that decision and remanded the case for an evidentiary hearing. Id. After the hearing, the state trial court again denied Bigelow's petition, the appellate court affirmed, and the state Supreme Court denied review. Id. at 567-68.

Bigelow filed a federal habeas corpus petition, which the district court initially denied. Id. at 569. On appeal, we affirmed the district court's decision regarding Bigelow's argument that — notwithstanding the state-court factual finding to the contrary — his lawyer knew of additional alibi witnesses but failed to call them for trial. Id. at 571. Yet, at the same time, we concluded that the district court failed to address Bigelow's contention that his trial counsel, Peter Rost, had failed to conduct a minimally adequate search for additional alibi witnesses before the trial. Id. at 576. On this basis, we returned the case to the district court so that it could consider this claim in the first instance. Id.

The district court referred the petition to a magistrate judge, who held a hearing on the effectiveness of Rost's representation. The magistrate recommended that the petition be denied, but the district court disagreed and granted the writ, concluding that Bigelow had demonstrated both that his counsel conducted an inadequate investigation and that the failure to search for additional alibi witnesses prejudiced him. The warden appeals.

II.

To prevail on an ineffective-assistance claim under the Sixth Amendment, a claimant must show that his counsel's performance was constitutionally deficient and that it prejudiced him, "render[ing] the trial unfair and the result unreliable." Hall v. Vasbinder, 563 F.3d 222, 237 (6th Cir. 2009). The Antiterrorism and Effective Death Penalty Act of 1996, Pub.L. No. 104-132, 110 Stat. 1214 ("AEDPA"), constrains our review of these issues. Under AEDPA, Bigelow may obtain relief only if he can show that the state court's decision was "contrary to, or involved an unreasonable application of, clearly established federal law," 28 U.S.C. § 2254(d)(1), or that the state court relied on an "unreasonable determination of the facts in light of the evidence presented in the State court proceeding," id. § 2254(d)(2).

A.

Was Rost's assistance constitutionally deficient? Yes. The initial question is whether his representation of Bigelow was "reasonable[] under prevailing professional norms." Strickland v. Washington, 466 U.S. 668, 688, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). To avoid the warping effects of hindsight, we "indulge a strong presumption that counsel's conduct falls within the wide range of reasonable professional assistance." Id. at 689, 104 S.Ct. 2052. And in the context of asking whether counsel adequately investigated his client's defense, the ultimate inquiry is whether the choice not to conduct additional investigation was "reasonable[] in all the circumstances, applying a heavy measure of deference to counsel's judgments." Id. at 691, 104 S.Ct. 2052.

Bigelow has met these requirements in showing that Rost did not reasonably investigate his alibi defense: that he was not in Toledo, Ohio on June 17, 1993, the day of the crime, but in Columbus, Ohio, 150 miles away. Before trial, Bigelow gave Rost the names of some people who might be able to confirm this alibi, and Rost contacted them. These contacts were unhelpful, however, because none of the individuals could remember whether Bigelow had been in Columbus on June 17, 1993 (although several confirmed that he had been there at some point during June).

Unsatisfied with this response, Bigelow tried to contact several other potential alibi witnesses in Columbus on his own, sending many of them letters about his situation. Bigelow, 367 F.3d at 568. Four days before the trial began, one letter bore fruit: Vernon Greenlee, an employee at Orkin Pest Control, contacted Rost, telling him that he had worked at the Columbus home of Gary Chasin on the day of the crime and that he had met Bigelow there as well. Id. Despite Bigelow's success in identifying Greenlee and in confirming that there was something to his alibi defense, Rost did not search for any other individuals who were present that day at the Chasin home to obtain corroborating testimony. Id. at 568, 572. As a result, the only alibi witness at Bigelow's trial was Greenlee, id., whom (at least without corroborating evidence) the jury found unconvincing.

In our prior run at this case, we observed that this sequence of events suggested that Rost had "abandon[ed] his investigation at an unreasonable juncture, making a fully informed decision with respect to trial strategy very difficult, if not impossible." Id. at 573 (internal quotation marks and alterations omitted). The question now is whether the facts developed at the subsequent evidentiary hearing support or contradict that suggestion.

