with whom MATTHEWS, Chief Justice, joins, dissenting in part.
1. INTRODUCTION
Because I disagree with this court’s resolution of the custody issue, I dissent from that portion of its opinion. I would remand because the record does not establish that the superior court made the findings the custody statute requires.
2. DISCUSSION
Alaska Statute 25.24.150(c) requires a superior court to consider eight enumerated factors in determining a child’s best interests before awarding custody.1
*1051Joseph argues that the superior court failed to consider two relevant statutory factors, the children’s preferences and the desirability of maintaining stability and continuity in their lives. He also argues that the superior court did not make an ultimate finding that it is in the children’s best interests that Kathey, rather than Joseph, be their sole custodian.
This court rejects those arguments. It does so after carefully reviewing both the record (to determine what statutory factors the parties disputed at trial) and the superior court’s findings (to determine whether it explicitly or implicitly addressed particular statutory factors). In light of the issues and findings thus perceived, this court holds that the custody award was consistent with the children’s best interests.2
I disagree with that approach for three reasons. First, it does not satisfy AS 25.24.150(c) and Alaska Civil Rule 52(a). Second, it rests on an implicit, but inappropriate, appellate assumption that the superi- or court either correctly resolved disputes about the two factors or determined that the factors were irrelevant. Important policy considerations should preclude such deferential treatment of custody awards. Third, even under the approach the court takes here, the facts require remand.
A. Explicit Findings Are Needed to Satisfy Statute and Rule.
Alaska Statute 25.24.150(c) imposes two mandatory duties on the superior court. It “shall determine custody in accordance with the best interests of the child”; and in determining the child’s best interests, it “shall consider” eight enumerated factors.3 Because such cases are tried without juries, our procedural rules require a trial court to “find the facts specially and state separately its conclusions of law thereon.”4
The statute mandates a two-part analysis. First, the trial court must find the extent to which each relevant statutory factor favors one parent or the other. Making these findings necessarily requires the trial court to determine the relevance of each statutory factor. Second, the trial court must weigh the relevant factors and make a best-interests finding before awarding custody.
The statute does not state that the trial court must memorialize its analysis and findings at each step. But because the trial court can satisfy the statute only by making findings of fact and conclusions of law, Civil Rule 52(a) effectively imposes a memoriaiization requirement because it demands that the trial court “find the facts specially” and “state separately” its legal conclusions.5 I believe that this demand requires trial courts to memorialize their custody findings and conclusions in writing or by minute order recorded electronically.
As Chief Justice Matthews observed in his dissent in Borchgrevink v. Borchgrevink6 the statute and rule collectively require the trial court to make both “integrative” findings and an “ultimate” finding.7 Chief Justice Matthews there identified four purposes served by a requirement that trial courts make find*1052ings in judge-tried cases. “This requirement (1) ‘aids the trial judge’s analytical process’; (2) ‘may be relevant for collateral estoppel purposes’; (3) ‘aids the appellate court on review; and (4) enables ‘the defeated party to determine whether the case presents a question worthy of consideration by the appellate court.’ ”8 To these I would add that proper findings set a baseline of circumstances that permit comparison when the noncustodial parent claims that changed circumstances justify a change in custody.
For these and other reasons, I am now convinced that the statute can best be satisfied by requiring expressly memorialized findings and conclusions after the trial court determines the relevant statutory factors, resolves disputes concerning those factors, and makes the ultimate best-interests finding by weighing those factors.
The court argues that such a requirement would elevate form over substance.9 And not long ago the court noted in dicta, “[w]e do not believe custody determinations should be overturned merely because a judge fails to tally the statutory factors like runs, hits and errors in a box score.”10 But I do not regard a requirement originating in this statute to be a mere formalism. Because a custody dispute is decidedly not a sporting event, because an appellate court cannot tally hits, runs, and errors, and because the best-interests determination ultimately turns on a complex weighing of statutory factors having unequal values in a given case, the trial court must explain what it did and how it did it.
Requiring a memorialization would demand nothing more than Civil Rule 52(a) now requires. Given this court’s presumption that the trial court has engaged in the analysis the statute demands, memorialization will add little extra time or effort. If this court’s presumption is incorrect, the memorialization requirement will encourage the proper statutory analysis. The trial court can easily preserve its findings in an oral minute order. The benefits will far outweigh any additional burden.
