Appeal from the United States District Court for the District of Massachusetts, Joseph L. Tauro, J. COPYRIGHT MATERIAL OMITTED COPYRIGHT MATERIAL OMITTED William L. Boesch with whom David A. Barry and Sugarman, Rogers, Barshak & Cohen, P.C. were on brief, for appellants.

Michael B. Flynn with whom Lori A. Wirkus and Flynn & Associates, P.C. were on brief, for appellees.

Before BOUDIN, Chief Judge, TORRUELLA and DYK,* Circuit Judges.

BOUDIN, Chief Judge.

In 1998, Frank Gillespie was working as a locksmith after many years in various building trades. In September of that year, he asked his employer to let him keep its 10-inch table saw at his house for work and personal use, and his employer agreed. The saw, which he received from coworker Michael Kane, had been designed and manufactured by Emerson Electric Co. and sold by Sears, Roebuck & Co. under the Craftsman brand. Kane did not give Gillespie the saw's owner's manual or its blade guard.

A standard table saw of the type used by Gillespie has a flat rectangular metal surface through which a rotary saw blade protrudes; the blade's height and angle can be adjusted for different kinds of cuts. The wood to be cut is fed into the blade by sliding the wood over the saw's table surface away from the user. For some cuts, the wood is kept in alignment by a metal guide called a rip fence on the saw's surface to one side of the blade. Behind the blade, there may also be affixed to the saw a spreader and blade guard as will be more fully described below.

On November 23, 1998, Gillespie used the table saw to trim an inch off each side of a 28-inch-wide door. After finishing the cut, he shut off the saw and set the door down. Six seconds after shutting off the saw,1 Gillespie turned and his hand came into contact with the blade before it had stopped. The still-spinning blade lacerated several of his fingers; his badly damaged right ring finger had to be amputated after several unsuccessful surgeries.

Gillespie and his wife, Deborah, brought suit in state court against Emerson and Sears (collectively, "Emerson"), who removed the case to district court. Gillespie's complaint, alleging that the accident was caused by defects in the saw's design, asserted claims of negligence and breach of implied warranty of merchantability under Massachusetts law; his theories of defect concerned (1) the design of the blade guard, (2) lack of a brake for the wheel, and (3) inadequate warning. Gillespie's wife claimed for loss of consortium.

Trial was held in June 2003, and ? after denying Emerson's motion for judgment as a matter of law ? the judge sent the case to the jury with a special interrogatory form asking seven questions: whether Emerson had breached the warranty of merchantability and this breach caused the accident; whether Emerson had been negligent and this negligence caused the accident; whether Gillespie was also negligent and if so the degree of fault attributable to each side; the amount of damages due to Gillespie; and two questions pertaining to Deborah's loss of consortium and damages due to her.

During deliberations, the jury sent two notes asking the purpose of the comparative negligence allocation. The judge responded (so far as is pertinent here) that the allocation of a percentage of negligence to Gillespie would reduce his recovery on the negligence claim by that percentage (unless his share of negligence exceeded 50 percent, in which case his recovery would be barred entirely). The response also noted that "no deduction ... will be made on the breach of warranty claim."

Shortly thereafter, the jury found that Emerson breached its warranty of merchantability; that Emerson was negligent but that Gillespie was 49 percent contributorily negligent; that Gillespie had suffered $750,000 in damages and his wife $100,000. Following the verdict, Emerson renewed its motion for judgment as a matter of law, and (in the alternative) sought a new trial. The district court denied both requests, and entered judgment for the full $850,000.

Emerson now appeals, asserting that there was insufficient evidence on the three defect theories to support the jury's verdict. Emerson also claims that the district court erred by refusing to instruct the jury on an "unreasonable use" defense to the breach of warranty claim. Finally, Emerson says that the district court erred in imposing discovery-related sanctions on Emerson and that these sanctions tainted the trial.

Sufficiency of the evidence. In assessing the sufficiency of the evidence, the question for the court is whether, viewing the evidence in the light most favorable to the verdict, a rational jury could find in favor of the party who prevailed. DaSilva v. Am. Brands, Inc., 845 F.2d 356, 359 (1st Cir.1988); see also Santos v. Sunrise Medical, Inc., 351 F.3d 587, 590 (1st Cir.2003) ("In so doing, we do not consider the credibility of witnesses, resolve conflicts in testimony, or evaluate the weight of the evidence" (internal quotation marks omitted)). Our review of the district court's denial of Emerson's motion for judgment as a matter of law is de novo. Santos, 351 F.3d at 590.

