delivered the opinion of the Court.
A jury found that M.P.A. committed sexual assault of a child based on the testimony of two witnesses who have now recanted, and sentenced M.P.A. to twenty years’ confinement after hearing false testimony by a State’s expert. The district court denied habeas relief and the court of appeals affirmed. M.P.A. asks us to reverse and hold that he is actually innocent, that the false testimony contributed to his sentence, and that his trial counsel rendered ineffective assistance.
We conclude M.P.A. is not entitled to relief on his claims of actual innocence or ineffective assistance of counsel. However, we hold false testimony by the State’s expert witness contributed to his sentence and he is therefore entitled to a new disposition (sentencing) hearing.
I. Factual and Procedural History
S.A. and A.A. accused their cousins M.P.A. and J.W.A. of sexually assaulting them. At the time of the alleged acts, S.A. was seven, A.A. was five, M.P.A. was fourteen, and J.W.A. was fifteen.1 M.P.A. and J.W.A. were charged with three counts of aggravated sexual assault of a child. J.W.A. entered a plea bargain and pleaded true to the allegations regarding S.A. M.P.A. pleaded not true and went to trial.
At trial, A.A. did not testify that M.P.A. had sexually assaulted him, but both S.A. and A.A. testified that M.P.A. sexually assaulted S.A. In addition, Alice Linder, a sexual assault nurse examiner who had examined S.A. and A.A. testified that they told her that M.P.A. and J.W.A. had sexually assaulted them. M.P.A. was the only defense witness and he testified that he did not sexually assault S.A. The trial court granted a defense motion for a directed verdict regarding the count that M.P.A. had sexually assaulted A.A. The jury found that M.P.A. had sexually assaulted S.A.
At the disposition phase, the State presented two witnesses: Dr. Frederick Wil-loughby, a licensed psychologist and registered sex offender treatment provider, and Kathie Lewis, a probation officer. Wil-loughby testified regarding an “Abel Assessment” that he had administered to M.P.A. Willoughby testified that the Abel Assessment is a test that predicts which people have an interest in particular sexes and age groups. One portion of the test consists of a questionnaire. M.P.A.’s answers to this portion of the test were “socially desirable.” The portion of the Abel Assessment at issue in this case consists of a series of slides that are shown to the subject. The slides depict individuals of various age and gender, and the subject’s sexual interest is measured by how long the subject looks at each slide. The results are computerized and sent to Atlanta, where the test is “scored.”
*282After the trial court overruled M.P.A.’s reliability objection to the Abel Assessment, Willoughby testified that M.P.A. was a “pedophile” who had a “significant sexual interest in eight to ten year-old females and two to four and eight to ten year-old males.” Lewis testified that probation and home supervision would be inappropriate for M.P.A. The only witness for M.P.A. was his mother, who testified that she would supervise M.P.A. if the jury assessed a sentence of probation. The jury sentenced M.P.A. to twenty years’ confinement.
A.A. recanted approximately nine months after the trial and S.A. recanted approximately twenty months after the trial. At the habeas court below, both S.A. and A.A. testified that they falsely accused their cousins because their mother, LaVon-na, told them to. J.W.A. also recanted his confession and testified at the habeas court that he did not sexually assault A A. and S.A. In addition, the evidence at the habe-as hearing showed that approximately four years after M.P.A.’s original trial, Wil-loughby entered into an agreed order with the Texas State Board of Examiners of Psychologists stating that he “misstated in his court testimony the research that had been conducted with respect to the Abel Assessment.”
M.P.A. filed the writ of habeas at issue in this case, arguing that he was actually innocent, that Willoughby’s false testimony contributed to his sentence, and that his trial counsel rendered ineffective assistance. The habeas court found that the recantations were not credible. In so finding, it relied on J.W.A.’s confession and the testimony from all the witnesses. It also found that Willoughby’s “misstatements, if any,” did not contribute to M.P.A.’s sentence, and that M.P.A.’s trial counsel was effective. The court of appeals affirmed and M.P.A. appealed to this Court.2
II. Actual Innocence
M.P.A. argues that he is entitled to relief based on the newly discovered evidence of S.A. and A.A.’s recantations. At the habeas hearing, they testified that La-Vonna told them to falsely accuse M.P.A. and J.W.A. S.A. testified that LaVonna told her this was necessary to keep La-Vonna out of jail. M.P.A. alleges that LaVonna’s motive was that these accusations would reflect badly on S.A. and A.A.’s father, Stephan, in their then-ongoing custody proceeding, and that the recantations are corroborated by the record.3
*283To prevail on his actual innocence claim, M.P.A. must show by clear and convincing evidence that no reasonable juror would have convicted him in light of the recantations. Ex Parte Elizondo, 947 S.W.2d 202, 209 (Tex.Crim.App.1996).4 Almost total deference is accorded to the trial court’s factual findings in habeas proceedings. E.g., Ex Parte Amezquita, 223 S.W.3d 363, 367 (Tex.Crim.App.2006). Here, the habeas court found that the recantation testimony was “not credible.” Because this finding has some support, we cannot grant M.P.A. relief on his actual innocence claim.5
In Keeter v. State, the Court of Criminal Appeals addressed a recantation of a juvenile’s sexual assault allegations in the context of a motion for a new trial. 74 S.W.3d 31, 33 (Tex.Crim.App.2002). The Court surveyed its cases and summarized the bases for disbelieving a recanting witness:
Such bases include, but are not limited to: evidence that the recanting witness was subject to pressure by family members or to threats from co-conspirators, evidence showing part of the recantation to be false, circumstances showing that the complainant recanted after moving in with family members of the defendant, and where an accomplice recants after being convicted.
