Lula B. MILLER, Appellee, v. MERCY HOSPITAL, INCORPORATED, D/B/A Mercy Hospital, Appellant

JAMES DICKSON PHILLIPS, Circuit Judge: This is a Title VII case in which Ms. Lula B. Miller, a black woman, sued Mercy Hospital, Inc. (Mercy), claiming discrimination on account of her race in Mercy’s failure to hire her as a nurse’s aide. Following a bench trial, the district court found Mercy liable as Miller had alleged and awarded Miller substantial monetary relief, costs, and attorney fees. On Mercy’s appeal, we conclude that the district court’s ultimate factual determination that Mercy intentionally discriminated against Miller was, on the whole record, clearly erroneous. We therefore reverse. I Lula B. Miller is a black woman who for a number of years prior to this litigation had been employed in the general field of nursing in Charlotte, North Carolina, where Mercy is located. The events leading to this litigation can be traced to her employment in 1968 by Presbyterian Hospital (Presbyterian), another Charlotte hospital, as a nurse’s aide (NA). In 1972, having been promoted by Presbyterian in the interim to Nurse Technician, she graduated from Central Piedmont Nursing School, and stood for licensure as a practical nurse by taking the state’s examination. At this *357 point Presbyterian — apparently in keeping with general custom among hospitals in the area — allowed Miller to perform the functions of a licensed practical nurse (LPN) pending receipt of the results of her licensing examination. Unfortunately, Miller did not pass this examination; nor did she pass it on three subsequent takings prior to this litigation. Miller’s employment in this capacity by Presbyterian nevertheless continued until July of 1973 when she voluntarily resigned under circumstances discussed later in this opinion. Within a month, according to her later testimony, Miller unsuccessfully applied to Mercy for employment in some nursing capacity. Around a year later, on August 14, 1974, Miller again applied for employment at Mercy in response to newspaper ads soliciting applications for LPNs and NAs. Miller’s application was processed in a personal interview with a Ms. Dillie Winchester whose routine function this was. A typical written employment application form was used to record Miller’s background and qualifications. It contained no formal entry for identifying the applicant’s race, nor did Miller’s completed form indicate her race by any special entry or by other manifest indicia. In a box marked “Type of Work Preferred,” Miller was invited to indicate her preference, and in response she herself entered “LPN.” In another box marked “Classification,” Winchester then entered “N.A.” reflecting, according to Winchester’s later testimony, her judgment as applicant interviewer that this was the position for which Miller’s application data revealed her to be qualified. According to Miller’s testimony, she had indicated to Winchester that she, Miller, would be interested in a nurse’s aide position as an alternative to her recorded preference for an LPN position. Following normal procedures, Winchester then forwarded the completed application form to Ms. Casmira Marciniszyn, the Director of Nursing at Mercy, whose authority it then was to make this type of hiring decision for Mercy. Again in keeping with usual procedures, Winchester made a telephoned request of Presbyterian for a refer-enee on Miller. The results of this reference call are a matter of critical import in this litigation. The only direct evidence on the point was provided by Winchester’s testimony. According to that testimony, Winchester entered the results of her report, immediately following its receipt, upon a standard reference form. The form, dated August 14, 1974, was introduced in evidence. As completed by Winchester, the form began “Lula W. (sic) Miller has applied to us for a position as ‘N.A.’ .... ” In a rating grid on the form Winchester indicated that Presbyterian evaluated Miller “very good” “as N.A.” but “Unsatisfactory” “as PN.” In a note section on the reference form, Winchester summarized her conversation with the Presbyterian referee as follows: “could not function as an LPN. Was unhappy when the hospptal] reported this on her 502 [North Carolina Employment Security Commission Separation Notice]— had lawyer contact hospital.” According to Winchester’s later testimony, not directly contradicted, her contemporaneous entries on the reference form reflected in full the substance of the reference: she was told no more about the reasons for Miller’s “unhappiness” with Presbyterian than appeared on the form, nor was she told any more about the nature of the lawyer’s “contact” with Presbyterian. Marciniszyn received Miller’s application in ordinary course. Within a week, under date of August 21, 1974, Marciniszyn made a “Not interested” entry upon the application and returned it to Winchester. This entry constituted Mercy’s decision not to hire Miller based upon the August 14, 1974 application. The rejected application form was filed by Winchester along with the reference form reflecting the Presbyterian reference report. At the time of her decision not to hire Miller, it was impossible for Marciniszyn to have determined Miller’s race solely from entries on the form. Whether she then knew Miller’s race from any other source is obviously a matter of critical import. It is disputed by the parties. Marciniszyn testified in this litigation that she did not then *358 know Miller’s race. This testimony is not contradicted by any direct evidence. No specific finding of fact on the point was made by the district court. We return to the point in later discussion of the court s findings. Neither is it apparent from any direct evidence of record whether Marciniszyn knew of the reference from Presbyterian at the time of her decision not to hire. Both Marciniszyn and Winchester later testified that they did not recall whether it was ever brought to Marciniszyn’s attention. No direct evidence contradicts this testimony, either as to the witnesses’ states of recall at trial or as to the fact itself. It is only clear that the reference form was at some point filed by Winchester with the rejected application form returned to her by Marciniszyn. Marciniszyn’s reason — hence Mercy’s — for rejecting Miller’s application was, according to Marciniszyn’s later testimony, a simple one: she considered the application to be one for employment as a LPN, Miller’s stated preference; Miller’s application revealed her not qualified for that position because not licensed; 1 hence Marciniszyn, as hiring authority, was “not interested” in interviewing Miller as an applicant for employment. No direct evidence contradicts this proffered reason. Whether other evidence —indirect, circumstantial — sufficiently disproved it is the disputed, dispositive issue in the case. Miller learned of the decision not to hire her sometime shortly after it was made, How she learned this and what she was told are not agreed between the parties. Miller’s version, accepted by the trial court, was that she learned of her rejection by making telephoned inquiry