OPINION

Patrico Ramonez (“Ramonez”) appeals the district court’s denial of his petition for a writ of habeas corpus naming his custodian, Mary Berghuis (“Berghuis”), as respondent. Ramonez is in custody pursuant to a conviction in Michigan state court. As he did on direct appeal, Ramonez argues that his trial counsel W. Frederick Moore (“Moore”) failed to investigate and call at trial three witnesses to the alleged crime, Charles Tames (“Charles”), Rene Tames (“Rene”) and Joel “Big Bun” Hack-ett (“Hackett”). Ramonez asserts that Moore’s performance was thus constitutionally deficient, prejudicing his defense in violation of the Sixth Amendment (as applied to the states through the Fourteenth).

Both the state trial court and the Michigan Court of Appeals found that Moore’s representation of Ramonez met the constitutional standard for effective assistance of counsel set out by Strickland v. Washington, 466 U.S. 668, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). Because it is clear that the Michigan Court of Appeals’ application of the Strickland standard for a defense counsel’s duty to investigate was unreasonable, we reverse the judgment of the district court and remand with instructions to grant a conditional writ of habeas corpus.

Background

On March 13, 2001 a Michigan state jury convicted Ramonez of third-degree home invasion (Mich. Comp. Laws § 750.110a(4)), ^[1] assault with intent to do[*484] great bodily harm (id. § 750.84) and aggravated stalking (id. § 750.411h). Those convictions arose from a complaint by Christina Fox (“Fox”),' Ramonez’s ex-girlfriend and mother of two of his children.

Fox testified at trial ^[2] that at 4:30 or 5 a.m. April 21, 2000 she and the two children were asleep in the living room of her house when she heard a knock on the door. After cracking the front door, Fox saw that it was Ramonez knocking. As her relationship with Ramonez had been violent and they had separated some years earlier, Fox became frightened upon seeing Ramonez and attempted to slam the door shut. Ramonez, however, forced the door open, knocking Fox to the ground in the process.

According to Fox, she then found herself lying on the floor between the foyer and the living room, where Ramonez pinned her down, began to strangle her and threatened her life. By kicking Ramonez, Fox was able to free herself and take off running for the front door before he punched her again, knocking her over. Fox was nonetheless able to make it out to her front porch. Once outside she encountered Charles, Rene and Hackett at the bottom of the stairs to her porch. Attempting to flee, Fox lost her footing and fell on the stairs. Ramonez again pinned her to the ground, while one of the other three covered her mouth to muffle her screaming. Finally the altercation ended after Fox saw lights go on nearby, and the four men let her go and drove away. ^[3] After the prosecution rested, Ramonez and Moore brought before the judge their disagreement as to Moore’s decision not to call any witnesses on Ramonez’s behalf and Ramonez’s intention to testify despite Moore’s advice. Ramonez wanted to call Charles, Rene and Hackett to testify to his story of what happened that day. Ramo-nez complained that he had told Moore about those witnesses months earlier, but Moore had failed to communicate with them. ^[4] Moore stated that it was his strategic decision not to call any witnesses, and the judge decided he was disinclined to interfere with counsel’s judgment.

With no other witnesses in his defense, Ramonez felt compelled to testify on his own behalf. Ramonez testified that on the day in question he did go to Fox’s house to check on his children. He claimed that Fox voluntarily invited him into her house, but once inside Ramonez did not see his kids and, believing that Fox appeared to be high, he became angry. Ramonez then pushed her against a wall — but he denied choking her — and Fox took off running out the front door. She then fell on the front steps, where Rene attempted to help her to her feet. Fox then ran off.

After the jury’s guilty verdict, the trial court sentenced Ramonez to concurrent terms of 2 to 10 years for the home invasion conviction, 12 to 20 years for the assault charge and 2 to 10 years for the aggravated stalking conviction. Ramonez appealed on ineffective assistance of counsel grounds, asserting Moore’s failure to investigate and to call Rene, Charles and[*485] Hackett. To facilitate the evaluation of that claim, the Michigan Court of Appeals remanded the case to the district court for an evidentiary hearing.

At the hearing Moore testified that he was aware of Rene, Charles and Hackett prior to trial, but he never made contact with them. Moore defended his decision not to call them based on what he characterized as his trial strategy to focus on the action inside Fox’s home. He believed that the three witnesses could not testify to that action because they were never inside the house. Moore’s plan had been to rely on cross-examination to point out discrepancies in Fox’s story and discredit her testimony. Even so, Moore did attempt to reach Charles (but only Charles) just three or four days before trial, but he did not succeed in speaking with him — the two simply exchanged phone messages.