Surveying the new evidence, we see no reasonable explanation for Rost's unwillingness to do more after learning that Greenlee could testify in support of Bigelow's alibi defense. At the evidentiary hearing, Rost testified that he knew of one additional witness — Chasin — who could not verify (or refute) that Bigelow was working at his home on the day of the crime, and who might have prejudiced Bigelow by telling the jury that Bigelow had been involved in a crime in Columbus. But the fact that one witness could not corroborate Greenlee's testimony does not mean that others could not have corroborated it. Rost added that Greenlee told him that "there were a total of three people there" — Greenlee, Chasin, Bigelow — which is why he did not think that additional witnesses would be available. JA 276-77. But, as we explained before, had Rost taken even "minimal additional investigative steps" after Greenlee contacted him — such as "confronting [Chasin] with the new information [and] asking [him] for records of the companies that helped with wedding preparations on the 17th" or "talking to Chasin's neighbors" — he would have learned that there were many others at the house that day, most of them working for Moonlighting Landscape. Bigelow, 367 F.3d at 573-74. Once evidence emerged supporting Bigelow's alibi defense, Rost's failure to take even these minimal steps to corroborate it was objectively unreasonable. See Ramonez v. Berghuis, 490 F.3d 482, 488-89 (6th Cir. 2007); Sims v. Livesay, 970 F.2d 1575, 1580-81 (6th Cir. 1992).

Resisting this conclusion, the State argues that, after Greenlee told Rost that there were only three people at Chasin's home on June 17 and after Bigelow failed to give him any contrary information, Rost had no reason to inquire further. But if "the duty of the lawyer to conduct a prompt investigation" exists regardless of "the accused's admissions or statements to the lawyer of facts constituting guilt," Rompilla v. Beard, 545 U.S. 374, 387, 125 S.Ct. 2456, 162 L.Ed.2d 360 (2005) (quoting 1 ABA Standards for Criminal Justice 4-4.1 (2d ed. 1982 Supp.)), surely the silence of a defendant about still more mitigating evidence — remember that Bigelow, not Rost, had identified the only useful witness (Greenlee) up to that point — does not vitiate that duty. An attorney's duty of investigation requires more than simply checking out the witnesses that the client himself identifies. And that is especially true here since Rost knew that Bigelow suffered from an "untreated mental illness," Bigelow, 367 F.3d at 568, and that his "recollection [was] not fully with him" regarding the June 1993 period because he was not "taking his medication at the time." JA 108. Rost had no reasonable basis for assuming that Bigelow's lack of information about still more witnesses meant that there were none to be found.

Nor did Greenlee's statement that there were only three people present at the Chasin house on June 17 excuse Rost's abbreviated investigation. Although counsel may "draw a line when they have good reason to think further investigation would be a waste," Rompilla, 545 U.S. at 383, 125 S.Ct. 2456, there was no such "good reason" here. Even if it was appropriate to accept Greenlee's statement at face value and thus assume that only three people were present at the Chasin home that day, Rost still had every reason to flesh out the details of Bigelow's time there. "Without making efforts to learn the details [of Bigelow's visit to Chasin's home], a convincing [alibi] argument . . . was certainly beyond any hope." Id. at 386, 125 S.Ct. 2456. A minimal investigation into the reason that Bigelow had been working there (surely a key component in developing a coherent alibi narrative) also would have brought to light the fact that Moonlighting Landscape was at the house that day. See JA 241-42 (Chasin testified that, if he had looked at his records at the time of the trial, he would have realized that Moonlighting Landscape was working at his home on June 17.); Wiggins v. Smith, 539 U.S. 510, 524, 534, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (assessing the reasonableness of counsel's refusal to conduct additional investigation in light of the extent and results of the earlier investigation).

In a case in which everything turned on the alibi defense — and in which the prosecution's ability to place Bigelow at the scene of the crime rested entirely on conflicting eyewitness testimony — Rost had ample reasons to do more than he did with the one alibi witness that his client found. "With every effort to view the facts as a defense lawyer would have done at the time, it is difficult to see how [Rost] could have failed to realize that" without seeking information that could either corroborate the alibi or contextualize it for the jury, he was "seriously compromis[ing] [his] opportunity to" present an alibi defense. Rompilla, 545 U.S. at 385, 125 S.Ct. 2456; see West v. Bell, 550 F.3d 542, 552 n. 3 (6th Cir. 2008) (noting that we may rely on Rompilla, despite AEDPA's strictures, because it "merely explain[s] Strickland" without establishing new law). In ruling to the contrary, the Ohio courts unreasonably applied Strickland.