of Winchester, and that Winchester told her the reason was the negative reference she received from Presbyterian. Winchester testified that she did not recall Miller’s having made any inquiry of her and that in any event she could not then have given Miller any specific reason because it was a matter known only to Marciniszyn to which Winchester was not then privy, In November of 1974, Miller, represented by private counsel, filed an EEOC charge alleging racial discrimination by Presbyterian in its forwarding of negative references to Mercy and Charlotte Memorial Hospitals and, arguably, also alleging discrimination by the latter two hospitals in acting upon the references. The charge against Presbyterian alleged that the negative references were given because of Miller’s complaint to Presbyterian of racially discriminatory treatment by that hospital. Mercy received n0 notice of this charge at the time of its filing. in March of 1975 the EEOC’s district director notified Miller that he had dismissed the charge as it might apply to Mercy, on the stated basis that the charge nowhere alleged that Mercy had any knowledge that any negative reference it received from Presbyterian was racially inspired, Mercy received no notice at that time of the dismissal of this charge. Perhaps significantly for the further course of this litigation, the district director, in an official cornmunication explaining to Miller’s counsel the basis for the dismissal, distinguished a case in which a negative employment reference had specifically identified a black charging party as “a troublemaker” who “was not averse to frivilously (sic) alleging racial discrimination.” The director’s letter concluded with the comment that “[w]e perceive no impediment to amending the charges to allege knowledge on the part of those respondents (Mercy and Charlotte Memorial) if your client believes that such knowledge did exist.” Following a formal request by Miller’s counsel for an opinion from the EEOC on the status of the charges against Mercy and Charlotte Memorial, the district director in April of 1975 advised Miller’s counsel that the charges had been *359 reopened and that “respondents” (presumably Mercy and Charlotte Memorial) would be notified. On February 6, 1976, Miller filed with the EEOC an “amended charge” alleging that Mercy had discriminated against her by failing to hire her in August of 1974. Mercy received formal notice of this charge on February 20,1976, some sixteen months after its August 1974 decision not to hire Miller. So far as the record indicates, this was the first indication Mercy had that it was the subject of any racial discrimination charges by Miller. 2 On November 26, 1976, while her EEOC charge against Mercy was pending, Miller returned to Mercy to apply for employment. Again she saw and was interviewed by Winchester, and again, except for agreement on the date of this second interview, the evidence of what transpired is in conflict. Miller’s version is that she again applied for employment as a NA and that in this connection she showed Winchester an EEOC probable cause determination letter finding that Presbyterian had indeed made a false entry, for racially discriminatory reasons, on Miller’s state Employment Security Commission separation form. By Miller’s later testimonial account, she had then pointed out to Winchester that this effectively removed any basis for Presbyterian’s negative reference and established Miller’s qualification for employment. Nevertheless, she was again not hired. Winchester’s version of the November 26, 1976, interview, based upon a contemporaneously prepared memorandum that was introduced in evidence, was at flat odds with the essence of Miller’s. According to Winchester, Miller flatly refused to apply for a position as NA — on the basis that she was qualified for a better position — and submitted no application. Winchester also testified that she had no recollection of Miller’s having shown her or spoken to her of an EEOC determination letter respecting Miller’s charge against Presbyterian. The district judge accepted Miller’s version in a specific fact finding to which we return in later discussion. Following exhaustion of the EEOC administrative procedures involving her charge against Mercy, Miller commenced this action on February 22, 1979. When commenced, the action included both Miller’s individual claim of racial discrimination by Mercy in failing to hire her when she applied on August 14, 1974, and a class action claim in behalf of a putative class of black applicants for nursing positions similarly denied employment by Mercy from and after May 15,1974. The class proposed was at first “conditionally certified” by the district court, but the class claim was later dismissed before trial when Miller failed as class representative to demonstrate the existence of a class. Following a period of discovery, Miller’s individual claim was tried to the district court in a one-day bench trial on January 4, 1982. In addition to a number of documentary exhibits, the evidence consisted solely of the live testimony of Miller, Winchester and Marciniszyn. On January 7, 1982, the district court entered a brief written memorandum of decision which announced its ruling that Mercy had racially discriminated against Miller as alleged and that appropriate relief should be granted. In the memorandum, plaintiff was requested to prepare appropriate findings of fact and conclusions of law and a proposed judgment. The gist of certain findings to be included was set out in the order. The order concluded with a determination that “Miller was denied employment because of her race, and specifically because she made complaints of racial discrimination when she was working at Presbyterian Hospital.” Counsel for Miller submitted proposed findings, conclusions and judgment. So far as the record reveals, counsel for Mercy was *360 not invited to comment upon nor object to the proposals either before or after their submission to the court nor to submit its own proposals, and did none of these. The proposed findings of fact included the substance of those suggested by the court and others originated by plaintiff’s counsel. Subject to a few non-substantive editorial and grammatical revisions and the inclusion of two additional findings of fact, one related to Mercy’s proffered explanation of its failure to hire Miller and the other to Miller’s qualifications for the positions sought, the court adopted the findings and conclusions as drafted by plaintiff’s counsel and entered judgment upon them for Miller. The judgment awarded Miller $27,174.40 in back pay, pre-judgment interest of $7,061.70, attorney fees of $9,500.00 and costs. This appeal followed. II Mercy has raised a number of issues on this appeal, 3 but because we find reversible error in the district court’s determination of liability vel non, we discuss only that completely dispositive issue. As tried to the district court, following dismissal of the class claim, this action had been reduced to a classic Title YII individual claim of disparate treatment in an isolated employment decision. As