Whether there was enough evidence depends upon what Gillespie was required to prove. Gillespie's first claim was for negligence. Massachusetts law, which controls this case, requires manufacturers to design and produce products with reasonable care to eliminate avoidable dangers arising from reasonably foreseeable uses of the product. Cigna Ins. Co. v. Oy Saunatec, Ltd., 241 F.3d 1, 15 (1st Cir.2001). Manufacturers must guard against such dangers, Colter v. Barber-Greene Co., 403 Mass. 50, 525 N.E.2d 1305, 1311 n. 9 (1988), taking account of foreseeable carelessness on the part of the user, DeMedeiros v. Koehring Co., 709 F.2d 734, 739 (1st Cir.1983).

In judging the adequacy of a product's design, pertinent factors include

the gravity of the danger posed by the challenged design, the likelihood that such danger would occur, the mechanical feasibility of a safer alternative design, the financial cost of an improved design, and the adverse consequences to the product and to the consumer that would result from an alternative design.

Colter, 525 N.E.2d at 1310 (quoting Back v. Wickes Corp., 375 Mass. 633, 378 N.E.2d 964, 970 (1978)) (internal quotation marks omitted). An essential element of such a design flaw claim is that there be a safer alternative design. See Kotler v. Am. Tobacco Co., 926 F.2d 1217, 1225 (1st Cir.1990), vacated on other grounds, 505 U.S. 1215, 112 S.Ct. 3019, 120 L.Ed.2d 891, and judgment reissued, 981 F.2d 7 (1st Cir.1992); see also Uloth v. City Tank Corp., 376 Mass. 874, 384 N.E.2d 1188, 1193 (1978).

Massachusetts law also treats manufacturers and sellers as warranting that their products are "fit for the ordinary purposes for which such goods are used." Mass. Gen. Laws ch. 106 § 2-314(2)(c) (2000). In most substantive respects (e.g., unreasonably dangerous design, reasonably foreseeable dangers, alternative design feasibility), the negligence and warranty inquiries are congruent. Oy Saunatec, 241 F.3d at 15-16; Back, 378 N.E.2d at 970; see also Allen v. Chance Mfg. Co., 398 Mass. 32, 494 N.E.2d 1324, 1326 (1986).2

Gillespie's first theory of design defect focused on the design of the blade guard. On Gillespie's Craftsman saw, as with most of the comparable table saws on the market, the guard ? a kind of elongated plastic cap fitting over the top of the blade ? is mounted on the spreader, a vertical metal plate directly behind the blade. The spreader is to be used for "through cuts" ? in which the blade cuts all the way through the wood's thickness to cut it into two pieces ? and keeps the cut open after the wood passes behind the blade so that the two separated sides do not clamp back together onto the blade.

For certain cuts, the blade is partly lowered to cut only a groove in the wood, rather than cutting all the way through the wood's thickness to produce two separate pieces. For such cuts ? called "non-through" cuts ? the spreader must be removed because it would otherwise block the wood as it passed over the blade. Gillespie's main expert (John Orlowski) testified that to conjoin the guard with the spreader ? which was removable and sometimes had to be removed ? invited users to take both off and then leave them off. Orlowksi said that the safer alternative design was to attach the blade guard to a cantilevered arm mounted by a bracket attached to the edge of the table, thus allowing the guard to be used with or without the spreader.

At trial Emerson introduced into evidence a cantilevered guard called the UniGuard which Emerson's expert (Jack Hyde, Jr.) described as "the standard example of that type of design." Gillespie's expert admitted that the cantilevered guard could not be used for all cuts. More important, Hyde ? a former Emerson engineer ? testified that the UniGuard could not have been used for the particular cut that Gillespie was doing when he was injured, because the 28-inch-wide door could not have fit between the UniGuard's arm and the saw's rip fence that keeps the wood aligned as it is being sawed.