Id. at 38 (citations omitted). The Keeter Court affirmed the trial court’s rejection of the recantation because, inter alia, the complainant recanted after moving into the residence of her mother, with whom the defendant had resided, and there was evidence that the defendant’s father had pressured the complainant to recant. Id. at 39.
Here, although substantial evidence corroborates the recantations, there is some evidence of pressure to recant by Stephan’s family and an investigator hired by Stephan. While we recognize that recantation of sexual assault in the context of custody litigation should be given serious consideration, there is evidence in this record to support the trial court’s rejection of the recantation testimony. S.A. first recanted when Stephan’s girlfriend told S.A. she did not like her because S.A. was a liar who put people in jail. A.A. and S.A. both testified that the investigator hired by Stephan did not pressure them to recant, but A.A. testified that the investigator gave him “encouragement” to recant and “kept asking” him if the sexual assault allegations were true.
In addition, J.W.A. confessed to the police that he had sexually assaulted S.A. and he subsequently pleaded true to sexually assaulting S.A. We note that J.W.A. recanted his confession and has testified that he did not understand the significance of his confession or plea. We further note that the Court of Criminal Appeals has explained that innocent defendants will sometimes plead guilty. Ex Parte Tuley, 109 S.W.3d 388, 393 n. 2 (Tex.Crim.App.*2842002) (citing United States v. Timbana, 222 F.3d 688, 718 (9th Cir.2000) (Kleinfeld, J., dissenting)). However, even if we were to accept J.W.A.’s explanations of his confession and plea as conclusive, his statements to M.P.A.’s trial attorney, Bobby Barina, would nevertheless support the rejection of the recantations.
Barina testified J.W.A. told him he and M.P.A. sexually assaulted S.A. Barina explained M.P.A. and J.W.A.’s family informed him J.W.A. would testify M.PA. did not commit the alleged assault. Therefore, Barina planned to call J.W.A. to the stand at the original trial to testify M.P.A. was innocent. However, when Barina met with J.W.A. to prepare for trial, J.W.A. implicated M.P.A. in the offense. J.W.A. said he himself “had done it and that he saw [M.P.A.] ‘do it.’ ” If J.W.A. testified M.P.A. did not commit the offense, it could amount to perjury. If J.W.A. testified M.P.A. committed the offense, it could damage M.P.A.’s defense. Barina decided not to call J.W.A. as a witness. This statement by J.W.A. is not negated by J.W.A.’s explanation of his confession to the police or his decision to plead true-neither his confession nor plea implicated M.P.A. in a sexual assault on 5.A. J.W.A. denies telling Barina he and M.P.A. sexually assaulted S.A. But the ha-beas court was entitled to believe Barina. Hence, JWA’s statement to Barina supports the habeas court’s decision.
M.P.A. cites several cases with factual similarities to the instant case. However, we defer to habeas courts’ credibility determinations and in those cases the habeas courts had credited the recantations. See Ex parte Calderon, 309 S.W.3d 64, 65 (Tex.Crim.App.2010); Ex Parte Thompson, 153 S.W.3d 416, 417 (Tex.Crim.App.2005); Elizondo, 947 S.W.2d at 210. In contrast, here the habeas court, with some record support, rejected the credibility of the recantations. Like the Calderon, Thompson, and Elizondo courts, we defer to the habe-as court’s evaluation of the recantations’ credibility because there is some evidence supporting that determination.6 Therefore, we conclude that M.P.A. has not established his right to relief on grounds of actual innocence.