in such cases generally, the dispositive issue as originally joined was the narrow motivational one of whether the defendant’s failure to hire plaintiff on an identified, single occasion was, in whole or part, on account of her race. 42 U.S.C. § 2000e-2(a). As the district court correctly saw, the course of proof at trial further narrowed the ultimate issue to whether racially discriminatory animus or the proferred nondiscriminatory reason of plaintiff’s lack of qualification for the position perceived to be the one sought was the actual reason for the failure to hire. United States Postal Service v. Aikens, — U.S. —, —, 103 S.Ct. 1478, 1482, 75 L.Ed.2d 403 (1983); Texas Department of Community Affairs v. Burdine, 450 U.S. 248, 256, 101 S.Ct. 1089, 1095, 67 L.Ed.2d 207 (1981). This ultimate motivational issue was one of fact. Pullman-Standard v. Swint, 456 U.S. 273, 285-90, 102 S.Ct. 1781, 1788-91, (1982). Upon that narrow factual issue, Miller as plaintiff bore the burden of persuasion — now merged with her continuing original burden to persuade the trier of fact that on the occasion in issue she had been the victim of intentional discrimination. Burdine, 450 U.S. at 256, 101 S.Ct. at 1095. Under controlling authority, this burden might be carried by proof by a preponderance of the evidence that the “discriminatory reason more likely motivated” Mercy, or that Mercy’s “proffered explanation [was] unworthy of credence,” being instead a pretext. Id. The district court plainly saw this as the critical, dispositive issue of fact in the case 4 and decided it in claimant’s favor in its factual finding on the ultimate liability issue: 11. The Defendant “articulated” a reason for not employing her (that they thought Plaintiff applied for only an LPN [licensed practical nurse] job); but the reason was pretextual. The Plaintiff was denied employment because of her race, and because of her “racial” problems at Presbyterian Hospital. She was unjustly treated as a black “trouble maker.” Ill This ultimate motivational finding is challenged on appeal. We review it, along with any subsidiary findings of fact upon which it is based, under the clearly errone *361 ous standard of Fed.R.Civ.P. 52(a). Swint, 456 U.S. at 290, 102 S.Ct. at 1791. For reasons that follow, we pause briefly to consider the special application of that standard to motivational issues in Title VII litigation. In the Supreme Court’s oft-cited elaboration of the “clearly erroneous” standard, we have been instructed that “[a] finding is ‘clearly erroneous’ when although there is .j evidence to support it, the reviewing court on the entire evidence is left with the definite and firm conviction that a mistake has been committed. United States v. United States Gypsum Co. 333 U.S. Ct. 525, 541, 92 L.Ed 746 (1948) Efforts at further refinement of this judgmental standard are not likely to give it much greater precision than does the Gypsum Co. formulation. Nevertheless because m this case we are left with a definite and firm conviction that a mistake has been committed” and because of the particular sensitivity of the standard’s application to ultimate motivational issues in Title VII litigation, see, e.g., Swint, 456 U.S. at 275-77, 102 S.Ct. at 1783-84, we elaborate briefly upon our understanding of the ways in which an appellate court may properly be “convinced” that a “mistake” in fact-finding has been made. We start with the proposition that such a conviction may not be based simply upon a perception derived from de novo review of the record that the “actual” facts are other than those found, see id. at 290-93, 102 S.Ct. at 1791-92; Zenith Radio Corp. v. Hazeltine Research, Inc., 395 U.S. 100, 123, 89 S.Ct. 1562, 1576, 23 L.Ed.2d 129 (1969). Closely related is the obverse proposition that the conviction of mistake need not rest upon any perception by the reviewing court that the “actual” facts are indeed different from those “found” (though it may obviously include such a subjective perrception). 5 Thus, the conviction of mistake ^ Profrly be based UP°° ? bha> Wltb°í rd to what the actual facts may be, the findings under review . / were induced by an erroneous view of the controlling legal standard, see, e.g., Uni t ed States v. Singer Manufacturing Co., 374 U.S. 174, 194 n. 9, 83 S.Ct. 1773, 1784 n. 9, 10 L.Ed.2d 823 (1963); MacMullen v. South Carolina Electric & Gas Co., 312 F.2d 662, 670 (4th Cir.1963); or are not supported by substantial evidence, Hodgson v. Fairmont Supply Co. 454 F.2d 490, 495 (4th Cir. 19 72 ); or were made without properly taking into account substantial evidence to the contrary or are against the clear weight of the evidence considered as a whole, Jones v. Pitt County Board of Education, 528 F.2d 414, 418 (4th Cir.1975); Sanders v. Leech, 158 F.2d 486, 487 (5th Cir.1946). In sum, these establish that “clearly erroneous” review is properly focused upon fact-finding processes rather than directly upon fact-finding results. The appellate function is to insure that the process shall have been principled^ the function is not authoritatively to find the “facts” first instance, 6 or to affirm or deny that the facts found by the trial court are the actual facts of the case. On this understanding, we are convinced that several mistakes in its fact-finding *362 process render the district court’s critical findings of fact clearly erroneous. Primarily, we are convinced that the court’s finding on the ultimate motivational issue as reflected in finding No. 11, is not supported by the requisite preponderance of evidence. 7 This is so whether in conceptual terms the court’s finding was that as between the two proffered reasons the discriminatory one was the “more likely,” or that the defendant’s proffered reason was “unworthy of credence,” i.e., “pretextual,” or — as is probable — both. See Burdine, 450 U.S. at 256, 101 S.Ct. at 1095. Our reasons are as follows. It is important at the outset of our review to identify the two opposing reasons for the challenged employment decision that were finally laid before the district court, hence the reason it accepted and the one it rejected in the end. The defendant’s proffered reason was identified by the court as being Marciniszyn’s understanding that Miller was interested only in an LPN position and her perception that for this position Miller was not qualified. For purposes of this appeal we can accept this as an accurate analysis of the nondiscriminatory reason advanced by the defendant. 8 More critical is the court’s identification of the specific discriminatory reason that it found established in conjunction with its rejection of the defendant’s proffered reason. The discriminatory reason was identified in the court’s critical finding No. 11: “Plaintiff was denied employment because of her race, and because of her ‘racial’ problems at Presbyterian Hospital. She was unjustly treated as a black ‘trouble maker.’ ” Though the “found” reason is stated con-junctively, first in general terms of “race” and then in specific terms of “racial troubles at Presbyterian Hospital,” the record is compelling that the reason actually found was the more specific one — that Miller was perceived by Marciniszyn, the decision-maker, to be a “black troublemaker.” Neither the race-alone nor the “race-plus” reason could properly be found on the record in this case. The record clearly reveals why the district court sought to rest decision on the more narrow ground identified in its concluding “black troublemaker” elaboration. Most critically, the evidence of record flatly negated any finding that race alone was the reason for the failure to hire Miller as a nurses aide. In statistical evidence intended to show a general climate of racial discrimination in Mercy’s employment practices, plaintiff succeeded in showing that blacks were freely employed by Mercy as nurses aides during Marciniszyn’s tenure, and particularly around the time in issue. In 1974, almost sixty percent of the nurses aides at Mercy were black. Other blacks were freely hired as NAs around the critical date. Perhaps ironically, but nevertheless importantly, this evidence — avowedly offered by Miller’s counsel to demonstrate that Mercy employed blacks mainly in lower-paying positions — effectively undercuts any claim by Miller that she was denied *363 employment in such a position on account of her race alone. Obviously aware of this development at the conclusion of the evidence, plaintiff’s counsel then deliberately staked out Miller’s narrow claim for the first time in these terms: “Judge, I’m now ready to tell you my theory of the case which I hadn’t told you before. My theory of the case is that Mercy Hospital refused to hire Ms. Miller because she was a black troublemaker.” The court’s ensuing findings of fact, culminating in ultimate Finding No. 11, make it plain that this was the specific factual theory of discriminatory motive accepted by the court. In summary, the court’s critical findings (express and implicit) of intentional discrimination ran as follows: Miller’s August 14, 1974, application for employment was for either an LPN or NA position; this was known both to Winchester and to Marcinisz-yn; Marciniszyn also knew when she rejected Miller’s application that Miller was a black who had experienced “racial troubles” in her prior employment; this led Marcin-iszyn to conclude that Miller was a “black troublemaker” and on that specific basis to deny her employment; this actual reason was then deliberately concealed from Miller by Winchester’s misleading response that the rejection was because of a negative reference of undisclosed nature from her former employer; the actual reason was further deliberately misrepresented by Mar-ciniszyn that it was because of her perception that the application was only for a LPN position for which Miller was not qualified; Mercy’s preferred reason, through Marciniszyn, was therefore “pretextual” in legal contemplation. The district court’s ultimate motivational finding must therefore be assessed on the basis of the “black troublemaker” theory. So assessed, it is clearly erroneous and cannot stand. To find that Mercy refused to hire Miller not simply because she was black but because she was perceived to be a “black troublemaker” required leaps of inference that could not be made under the legal constraints imposed by applicable proof burdens and the requirements of demonstrable rationality in the fact-finding process. 9 The specific motivation in issue is indisputably that of Marciniszyn, acting with Mercy’s authority upon Miller’s application of August 14, 1974. It is to Marciniszyn, therefore, that the “black troublemaker” perception and motivational bias must be ascribed in order to justify this motivational finding. The only evidence arguably supporting such an ascription of knowledge and personal motivation is but distantly circumstantial. Motivation in this or any context may of course be proven circumstantially, and is usually only provable in this way in Title VII litigation, but here the circumstantial evidence supporting the inference is simply too attenuated when considered in light of the evidence as a whole to allow the inference. 10 *364 There are two major gaps that in logic had to be leaped in order to make the ultimate inference. The first had to do with Marciniszyn’s knowledge in the first place that Miller was black — whether or not a “troublemaker.” As indicated in our recital of the factual and procedural background, Marciniszyn denied in sworn testimony that she knew Miller’s race at the time of her decision. Tending to support this is the fact that the application form did not specifically identify Miller by race nor give any manifest clues of race to anyone without special private knowledge that would allow the deduction from collateral indicia on the form. Further supporting Marciniszyn’s testimony is the testimony of Winchester, who did of course know Miller’s race, that she had no occasion to and did not convey her knowledge to Marciniszyn. The district court made no specific finding that, despite this testimony, Marciniszyn did know Miller’s race, but it is of course necessarily implicit in the ultimate “black troublemaker” finding. This implicit finding therefore required an inference that despite the testimony of these two Mercy employees, one or both was either deliberately lying under oath or had simply now forgotten that in fact Marciniszyn did know. In logic, Mar-ciniszyn could only have known from unrevealed private inquiry outside the application form, by deducing Miller’s race from other indicia on the form by reason of private knowledge making this possible, or from being told by Winchester or some other person. No direct evidence suggested any such source. More critical is the gap respecting any perception by Marciniszyn that Miller was a black “troublemaker.” The most relevant circumstantial evidence is Winchester’s knowledge, derived from Presbyterian, that on one occasion, being unhappy with that employer, Miller had “contacted a lawyer” to pursue a grievance. That this grievance involved allegations of racial discrimination was, by Winchester’s contemporaneous notes, not revealed to her. Winchester’s testimony on trial was that it was not. Whether Marciniszyn even knew of this one episode of Miller’s pursuing a grievance with a prior employer is not clear. For purposes of the appeal we can accept — as the district court must implicitly have found — that Marciniszyn did have access to the reference report prepared by Winchester. But if Marciniszyn knew more than the report revealed — if she knew that the unspecified grievance was racial — she could only have learned this from a source of whose existence again there is no direct evidence. Even if such knowledge be nevertheless inferred, there remained yet a further necessary inference: that this knowledge caused Marciniszyn to perceive Miller as a “troublemaker” whose race in conjunction with her troublemaking propensity made her undesirable as an employee. To make the required inferences despite the directly uncontradicted opposing evidence, the district court relied essentially upon credibility assessments in which it simply rejected the critical portions of Winchester’s and Marciniszyn’s testimony