B.

Was Bigelow prejudiced by the inadequate representation? Yes. To demonstrate prejudice, Bigelow must show that there is "a reasonable probability that, but for counsel's unprofessional errors, the result of the proceeding would have been different." Strickland, 466 U.S. at 694, 104 S.Ct. 2052. Bigelow made that showing here.

When we considered this issue the first time, we tentatively observed that "it seems . . . that Rost would have uncovered the Moonlighting witnesses had he investigated further after learning of Greenlee," and that the three witnesses "would have bolstered Bigelow's defense and [would have been] anything but cumulative." Bigelow, 367 F.3d at 574-75. One of the reasons we sent the case back to the district court for an evidentiary hearing was to test this hypothesis.

In the aftermath of that hearing, we now have additional evidence, and it supports the view that Rost's inadequate counsel prejudiced his client. In considering the likelihood that new evidence would have affected the jury verdict, we place considerable weight on the strength of the evidence that led to the conviction. See Strickland, 466 U.S. at 696, 104 S.Ct. 2052; Clinkscale v. Carter, 375 F.3d 430, 445 (6th Cir. 2004). That evidence, as we noted before, was far from overwhelming: two witnesses, each with obscured views, placed Bigelow at the scene of the crime, and there was no forensic or other evidence supporting their testimony. Bigelow, 367 F.3d at 575. Schrier, for example, identified Bigelow in court, but her recollection rested on a shallow foundation: She had only two fleeting opportunities to view her assailant, and she gave varying descriptions of him at different times. Id. The perspective of the other eyewitness, Thomas Mermer, a bystander, "was even weaker," as he saw the assailant "only from the back and side" and linked Bigelow to the crime only after the highly suggestive event of seeing his face on television in connection with a story about the crime. Id. at 575-76.

When placed on the other side of the scale along with Greenlee's alibi testimony, the Moonlighting witnesses easily could have tipped the balance. Christine Ceresna-Patridge would have testified to the following facts: Several Moonlighting Landscape employees engaged in landscape-lighting work at the Chasin home from June 16 through June 18, 1993. She visited the work site, and remembers seeing Bigelow working there on all three days. She says that Bigelow was pruning shrubs in the backyard in preparation for the upcoming wedding of Chasin's daughter. She also recalls a conversation that Bigelow had with Chasin regarding lawn furniture that needed to be painted. When added to Greenlee's, this testimony would have offered a potent counterweight to the defense: two reasonably confident alibi witnesses, with no connection to Bigelow and no axe to grind, who had time to observe him under ordinary conditions, versus two prosecution witnesses who caught no more than a glimpse of the perpetrator's face.

The State questions Ceresna-Patridge's value as an alibi witness, pointing out that the magistrate judge who held the hearing did not find her persuasive because, among other reasons, her testimony was not consistent with her earlier testimony at the state court evidentiary hearing. But the district court, engaging in a fresh review of all aspects of the magistrate's recommendation, see 28 U.S.C. § 636(b)(1); Rules Governing § 2254 Cases, at R. 11; Fed.R.Civ.P. 72(b)(3), rejected this conclusion — and had ample bases for doing so.

It explained that what appeared to be an inconsistency vanished upon closer examination: Ceresna-Patridge testified at the 1999 state evidentiary hearing — and then again at a 2004 deposition — that she had seen Bigelow working at Chasin's house "on the same day that [Moonlighting employee Victor] Timler worked there," JA 182-83, and some business records showed that Timler was there only on the 16th, not the 17th. (It is unclear, on this record, whether Timler was at the Chasin home on the 16th or the 17th; he himself maintains, based upon his own records, that he was at the house on the 17th.) But Ceresna-Patridge never testified that Bigelow was there on only one day, and in point of fact she has now explained the apparent discrepancy: Bigelow was there on the 16th, as well as on the 17th (when Timler may or may not have been absent).

The State nonetheless urges us to reject this conclusion because Ceresna-Patridge never previously testified that Bigelow had been at the Chasin home on multiple days. But the State has not shown that anyone ever asked Ceresna-Patridge that question, making the hoped-for inference an illusory one. Although we must "give due regard to the [magistrate's] opportunity to judge [Ceresna-Patridge's] credibility," Fed.R.Civ.P. 52(a)(6), the magistrate never said that she rejected Ceresna-Patridge's testimony based on her demeanor as a witness. Her decision instead was based on the apparent — but since explained — inconsistency between Ceresna-Patridge's state-court testimony and her federal-court testimony.