III. False Testimony
A. Background
As a threshold matter, we must determine whether M.P.A. may bring his false testimony claim via habeas. Defendants are ordinarily barred from raising claims on habeas that could have been raised at trial or on direct appeal. Ex parte Napper, 322 S.W.3d 202, 228 (Tex.Crim.App.2010) (citing Ex paHe Pena, 71 S.W.3d 336, 337-38 (Tex.Crim.App.2002)). However, while the Court of Criminal Appeals has found that a defendant’s failure to complain about false testimony by the State before a habeas proceeding affects the threshold for showing harm, it did not *285find that the claim was barred. Ex parte Fierro, 934 S.W.2d 370, 374-75 & 374 n. 10 (Tex.Crim.App.1996), see also Ex parte Ghahremani, 332 S.W.3d 470, 481 (Tex.Crim.App.2011) (citing Fierro, 934 S.W.2d at 371-74, and noting that, in Fierro, evidence showing that testimony was false was available at suppression hearing but not used until habeas).
The State knowingly used false testimony in Fierro, 934 S.W.2d at 371-73, but no evidence shows that the State knowingly used the false testimony in the instant case. However, the Court of Criminal Appeals does not distinguish between knowing and unknowing usage of false testimony for purposes of harm analysis. Ex parte Chabot, 300 S.W.3d 768, 771 (Tex.Crim.App.2009).7 Therefore, M.P.A.’s false testimony claim is properly before us.
A false testimony claim is subject to harmless-error review. Chabot, 300 S.W.3d at 771; Fierro, 934 S.W.2d at 374. A habeas applicant must establish by a preponderance of the evidence that the error contributed to his conviction or punishment. Chabot, 300 S.W.3d at 771; Fierro, 934 S.W.2d at 375. This framework applies to the trial’s disposition phase as well as the adjudication phase. Ghahremani, 332 S.W.3d at 477 (citing Estrada, 313 S.W.3d at 287-88). Applying this standard here, M.P.A. must show that Willoughby’s testimony would have been excluded had it been truthful and that M.P.A. would have received a different sentence absent Willoughby’s testimony.8 We hold that M.P.A. has met this burden and is entitled to a new disposition hearing.
B. Admission of the Willoughby Testimony
Willoughby testified as an expert in this case. A party offering scientific expert testimony must show by clear and convincing evidence that the science is reliable. Kelly v. State, 824 S.W.2d 568, 573 (Tex.Crim.App.1992); see also In re D.W.P., No. 06-07-00113-CV, 2008 WL 53211, at *1 (Tex.App.-Texarkana Jan. 4, *2862008, no pet.) (“Even though appeals of juvenile court orders are generally treated as civil cases, we believe the criminal standard for the admission of scientific evidence should apply in light of the quasi-criminal nature of juvenile proceedings.” (footnote and citation omitted)).9 “ ‘Unreliable ... scientific evidence simply will not assist the [jury] to understand the evidence or accurately determine a fact in issue; such evidence obfuscates rather than leads to an intelligent evaluation of the facts.’ ” Kelly, 824 S.W.2d at 572 (alterations in original) (quoting Kenneth R. Kreiling, Scientific Evidence: Toward Providing the Lay Trier With the Comprehensible and Reliable Evidence Necessary to Meet the Goals of the Rules of Evidence, 32 Ariz. L.Rev. 915, 941-42 (1990)).
Kelly governs the reliability determination and specifies several nonexclusive factors to guide the inquiry.10 824 S.W.2d at 571-73. Two of these factors, the potential error rate and the existence of supporting literature, are the primary issues in M.P.A.’s false testimony claim and the subjects on which Willough-by testified falsely.
Willoughby testified regarding the Abel Assessment outside the presence of the jury. When asked about the Abel Assessment’s error rate, he stated that “[f]or classifying people who have significant sexual interest in female children under the age of fourteen, the accuracy rate is 85 percent.” This is particularly significant because at the time of the alleged offense, S.A. fell into this category. In addition, in response to a question regarding the existence of literature supporting or rejecting the Abel Assessment, Willoughby stated that “[t]here is [sic] a number of articles out by Gene Abel and his colleagues. Also researchers at Brigham Young University have established the reliability of the instrument and the classification accuracy of the instrument.”
Much of this testimony was false. In 1998, the accuracy rate of the Abel Assessment, according to Abel and his colleagues, for classifying people with a significant sexual interest in female children under fourteen was only 65%, not 85%. This weighs against the reliability of the Abel Assessment.11
*287Furthermore, contrary to Willoughby’s testimony, the Brigham Young University (BYU) studies failed to establish the Abel Assessment’s reliability as applied to adults and actively established that it was unreliable as applied to adolescents. Regarding adults, they found that it was a “promising instrument based on a sound idea,” but concluded that “the evidence of its reliability and validity for use with adults is weak as of yet,” labeled it a “nonvalidated instrument,” and called for “further research” and “refinement.”
Regarding the application of the Abel Assessment to adolescents, they found that no research other than their own had been done and that Abel’s initial study only included two adolescents. Their own research led them to conclude that data did “not support the reliability of [the Abel Assessment] for use with adolescents,” “that the ability of [the Abel Assessment] to discriminate adolescent offenders from nonoffenders was not significantly better than chance,” and the Abel Assessment’s “ability to screen or diagnose adolescent perpetrators reliably has not been demonstrated.”