respecting their dealings with Miller’s application. This was based upon the court’s stated perception of the uncertainty of their testimonial recall of events, and the perceived internal ambiguity and logical implausibility of their accounts. In assessing the opposing versions of what was intended, what understood, and what communicated about the reasons for rejection of the August 1974 application, the court explicitly accepted Miller’s version, finding that her “testimony was direct, straightforward and believable.” In contrast, the court noted that both Winchester and Marciniszyn had “no recollection” at trial of the exact circumstances of *365 the application’s processing, and that Win-Chester had “little recall” of the follow-up “occasion” when, according to Miller, Win-Chester had reported to her that the application was rejected because of Presbyterian’s negative reference. To substantiate its rejection of Winchester’s versions of these events the court pointed to her “vacillation” in testifying about her understanding of the position for which the application showed Miller to be applying. To substantiate its rejection of Marciniszyn’s testimony concerning her decision to reject the application-that she understood it to be for an LPN position for which Miller was not qualified by license — the court pointed to evidence of three other hiring episodes involving Marciniszyn that, for the court, tended to give the lie to Marciniszyn’s proffered explanation. In one episode, two days after Miller’s rejection, Marciniszyn had hired a white nurses aide whose application had no entry in either the “Classification” or “Type of Work Preferred” spaces. For the court, this undercut Marciniszyn’s purported reliance upon the LPN entry in the “Position Preferred” space on Miller’s application. In one of the other episodes Marciniszyn had allegedly declined, a year after Miller’s rejection, to hire another black applicant whose former employer had indicated on a reference check: “first black hired in office, caused some tension”; while in the other, three years before Miller’s rejection, she had hired a white applicant with known performance and disciplinary problems, For the court, these two episodes in conjunction suggested a general bias on Mar-ciniszyn’s part against black as opposed to white “troublemakers” that supported its ultimate finding that this was the narrow basis upon which she rejected Miller’s application. With all respect, and after according the district judge’s better vantage point the deference we must, we are convinced that the critical findings of intentional discrimination on the “racial troublemaker” theory cannot be allowed to stand. More specifically, we are convinced that to find as the district court found on this ultimate issue of fact — an issue upon which plaintiff had the burden of persuasion — required a process of speculation or intuition rather than of legally justifiable inference from the evidence. We are therefore left with a definite and firm conviction that a mistake in the fact-finding process has been made which renders the finding clearly erroneous. Because so to conclude is serious business, not lightly to be done, we explore the record in some detail to explain the basis for our conviction. , We look first to the courts ^edibility assessments. We do so because it is essenon that stated basis that the court s ultimate finding is based. In the final anal-ys^s the inferential gaps are leaped by rejccting Marciniszyn s testimony as pretextual”: as either deliberate lie under oath or honest memory somehow twisted 180 degrees from actual fact. Giving “due regard ... to the opportunity of the trial court to judge of the credibility of the witnesses,” Fed.R.Civ.P. 52(a), we simply do not believe this credibility assessment can serve as a rational basis for the critical fact findings here. The special advantage had by the trial court as opposed to ours in review of credibility assessments is in relation to the opportunity to observe witnesses’ demeanor, But, of course, “[credibility involves more than demeanor and comprehends an overall evaluation of testimony in the light of its rationality or internal consistency and the manner in which it hangs together with other evidence.” 9 C. Wright & A. Miller, Federal Practice and Procedure: Civil § 2586, pp. 736-37 (1971). The district court suggested its reliance upon demeanor as a major component of its credibility assessment by contrasting the testimony of Marciniszyn and Winchester unfavorably with that of Miller in terms of their relative degrees of precision and assurance. Miller’s testimony is characterized as “direct, straightforward and believable.” That of Marciniszyn and Winchester on the other hand is characterized variously as “vacillating,” as indicating “no recall” or “no recollection” of the critical events in 1974, as internally inconsistent and as inconsistent with other evidence. *366 We have carefully reviewed the record— including significant portions not included in the joint appendix — and are persuaded that a more accurate characterization of the opposing testimony is that it reveals remarkably similar failures of recall and ambiguity — with the greater degree of both on Miller’s side. To characterize Miller’s testimony as “direct, straightforward and believable” (a characterization, incidentally, that was drafted by plaintiff’s counsel for inclusion in the court’s findings) becomes highly questionable when critical portions of her testimony at trial and by deposition 11 —portions not alluded to in the court’s findings — are considered. An example is Miller’s hopelessly confused and contradictory testimony about an earlier application for employment at Mercy than the one in issue. At length on her deposition, but only in a passing reference on trial, Miller testified that she had actually applied unsuccessfully at Mercy a full year before the August 1974 application at issue in this case. If her deposition testimony is accepted, it was on this earlier occasion that Winchester first suggested that Presbyterian had given a negative reference that prevented Miller’s employment by Mercy and that should be “cleared up.” The improbability that, as Miller later testified, this experience with Winchester was repeated in all its essentials less than a year later, with no apparent recognition by Winchester of an earlier encounter, is manifest. That no more is involved than a simple confusion of dates is belied by Miller’s firm insistence on deposition that there were two distinct encounters of this kind, the second being the object of this action. Neither the earlier encounter nor the ambiguities in Miller’s testimony respecting it are noted in the district court’s findings. Another example appears in Miller’s testimony respecting her final encounter with Winchester, an occasion that, on the record, indisputably occurred some two years following the August 1974 rejection of her application, on November 26,1976. On this occasion, Miller testified, she exhibited to Winchester, who relayed to Marciniszyn, an EEOC determination letter finding, against Presbyterian, that that institution had indeed made a false official report concerning the circumstances of Miller’s resignation from that institution and had done so for discriminatory reasons. 