Patricia Pries lost control of her 1988 Honda CRX automobile, which rolled over. She was thrown clear of the car, broke her neck, and became a quadriplegic. Pries sued the car's manufacturer and distributor (collectively Honda) under the diversity jurisdiction, contending that the car was defective because the seat belt mechanism permitted the belt to become slack when the car rolled over. Indiana, whose law governs, permits owners of automobiles to recover when defective safety apparatus increases the injury suffered in a crash. Miller v. Todd, 551 N.E.2d 1139 (Ind.1990). See also ALI, Restatement of the Law--Torts: Products Liability Sec. 6 (Tent.Draft No. 1, 1994). The district court granted summary judgment for Honda in this diversity case, ruling that Pries had not been wearing a seat belt and therefore could not complain about the design of the seat belt mechanism.

During her deposition, Pries testified that, when asked at the scene whether she had been wearing a seat belt, she answered "no." This answer deserves commendation--the plaintiff provided strong evidence against herself even though none of the rescuers could remember asking that question of her. Having given an honest answer, Pries supplied a reason: "I told them no, because I didn't think I could have got out of the car if I had it on". Pries lacks an independent recollection whether she buckled her seat belt that day. There was another potent bit of evidence from the accident scene: a deputy sheriff testified that he observed the driver's seat belt retracted against the door pillar as if it had not been used. An emergency medical technician agreed that the seat belt was "just there hanging". This evidence led the district judge to conclude that Pries was not using her belt.

Against this evidence, Pries submits testimony that she habitually fastened her seat belt. See Fed.R.Evid. 406. Her children testified that she insisted that everyone in the car buckle up. Wear on the seat belt assembly suggests that the belt was used frequently. A physician testified that some of Pries' injuries were consistent with the kind of bruises caused by seat belts during rapid deceleration. The district court discounted this evidence: how can habit overcome the hard fact that the belt was found in a retracted position? That Pries used the seat belt 99% of the time does not prove that she fastened it on the day of the accident. Beliefs and memories cannot contradict physical facts, the judge believed. But the position of the seat belt when the car was found is not one of the laws of physics. We have two reports of observations about the condition of the belt, but these observations may be incorrect. The deputy sheriff and medical technician may have misunderstood (or misremembered) what they saw. Memories play tricks on the best observers.

The car itself may be the best witness about conditions at the time of the accident. Strong forces leave telltale signs in physical objects, signs that can be read by people who know what to look for and have the right instruments. Gerald Rosenbluth, a consultant in industrial design, testified by deposition that the seat belt bore signs of the stress that use during an accident would have produced:

As I recall, the pattern was indicative of the loading, of high frictional coefficient, and was consistent with the belt web fabric provided enough pressure, generating enough heat, to form the striated pattern in the D ring.... I construe these to be indicative and consistent with loading; therefore, I would concur that it was proof that [Pries] was wearing the seat belt at the time of this incident.

Later Rosenbluth testified that the striations in the belt are inconsistent with normal use. The district court did not mention this testimony, which is stronger "physical" evidence of seat belt use than the witnesses' observations are of non-use. It might be possible to reconcile the observations if the car had been in another accident, but there is no evidence of that. We conclude, therefore, that there is a bona fide factual dispute about whether Pries was wearing a seat belt when her car rolled over. Reasonable jurors could decide either way--which means that the jurors must choose, after trial. Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 249-51, 106 S. Ct. 2505, 2510-11, 91 L. Ed. 2d 202 (1986); Celotex Corp. v. Catrett, 477 U.S. 317, 106 S. Ct. 2548, 91 L. Ed. 2d 265 (1986). Summary judgment was improper.

The district court found otherwise in part because he believed that Pries should not be allowed to have things both ways. After conceding in her deposition that she did not fasten her seat belt, the judge believed, Pries should not be allowed to take it back. We have held that a conflict between a party's deposition and a later affidavit does not create a material dispute requiring a trial. E.g, Lovejoy Electronics, Inc. v. O'Berto, 873 F.2d 1001, 1005 (7th Cir.1989). A party may not wriggle out of a concession after its legal consequence becomes clear. But that is not what occurred here. Pries has said from the outset that she cannot remember whether she fastened her belt at the beginning of her fateful drive; she recounts that in a dazed state she told her rescuers that she was not belted only because she could not imagine how she could have been thrown from the car if she had worn one. Now she believes she knows how that could happen. New knowledge is a good reason for new views. Anyway, the principal evidence that Pries wore her seat belt is physical, and the striations in the webbing need not explain why they contradicted the driver's remarks.

Although Pries' principal theory was that she was belted at the time of the accident and slipped out when the seat belt apparatus allowed undue slack, she presents several other issues. Because these may come up again, we discuss them briefly.