Each of Charles, Rene and Hackett testified at that hearing. Charles is Ramo-nez’s son, Rene his stepson and Hackett an acquaintance of the three other men from work. Each testified that the three were driving in a car with Ramonez on the night in question, ultimately arriving at Fox’s house. There was some inconsistency in their testimony: Rene remembered stopping at a bar first, while Charles did not remember if they made any other stops and Hackett’s recollection was that they did not make any other stops. Each then testified that he observed Ramonez go up to Fox’s front door, where Fox then invited him inside (each affirmatively stated that Ramonez did not force his way into the house).

All three witnesses testified that even from their vantage point in the car they could see the interaction between Ramonez and Fox inside the house through the doorway — only Hackett allowed that he may have lost sight of them for some two minutes. Each said that he witnessed yelling and a physical altercation between Fox and Ramonez, but that Ramonez did not choke or punch Fox. All testified that the three men got out of the car and went up to Fox’s porch where the altercation was continuing outside the house. Their stories differed somewhat as to what then occurred outside the house, with Hackett simply saying that Fox ran off, Rene remembering Fox falling and one of them helping her up and Charles remembering having words with Fox and attempting to help her off the ground before she fled. All three affirmed that they would have been willing to testify even if the prosecutor had threatened to charge them as accessories to the crime.

At the conclusion of the hearing the trial court denied the motion for a new trial on ineffective assistance of counsel grounds. Applying state law that mirrors the two-step Strickland analysis, the court found Moore’s decision not to call the witnesses reasonable because of his trial strategy to focus on what occurred inside the house and his expectation that the witnesses could not offer competent testimony in that regard, due to what he thought was their limited vantage point outside the house. And as to prejudice, the trial court found that Hackett “was not a particularly helpful witness” and that Rene was an “incredible witness,” so that the witnesses’ testimony could not have changed the outcome of the trial.

As to Ramonez’s ineffective assistance of counsel claims, the Michigan Court of Appeals affirmed the trial court’s judgment on essentially the same reasoning (other claims made at the state level are not the subject of this federal habeas effort). Ra-monez’s application for leave to appeal to the Michigan Supreme Court was denied, making that intermediate appellate court opinion the final state court decision on his claims. On the ensuing collateral attack[*486] under 28 U.S.C. § 2254, ^[5] the district court below rejected Ramonez’s petition for a writ of habeas corpus, finding that the Michigan Court of Appeals’ decision was at least a reasonable application of Strickland. On August 14, 2006 the district court granted Ramonez’s motion for a certificate of appealability to this Court.

AEDPA ^[6] Review

We review de novo a district court’s denial of a writ of habeas corpus (Dando v. Yukins, 461 F.3d 791, 795-96 (6th Cir.2006)). And where as here the district court has reviewed only trial transcripts and other court records, any factual determinations by the district court are also reviewed de novo (id.).

Federal court examination of a habeas petition by a prisoner in custody pursuant to a state court judgment is circumscribed by AEDPA, in this instance more specifically under this part of Section 2254(d):

An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim—

(1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States....

Under that first “contrary to” clause of Section 2254(d)(1), we may grant the writ only if the state court decision was based on a conclusion of law opposite to that reached in Supreme Court precedent (Dando, 461 F.3d at 796). And as to the other alternative, Williams v. Taylor, 529 U.S. 362, 413, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000) teaches:

Under the “unreasonable application” clause, a federal habeas court may grant the writ if the state court identifies the correct governing legal principle from [the Supreme] Court’s decisions but unreasonably applies that principle to the facts of the prisoner’s case.

To issue a writ on that ground, the federal court must find the state court’s application of Supreme Court precedent “objectively unreasonable,” not merely “incorrect or erroneous” (Wiggins v. Smith, 539 U.S. 510, 520-21, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003)).

Finally, Section 2254(e)(1) requires us to presume that state court fact determinations are correct. To overcome that presumption, the statute requires the petitioner to demonstrate any state court error by clear and convincing evidence.

Michigan Court of Appeals’ Application of Strickland

Strickland, 466 U.S. at 687, 104 S.Ct. 2052 sets forth the familiar two-prong test for evaluating a claim of ineffective assistance of counsel: To prevail the petitioner must establish both (1) that defense counsel’s performance was constitutionally deficient and (2) that the deficient performance prejudiced the defense sufficiently to undermine the reliability of the trial (id.). That first element requires the petitioner to “show that counsel’s representation fell below an objective standard of reasonableness” (Strickland, 466 U.S. at 688, 104 5.Ct. 2052), for which purpose we must (id. at 689, 104 S.Ct. 2052)(internal quotation marks omitted):

[*487] indulge a strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance; that is, the defendant must overcome the presumption that, under the circumstances, the challenged action might be considered sound trial strategy.