Hyde's testimony on this point was not contradicted and it is fatal to Gillespie's theory of design defect. Even if it were negligent or a breach of warranty to supply the spreader-mounted guard rather than the less common cantilevered guard, Gillespie cannot prevail, as to this theory, without showing that the "better" design would have prevented his injury. See, e.g., Solimene v. B. Grauel & Co., K.G., 399 Mass. 790, 507 N.E.2d 662, 667 (1987); Fidalgo v. Columbus McKinnon Corp., 56 Mass.App.Ct. 176, 775 N.E.2d 803, 808 (2002). The cantilevered guard might be better much of the time but, if it could not have been used for the cut that injured Gillespie, its omission was not the cause in fact of his injury.

Hyde's testimony was not quite conclusive. He said that the cut Gillespie made could not have been done with the UniGuard in place but he said nothing about other brands on the market. But the UniGuard was dubbed "standard" and Gillespie presented no other product to depict the "better" cantilevered guard. And absent rebuttal testimony, it would have been pure speculation for the jury to have supposed that some other model of cantilevered guard ? never offered in evidence or described ? could have been used for the cut in question. See Quintana-Ruiz v. Hyundai Motor Corp., 303 F.3d 62, 76-77 (1st Cir.2002).3

Hyde also said that Emerson had previously tested cantilevered guards and found them less satisfactory than spreader-mounted guards; each type was compatible with some cuts but not all and, according to Hyde, the cantilevered guard was more prone to block wood unexpectedly and so to cause unexpected twisting of the wood and possible injury. Gillespie's expert gave no direct answer to this testimony. Given the lack of causation, however, we need not decide whether the jury could reasonably find the cantilevered guard superior.

Gillespie's second theory of defect was that the saw blade should have had a brake that would decrease its "coasting" time once the power was shut off. Here, Gillespie had two experts; the thrust of their testimony was that the normal coasting time for the saw at issue ? 14 to 15 seconds after power shut-off ? could be reduced to a couple of seconds by an automatic electric-current brake available for use on power equipment. This longer-than-necessary coasting time, Gillespie claimed, constituted a foreseeable (and unreasonable) hazard that was compounded by the risks of removal of the guard and user inattention after the machine was switched off.

Hyde said that he had tested such a brake for Emerson in 1986 but found that the braking effect tended to loosen the arbor nut holding the blade to the shaft, creating separate dangers of its own.4 Also, Hyde said that in 2003 he had tested for litigation purposes three brands of table saw equipped with brakes. The brakes did stop the blades in 2 to 4 seconds; but one saw's nut spun loose after nearly a thousand power cycles over the course of three hours. Over a longer period, the other two saws' nuts actually tightened.

Much of the testimony of the dueling experts concerned the likelihood that the brakes would loosen the nuts and the consequences that would ensue if a nut did loosen. There was also conflicting testimony as to whether the Emerson saw, lacking a brake, met applicable ? but relatively vague ? standards issued by American National Standards Institute as well as similar OSHA regulations. It appears that such brakes were once rarely sold with table saws but are becoming somewhat more common.

In principle, it appears that the braking action can increase the risk of loosening the nut, especially where the braking is severe.5 But, as we read the testimony in this case, a jury could have concluded that a brake of moderate force should have been employed. Such a brake could arguably have halted the saw in the 4 to 6 second range while leaving the start-up torque (which tightens the nut) sufficiently great compared to the braking torque (which loosens the nut) to ensure an overall tightening effect and minimal risk of arbor-nut loosening. Even in the 2 to 4 second range, Gillespie's expert said that the single failure in Hyde's test could have been caused by machine vibration or other causes unrelated to the brake.

Emerson says on appeal that the brake was not needed if its guard was used, but the jury had adequate evidence that guards were often detached and that this was foreseeable. Nor is there a causation problem: the jury could also have found that Gillespie turned back to the still spinning blade only six seconds after he switched the machine off. Gillespie's earlier versions of the accident's timing might have made the brake causally irrelevant, but his final version, if believed, brought the accident into the time window where a potentially safe brake would have mattered.

The claim of design flaw was debatable. The use of a brake for table saws was relatively rare even at the time of trial. A brake lessened but did not eliminate coasting, and it could loosen the arbor nut, creating additional dangers. Emerson had considered the balance of risks and benefits and consciously decided against adopting the brake for table saws. Many would think this a defensible judgment.

