This appeal arises from the claims of a fireman, Winston Nash, as an individual, that the City of Jacksonville (City or Jacksonville) discriminated against him by failing to promote him to Fire Lieutenant (lieutenant) in the Combat Division because he did not pass a 1981 promotion examination. The United States District Court for the Middle District of Florida held that Nash did not prove discrimination under Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e et seq., under the “disparate impact” theory, even though it can be unintentional on the employer’s part. (Nash also alleged claims under 42 U.S.C. §§ 1981 and 1983, but the court held that any recovery under these laws would require dis*1395criminatory intent which had not been shown. Nash has not appealed the decision on these issues.)
This court has jurisdiction under 28 U.S.C. § 1291. We conclude that the district court’s ultimate finding, that discrimination was not established, is clearly erroneous being based on a mistake of law, and we remand for further consideration.
I
District Court Proceedings
As established at trial, the facts are as follows: Winston Nash, a black fire engineer employed by the City of Jacksonville since November 1972, has applied for promotion to the position of lieutenant three times. A written examination is used by the Fire Department for promotion to lieutenant; Nash scored an 84 on the examination in 1976, a 69 in 1978, and a 69.072 in 1981. The Civil Service Board has set an automatic cut-off score of 70 for all employment examinations given by Jacksonville. Thus Nash was placed on the promotion eligibility list of the 1976 examination only. Those who had attained the eligibility list were then ranked based on a composite score which is the sum of the test score and seniority points. Nash was not promoted after the 1976 test since he did not place high enough on the list to be selected prior to the list’s supersession.
Evidence was adduced at trial regarding discrimination at the Fire Department and the manner in which the 1981 test was prepared. It was established that there have been no blacks in the positions of lieutenant or above in the Combat Division. The trial judge found that only three of the 20 blacks who had taken the test had passed, while 57 of the 100 participating whites had passed. None of the three blacks placed on the list were actually promoted to the 19 vacancies which were ultimately filled; the blacks were not high enough on the list to be placed. Under the four-fifths rule, which states that “[a] selection rate for any race, sex, or ethnic group which is less than four-fifths (4/s) (or eighty percent) of the rate for the group with the highest rate will generally be regarded by Federal enforcement agencies as evidence of adverse impact,” 29 C.F.B. § 1607.4(D) (EEOC guidelines), as well as statistical evidence produced by Nash’s expert witness, the court found that a case of prima facie discriminatory impact had been stated. The court then considered evidence that there was a legitimate justification for the discriminatory effect. Turning to whether the test was content-valid, and thus job-related, the court found that Jacksonville’s preparation of the test was acceptable. The test was prepared under the supervision of a personnel examiner who had been employed by Jacksonville for approximately 7 years and who was familiar with the EEOC guidelines. The examiner, Ms. Horne, testified that the test measured knowledge based upon work behaviors. Ms. Horne explained that a committee was formed to prepare the examination, which was composed of two chiefs and one captain. Ideally a lieutenant would have been included in the committee, but none were available; however, the committee members had all once been lieutenants. The court found that each of those in the committee did possess knowledge regarding the lieutenant position.
The committee developed questions for the examination by listing work behaviors and knowledge they deemed necessary for the position; the behaviors were then compared to the job specification for the lieutenant position to determine if the specification was up-to-date. The behaviors were analyzed to determine which knowledges were necessary for a particular behavior, and questions were developed to test these areas of knowledge. The committee composed the questions using the fire manuals and the study materials. Although the committee noted that supervisory ability was necessary for the position, no questions in this area were prepared since the committee felt that supervision could not be tested by a written examination. As questions were prepared, each committee member rated each potential question as to the importance of the tested knowledge to *1396the job. After the final questions were selected based on their job-relatedness, the examiner reworded the questions so they would be simply stated and grammatically correct. Evidence was also produced, through Nash’s testimony, that the questions were aimed at job-relatedness but, at least in Nash’s opinion, missed the point because they were poorly constructed.
The court determined that the testimony of Dr. Shapiro, Nash’s expert, did not bear on the issue of job-relatedness. That expert correlated the pass-fail rate of whites versus blacks, and analyzed the pattern of participants’ answers as a whole. The expert concluded that the test was a poor one as some of the questions showed a disparate impact. The expert also found it particularly revealing that some questions were correctly answered by almost all, and some questions were predominantly answered correctly by those who scored poorly on the test as a whole. Thus, the expert concluded the test could not be job-related because it did not actually test anything. The court considered the expert’s testimony and found it probative evidence of the disproportionate impact of the test, but not relevant to job-relatedness because the expert had not seen the examination and had never correlated the test with successful job performance. The questions asked, or samples thereof, were not in the record. While Jacksonville did not produce direct evidence that the test scores correlated with successful job performance, several witnesses stated that it was their belief, based on their experience supervising and their experience composing the test, that there was a correlation. The method of test preparation was also indirect evidence of job-relatedness. (The City does have a probationary period for those promoted; during this time the Fire Department checks to see that those promoted are performing well. The Fire Department appears satisfied with the performance of those promoted. It has not, however, determined directly whether those not promoted could have performed.)