And as to the prejudice element, Strickland, id. at 694, 104 S.Ct. 2052 instructs:

The defendant must show that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different. A reasonable probability is a probability sufficient to undermine confidence in the outcome.

For those purposes Combs v. Coyle, 205 F.3d 269, 278 (6th Cir.2000) has held that “[b]oth the performance and prejudice components of the ineffectiveness inquiry are mixed questions of law and fact.” They are thus not findings of “historical facts” (McGhee v. Yukins, 229 F.3d 506, 513 (6th Cir.2000)) that are subject to the Section 2254(e)(1) presumption of correctness for state court factual findings.

As Ramonez does not dispute, the Michigan Court of Appeals correctly articulated the Strickland standard, so there is no issue under AEDPA’s “contrary to” Supreme Court precedent prong. Hence the only question is whether that court applied the Strickland standard reasonably in coming to its judgment that Moore’s investigation leading to his decision not to call the three witnesses, and that decision itself, were sufficient for constitutional purposes.

On that score Ramonez asserts that Moore’s decision to limit (or more accurately, not to pursue at all until it was too late) any investigation regarding the three potential witnesses was objectively unreasonable, leading to an uninformed and therefore unreasonable decision not to call those witnesses at trial. Strickland, 466 U.S. at 690-91, 104 S.Ct. 2052 set forth a defense counsel’s duty to investigate:

As the Court of Appeals concluded, strategic choices made after thorough investigation of law and facts relevant to plausible options are virtually unchallengeable; and strategic choices made after less than complete investigation are reasonable precisely to the extent that reasonable professional judgments support the limitations on investigation. In other words, counsel has a duty to make reasonable investigations or to make a reasonable decision that makes particular investigations unnecessary. In any ineffectiveness case, a particular decision not to investigate must be directly assessed for reasonableness in all the circumstances, applying a heavy measure of deference to counsel’s judgments.

And Towns v. Smith, 395 F.3d 251, 258 (6th Cir.2005) dispels any doubt that a lawyer’s Strickland duty “includes the obligation to investigate all witnesses who may have information concerning his or her client’s guilt or innocence.”

In its effort to apply that standard, the district court stated that while it would have found Moore’s performance deficient due to his failure even to speak with the three witnesses before trial, it could not say that the Michigan Court of Appeals’ contrary conclusion was an unreasonable application of Strickland, thus making it unchallengeable under AEDPA. In so finding the district court considered it within reason for the Michigan court to conclude, based on the information available to Moore before trial and in conformity with his chosen trial strategy, that Moore made a reasonable professional call not to interview the witnesses and not to call them at trial.

[*488] Even given the required deference to the Michigan Court of Appeals, the district court’s restraint in those terms does not withstand analysis. As the district court’s opinion makes clear, the state court focused largely on the notion that Moore’s decision not to call the three witnesses was rooted in what it perceived to be a reasonable trial strategy — one of focusing on undermining Fox’s credibility as to what happened between her and Ramonez inside the house — and because the three witnesses were never inside the house, Moore thought that they could add little to that effort. But that belief was grounded on a fatally flawed foundation, for if Moore had only engaged in the minimal — and essential — step of interviewing the witnesses, he would have learned that they could testify as to what took place in the house, and that their testimony would have supported Ramonez’s version of events.

That being so, the state court ignored the central teaching of Strickland, as reaffirmed by Wiggins, 539 U.S. at 522-23, 123 S.Ct. 2527, that the investigation leading to the choice of a so-called trial strategy must itself have been reasonably conducted lest the “strategic” choice erected upon it rest on a rotten foundation. Towns, 395 F.3d at 258(internal quotation marks omitted) made that same point:

A purportedly strategic decision is not objectively reasonable when the attorney has failed to investigate his options and make a reasonable choice between them.

In evaluating Moore’s decision to limit his investigation as to the three witnesses (or more precisely, not to investigate them at all), Berghuis contends that the decision not to interview the witnesses was reasonable in light of Fox’s testimony in the preliminary hearing that the assault occurred inside the house. On that basis she urges that Moore was justified in concluding, without bothering to speak to the witnesses, that they would have nothing to add to his case.