Nevertheless, Gillespie offered two experts, one of whom made such brakes, both saying the judgment was wrong. At Emerson's behest we have reviewed their credentials and testimony and we cannot say that they were unqualified to give their testimony. Admittedly, Hyde was peculiarly qualified to testify to the contrary ? he had been involved in Emerson's safety engineering and had done the testing in 1986 and 2003 ? but he (no less than the experts for Gillespie) had his own kind of self-interest in the outcome. The blade-brake theory was for the jury to resolve.

Gillespie's final theory rested on the duty to warn adequately of foreseeable dangers, which is a legitimate theory for liability under both negligence, see Slate v. Bethlehem Steel Corp., 400 Mass. 378, 510 N.E.2d 249, 251 (1987), and breach of warranty, see Vassallo v. Baxter Healthcare Corp., 428 Mass. 1, 696 N.E.2d 909, 922 (1998); Bavuso v. Caterpillar Indus., Inc., 408 Mass. 694, 563 N.E.2d 198, 201 (1990). Here, Gillespie's evidence was thin as to both causation and supposed inadequacy of the warning; but it was perhaps just barely possible for a rational jury to find in Gillespie's favor on both points.

Gillespie's main failure-to-warn theory appears to have been that the warning labels on the side of the machine should have been placed on the saw's top surface and had better contrast and larger lettering. These warnings did advise users that the saw was dangerous and referred to the coasting danger (at least indirectly) by warning users to wait for the blade to stop "before adjusting or servicing." Hyde answered that a warning on the surface would be covered by the wood and eventually rubbed off. However, based on the testimony and depictions in the user's manual, we think the jury could have concluded that the label's coloring, contrast, size, and perhaps explicitness were inadequate.

Causation would be lacking if the coasting-blade danger was obvious or otherwise known to the plaintiff. See Bavuso, 563 N.E.2d at 201-02; Colter, 525 N.E.2d at 1312. Gillespie admitted that the still-spinning blade was visible and audible, and that he "understood" that blades coasted, as any experienced user would assuredly know, see Bavuso, 563 N.E.2d at 202. The jury might have concluded from Gillespie's further testimony ? that he didn't "consider" the coasting hazard ? that he was not "fully aware," Morrell v. Precise Eng'g, Inc., 36 Mass.App.Ct. 935, 630 N.E.2d 291, 293 (1994), of the duration of the danger and that a more explicit or conspicuous warning would have heightened his awareness and prevented the accident. This is enough, although just barely so, to take the claim to the jury.

Nevertheless, a new trial is required because the blade-guard theory fails. In this circuit, the rule in civil cases is that "a new trial is usually warranted if evidence is insufficient with respect to any one of multiple claims covered by a general verdict." Kerkhof v. MCI Worldcom, Inc., 282 F.3d 44, 52 (1st Cir.2002); see Lattimore v. Polaroid Corp., 99 F.3d 456, 468 (1st Cir.1996). This rule applies not only to general verdicts encompassing multiple causes of action, but to special verdicts where a single verdict question encompasses multiple theories, one of which is defective. Lattimore, 99 F.3d at 468.

Although not all circuits follow our practice,6 it does accord with the Supreme Court precedents that require remand in civil cases when one of several claims or theories encompassed in a general verdict was flawed (e.g., because of a mistaken jury instruction). The only Supreme Court case squarely dealing with insufficient evidence on one theory, however, is a hundred years old, see Wilmington Star Mining Co. v. Fulton, 205 U.S. 60, 78-79, 27 S.Ct. 412, 51 L.Ed. 708 (1907), and the Court follows the opposite rule in criminal cases where the evidence is adequate on one theory but not another.7 The reality is that the degree of confidence that the jury picked a theory with adequate evidentiary support varies along a spectrum of situations.

Our own approach is by no means rigid. Recognizing that a jury is likely to prefer a better supported theory to one less supported, we have generously applied the harmless error concept to rescue verdicts where we could be reasonably sure that the jury in fact relied upon a theory with adequate evidentiary support. See Davis v. Rennie, 264 F.3d 86, 106 (1st Cir.2001), cert. denied, 535 U.S. 1053 , 122 S.Ct. 1909, 152 L.Ed.2d 820 (2002); Brochu v. Ortho Pharm. Corp., 642 F.2d 652, 662 (1st Cir.1981). But in this case, none of the theories was strongly supported by the evidence; this, indeed, illustrates the danger in automatically presuming that the jury adopted the "stronger" ground.