The State argues that the following evidence supports the admission of Willough-by’s testimony:
• The statement in one of the BYU articles that “approximately 300 therapists in 86 states and two foreign countries, as well as 8 states’ judicial systems” used the assessment;
• Abel’s study of the Abel Assessment;
• Four independent studies supporting the theory underlying the Abel Assessment;
• The inability of M.P.A. and J.W.A.’s attorneys to find an expert to attack the Abel Assessment.
With the exception of Abel’s own study, the State did not present this evidence to the trial court. Nor would this evidence have been presented to the trial court had Willoughby testified truthfully regarding the Abel Assessment’s error rate and the BYU studies’ reliability findings. Therefore, we do not consider it in our determination of whether the trial court would have found the Abel Assessment reliable absent Willoughby’s false testimony.12
The State argues that we should consider the four independent studies because the State would have used them to rebut the criticisms in the BYU studies if Wil-loughby had testified truthfully about the BYU studies.13 The State’s framework would require that we assume Willoughby was aware of these studies and speculate as to how he would have testified about them. We reject this approach and do not consider the four studies. See, e.g., Graves v. Cockrell, 351 F.3d 143, 156 (5th Cir.2003) (referencing the largely speculative nature of allegations of what an uncalled *288witness would have testified to as a reason why complaints of uncalled witnesses are not favored).14
The State additionally argues that we should apply the less stringent standard from Nenno v. State to this case. 970 S.W.2d 549, 561 (Tex.Crim.App.1998), overruled on other grounds by State v. Terrazas, 4 S.W.3d 720, 727 (Tex.Crim.App.1999). Nenno held that Kelly’s reliability requirement applies with less rigor to fields of study aside from the hard sciences. Id. Nenno noted that “hard science methods of validation, such as assessing the potential error rate or subjecting a theory to peer review, may often be inappropriate for testing the reliability of fields of expertise outside the hard sciences.” Id.
This case stands in sharp contrast to Nenno. There, an expert testified regarding future dangerousness based on his experience studying eases. Id. at 562. That expert “did not contend that he had a particular methodology,” Id. Here, the Abel Assessment was subject to peer review and testing of its accuracy rate. Therefore, we consider those factors. See Mendoza v. State, No. AP-75213, 2008 WL 4803471, at *22 n. 62 (Tex.Crim.App. Nov. 5, 2008) (applying peer review factor in the soft science context of predicting future dangerousness because expert claimed to have a methodology, and contrasting Nenno); Nenno, 970 S.W.2d at 561 & n. 9 (stating that Nenno does not preclude employing the error rate and peer review factors in appropriate cases).15
In sum, had Willoughby testified truthfully, the trial court would have been faced with testimony regarding a test that had only a 65% accuracy rate as applied to this case, was subject to at least some criticism in the literature as applied to this case, and had no support from independent studies as applied to this case. The only *289evidence to support admission of the testimony regarding the Abel Assessment would have been a study by its creator that did not address the assessment’s application to this case. Given the evidence regarding the Abel Assessment’s application to adolescents, had Willoughby testified truthfully, the State would not have established the assessment’s reliability under Kelly. Therefore, we hold that the trial court would have excluded Willough-by’s testimony.
C. Harm Analysis
In order to obtain a new sentencing hearing, M.P.A. must prove by a preponderance of the evidence that Wil-loughby’s testimony contributed to his sentence. Ex parte Williams, 65 S.W.3d 656, 658 (Tex.Crim.App.2001).16 We review a trial court’s legal conclusions de novo, but defer to its fact findings if they are supported by the record. See Reliance Nat’l Indem. Co. v. Advance’d Temporaries, Inc., 227 S.W.3d 46, 50 (Tex.2007) (“Appellate courts review legal determinations de novo, whereas factual determinations receive more deferential review based on the sufficiency of the evidence.”). Applying this standard to the instant case, we conclude that the State’s use of Willoughby’s testimony throughout its closing argument contributed to M.P.A.’s sentence.
The State argues that the testimony of M.P.A.’s trial counsel, Bobby Barina, supports the habeas court’s finding that Wil-loughby’s testimony likely did not sway the jury. Barina stated in his affidavit that Willoughby’s testimony had “zero impact” on the jury. At the habeas hearing, he explained that Willoughby’s testimony was “boring.” He stated that it “didn’t provide any insight to anybody,” but did not remember that Willoughby likened M.P.A. to a pedophile. Barina also described Wil-loughby as “arrogant” and stated that the jury did not take “much consideration to anything Dr. Willoughby told them ... just because of the nature of Willoughby.”