But that argument is at odds with other crucial parts of the record. Months before trial Ramonez began insisting to Moore that the three witnesses were present at the time and could tell him what really happened at Fox’s house. At the same preliminary hearing on which Berghuis relies, Fox testified that Ramonez kicked in her door to gain entry to the house. Such asserted action was clearly not inside the house, such as to render the observers outside the house unable to verify or dispute the parties’ divergent versions. Moreover, Fox also testified at that same hearing that three men witnessed the assault continue outside on her front porch (even stating that one of them assisted Ramonez by covering her mouth to keep her from yelling). If Moore was seeking to show that Fox was embellishing her story of the altercation with Ramonez, why would he not also have found it useful to look into this part of her tale? With such information available to him, how could he rationally have concluded that neither Charles nor Rene nor Hackett could possibly have anything to add to Ramonez’s case?

Of course the answer is he didn’t — at least not entirely. Instead, believing that Charles (at least) might be able to “shed light on” some fact issues, Moore did attempt to reach him — but only a few days before trial. Despite months of lead time, Moore just put off the effort until he did not leave himself enough time to actually reach Charles. At trial Moore had to concede that Charles, Rene and Hackett could have had something to add:

I’m not going to say the nature of their testimony would not add. Because that suggests they don’t have anything at all to testify to. It’s my opinion that they[*489] could potentially add some information which would contradict what the complainant has testified to.

Having thus recognized the possibility that the three witnesses could provide testimony beneficial to Ramonez, it was objectively unreasonable for Moore not to interview them (or at least make reasonable efforts to interview them) before coming to his ultimate choice of trial conduct (see Towns, 395 F.3d at 259). In sum, the point is this: Constitutionally effective counsel must develop trial strategy in the true sense — not what bears a false label of “strategy” — based on what investigation reveals witnesses will actually testify to, not based on what counsel guesses they might say in the absence of a full investigation. Moore’s performance fell well on the wrong side of that line.

Having thus successfully demonstrated that Moore’s decision to limit his investigation into the potential testimony of the three witnesses was constitutionally deficient (and that the state court’s conclusion to the contrary was objectively unreasonable), Ramonez must show a reasonable probability that but for that deficiency the outcome of the trial would have been different (Strickland, 466 U.S. at 694, 104 S.Ct. 2052). To that end Ramonez points to multiple aspects of the three witnesses’ testimony at the ineffective assistance of counsel hearing to show that each could have benefitted his defense materially.

First, each would have testified that rather than Ramonez forcing his way into Fox’s home, Fox opened the door and voluntarily invited Ramonez into the house. Such evidence as to permission to enter the premises would have gone directly toward negating the “breaking and entering” element of the home invasion crime of which Ramonez was ultimately convicted (see Mich. Comp. Laws. Ann. § 750.110a(l)(c) and 110(a)(4); see also People v. Brownfield, 216 Mich.App. 429, 548 N.W.2d 248, 249-50 (1996)). Attacking that element would have been a vital point to focus on at trial. Indeed, Ramonez points out that the jury likely questioned the breaking and entering issue: It sent a note to the judge requesting testimony on whether or not Ramonez forced his way into the house, then a later note stating that it was deadlocked on the home invasion count. ^[7]

Second, each of the three witnesses testified that despite his position outside the house, he was able to see most (if not all) of the action between Ramonez and Fox that occurred just inside the open front door. And third, each of the witnesses was in a position to challenge Fox’s testimony as to the physical altercation between herself and Ramonez outside the house on the front porch. With the trial in principal part boiling down to a credibility contest between Fox and Ramonez, there is at least a reasonable probability that the witnesses’ corroboration of Ramonez’s story and their contradiction of Fox’s on those points could have influenced the jury.

In that respect the Michigan Court of Appeals simply observed in perfunctory fashion that “the three witnesses would not have provided testimony that would have changed the trial’s outcome.” In spite of the obviously helpful testimony of the three witnesses (if believed by a jury), Berghuis contends that Section 2254(e)(1) demands that we defer to the Michigan trial court’s assessment of the lack of credibility and helpfulness of the three wit[*490] nesses, assertedly undercutting any reasonable probability that the jury would have altered the verdict based on their testimony. But Section 2254(e)(1) does not support that proposition.

Initially, we note that the state trial court made an explicit adverse credibility finding only as to one of the three witnesses, Rene. Its statement that Hackett was “not a particularly helpful witness” reads more like an observation on the substance of Hackett’s testimony than a statement about his credibility. And the court did not even speak about Charles. Nothing more in the state court opinions can be characterized as factual findings (see Wiggins, 539 U.S. at 530-31, 123 S.Ct. 2527). In sum, it cannot be said that the state court made a factual determination that would arguably demand deference from this court as to all three witnesses.