The uncertainty in this case could have been avoided if either side had pressed, and the district court had agreed, to have the jury decide specifically which theories succeeded and which failed. Defendants may have an incentive not to ask for special verdicts, thus saving a ground for appeal if any theory was defective. But it is a mystery why the plaintiff would not request, or the judge insist upon, a breakdown by theory ? especially where one or more of the theories is at best a close call.

A few circuits have held, although inconstantly, that an appellant who did not request such special verdict has "waived" a claim of error that such a verdict form could have isolated.8 This court has never adopted such a rule and has regularly vacated general verdicts where the award might have been rescued if either side had sought and obtained an adequate special verdict, see, e.g., Lattimore, 99 F.3d at 468; Levinsky's, Inc. v. Wal-Mart Stores, Inc., 127 F.3d 122, 134 (1st Cir.1997); cf. Fleet Nat'l Bank v. Anchor Media Television, Inc., 45 F.3d 546, 551, 557 (1st Cir.1995); but recently we took note of several cases applying a waiver theory and said that we might consider adopting such a requirement, Davis, 264 F.3d at 106.

On balance, we conclude that we should not adopt a waiver rule. The term "waiver" is misleading; either side in this case could have asked for the special verdict to further break down the negligence and warranty claims by theory of defect. The issue is one of policy, i.e., whether the court should create a rule that forfeits claims by an appellant that could have been isolated if either side had requested a better breakdown. The efficiency argument for such a rule is obvious; but the objections are also real.

At first blush, it looks like mandating a special verdict would be easy; but, as this case illustrates, there are multiple levels of detail. Here, dividing by causes of action alone did not solve the problem: a further division by theories of defect was required. And theories, unlike "causes of action," are not pre-existing formal categories derived from statutes or common law but are endlessly variable, depending upon the facts of the case. Gillespie's warning theory alone could itself have been subdivided in several ways (content, placement, size and color).

In some cases special verdicts make sense but there may be others where using them, to a sufficient level of detail, is infeasible or otherwise undesirable. Where feasible and desirable, one of the parties ordinarily has an incentive to ask for such a special verdict; and, if no one does, the judge can insist. In sum, we think that a uniform obligation to ask for a special verdict, or have relevant claims of error forfeit by appellant, is not warranted.

Some courts have treated the appellate consequences of general verdicts encompassing multiple claims or theories as a procedural issue to be determined by federal law even in a diversity case; others have assumed usually without discussion that state law controlled.9 Our own tentative view is the former, see Levinsky's, 127 F.3d at 134 (implicitly so assuming); but anyway Massachusetts follows the First Circuit practice on this issue. See Slate, 510 N.E.2d at 253 (insufficient evidence on one claim).

Accordingly, as the verdict may have rested solely on the blade guard theory, a new trial is required. Because the evidence on the blade guard theory was inadequate on causation grounds and Emerson was entitled to judgment on that issue, Gillespie's defect theories at the new trial are confined to the blade brake and inadequate warning claims of defect. See Lattimore, 99 F.3d at 468.

Unreasonable use instruction. Under Massachusetts law, there is a comparative fault defense for negligence claims; for warranty claims, the closest analogue is an "unreasonable use" defense. Oy Saunatec, 241 F.3d at 16-17. Unreasonable use provides a complete defense to warranty claims (the plaintiff might still recover for negligence). See id. at 17. Emerson sought an unreasonable use instruction but the district court declined to give it. Emerson now claims that this was error.

A party is entitled to an instruction on a legally available defense if sufficient evidence is present to permit a jury to find in favor of the defense. Sullivan v. Nat'l Football League, 34 F.3d 1091, 1107 (1st Cir.1994), cert. denied, 513 U.S. 1190 , 115 S.Ct. 1252, 131 L.Ed.2d 133 (1995). Under Massachusetts law, the unreasonable use defense required Emerson to prove that Gillespie "knew of the product's defect and its danger, that he proceeded voluntarily and unreasonably to use the product and that, as a result, he was injured." Allen, 494 N.E.2d at 1326; see also Correia v. Firestone Tire & Rubber Co., 388 Mass. 342, 446 N.E.2d 1033, 1040 (1983).