The case for prejudice, at any rate, does not depend solely on Ceresna-Patridge's credibility as a witness and the value of having her corroborate Greenlee's testimony. Additional Moonlighting witnesses reinforced parts of her testimony: Based upon his own business records, Timler testified that he was at the work site on June 17th and that he met Bigelow there while Bigelow was "pruning the boxwood." JA 714. He stated that he was "80 percent" sure, and later "100 percent sure," that he saw Bigelow there. JA 711-12. This testimony, we recognize, is partially undercut by a handwritten note suggesting that Timler may have been there on the 16th, not on the day of the crime, but because his own records were to the contrary, Timler's account on balance reinforces Ceresna-Patridge's (and Greenlee's) testimony that Bigelow was in Columbus, not Toledo, on the day of the assault.

There is more. Another Moonlighting employee, Jay Loyzelle, recalled meeting Bigelow at the Chasin home after Bigelow had "cut his hand and asked if I had a Band-Aid." JA 731. He remembered being at the Chasin home on the same day as Timler, which reinforces Timler's testimony. No doubt, the handwritten note that partially undercut Timler's testimony does the same to Loyzelle's. But the fact remains that we have three witnesses — all "completely disinterested," Bigelow, 367 F.3d at 575 — who claim to have interacted with Bigelow in Columbus on the day of the crime and who directly supported Greenlee's alibi testimony.

In the past, we have found it reasonably likely that a jury would acquit if they had heard the testimony of three interested witnesses whose recollections were inconsistent with the prosecution's theory of the case, even when substantial "inconsistenc[ies]" clouded "their testimony." Ramonez, 490 F.3d at 485. Bigelow's case is stronger: The three witnesses who corroborated his alibi were strangers to Bigelow, to the victim and to the crime and thus had no reason to perjure themselves one way or another. And all of this must be weighed against the flaws in the prosecution's case. Had the State linked Bigelow to the scene of the crime through forensic evidence, it would be hard to say with confidence that adding more alibi witnesses would have affected the jury's verdict. But when the only evidence linking Bigelow to the scene of the crime was two far-from-ideal witnesses, it is fair to conclude that these three alibi witnesses — when added to Greenlee's previously un-corroborated alibi testimony — readily could have altered the balance. See Ramonez, 490 F.3d at 489-90 (where a trial "boil[s] down to a credibility contest," there is a "reasonable probability that the jury would have altered the verdict" based upon the testimony of additional corroborating witnesses for the defense); Griffin v. Warden, 970 F.2d 1355, 1359 (4th Cir. 1992) (noting that eyewitness evidence is "precisely the sort of evidence that an alibi defense refutes best").

Nor — as we found previously — does the state trial court's contrary conclusion bar Bigelow from obtaining relief. See Bigelow, 367 F.3d at 576; see also Ramonez, 490 F.3d at 490-91 (holding that whether a witness was credible enough to affect the outcome of a trial is a mixed question of law and fact, which we review under the "unreasonable application" standard, rather than a finding of fact that we presume to be correct under 28 U.S.C. § 2254(e)(1)). As we said before, the inconsistencies identified by the state court as its basis for concluding that the witnesses would have been "unconvincing" were few and insignificant. Neither "typographical mistakes as to the year that Bigelow was at the Chasin home," nor an inconsistency concerning whether Ceresna-Patridge made contact with Rost at some point nor "Loyzelle's testimony that he left the property between 5:00 p.m. and 5:30 p.m., not at 4:15 p.m.," undermine "the pivotal facts established" by these witnesses. Bigelow, 367 F.3d at 576.

The State adds that the state court's conclusion may have turned on the fact that, although the Moonlighting witnesses claimed to have reviewed company records before the proceeding, "[n]o documents of any sort were presented at the [state court] evidentiary hearing." JA 83. Neither the state court nor the State, however, has cited any authority suggesting that a petitioner's failure to produce corroborating documents leads inexorably to the conclusion that the reasonably consistent alibi testimony of three independent witnesses could not affect a jury's verdict. Bigelow, in brief, has shown that his counsel's constitutionally inadequate alibi investigation prejudiced his defense, and that the state court's conclusion to the contrary unreasonably applied Strickland.

III.

For these reasons, we affirm.