In light of .the statute and Rule 52(a), I conclude that we should alter our current approach in reviewing such cases. It is one thing to review an appellate record to determine whether the superior court has made the appropriate findings where none appear there plainly.11 But we should no longer sift through appellate records to determine which statutory factors seem to have been irrelevant in particular cases and which factors the parties have chosen not to dispute.12 Nor should we attempt to discern from the record whether particular findings were meant to dispose of statutory factors the trial court failed to discuss explicitly.13 Our current practice does not remedy a failure to make findings and memorialize them in the record. It does not satisfy the first, second, and fourth purposes Chief Justice Matthews identified in Borchgrevink,14 Even as to the third purpose, careful scrutiny of the record is a poor substitute for the assistance that express findings would provide during appellate review. A trial court’s failure to memorialize its reasoning and findings gravely impedes appellate review and impairs its accuracy. And if a change of circumstances is alleged, reconstructing the circumstances the trial court originally considered when it first awarded custody is neither effective nor efficient.
B. The Drawbacks of Our Current Approach Outweigh Any Possible Benefits.
This brings me to my second concern. We are a court of review. We do not try or retry fact disputes committed to the trial court. To review custody awards on appeal, *1053we need to know how the trial court ruled. When the trial court fails to address a statutory factor, is it because the court thought that factor irrelevant or because the court simply overlooked it? When the trial court awards custody without making an ultimate best-interests finding, has it applied the correct statutory standard? Has it weighed any factors not mentioned in its findings? If a factor is relevant and in dispute, did the trial court consider it in deciding the children’s best interests when there is no indication what it considered? When the record is silent or ambiguous as to a given factor, this court’s approach necessarily rests on an assumption of regularity, i.e., an assumption that the superior court did not overlook any factors and awarded custody based on the children’s best interests.
In my view, it is not enough that it seems probable in a given ease that the trial court reached a permissible result, because there can be more than one permissible result. And the outcome may hinge on a factor the trial court may have overlooked. When the trial court does not say what it did, we should not have to assume that it applied the correct statutory analysis and made a best-interests determination. Nor should the parties. A more rigorous approach will reassure parties (and appellate jurists) that the trial court engaged in the right analysis.
This reassurance should not be lightly ignored. It is important that the losing party have confidence that the trial court resolved the dispute on its merits, engaged in the right analytical process, and overlooked no relevant statutory factors. If the trial court makes no memorialized findings on specific factors or makes no memorialized best-interests finding, and does not explain why it does not, the losing party can fairly wonder whether justice was done. That party is not likely to be reassured by an appellate process that assumes that the trial court followed the correct process even though it is not clear what process the trial court followed.
This raises the related problem of possible appellate inaccuracy. I think that it is not adequate for us to resolve silence or ambiguity by scrutinizing the parties’ trial court arguments. There are two reasons. First, the children themselves are rarely represented, and the parents’ needs may not align with their children’s best interests. The parents’ failure to address pertinent factors cannot excuse a superior court’s failure to consider those factors. Second, rarely is the record so unambiguous that we can definitively determine that the parents waived particular factors. Testimony and argument about custody often do not clearly separate one factor from another. Cases involving pro se parents are even more problematic because they are more likely to overlook relevant factors. Consequently, appellate records are often sufficiently unclear that reasonable jurists can permissibly draw different conclusions about the evidence, the parties’ arguments, and what the trial court intended. Our approach, therefore, may cause us to draw inaccurate conclusions that do not reflect what the trial court really did.
Policy reasons should also foreclose our willingness to review the record to reconstruct the superior court’s analysis. Our approach adversely affects the finality of custody awards by encouraging appeals on the avoidable issue of the adequacy of the findings. It increases litigation costs by creating avoidable appellate issues and exposes the parties to the expense and delay of remand if findings prove insufficient.15 It decreases predictability because the appellate result turns on what a majority of this court, after examining the record, decides the trial court meant to do absent memorialized findings. It adversely affects consideration of custody modification motions because the baseline circumstances may have to be reconstructed. And if we misinterpret the record, and remand for a different result, we may interfere with what might have been an allowable exercise of trial court discretion.
It also interferes with resolving custody disputes expeditiously. Appeals in these cases are especially time-consuming, because our approach encourages each justice to engage in a fact-intensive review of the record, and often provokes separate opinions. Sepa*1054rate opinions were issued in Park v. Park,16 Duffus v. Duffus,17 and Borchgrevink,18
The resulting delays harm the parties. And the time individual justices spend reviewing the records in these cases trying to determine what the trial court must have intended delays our consideration of other worthwhile, but nonexpedited, appeals.