1. An alternate theory of how Pries got out of her seat belt is that the latch opened during the accident. Rosenbluth testified that the latch was properly designed and functional. Anthony Sances, Professor of Biomechanics at Marquette University and Chairman of Biomechanics at the Medical College of Wisconsin, opined that the latch was defective because it was possible for objects (including flailing limbs) to strike and open it during an accident. This submission misunderstands the legal standard of "defect." The question is not whether it is "possible" for something untoward to occur during an accident but whether "the design creates unreasonable danger" according to "general negligence principles". Miller, 551 N.E.2d at 1141. See also Bammerlin v. Navistar International Transportation Co., 30 F.3d 898 (7th Cir.1994). Counsel for plaintiffs performed a "test" of a similar latch, dropping it on a hard surface to see whether it would open; on occasion, it did. Sances was asked what forces had brought this about and whether these were commonly achieved in a crash; he did not know. Evidence of this kind is not scientific and does not satisfy Fed.R.Evid. 702. See Daubert v. Merrell Dow Pharmaceuticals, Inc., --- U.S. ----, 113 S. Ct. 2786, 125 L. Ed. 2d 469 (1993). No evidence in this record suggests that the latch was defectively designed, and Honda was entitled to prevail on this theory.

2. Pries believes that even if she was unbelted, she would not have suffered such severe injuries had the car been designed to prevent her ejection. She left in one of two ways, through a window or through an open door, and contends that each could have been redesigned to contain her.

The window was made of tempered glass. Rosenbluth testified that this is "state of the art." Sances testified that a window of laminated glass would have a greater chance of restraining Pries. We may assume that this is true, but again it does not imply the existence of a defect. For all this record reveals, laminated glass would have posed even greater problems in other kinds of accidents. To demonstrate a defect, the plaintiff must compare the costs and benefits of alternative designs. Miller, 551 N.E.2d at 1141-42; Products Liability Sec. 2(b) and comment c. See also Barron v. Ford Motor Co., 965 F.2d 195 (7th Cir.1992), holding under North Carolina law that windows and sunroofs need not be designed as part of the passenger-restraint system. "Common sense says that sunroofs meant to be opened need not be built to restrain an occupant who declines to fasten her seatbelt." Id. at 202. Because this record is silent on the costs and benefits of alternative glazing materials, and contains nothing to suggest that a reasonable designer would choose laminated glass for a window designed to be rolled down, Honda was entitled to prevail on this theory.

Rosenbluth believed that Pries fell out of the car when the door opened during the rollover. He thought the door's design made such opening dangerously likely--a danger that he testified could readily have been avoided by alternative designs. Honda does not deny that this testimony created a material dispute under Indiana law about the adequacy of the door's design. It denies, however, that any evidence shows that Pries left through the door. The car's door was closed when the rescuers arrived; on this there is no dispute. It is also undisputed that the door does not show signs of the kind of damage apt to occur when a door is open during a rollover. So Pries needed a theory that could not only open the door but also get it closed again, undamaged. Rosenbluth's favored theory, that the door popped open when the frame was squeezed during the accident, seems inconsistent with finding the door closed again. If the frame was sprung, the door could not close normally.

Sances believed, based on his study of Pries' injuries, that she was ejected through the window. This opinion, which accords with the view of Honda's experts, is within Sances' area of expertise--but, if it is correct, then our conclusion that the glazing was not defective forecloses all of Pries' arguments other than her attack on the design of the seat belt mechanism. Rosenbluth, who believed that Pries was ejected through an open door, has a theory of defect but needs to explain how the door could have closed again consistent with the mechanics of its opening. We cannot say that the record absolutely precludes additional efforts to show a defect along these lines, but they seem unlikely to do much beyond distracting the jury from the principal theory of defect.

Lest there be any doubt in light of the district judge's antipathy to inconsistent theories, we add that it does not count against Pries that her experts do not agree about how she came to be outside the car. Litigants are not bound by their witness's statements; the voucher theory is long gone. We find it comforting to see experts with analyses derived independently, rather than parroting a consistent (but often bogus) theory concocted by counsel. Pries left by either the door or the window; that experts cannot agree on which route she took hardly demonstrates the lack of an injurious defect. Cf. Fed.R.Civ.P. 8(e)(2).

3. Our concentration on the question whether Pries fastened her seat belt does not imply that if she fastened the belt, and nonetheless slipped out during an accident, the assembly must have been defectively designed. As we have emphasized here and in Bammerlin, Indiana requires the plaintiff to show that another design not only could have prevented the injury but also was cost-effective under general negligence principles. Rosenbluth testified that particular additional devices would have kept the belt tight in a rollover, but he conceded that no car in production anywhere in the world in 1988 used the combination of devices he favored. This, coupled with the absence of data about either the costs of additional precautions or the aggregate injuries avoidable by using them, raises a serious question whether failure to adopt such a combination was negligent. Rosenbluth is not an engineer; his education stopped with a master's degree in industrial design. Without the aid of an engineer or statistician to set out the factors relevant to negligence (on which see Bammerlin and, e.g., United States v. Carroll Towing Co., 159 F.2d 169, 173 (2d Cir.1947)), Pries may find it impossible to prevail. But decision on this subject lies ahead.

REVERSED AND REMANDED.