Barina’s observations do not address the State’s use of Willoughby’s testimony to refer to M.P.A. as a pedophile throughout its closing argument. See Serv. Corp. Inti v. Guerra, 348 S.W.3d 221, 236 (Tex.2011) (stating that determination of whether error is harmful includes evaluating closing argument and counsel’s emphasis of erroneous evidence); Mathis v. State, 67 S.W.3d 918, 929 (Tex.Crim.App.2002) (Johnson, J., concurring) (separately analyzing admission of testimony and State’s use of that testimony during closing statements); LaPoint v. State, 750 S.W.2d 180, 192 (Tex.Crim.App.1988) (analyzing whether the State exploited erroneous instruction during closing argument). Here, the State argued:
• “He’s been diagnosed as a pedophile by an expert. He is at a high risk to re-offend.”
• “[Y]ou’ve heard the psychologist tell you he is a pedophile. He is at a high risk to reoffend.”
• “You now know he’s been classified as a pedophile by an expert. You now know that he is interested in children, *290interested in children, in fact, in the same age group as little [S.A.]. Think about her and think about that.”
These references to Willoughby’s testimony bolstered the State’s closing theme of protecting the community:
• “fl]f you put him on probation, we’ve already seen that just allows for victims.”
• “Our community simply cannot take that chance by releasing him back in that home. It’s a tough decision to make, but it’s a decision that’s backed up by the evidence and the testimony.”
• “How are you going to protect the public? The evidence has shown that the only way you’re going to be able to do that is by putting him away for some time. Because you’re going to have to protect other children. And with your verdict, you can at least keep him out of your community for a while.”
• “[Yjou’re also telling him, Tf I put you on probation, I’m going to walk right out this door with you.’ He could be next to you in the parking lot today and in your neighborhood tomorrow. Think about that.”
In sum, the State utilized Willoughby’s testimony throughout its closing theme of protecting the community. In addition, the State emotionally appealed to the jury to think about Willoughby’s classification of M.P.A. as a pedophile with a specific interest in S.A.’s age group. Indeed, the State’s closing argument made more express references to Willoughby’s testimony than to any other testimony in the case. Therefore, we conclude that the State’s use of Willoughby’s testimony at closing contributed to M.P.A.’s sentence.
IV. Ineffective Assistance of Counsel
Juveniles are entitled to effective assistance of counsel in adjudication proceedings. E.g., In re R.D.B., 20 S.W.3d 255, 258 (Tex.App.-Texarkana 2000, no pet.). To obtain habeas relief on a claim of ineffective assistance of counsel, M.P.A. must show that Barina’s performance was objectively unreasonable and Barina did not function as the “counsel” guaranteed by the Sixth Amendment. Strickland v. Washington, 466 U.S. 668, 687-88, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984). M.P.A. must make this showing by a preponderance of the evidence. Thompson v. State, 9 S.W.3d 808, 813 (Tex.Crim.App.1999).
Courts measure reasonableness by prevailing professional norms, Strickland, 466 U.S. at 688, 104 S.Ct. 2052, and give a heavy measure of deference to counsels’ judgments, id. at 691, 104 S.Ct. 2052. We look to “the totality of the representation and the particular circumstances of each case” in evaluating effectiveness. Thompson, 9 S.W.3d at 813. There is “a strong presumption that counsel’s conduct falls within the wide range of reasonable professional assistance.” Strickland, 466 U.S. at 689, 104 S.Ct. 2052.
The question of Barina’s effectiveness is one of mixed law and fact, id. at 698,104 S.Ct. 2052, and we therefore defer to the trial court’s factual determinations but review its legal determinations de novo, Perry Homes v. Cull, 258 S.W.3d 580, 598 (Tex.2008). We conclude that Barina’s performance was not deficient under Strickland.
M.P.A. argues that Barina rendered ineffective assistance at the adjudication phase because he failed to investigate or advance arguments that no sexual abuse occurred. However, at the time of trial, J.W.A. had confessed, which was inconsistent with an argument that no sexual assault occurred. Therefore, Barina could reasonably have believed that a strategy of affirmatively disproving the existence of a *291sexual assault would not have been effective.
M.P.A. specifically argues that Barina failed to interview a member of Catholic Charities who had noted that S.A. and A.A. denied abuse and showed no definitive indicators of abuse. However, the statement that the children denied abuse was not made until after trial. In addition, Barina stated that he had reviewed a report by Catholic Charities at the time of trial, which stated that there was no outcry but also discussed reasons why children typically might not volunteer information. Furthermore, pursuing this issue likely would have led the State to put on evidence that S.A. and A.A.’s behavior was consistent with abuse.