For these reasons, I feel parties will be better served if we follow the approach Chief Justice Matthews advocated in Borchgrevink and further require the trial court to memorialize its findings.
C. Even under the Current Approach, this Custody Dispute Should Be Remanded.
But even under the approach the court follows here, I think remand is necessary. First, the superior court did not address the children’s preferences although two children were arguably old enough to express preferences. The court seems to interpret the superior court’s comments to indicate that it may have reasoned that the children’s preferences would not have been reliable.19 But that interpretation supposes that the superi- or court actually made an unreliability determination when it did not. This court also reasons that the issue was not really disputed at trial anyway.20 Given the statutory mandate, I am not convinced that a failure of the parents to raise the preference issue should be determinative. But in any event, the issue was sufficiently raised here that the superior court should have memorialized any such findings. At least the trial court should have stated that the factor was not relevant, or that the children’s preferences would have been unreliable, if that is what it indeed found.
This court also reads the record to suggest that the statutory stability and continuity factor was not in real dispute, and that in any event the evidence on this factor favored Kathey.21 But the initial relevance determination is best made by the trial court because it is most familiar with the evidence and the parties’ arguments. Imposing that duty aids the analytical process that the trial court must follow anyway. Likewise, it is for the trial court to make findings about the evidence, and I do not feel comfortable reasoning that the facts could only be read to favor Kathey on this issue.
I am also uncomfortable with assuming that the disputed factors favored Kathey and that the other factors were at best neutral, and that the superior court must have found the children’s best interests before awarding Kathey custody.22 It is circular to reason that the court applied the right analysis because the factors justified the award made. If only one outcome were possible, perhaps that would be an appropriate way to reason. But even if all factors but one favor one parent, in a given case that one factor may outweigh the others.
3. CONCLUSION
The court has engaged in an admirable but time-consuming attempt to determine whether the superior court conducted the analysis the legislature requires. I think it illustrates why we must change the way we review such disputes. But even applying the approach we have followed in the recent past, I would reverse and remand the custody issue in this case for consideration of the preference and continuity factors and for memorialization of the necessary findings.
. AS 25.24.150(c) (1998) provides:
The court shall determine custody in accordance with the best interests of the child under AS 25.20.060 — 25.20.130. In determining the best interests of the child the court shall consider
(1) the physical, emotional, mental, religious, and social needs of the child;
(2) the capability and desire of each'parent to meet these needs;
(3) the child’s preference if the child is of sufficient age and capacity to form a preference;
(4) the love and affection existing between the child and each parent;
*1051(5) the length of time the child has lived in a stable, satisfactory environment and the desirability of maintaining continuity;
(6) the desire and ability of each parent to allow an open and loving frequent relationship between the child and the other parent;
(7) any evidence of domestic violence, child abuse, or child neglect in the proposed custodial household or a history of violence between the parents;
(8) evidence that substance abuse by either parent or other members of the household directly affects the emotional or physical well-being of the child;
(9) other factors that the court considers pertinent.
.See Op. at 1047.
. See AS 25.24.150(c).
. Alaska R. Civ. P. 52(a).
. See Alaska R. Civ. P. 52(a).
. Borchgrevink v. Borchgrevink, 941 P.2d 132 (Alaska 1997).
. See id. at 142. I authored the court's opinion ' from which Chief Justice Matthews dissented. But our opinion in Borchgrevink did not reject his reasoning concerning the purposes of integrative and ultimate findings. It instead concluded that those purposes had been served in that case, that the superior court had considered the relevant factors, and that its integrative findings were sufficient. Id. at 141.
. Id. at 142 (Matthews, J., dissenting) (quoting In re D.C., 715 P.2d 1 (Alaska 1986); Merrill v. Merrill, 368 P.2d 546 (Alaska 1962); Rogge v. Weaver, 368 P.2d 810, 814 n. 7 (Alaska 1962)).
. See Op. at 1047.
. Julsen v. Julsen, 741 P.2d 642, 649 n. 10 (Alaska 1987).
. See, e.g., Duffus v. Duffus, 932 P.2d 777, 779-80 (Alaska 1997).
. See Borchgrevink, 941 P.2d at 140-41.
. See Park v. Park, 986 P.2d 205, 206 (Alaska 1999).
. See Borchgrevink, 941 P.2d at 142.
. See Park, 986 P.2d at 211.
. 986 P.2d at 211.
. 932 P.2d at 781.
. 941 P.2dat 141.
. See Op. at 1045-1046.
. See Op. at 1045-1046.
. See Op. at 1046-1047.
. See Op. at 1046-1047.