There was surely enough evidence on the objective element of the defense, namely, "unreasonable use." See Allen, 494 N.E.2d at 1326. Gillespie admittedly knew that table saws came with guards and knew that this machine had no guard attached. Yet he made no effort to obtain and employ the guard; the evidence showed that the guard for this saw had been taken off and remained with the coworker who transferred the saw to Gillespie. The existing spreader-mounted guard would apparently have worked for the cut that Gillespie was attempting at the time of the accident and so would have prevented the accident.

Whether the failure to use the guard was unreasonable under all the circumstances is perhaps a matter on which a jury could take different views, given that no guard was readily available to Gillespie. The fact that the jury found Gillespie 49 percent negligent does not prove that it would have found the failure to use the guard unreasonable use; no one knows what act the jury found negligent (it could have been Gillespie's inattention). But a jury could rationally have found objectively unreasonable use if it wanted to do so.

The more difficult element of the defense concerns Gillespie's knowledge. Under Massachusetts law, it is not enough that the plaintiff be aware that the machine is dangerous; rather, he must have a reasonably specific awareness both that the product is defective, Velleca v. Uniroyal Tire Co., 36 Mass.App.Ct. 247, 630 N.E.2d 297, 298 (1994), and the extent of the harm to which this exposes him. Allen v. Chance Mfg. Co., 873 F.2d 465, 473 (1st Cir.1989); see also Goulet v. Whitin Mach. Works, Inc., 30 Mass.App.Ct. 310, 568 N.E.2d 1158, 1161-62 (1991). When the alleged unreasonable use is the failure to use a safety device,

plaintiff must at least have perceived that [the] safety device is standard equipment or regarded as necessary to safe operation, so that removing it or using the equipment without the device is dangerous.

Downs v. Gulf & W. Mfg. Co., 677 F.Supp. 661, 665-66 (D.Mass.1987), quoted in Allen, 873 F.2d at 473.

Here, Gillespie admitted that he knew saws were sold in stores with blade guards on them and that he had seen one used with a blade guard on a home repair television show. Based on these admissions, we think the jury could permissibly have inferred that he knew that guards were standard safety equipment, that their absence meant that something was wrong in a way that increased the risk of injury, and that (as a matter of common sense) working with a high-speed cutting tool with an unguarded blade could cause serious injury up to and beyond what Gillespie suffered.

There is a counter-argument. Gillespie testified in detail that he "never considered a blade guard on a blade and what it will do," had not seen one used except on television, didn't think he was taking a risk by using the blade without the guard, and even continued using the saw without the blade after the accident. Such a view might, if actually held, negate the subjective awareness element of the defense as prescribed by Massachusetts case law.

It can sometimes be irrational for a jury to reject "uncontradicted" testimony, cf. Quintana-Ruiz, 303 F.3d at 75-76, especially about the witness's state of mind; but a jury may disbelieve uncontradicted testimony if it is improbable. See id. at 76; see also EEOC v. G-K-G, Inc., 39 F.3d 740, 743 (7th Cir.1994). The jury might have concluded that it is improbable that Gillespie ? working around power equipment and knowing that saws were sold with guards ? would not have appreciated that the lack of a guard meant that something was wrong with the saw and that the absence of the guard increased the risk of serious injury. That Gillespie chose to take the risk does not preclude knowledge of it.

Further, Gillespie had a strong financial interest in negating the unreasonable use defense; in this case doing so doubled Gillespie's damages because his recovery on negligence alone would have been reduced by 49 percent. The jury was free to consider Gillespie's self-interest and also that Gillespie's somewhat improbable testimony dovetailed rather neatly with the counter to a rather technical defense ? which his lawyer may well have discussed with him. The jury surely did not have to accept Gillespie's gloss on his own thinking.

Thus, the unreasonable use instruction should have been given. Although the blade guard theory is not to be retried, we see no reason why the instruction should not apply to brake and warning theories as well. If it was unreasonable use of the saw to operate it without a blade guard for a cut where the guard was feasible, then the unreasonable use causally enabled the injury whatever the defect. Even if the jury accepts such a defense to defeat the warranty claim, it can still award damages for negligence (adjusted for Gillespie's own contributory negligence).