While not specifically discussed by the trial judge in his opinion, the City also produced evidence to rebut the expert’s conclusions. A former personnel examiner for the City, whose duties had included reviewing the computer-produced statistical data from the 1981 test, testified that the statistics produced might not accurately reflect nonjob-relatedness because they were formulated based on a sample size of only twenty, the twenty blacks who participated in the test. The employee testified that the statistics might be useful to flag questions with possible problems, but consideration of the questions was necessary to determine if they were nonjob-related. Mr. Rodeheaver, the employee, testified that he reviewed Nash’s answers and found the answers indicated that Nash’s mistakes were generally in specific areas of knowledge. (For example, Nash did well on questions pertaining to water distribution and radio signals, but poorly on questions pertaining to first aid.) Mr. Rodeheaver also testified that many of the questions Nash missed were ones which did not show any adverse impact, and that Nash, like the majority of the participants, often chose a commonly selected wrong answer. Mr. Rodeheaver also testified that there was a quirk in the statistical methodology of Nash’s expert.
The court concluded that the City had shown that the test was content valid and therefore job-related and then found that Nash had not shown there were less discriminatory ways to meet the City’s business necessity. The court considered the alternatives that Nash had suggested during his case-in-chief and found that they would afford the City with a greater opportunity to discriminate than present with the test. The trial judge concluded that ultimately Nash had not proven by a preponderance of the evidence that the City had discriminated against him.
II
Discussion
The finding on discrimination is one of ultimate fact, and this court will reverse the finding only if it is clearly *1397erroneous, or if reversible error occurred in applying the law. In this case, consideration of the court’s subsidiary findings is made somewhat more difficult as the trial judge applied some ease law developed for disparate treatment cases in analyzing his disparate impact problems. Thus the legal analyses of these theories are intermingled. Still, when a trial has taken place and the parties have had the full opportunity to provide their evidence and arguments, this court will not reverse for harmless errors and will review the ultimate question of discrimination vel non. United States Postal Service Board of Governors v. Aikens, 460 U.S. 711, 103 S.Ct. 1478, 75 L.Ed.2d 403 (1983). All in all, we think that even so, a remand is necessary so the judge may render fact findings so stated that they are not infected by incorrect assumptions as to law.
Appellant states that he has a claim both under the “disparate impact” and the “disparate treatment” standards. The former test, under Title VII, does not require a specific intent to discriminate and therefore is easier to apply and more appropriate when, as here, the employer has a facially neutral method of selection which, as applied, has a discriminatory “impact.” Connecticut v. Teal, 457 U.S. 440, 446, 102 S.Ct. 2525, 2530, 73 L.Ed.2d 130 (1982). The latter test, when under Title VII, may be more appropriate when specific intent to discriminate is shown either directly or by inference. In reality, as we view the case, the only showing of intent is the inference that people intend the inevitable consequences of their acts, i.e., that in using a method of selection that was invariably fatal to black candidacies, appellee must have intended no black to succeed. In the circumstances of this case, such an inference could be drawn only if the “disparate impact” was too obvious to be overlooked, and in that event, inquiry into intent would be superfluous. Appellant therefore rightly states that “impact” is the principal issue, and we regard it as being for practical purposes the only one. The court, in any event, made a specific finding as to “intent” only under the §§ 1981 and 1983 claims which are not before us.
Appellant attacks the holding below on the ground that the court stated (R. 170) that in an “impact” case the employer need only “articulate some legitimate, non-discriminatory reason for its actions,” citing Texas Department of Community Affairs v. Burdine, 450 U.S. 248, 101 S.Ct. 1089, 67 L.Ed.2d 207 (1981); Albemarle Paper Co. v. Moody, 422 U.S. 405, 95 S.Ct. 2362, 45 L.Ed.2d 280 (1975). But this is not the rule of the above cases, or any others, in an “impact” situation involving employment tests under Title VII. It is the rule in intent cases only. The Court in Burdine, 450 U.S. at 252 n. 5, 101 S.Ct. at 1093 at n. 5, expressly distinguishes the “impact” situation. In an “impact” case the employer must do more than “articulate.” He bears the burden of proving or “persuasion,” that the examination was a “business necessity.” Griggs v. Duke Power Co., 401 U.S. 424, 91 S.Ct. 849, 28 L.Ed.2d 158 (1971); Walker v. Jefferson County Home, 726 F.2d 1554, 1558 (11th Cir.1984); Jackson v. Seaboard Coast Line Railroad Co., 678 F.2d 992 (11th Cir.1982).