M.P.A. additionally argues that Barina failed to follow up on S.A. having to stop an interview in order to practice answering questions with LaVonna. The Iowa Department of Human Services had interviewed S.A. and A.A. at the request of Iowa and Texas police departments. A report from a police detective who was present noted that S.A. wanted to practice the questions with LaVonna. A social worker later interpreted this as S.A. wanting LaVonna to tell her what to say, which Barina viewed as an exaggeration. Barina interpreted the tape as S.A. being nervous and wanting her mother in order to calm down. Barina further stated that the tape referenced other sexual assaults not alleged by the State, so he did not want it admitted into evidence.
M.P.A. also asserts that Barina failed to adequately cross-examine the State’s medical expert, Dr. Green. Dr. Green had testified that the physical evidence was consistent with a possible suspicious finding. Barina testified that he thought Green’s testimony was as tentative as could have been hoped for, so there would be little benefit to aggressively cross-examining Green and allowing redirect examination to refute whatever he could have accomplished via cross-examination.
Finally, M.P.A. asserts that Barina should have obtained an expert to refute Green’s findings. The Second Circuit Court of Appeals granted relief on a similar claim in Gersten v. Senkowski, 426 F.3d 588 (2d Cir.2005). In Gersten, the State’s case “rested centrally on the alleged victim’s testimony and its corroboration by the indirect physical evidence as interpreted by the medical expert.” Id. at 608. The petitioner’s trial counsel had “essentially conceded that the physical evidence was indicative of sexual penetration without conducting any investigation to determine whether this was the case.” Id. The court noted that “medical expert consultation or testimony is particularly critical to an effective defense in sexual abuse cases where direct evidence is limited to the victim’s testimony.” Id. The petitioner’s trial counsel unsuccessfully pursued a trial strategy of cross-examining the State’s expert. Id. at 609-10. The counsel thought that disproving sexual activity had occurred would be fruitless because of hearsay from the petitioner that the victim engaged in sexual conduct with her ex-boyfriend. Id.
The general premise of Gersten is inapplicable here for two reasons. First, in Gersten, the victim had accused the petitioner of “‘continuing rape and sodomy over a period of years.’ ” Id. at 608 (quoting Gersten v. Senkowski, 299 F.Supp.2d 84, 91 (E.D.N.Y.2004)). Therefore, “the medical testimony was central not only because it constituted the most extensive corroboration that any crime occurred, but because to undermine it would undermine the alleged victim’s credibility and thus the entire prosecution case as to all charges.” Id. at 608. Although an expert for M.P.A. *292could have shown that the evidence did not corroborate the State’s claim, it would not have rendered the victim’s testimony “demonstrably incredible.” Id. Second, J.W.A.’s confession and statement to Bari-na that he had sexually assaulted S.A. provided Barina greater reason to think an exam would support the State’s theory than the hearsay from the client in Ger-sten. Because the petitioner’s trial counsel in Gersten had less reason to think challenging the State’s expert would be fruitless, and a successful challenge would have been more powerful in that case, we decline to follow Gersten’s finding of ineffective assistance.
In sum, looking at the totality of the representation, Gersten does not apply here. Given the heavy discretion afforded trial counsel, we cannot say that M.P.A. has met his burden to show that Banna’s representation was objectively unreasonable.
Because we have determined that M.P.A. is entitled to a new disposition hearing, see Part III, supra, we do not address M.P.A.’s claim of ineffective assistance of counsel at the disposition phase.
V. Conclusion
We hold that M.P.A. has not established his right to relief on his claims of actual innocence or ineffective assistance of counsel at the adjudication phase. However, M.P.A. is entitled to a new disposition hearing because Willoughby’s false testimony contributed to his sentence. We remand this cause to the district court to grant M.P.A.’s writ of habeas corpus in accordance with this opinion.
. S.A. and A.A. are sister and brother and M.P.A. and J.W.A. are brothers. The four children are related through their fathers, who are brothers. Thus, M.P.A. and J.W.A.'s father is S.A. and A.A.’s uncle, and S.A. and A.A.'s father is M.P.A. and J.W.A.'s uncle.
. "Although quasi-criminal in nature, proceedings in juvenile court are considered civil cases; thus, this Court, rather than the Court of Criminal Appeals, is the Texas court of last resort for such matters.” In re Hall, 286 S.W.3d 925, 927 (Tex.2009). "[T]he Court of Criminal Appeals has concluded that it lacks jurisdiction to issue extraordinary writs in such cases, even those initiated by a juvenile offender who has been transferred to the Texas Department of Criminal Justice because he is now an adult_[I]t is the applicant’s age at the time he commits the delinquent acts that determines jurisdiction, rather than his age when applying for habeas corpus.” Id. at 927 (citing Ex Parte Valle, 104 S.W.3d 888, 889 (Tex.Crim.App.2003) (internal citations omitted)).
