POSNER, Circuit Judge.

Gary Allord is embarked on a quest, thus far unsuccessful, for social security disability benefits. He is a decorated Vietnam combat veteran conceded to be suffering from severe post-traumatic stress disorder, but because he lost social security coverage in 1992 he cannot obtain benefits unless he proves that his PTSD was disabling then, as it is now.

He was honorably discharged from the Marine Corps in 1987 and had difficulty adjusting to civilian life; he was also getting divorced. A woman who had known him since 1982, Melissa Chappell-White, observed his condition and suggested that *820 he consult a psychologist whom she knew, Dr. Foster Hutchinson, and he did so; this was in 1987, shortly after his discharge. Dr. Hutchinson, in his notes of the consultation, described Allord as pleasant, controlled, and articulate, though sad, and diagnosed him as suffering from an “adjustment disorder with depressed mood” and “compulsive personality disorder” and recommended that he undergo weekly therapy; Allord declined. But in 1993-four months after his social security coverage lapsed — he consulted a psychologist at a veterans’ medical center, Aphrodite Mat-sakis. A specialist in PTSD, she diagnosed Allord’s condition as severe and chronic PTSD. Again he declined treatment, this time until 1996. But between then and 1999 he received 200 hours of therapy from Dr. Matsakis.

In 1997, the Department of Veterans Affairs determined that Allord was 100 percent disabled by PTSD. Determinations of disability by other agencies do not bind the Social Security Administration, however, 20 C.F.R. § 416.904, and anyway the Department of Veterans Affairs requires less proof of disability than the Social Security Administration does. Compare Ortiz v. Principi, 274 F.3d 1361, 1364 (Fed. Cir.2001), with Jones ex rel. Jones v. Chafer, 101 F.3d 509, 512 (7th Cir.1996). We have said that SSA should give the VA’s determination of disability “some weight.” Davel v. Sullivan, 902 F.2d 559, 560-61 n. 1 (7th Cir.1990). The Ninth Circuit says “great weight,” McCartey v. Massanan, 298 F.3d 1072, 1075-76 (9th Cir.2002), but this disregards the substantial difference between the criteria used in the two programs. The administrative law judge gave the VA’s determination that Allord was disabled no weight because the determination had been made almost four years after Allord had lost coverage. The judge should have given the VA’s determination some weight, in light of the evidence we discuss below that Allord’s PTSD was of long standing.

At Allord’s hearing before the administrative law judge, Dr. Matsakis opined that Allord had begun having symptoms of PTSD in the early 1970s, and that by 1987 (six years before his coverage expired) he was totally disabled from gainful employment. She described his condition as follows:

[I]f you run down the criteria for post traumatic stress, they are all there. The hyper alertness, the constant vigilance. The fact that he was startled and almost ready to assault an old lady who he cherished.... He’s triggered by everything. The fact that he’s dissociating, memory problems.... The fears he has about being with people. The anger, the distrust, the state of paranoia, the withdrawal. If he’s successful at anything, at the teeniest hint, which he doesn’t even dream of that, even if he was he would be frightened because of triggers, which we don’t have time to go into which would take probably a year to explain. But basically anything that anyone does will set him off. Being in a closed room will set him off, being in open spaces. Noise sets him off. Silence sets him off. Kindness sets him off. Hostility sets him off and sends him off into either no man’s land where he can’t think or you see that angry look in his eye and he’s doing everything to control it. If the scratches and bruises I’ve seen on him a couple times are evidence that he sometimes can’t control himself.... The only work he does is staying alive, which is amazing that he’s even here. No he can’t do anything.

Another psychologist specializing in PTSD, also employed by the Department of Veterans Affairs, Jonathan Shay, seconded Matsakis’s testimony on the basis of an *821 evaluation of Allord that Dr. Shay conducted in 1997.

Chappell-White testified that she had observed Allord’s mental deterioration during the 1980s; that he could not communicate, could not carry out instructions (he did some handyman work for her), and failed to show up for work on a regular basis. Her testimony was important because neither Dr. Matsakis nor Dr. Shay had observed Allord in the critical period, that is, before his coverage lapsed. Not that Chappell-White’s testimony by itself established that Allord was totally disabled; but it supported the psychologists’ belief that Allord’s disabling PTSD was of long standing.