Where, as here, there exist glaring weaknesses in the cases of both sides, the difference between a burden of “articulation” and of “persuasion” may be a momentous one, and it is not just a question of who goes forward, which may cease to be important when all the testimony is in. The observations of the trial court repeatedly in its opinion indicate that the defendant need only say something: “if he simply explains what he has done” (R. 179). It is also said in the same breath that defendant “successfully refuted plaintiff’s prima facie case by showing the test was valid,” but in context this may only mean that an “articulation” sufficed to make a successful refutation. The court says (R. 179), plaintiff at all times retains the burden of persuasion. It is true that defendant, in its testimony, went much beyond mere “articulation,” and the court recited this testimony with every indication of believing it, yet the court at 180-181 re*1398peatedly states that plaintiff has failed to persuade the court “by a preponderance of the evidence.” The idea seems to be that mere “articulation” shifts the burden back to plaintiff to persuade that the test is discriminatory, but this is contrary to the authorities in this circuit as to “impact” employment test cases that we have cited.
Had the court made numbered separate findings of fact and conclusions of law, we might not be in such perplexity. The discursive mode it adopted, while permissible under the rules, leaves the reader frequently in the dark as to the legal assumptions that may underlie any particular fact assertion. We have considered whether we could hold that the “articulation” assertions were harmless error because they could be severed from the most relevant fact findings. We conclude that this mode of rescuing the decision below is not available.
We further express wonder that any court should attempt to determine whether a written examination was “job-related” and “content-valid” without having before it the questions asked of the applicants. The court, though accepting as true the testimony of Dr. Shapiro, plaintiffs expert, as to discriminatory “impact,” rejected it as to “content-validity,” in part because Shapiro too had not seen the examination. It would indeed seem that it is equally as hopeless for an expert to try to form an opinion in such a matter, without this necessary aid, as it is for the judge to adjudicate. Had the burden of persuasion as to “content-validity” been placed where it should have been, on defendant, we think defendant might have felt it incumbent to produce the examination questions. They were successfully withheld from discovery on the ground of secrecy, in rulings not before us for review, but we hope that upon our remand, the questions will somehow find their way before the court.
Appellant finally says that the finding that Nash has failed to sustain his burden under Title VII is clearly erroneous where the district court was already predisposed against “these kinds of cases.” The repeated use of quotation marks suggests the court actually used these words, but it did not. We read this argument to allege bias against Title VII plaintiffs on the part of the district judge, though this complaint is shrouded in a decent obscurity.
Counsel for appellants generally do better to confine themselves to actual error in the record rather than broad general charges of bias. The suggestion here is, in effect, we should supersede the “clearly erroneous” standard and, in effect, determine that all findings were “clearly erroneous” and retry the case de novo on the record. This kind of suggestion is not very helpful, because it lacks support in the federal rules. It reads like a disgruntled loser cursing the court, as all losers are entitled to do, but on the courthouse steps rather than in appellate briefs. If counsel really and truly believes the trial court is biased, there are measures counsel can and should take, but they were not taken here.
The incident which of record establishes, in counsel’s eyes, that the court is biased, not only against Title VII plaintiffs, but against her personally, occurred at a pretrial conference shortly before the actual trial started. She was seeking a continuance for further discovery. The judge denied the continuance and in doing so, was sharply critical of her conduct of this case and other cases. He said she made a practice of filing suit whenever a black person was “discharged or doesn’t get a job” without adequate or any investigation of the merits, hoping to establish the facts later and, presumably, dismiss if the facts were insufficient. He thought she should “have a good case before you file the case.” In this particular suit she had had anyway five months for discovery after filing and not used them, he said.
This view of the ethical duties of counsel filing a lawsuit is not before us for review and we do not pass judgment upon it. It is, however, an understandable view to be entertained by anyone dismayed at the flood of litigation in the federal courts. It would not reflect the existence of bias or prejudice here unless the judge applied it only to some particular category of plain*1399tiffs or attorneys and there is nothing to convince us that he did. He could perhaps have reflected more on the likelihood that what he said would be misunderstood by the target of his disapprobation, and by spectators. He, just as counsel, had his established and proper ways to take action in case he perceived unethical conduct before him, and, he resorted to an open court tongue-lashing, a mode which, justified or not justified, still is adopted at some peril.
Accordingly, we decline the invitation to hold all the fact findings “clearly erroneous” and find in it no basis for disturbing the judgment below, which, however, cannot be sustained for other reasons.
Conclusion
The judgment below is vacated and the cause is remanded for further proceedings in accordance with this opinion.
VACATED AND REMANDED.