. Evidence corroborating the recantations includes: evidence that the allegations were made after LaVonna left Texas with S.A. and A.A. in violation of a court order and having stolen money from her employer; evidence that the allegations of abuse in the present case coincided with key events in the custody dispute between S.A. and A.A.’s parents; evidence that LaVonna had made false allegations of sexual abuse of children on other occasions; testimony that LaVonna had said she would do anything to maintain custody and that she would hurt M.P.A. and J.W.A.’s family; an affidavit from LaVonna’s sister stating that LaVonna would lie about sexual abuse of children to gain an advantage in a custody proceeding; and evidence attacking the inculpatory medical evidence.
. An applicant may also present an actual innocence claim that is tied to a showing of constitutional error at trial. To prevail on this type of claim, “[a]n applicant must show that the constitutional error probably resulted in the conviction of one who was actually innocent.” Ex parte Spencer, 337 S.W.3d 869, 878 (Tex.Crim.App.2011) (citing Schlup v. Delo, 513 U.S. 298, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995)). M.P.A. has not argued that he is entitled to relief based on a Schlup-type claim.
. In the context of a motion for a new trial, the Court of Criminal Appeals has noted that it is unclear whether a trial court's rejection of a recantation requires record support. Keeter v. State, 74 S.W.3d 31, 37-38 (Tex.Crim.App.2002). Because we find that there is record support for the habeas court's rejection of the recantations, we do not address this issue.
. The Texas Constitution prohibits our Court from engaging in factual sufficiency review. TEX. CONST, art. V, § 6(a) (stating that courts of appeals' decisions are conclusive on all questions of fact). In the habeas corpus context, we are bound by the trial court’s credibility determination if there is record support for that finding. Here, the trial court determined that the recantation testimony was not credible. The trial court heard all the testimony and is best suited to make credibility determinations. See Wainwright v. Witt, 469 U.S. 412, 428, 105 S.Ct. 844, 83 L.Ed.2d 841 (1985) ("such a finding is based upon determinations of demeanor and credibility that are peculiarly within a trial judge’s province”). We cannot speculate as to how we would hold regarding the credibility of the recantation testimony had we been in the trial court’s position. However, because there is some record support for the trial court’s determination, M.P.A. has not established his right to relief on actual innocence grounds.
. The Court of Criminal Appeals has allowed for the possibility of a lesser threshold for the showing of harm when the State knowingly used false testimony and the defendant did not have the opportunity to discover and raise that claim at trial or on direct review. Napper, 322 S.W.3d at 242-43 (citing Chabot, 300 S.W.3d at 770-72; Fierro, 934 S.W.2d at 372-75 & 375 n. 10). This implicitly allows for the possibility of a greater threshold to show harm — or theoretically a stricter preservation of error standard — when there is unknowing use of false testimony and the applicant had an opportunity to raise the issue at trial or on direct review. However, we will not find waiver based on a potential implication in the case law when a plain reading of Fierro and Chabot indicates that M.P.A. can bring his false testimony claim.
We also note that the Court of Criminal Appeals has stated that there is a question about forfeiting a false testimony claim when an applicant made no complaint about allegedly false testimony at trial. Napper, 322 S.W.3d at 241 (citing Estrada v. State, 313 S.W.3d 274, 288 (Tex.Crim.App.2010)). In Napper, the Court of Criminal Appeals addressed the merits “[w]ithout expressing any opinion on the preservation of error question.” Id. Regardless, its current precedent allows M.P.A. to bring the false testimony claim in habeas and we will not find waiver by anticipating future potential decisions. This is consistent with the Legislature's statutory directives that its provisions relating to habeas "shall be most favorably construed in order to give effect to the remedy, and protect the rights of the person seeking relief under it,” Tex.Code Crim. Pro. art. 11.04, and that the purpose of a habeas appeal is "to do substantial justice to the parties,” Tex.R.App. Proc. 31.2.
. Alternatively, M.P.A. could prevail by showing that, even had Willoughby's testimony been admitted, if it was truthful the jury would have handed down a lesser sentence. We do not reach this issue.
. Kelly applies to all scientific evidence, regardless of whether or not it is novel. Nenno v. State, 970 S.W.2d 549, 560 (Tex.Crim.App.1998).
. M.P.A. briefed this issue under E.I. du Pont de Nemours and Co. v. Robinson, which controls the reliability inquiry in civil cases. 923 S.W.2d 549, 552-58 (Tex.1995). Because juvenile proceedings are quasi-criminal in nature and criminal law evidentiary rules govern, In re M.A.F., 966 S.W.2d 448, 450 (Tex.1998), we use the language from Kelly rather than Robinson. However, neither party suggests that the tests are substantively different as applied to this case. See also Merrell Dow Pharms., Inc. v. Havner, 953 S.W.2d 706, 731 (Tex.1997) (Gonzalez, J., concurring) (calling Kelly the Court of Criminal Appeals' "equivalent of Robinson ”). Reliable evidence should be admitted in both the civil and criminal context, and unreliable evidence or "junk science” should be kept out of court in either context.