Expert discovery. In July 2002, Emerson's expert, Hyde, provided an expert report, duly made available to Gillespie, and then in January 2003 was deposed by Gillespie in the conventional fashion. Thereafter, plaintiff's main expert, Orlowski, during his deposition in February 2003, noted that another manufacturer (Makita) was selling a table saw with an electric brake as a standard feature. In response to this information, Hyde (as already noted) in April 2003 conducted further tests on a Makita saw and two others similarly equipped with brakes.

Hyde's findings from these tests were summarized in a supplemental report dated May 2, 2003, which was turned over to Gillespie on May 8. See Fed.R.Civ.P. 26(e). This report said that Hyde had tested the three saws and then described in detail the test performed on the Makita and the fact that after the described hours of testing the blade spun loose on the shaft (allegedly because the braking action had loosened the nut). The report noted much more briefly that the other two saws tested through repeated cycles had not failed, but the report expressed Hyde's view that such failures could be expected to occur unpredictably in all three of the saws.

The supplemental report when turned over to Gillespie was accompanied by a letter saying that a DVD containing a "relevant excerpt" of the testing ? a short video of the Makita's failure-would be used in evidence and was available for review by Gillespie's counsel. What was not furnished with the May report, nor explicitly mentioned, were the existence of Hyde's working notes on the tests (including an "event record" describing the tests) and the existence of videotapes of all three tests.

On May 30, 2003, the Friday before the scheduled Monday, June 2, trial, Gillespie's counsel asked for all documents and videos generated during the testing; and on Sunday, June 1, Emerson's counsel faxed Gillespie's counsel saying that the extensive videotapes and other materials were being copied and would be delivered shortly. On June 2, the first day of trial, the district judge told Emerson to "turn over everything" immediately, saying that he had earlier ordered production of all relevant evidence.

In all events, Emerson produced the additional Hyde material on June 3. The next day, the district court nevertheless ruled that Emerson had violated "my order," directed that Hyde be made available for a mid-trial deposition, and promised wide latitude for Gillespie's cross-examination of Hyde. Gillespie took advantage of this in closing argument (over Emerson's objection) and in cross-examination to suggest that Hyde had deliberately withheld information in violation of court order.

On this appeal, Emerson complains that it was prejudicial error for the judge to permit such cross-examination and closing argument because there was no violation of any valid order. Gillespie makes little effort to defend the premise that there was such an order, beyond describing the court's vague warnings and urgings about the importance of producing all relevant documents and asserting that Hyde conceded that he had been ordered to produce the material. Hyde did concede this, under cross-examination at the trial; but this was after the district court had ruled that it had so ordered, so what else was a lay witness to say?

Emerson says that the district judge never issued such an order. In fact, there was considerable confusion as to whether the parties were required to produce automatically only such evidence that they might use to support its case or all "relevant" evidence.10 It is arguable that Emerson should eventually have understood the judge to be taking the latter view; but nothing suggests that Hyde personally should have thought that he had any obligation to produce his working notes or do anything beyond what was required by the specific rule governing expert discovery.

Gillespie claims that this expert discovery rule itself required the working notes because it requires a report stating the expert's opinions and reasoning and "the data or other information considered by the witness in forming the opinions." Fed.R.Civ.P. 26(a)(2)(B). But this language does not require that the expert report contain, or be accompanied by, all working notes or recordings, nor does it appear that even Gillespie's expert Orlowski so construed it. The other rule cited by Gillespie, requiring supplementing of discovery, Fed.R.Civ.P. 26(e), was complied with by delivering Hyde's report.11

The district court was free to order a mid-trial deposition for any good reason; but we see no basis in Rule 26 or anything else in this record to justify the assertion to the jury that Hyde withheld notes and tapes that he knew or should have known he was required to produce. And prejudice may have occurred: Hyde was a critical and well-qualified witness for Emerson, dishonesty is a serious charge, and Gillespie's case was at best very debatable on liability.

We need not resolve the prejudice issue because a new trial is necessary in any event. But we have given abbreviated attention to this discovery imbroglio for two reasons. One is to make clear that on retrial there should be no charge that Hyde violated any automatic discovery rule or order, although he can certainly be cross-examined about whether his supplemental report was sufficiently complete and balanced. The other reason for discussion concerns the scope of retrial, to which we turn next.