12 The purpose of this testimony was presumably to confirm Miller’s contention that Mercy’s responsible employees knew all along that her troubles at Presbyterian had been racially inspired. Whatever the intended purpose of this testimony, it was internally suspect: the EEOC determination letter purportedly shown Winchester and Marciniszyn on November 26, 1976, bore a date indicating that its date of issue was December 29, 1976. 13 There are other examples that tend to draw in question the essential accuracy of the court’s characterization of Miller’s testimony as “direct” and “straightforward” by way of contrast to that of Winchester and Marciniszyn. We mention only one. It deals with a rather critical issue. Miller testified that she told Winchester (in both her 1973 and 1974 encounters) of the nature of her racial difficulties at Presbyterian that led to her 1973 resignation. But on her 1974 application to Mercy, an application that, by Miller’s account, she completed during her interview with Winchester, Miller entered as the reason for her leaving *367 Presbyterian that she was having a baby, a fact verified in her other testimony. We explore these instances in this much detail not to make, de novo, a contrary credibility assessment that rejects all of Miller’s testimony; but simply to indicate our inability to accept the district court’s stated perception that the inherent credibility of Miller’s testimony provided a sufficient basis for accepting all of its critical content. We think the only fair characterization of Miller’s testimony is that it was render© at least as suspect as that of Winchester and Marciniszyn by virtue of demonstrably failed recall, internal inconsistency, and contextual ambiguity. Indeed, to the extent it purported to be based upon confident, detailed recall of critical events occurring six to eight years earlier, its demonstrable internal inconsistencies and ambiguities render it more suspect than Winchester’s and Marciniszyn’s flat concessions of inability to recall those events with perfect fidelity. There is of course always the possibility that protestations of inability to recall events are in fact merely the “stonewalling” technique at work. Trial judges are assuredly better situated to detect this special technique of perjury than are appellate courts sitting in review. But where, as here, the failures of recall asserted on one side are matched by unconceded but demonstrated failures of equal or greater magnitude on the other and where, as here, a long lapse of time and the banality of the events as they occurred may well explain both sets of recall failures, it would seem a most questionable fact-finding process to reconcile them by completely forgiving one set of failures and ascribing the other to perjury by stonewalling. To the extent that this is what underlies the district court’s credibility assessments here, we are convinced that a mistake in the fact-finding process has been made. Nor do we see how the ambiguities and internal inconsistencies in the testimony of Winchester and Marciniszyn specially relied upon in the district court’s findings have the negative weight ascribed them. Reliance is placed upon Winchester’s “vacillation” (we repeat, a characterization that originated with plaintiff’s counsel) in describing her understanding of the position that Miller was applying for. We have carefully reviewed Winchester’s testimony on this point and are persuaded that this is not a fair characterization. A fairer characterization would be of understandably uncertain responses to a protracted course of cross-examination by Miller’s counsel largely devoted to semantic quibbling about a point of only marginal relevance — Winchester’s perception, not Marciniszyn’s, of what position Miller had applied for in the August 1974 application form. Fairly assessed, Winchester’s testimony never evaded concession of the one relevant fact finally elicited: that she assumed that Marcin-iszyn would interpret the application as showing Miller qualified for a nurses aide position and, presumably, as indicating a desire to be considered for it. The impression is inescapable that it simply took Miller’s counsel more questioning than it should have to elicit the point and that this, rather than Winchester’s vacillation, explains any difficulty experienced. It is at this point that the greatest contextual ambiguity in the testimony of Mercy’s witnesses occurs. Marciniszyn’s stated perception that Miller’s application should only be considered as one for LPN is at odds — though not in direct conflict — with Winchester’s stated assumption that Mar-ciniszyn would or might consider it for a NA position. No effort was made by these witnesses to reconcile the difference between the one’s assumption and the other’s perception. The evidence was left to speak as it lay — to be accepted as a simple example of bureaucratic slippage or as evidence of Marciniszyn’s deliberate falsification or completely skewed memory of her actual state of mind. This shred of ambiguity is in fact the strongest bit of circumstantial evidence available to prove that Marcinisz-yn’s actual motive was different from her professed motive, and, incidentally, to establish her as a racist and perjurer or as a racist with wholly failed memory — the necessary implication of the ultimate finding *368 of “pretext.” We are convinced that it could not rationally be given that great probative force. Finally, we think the other-episode evidence relied upon to establish Marciniszyn’s narrow, special bias — against “black troublemakers” — is simply too meager to tip the scales on that motivational issue. In the first place, three hiring episodes, over a period of four years, out of the great number in which Marciniszyn was demonstrably involved, is a treacherously small sample from which to deduce a general mind-set of racial prejudice to be used in turn as inferential proof of specific discriminatory intent on the occasion in issue. Cf. Furnco Construction Corp. v. Waters, 438 U.S. 567, 580, 98 S.Ct. 2943, 2951, 57 L.Ed.2d 957 (1978) (evidence relevant to establish general pattern of discriminatory conduct). Particularly is this true when, as here, the uncon-tradicted evidence establishes a general pattern of hiring by Marciniszyn that belies any general bias against hiring blacks for the position in question. Even more critically, the inferential force of the three episodes offered to establish the mind-set here is, upon inspection, so attenuated as to be of questionable relevance, not to say probative force. Marcin-iszyn’s hiring of an applicant whose application form contained no entries in the “Classification” or “Type of Work Preferred” spaces was easily and spontaneously explained during her cross-examination on a basis perfectly consistent with her stated reason for rejecting Miller’s. The two purportedly contrasting episodes involving one black “troublemaker” and one white are of even more questionable force. There is, in the first place, no evidence, direct or circumstantial, that Marciniszyn or anyone at Mercy had ever seen the employment reference on the other supposed “black troublemaker.” 