.See United States v. Birdsbill, 243 F.Supp.2d 1128, 1135-36 (D.Mont.2003) (describing later version of the Abel Assessment's error rates of 21-22% and 32% as "poor” and finding test unreliable); United States v. White Horse, 177 F.Supp.2d 973, 975 (D.S.D.2001) (stating that 24% false negative on Abel Assessment “does not assist the jury” and finding test inadmissible); State v. Ericson, 13 A.3d 777, 781-82 (Me.2011) (affirming trial court’s exclusion of testimony regarding Abel Assessment and stating that error rate between 21% and 32% raises significant concern about reliability); Figueroa v. State, No. 04-08-00452-CR, 2009 WL 2183460, at *3-4 (Tex.App.-Amarillo July 22, 2009, no pet.) (citing Birdsbill, 243 F.Supp.2d at 1135, and referring to Abel Assessment’s “significant potential for error” and affirming exclusion of *287testimony regarding Abel Assessment); cf. United States v. Robinson, 94 F.Supp.2d 751, 753-54 (W.D.La.2000) (finding Abel Assessment reliable where evidence at hearing indicated 78% or 79% accuracy).
. For the same reason, we would not consider the numerous published attacks on the Abel Assessment's reliability and cases rejecting the Abel Assessment which were not raised in the trial court but are now cited by M.P.A. and amicus Innocence Project of Texas, even if they were available at the time of trial. To do otherwise would convert a claim attacking false testimony into a general retrial of the reliability of the Abel Assessment. Indeed, the false testimony would be irrelevant under this approach.
. These studies were papers presented at a conference and are not in the record before us. See Gene Abel, The Importance of Meeting Research Standards: A Reply to Fischer and Smith’s Articles on the Abel Assessment for Sexual Interest, 12 Sexual Abuse: J. Res. & Treatment, 155, 161 (No. 2, 2000) (citing papers).
. Furthermore, the State does not specify their content or what aspect, if any, of the BYU studies they would have been responsive to. Although they were cited in an article by Abel that was published after M.P.A.’s trial and attacked the BYU studies, he cited the papers referred to by the State in his introduction as studies coming to a different conclusion from the BYU studies. He did not suggest the papers cited by the State were themselves critical of the BYU studies.
Moreover, the titles of at least two of the papers, A Comparison of the Penile Plethysmo-graph with The Abel Assessment for Sexual Interest on Incarcerated Military Sex Offenders, and Comparing Outcomes of Plethysmographic Assessment with The Abel Assessment in a Prison Based Sex Offender Sample, indicate they did not address the use of the Abel Assessment on adolescents.
Finally, we note that other courts found the Abel Assessment unreliable and criticized what appear to be the same studies. See Birdsbill, 243 F.Supp.2d at 1134-35 (criticizing studies on multiple grounds and finding they do not support the reliability of a later version of the Abel Assessment); Ready v. Massachusetts, No. Civ.A. 00-10390 SDP, 2002 WL 1255800, at *8-12 (Mass.Super.Ct. May 17, 2002, pet.denied) (same).
. Kelly frames the reliability inquiry as addressing: (1) the validity of the underlying scientific theory; (2) the validity of the technique applying the theory; and (3) the correct application of the technique on the occasion in question. 824 S.W.2d at 573. Nenno frames the inquiry as addressing: (1) whether the field of expertise is legitimate; (2) whether the subject matter of the expert's testimony is within the scope of the field; and (3) whether the expert's testimony properly relies on or utilizes the field's principles. 970 S.W.2d at 561. Because the peer review and error rate factors failed to establish the Abel Assessment’s reliability, the differences between the tests are immaterial here. See State v. Medrano, 127 S.W.3d 781, 785 (Tex.Crim.App.2004) (stating that "because the objective of both Kelly and Nenno was to ensure the reliability of expert testimony and scientific evidence, Nenno ‘[did] not categorically rule out employing [the Kelly \ factors in an appropriate case’ ") (alterations in original) (quoting Nenno, 970 S.W.2d at 561 n. 9).
. M.P.A. argues that we should review this determination de novo and cites Johnson v. Cain, 215 F.3d 489, 495 (5th Cir.2000). However, there is no indication that the Johnson court reviewed findings of fact de novo. M.P.A. also argues that the Court of Criminal Appeals has stated that it "must independently examine the record.” Burnett v. State, 88 S.W.3d 633, 638 (Tex.Crim.App.2002). We note that Burnett was in the context of a direct appeal, where there were no relevant findings regarding harmful error to defer to. In contrast, in the context of harmless error analysis in a habeas case, the Court of Criminal Appeals has stated that it generally defers to the lower court’s findings of fact and conclusions of law when they are supported by the record. Chabot, 300 S.W.3d at 772.