More importantly, a state court’s blanket assessment of the credibility of a potential witness — at least when made in the context of evaluating whether there is a reasonable probability that the witness’s testimony, if heard by the jury, would have changed the outcome of the trial — is not a fact determination within the bounds of Section 2254(e)(1). After all, what the state court has really done is to state its view that there is not a reasonable probability that the jury would believe the testimony and thus change its verdict. And in that regard Barker v. Yukins, 199 F.3d 867, 874 (6th Cir.1999) has made it clear that our Constitution leaves it to the jury, not the judge, to evaluate the credibility of witnesses in deciding a criminal defendant’s guilt or innocence. In the context of deciding whether a defective self-defense jury instruction was harmless error, Barker, id,., held that the state court crossed that line when it found the defective instruction had no consequence because the jury would not have believed the self-defense testimony of the defendant anyway. Whether to believe the defendant’s testimony on that score was an issue for the jury and not the judge.

Our later decisions help demonstrate the difference between such credibility determinations that are for the jury and those appropriately made by state judges entitled to the Section 2254(e)(1) presumption. Examples of such deference to a judge’s assessment of credibility of witnesses include instances where credibility determinations are within a judge’s proper role, such as in assessing a juror’s impartiality at voir dire (Dennis v. Mitchell, 354 F.3d 511, 520 (6th Cir.2003)) or in factual determinations at a Miranda suppression hearing ( Hill v. Brigano, 199 F.3d 833, 840-41 (6th Cir.1999)). Or in the context of a Strickland evidentiary hearing, it is for the judge to evaluate the credibility of the criminal defendant and the former defense counsel in deciding what advice counsel had in fact given to the defendant during his trial, and such findings are entitled to the Section 2254(e)(1) presumption (see Sophanthavong v. Palmateer, 378 F.3d 859, 867 (9th Cir.2004)).

Those examples, involving credibility determinations within the judge’s province, are different in kind from a finding that a jury would not believe a witness’s testimony — what the state court effectively did here. While there would have been plenty of grist for the cross-examination mill as to Ramonez’s three witnesses, the question whether those witnesses were believable for purposes of evaluating Ramonez’s guilt is properly a jury question. As Matthews v. Abramajtys, 319 F.3d 780, 790 (6th Cir.2003)(emphasis added) has stated:

The actual resolution of the conflicting evidence, the credibility of witnesses, and the plausibility of competing explanations is exactly the task to be performed by a rational jury, considering a[*491] case presented by competent counsel on both sides.

In the end, weighing the prosecution’s case against the proposed witness testimony is at the heart of the ultimate question of the Strickland prejudice prong, and thus it is a mixed question of law and fact not within the Section 2254(e)(1) presumption. Even though the jury could have discredited the potential witnesses here based on factors such as bias and inconsistencies in their respective stories, there certainly remained a reasonable probability that the jury would not have. Ramo-nez’s case was therefore prejudiced where their testimony would have helped corroborate his testimony and contradict that of complaining witness Fox (see Workman v. Tate, 957 F.2d 1339, 1346 (6th Cir.1992)), but where counsel’s default in carrying out his constitutional obligations resulted in that testimony not being introduced at trial. All it would have taken is for “one juror [to] have struck a different balance” between the competing stories (Wiggins, 529 U.S. at 537, 120 S.Ct. 1620). ^[8]

Berghuis’ remaining arguments in claimed support of the reasonableness of the state court’s prejudice conclusion may be quickly set aside. Any assertion that Charles, Rene and Hackett would not have testified on Ramonez’s behalf in any event due to the prosecution’s threat to charge them as accessories if they showed up to testify is flatly belied by their affirmations under oath that they would have testified on Ramonez’s behalf — even under threat of prosecution. Finally, despite the weight of evidence supporting the prosecution’s case, it is simply unreasonable to say that leaving Ramonez hamstrung without the support of available corroborating witnesses in the swearing contest between himself and Fox did not prejudice his case. In short, we conclude that the state court’s judgment to the contrary was an unreasonable application of Strickland.

Conclusion

Because the Michigan Court of Appeals’ application of the Strickland standard was objectively unreasonable, the district court erred in failing to grant the writ. We therefore REVERSE the decision below and REMAND this case to the district court with instructions to grant a conditional writ of habeas corpus, giving the State of Michigan 120 days within which to provide Ramonez a new trial or, failing that, to release him.