The administrative law judge disbelieved Chappell-White’s testimony, however, and ordinarily a trier of fact’s credibility finding is binding on an appellate tribunal. But not if the finding is based on errors of fact or logic. E.g., Hanaj v. Gonzales, 446 F.3d 694, 700 (7th Cir.2006); Georgis v. Ashcroft, 328 F.3d 962, 970 (7th Cir.2003); Cao He Lin v. U.S. Dept. of Justice, 428 F.3d 391, 403 (2d Cir.2005). The administrative law judge said that a lay witness’s testimony about the mental illness of an acquaintance cannot be believed unless the witness acted on her observation, as by referring the acquaintance for a psychiatric examination. The premise is dubious, to say the least, but in any event Chappell-White did act on her observation — she referred Allord to Dr. Hutchinson. The administrative law judge also reasoned that if ChappellWhite had thought Allord unable to work, she wouldn’t have recommended him to people who needed the kind of work he did. But her unchallenged testimony was that she made no such recommendations. The administrative law judge further disbelieved her testimony because she’d been friendly with Allord (there is no suggestion, by the way, of any romantic attachment). But a friend would be more likely than a stranger to observe symptoms of mental illness, yet less likely to refer the mentally ill person to a psychologist on frivolous grounds; one doesn’t casually conclude that one’s friends are mentally ill.

The government argues that these errors were harmless because the administrative law judge could have discredited Chappell-White’s testimony on other grounds, for example its apparent inconsistency with Dr. Hutchinson’s contemporaneous evaluation of Allord’s condition. The argument misunderstands the nature of a credibility determination. Such a determination is a judgment call based on whatever circumstances bear on a witness’s credibility. The administrative law judge based his judgment call on a variety of considerations but three of them were mistaken. Whether he would have made the same determination had he not erred in these respects is speculative.

The only situations in which an error in the factors considered by the trier of fact in making a credibility determination can confidently be thought harmless are when a contrary determination would have to be set aside as incredible or when the trier of fact says that he would have made the same determination even if the questioned circumstances had been different from what he thought them to be and he gives an adequate reason for that back-up position. In either of those situations it would be inconceivable for the trier of fact to have made a different finding on credibility, Frank v. Barnhart, 326 F.3d 618, 621-22 (5th Cir.2003), and if so the finding stands despite the flaws in his analysis.

So only if no reasonable trier of fact could have believed Chappell-White’s testimony (for example, if she’d testified that she had first met Allord before she was born), or if the administrative law judge *822 had said that even if Chappell-White had acted on her observations of Allord’s mental condition, had not referred him for work to other people, and had not been a friend, he would have disbelieved her for sufficient other reasons, would the errors that we have identified have been harmless. Neither condition is satisfied.

The administrative law judge rejected Dr. Matsakis’s opinion regarding Allord’s condition in the critical period on the ground that there must be contemporaneous medical evidence of the applicant’s condition. We said in Wilder v. Apfel, 153 F.3d 799, 802 (7th Cir.1998) (emphasis in original) that “what is required [to establish a retrospective diagnosis] is contemporaneous corroboration [contemporaneous with the period of coverage, that is] of the mental illness, ... not necessarily contemporaneous medical corroboration.” “Retrospective diagnosis of an impairment, even if uncorroborated by contemporaneous medical records, but corroborated by lay evidence relating back to the claimed period of disability, can support a finding of past impairment.” Newell v. Commissioner of Social Security, 347 F.3d 541, 547 (3d Cir.2003); see also Loza v. Apfel, 219 F.3d 378, 396 (5th Cir.2000). Contemporaneous corroboration was provided in this case by Chappell-White, whose testimony the administrative law judge, as we have just seen, improperly discredited.

Moreover, contemporaneous corroboration is not always required — just usually. Likes v. Callahan, 112 F.3d 189, 191 (5th Cir.1997); Jones v. Chater, 65 F.3d 102, 104 (8th Cir.1995). A disease might have a well-understood progression, so that a physician examining a patient at time t might have a good idea of what the patient’s condition had been at time t — n, where n was the number of years, prior to the examination, by which time the patient would have had to be completely disabled to be entitled to benefits. But that will be an issue in this case only if, on remand, the administrative law judge determines once again, but this time on acceptable grounds, that Chappell-White is not a credible witness.

The administrative law judge also gave too much weight to Dr. Hutchinson’s 1987 description of Allord’s condition. Hutchinson’s examination of Allord was much farther in time from the critical date on which Allord’s social security coverage ended than Dr. Matsakis’s — more than five years versus four months. More important, Matsakis is an expert in PTSD; Hutchinson, so far as appears, is not. Apparently in the 1980s, when Hutchinson examined Allord, the condition was not yet well understood. See Matthew J. Friedman, “Posttraumatic Stress Disorder: An Overview,” July 20, 2006 (Dept, of Veterans Affairs, National Center for PTSD), http:// www.ncptsd.va.gov/facts/general/fs_over-view.html. If Hutchinson was not familiar with it, his failure to diagnose it in Allord would have little if any significance. The administrative law judge missed this point too.

Whether Allord is entitled to benefits remains to be determined. The evidence on which he relies contains infirmities; these include discrepancies between Dr. Matsakis’s 1993 intake notes when she first examined Allord and her diagnosis and testimony based on her treatment of him that began in 1996. But the flaws in the administrative law judge’s reasoning are too serious to allow us to uphold the Social Security Administration’s decision. The case must be returned to the Administration for further proceedings consistent with this opinion.

Reveksed And Remanded.