14 If it were nevertheless accepted that Marciniszyn had seen it — perhaps by inferring that in ordinary course she would have — the reference was utterly ambiguous on the point. The report was merely that the hiring of the black had “caused some tension,” with no inkling that this was because of troublemaking propensities of the black rather than racial prejudice of other employees. The episode involving the hiring of one white with known prior employment problems is of even more dubious relevance, particularly in view of the district court’s total disregard of Mercy’s countering evidence of the hiring within the critical period of two blacks with known prior employment problems of comparable quality. We conclude this generally unwanted and always somewhat treacherous review of trial court fact-finding processes with a particularly distasteful, but necessary, reference to an aspect of the processes employed here that bolsters our conviction of mistake. Mercy, as appellant, has legitimately complained on appeal of the practice followed by the district court in formulating its findings of fact. We have earlier outlined the procedure: the court announced its general decision finding liability; requested plaintiff’s counsel to prepare proposed findings, conclusions and judgment; then adopted those proposed with minor revisions and two additions and with no formal opportunity given opposing counsel either to submit proposed alternatives or specifically to object to those proposed before their adoption. In a series of decisions, most recently in Anderson v. City of Bessemer City, 717 F.2d 149 (4th Cir.1983) and Lilly v. Harris-Teeter Supermarket, 720 F.2d 326 (4th Cir.1983), we have expressed varying degrees of disaffection with and disapproval of the general practice. See also EEOC v. Federal Reserve Bank of Richmond, 698 F.2d 633, 639-41 (4th Cir.1983) (citing cases). Whether, in light of our most recent pronouncement in Lilly, 720 F.2d at 331, we would be justified on this basis alone in rejecting the findings *369 here as clearly erroneous, we need not decide. As indicated, we think there are more fundamental reasons for doing so here. Nevertheless, we are bound to note the strong possibility that the practice as followed here may have contributed significantly to the mistakes in the fact-finding process we have identified. The special vice of this practice is not so much that it may actually induce a wholly blind, unreviewed adoption of proposed findings by trial judges. We doubt that this frequently occurs — if ever it does. Certainly it did not occur here, as is evidenced by the district court’s obvious care in editorial revisions to the proposed findings, and in adding a critical finding on the ultimate motivational issue. The more realistic danger — which we think fully exemplified here — is that the practice tends to deflect the court’s attention from, or actually to obscure, the more difficult factual issues and credibility problems presented by the evidence. The natural tendency of counsel given an opportunity free of adversary constraints to shore up weak points, to gloss over evidence or credibility problems at odds with necessary findings, and to argue inferences in the guise of “findings,” is obvious. In effect, the practice gives to one side but not the other a final, second opportunity to argue the case to the fact-finder, free of rebuttal, and essentially ex parte. Though tentative decision may by then have been reached, and may at the time reflect a firm judicial conviction, it is still at this point tentative — the decisional process is still in progress. Though it still lies with disfavored counsel to note formal objections to findings after their adoption, Fed.R.Civ.P. 52(b), this, as every litigator knows, is but a poor substitute. At this point the judge’s decision, not the adversary’s proposed findings and conclusions, must be challenged, and any fair opportunity to influence the decisional process in the trial court has in practical terms been lost. 15 Here we are satisfied that this potential vice of the practice almost certainly skewed the fact-finding process. As our discussion has indicated, the findings as drafted by plaintiff’s counsel and adopted by the district court were highly argumentative in form and notably selective in overemphasizing the probative force of some evidence while glossing over or wholly disregarding unfavorable evidence of obviously serious import. Of course we cannot know but what the district court would independently have come to the same findings by the same essential processes charted in the proposed findings, accepting exactly the evidence emphasized in counsel’s draft proposals and rejecting all that disregarded in those proposals. We can only say with assurance that had defendant’s counsel been given an equal opportunity before final decision was reached to force consideration of evidence opposing the findings adopted or to demonstrate the rationality of contrary findings, we are satisfied that the district court would have been forced to critical reexamination of portions of ‘the findings actually made. We close as we began with the observation that the purpose of our review of the district court’s dispositive findings of fact in this or any case is not to affirm or to deny that the “actual” facts of this controversy are as that court “found” them to be. Even as we adjudge the findings here to be clearly erroneous, we must be prepared to concede — and do — that by some process this experienced, sensitive, justly respected trial judge may indeed have “found” the “true” facts. We only have the power, and the responsibility, to say that on the evidence of record he could have done so only on the basis of an intuition or insight whose probable accuracy lies beyond the capacity of an appellate court to review on any principled basis. Assessed according to legal standards of rationality in drawing inferences of motivation from raw historical facts in evidence and under controlling burdens of proof, the ultimate finding of discriminato *370 ry motivation in this case must be rejected as clearly erroneous. REVERSED. Footnotes [1] It is not disputed that though Mercy’s policy at the time was to allow its own employees to function as de facto LPNs pending the results of licensing examinations, it did not permit the new hiring of unlicensed persons to' perform those functions. It is therefore not disputed that Miller was not “qualified” for an LPN position under Mercy’s general policy. Her claim accordingly was treated and decided solely 0n the basis of Mercy’s failure to hire her as a nA. [2] Though Mercy raises issues on this appeal respecting the administrative stages of Miller’s proceeding, we decide the case on other grounds. Our extended recitation of the course of the protracted administrative process is included only because of its bearing upon critical credibility assessments made by the district court, particularly as its confusion and duration may suggest the basis for confusion and failures of recall by witnesses on both sides testifying much later to the critical events in issue. [3] 1) Failure to file a timely EEOC charge; 2) clearly erroneous factual finding of discrimination; 3) error in awarding backpay; 4) excessive award of attorney fees. [4] It was obviously to sharpen and to focus upon this as the ultimately dispositive issue that the court added to the findings proposed by plaintiffs counsel the specific one quoted in text as Finding No. 11. Whether, as the district court found, plaintiff had initially made out a prima facie case is not before us, given the course of trial. See Aikens, — U.S. at —, 103 S.Ct. at 1480. We think it an exceedingly close question. [5] Though in technical contemplation clearly erroneous review may rightly proceed with a completely neutral attitude toward what the “actual” facts may be, it is of course inevitable that a reviewing court’s conviction of mistake may sometimes include a conviction that the actual facts are other than those “found.” See, e.g., Sanders v. Leech, 158 F.2d 486, 487 (5th Cir.1946) (“testimony considered as a whole convinces that the finding is so against the great preponderance of the credible testimony that it does not reflect or represent the truth and the right of the case”) (emphasis added). Though occasionally they must exist, such convictions about the “truth” or “right” of a case are merely incidental to, not necessary predicates for, a reviewing court’s conclusion that a finding is clearly erroneous. It is enough that the finding is “against the great preponderance of the evidence” in the record presented for review. A reviewing court’s ability to discern the “truth” and “right” of a case presupposes that the record is also “true” and “right” (and complete) and this of course is beyond the court>s a5ility and need. [6] Except perhaps m reviewing constitutional fact-finding, m taking judicial notice and very occasionally and cautiously, when facts not found are manifest on the record, [7] This puts somewhat obversely the more common way of stating this particular basis for holding that a finding is clearly erroneous: that the finding is against the clear weight of the evidence. See, e.g., Jackson v. Hartford Accident & Indemnity Co., 422 F.2d 1272, 1275-78 (8th Cir.1970) (Lay, J., concurring). Where, as here, the finding was in favor of a party having the burden of persuasion upon the issue, we believe the statement in text more accurately identifies — and in appropriately minimal terms — the precise mistake in the fact-finding process that is involved. To find mistake in this respect contemplates that there may have been substantial evidence — if credible — to support the finding. Whether there was such evidence in this case is a close question that we need not address in view of our holding stated in text. See id. [8] The district court expressly found that Miller was “qualified and experienced” for both the LPN and NA positions, and had applied for both. There is no question, however, that Miller was not qualified, under Mercy’s general policy, to be hired as a LPN. This is borne out by the court’s determination that Mercy’s violation was its refusal to hire Miller for the lower-paying NA position. Marciniszyn’s proffered reason for failing to hire was, therefore, a valid one if, as is disputed, she perceived the application to be one only for a LPN position. [9] We accept, for purposes of this appeal, the theoretical possibility of proving intentional racial discrimination on this narrow “race-plus” basis under circumstances conclusively shown to be free of any general racial bias in making comparable employment decisions. But we confess grave misgivings about the ability of courts fairly and rationally to assess the existence of such an amorphous special type of racial bias imbedded in a generally nondiscriminatory pattern of employment decisions. “Troublemaking” in the employment context is of course a many-faceted, perfectly justifiable basis for making negative employment decisions — even bad ones. The attempt to discern troublemaking propensities made disqualifying only because of superimposed racial considerations may well tax judicial fact-finding processes beyond their capacity for fair and rational adjudication. [10] The normal principles that control .the weighing of evidence by fact-finders apply in Title VII litigation. That Title VII is quintessentially remedial legislation, and that proving intentional race discrimination may be increasingly difficult as its more overt forms have been driven underground by the first wave of major Title VII litigation victories, does not justify ad hoc judicial relaxations of the principles of rational proof that apply in civil litigation generally. The “liberal interpretation” that is appropriate because legislation is “remedial” in purpose is properly reflected only in substantive interpretation and in general rules easing normal proof burdens borne by litigants *364 for whose benefit the legislation is enacted. The general proof scheme adopted by the Supreme Court in McDonnell Douglas Corp. v. Green, 411 U.S. 792, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973), featuring in its first stage a rebut-table presumption easing the normal production burden borne by Title VII disparate treatment claimants, is of course an example of the latter. That general rule represents the present limit of authoritatively approved relaxation of the normal burdens of production and persuasion borne by claimants in this type case. [11] Miller’s full deposition was introduced in evidence by Mercy. [12] Miller’s charge against Presbyterian was resolved by a settlement in which she was reinstated in 1980 and received backpay from that institution. [13] There may be an explanation for this apparent testimonial error — e.g., an erroneous date or an earlier formal or informal letter or notice — but if there is, it does not appear of record. By written brief, plaintiff’s counsel specifically identified the letter dated December 29, 1976, as the one shown by Miller to Winchester on November 26, 1976. If there is an explanation, its absence on the record suggests, in any event, a failure by the fact-finder to note or to elicit an explanation of an apparent testimonial error of considerable import. Perhaps significantly, the finding on this matter was not one of those specifically directed by the district court; it originated with and was drafted in its final form by plaintiffs counsel. [14] The evidence of this particular reference is of the most dubious relevance. The record does not indicate anything about the processing by Mercy of the application in question; it does not negate the possibility that employment was offered, nor indicate the ostensible reason for denying it if that was the outcome. [15] Though post-finding objections or motions to amend may be made, they are not required in order to challenge findings on appeal. Fed. R.Civ.P. 52(b). This presumably reflects a concession of their usual futility.