The scope of retrial."An appellate court has broad discretion to remand for a new trial on all, or only some, of the issues in the case." Dopp v. HTP Corp., 947 F.2d 506, 518 (1st Cir.1991). "A new trial may not, however, be limited to fewer than all the issues unless it clearly appears that the issues to be retried are so distinct and separable from the other issues that a trial of those issues alone may be had without injustice." La Plante v. Am. Honda Motor Co., 27 F.3d 731, 738 (1st Cir.1994).

Here, for reasons already explained (primarily lack of causation evidence for the defective guard theory) a new trial is required as to Emerson's liability to Gillespie and to his wife (her claim depends on liability to him). Similarly, the comparative negligence issue must be retried; the jury may have been comparing Gillespie's negligence with conduct (the lack of a cantilevered guard) that had no causal role in the case. See La Plante, 27 F.3d at 738.

It is a closer call whether damages, for Gillespie or his wife, must also be retried. Emerson asks for a retrial on all issues; Gillespie's position is less clear. But it is often in a plaintiff's interest to have damages evidence known to the jury while it is considering liability ? unless the plaintiff thinks he did unduly well the first time around. There is some cost in retrying the damages issue, but ordinarily much of the evidence is the plaintiffs' own testimony and records.

Technically, the damages verdicts are not directly undermined by any of the errors we have discussed ? the causation issue, the omitted instruction or the attempted discrediting of Hyde. About the most one can say for sure is that giving the unreasonable use instruction and subtracting the unfair critique of Hyde's honesty would have improved the atmosphere for the defense as to damages as well as liability. So the scope of the remand is a close issue.

On balance, we think there should be a new trial on damages as well as liability unless both sides stipulate otherwise. The case law suggests a preference for full trial unless a partial one will clearly do justice. Here, the jury ? through its questions concerning the role of comparative negligence in damages awarded ? did show some interest in the link between liability and damages. Given at least a possibility of spill-over effects from the errors, the broader retrial is appropriate.

The judgment of the district court is vacated and the case remanded for further proceedings consistent with this decision.

It is so ordered.

Notes:

* Of the Federal Circuit, sitting by designation

1 Gillespie had earlier reported delays of four seconds, three to four seconds, and "instantaneously." His ultimate estimate of six seconds derived from a post-accident reenactment that he performed. On sufficiency-of-the-evidence review we resolve these ambiguities in favor of the jury verdict and, therefore, use Gillespie's six-second estimate

2 Although warranty is sometimes described as a no-fault claim and "focuses on product characteristics rather than on the defendant's conduct" as in negligence,Oy Saunatec, 241 F.3d at 15 (quoting Back, 378 N.E.2d at 970) (internal quotation marks omitted), the use of foreseeability and unreasonable dangerousness tests makes the warranty standard similar to negligence, although various incidents of the two claims are different (e.g., the consequences of user fault, the circumstances in which sellers in addition to manufacturers are liable).

3 At oral argument, Gillespie argued for the first time that the door being cut could simply have been flipped over to avoid its hitting the arm of a cantilevered guard. There is no indication that any such evidence was offered to the jury; and it appears from Hyde's description and the owner's manual that the rip fence would have prevented any such re-alignment of the wood. Although the rip fence is removable, and should be removed for some cuts, testimony and the saw's manual indicate that safety and accurate cutting required its being in place for the cut that Gillespie was making

4 The brake is not mechanical; rather when the shut-off switch is hit, an electric current is sent into the motor creating a magnetic field that counters the motor's spinning motion and so enhances the motor's deceleration. But because the saw is designed so that the motor's start-up torque automatically tightens the arbor nut, the reverse braking torque apparently tends to loosen it, and the net effect on the nut depends on the balance between the magnitudes of the start-up acceleration and the brake-induced deceleration

5 Hyde's 1986 test had produced a failure where the pre-set stopping time was precipitous ? almost as short as the 1 to 1.5 second start-up time. In Hyde's 2003 test, a less severe braking time of 2 to 4 seconds produced a loosened nut but only on one of three machines tested; on the other two the nuts were tighter after the test

6 See, e.g., E. Trading Co. v. Refco, Inc., 229 F.3d 617, 621-22 (7th Cir.2000) (requiring an affirmative showing that the jury "probably was confused" by being instructed on a claim or defense for which the party had presented no evidence). Compare Traver v. Meshriy, 627 F.2d 934, 938-39 (9th Cir.1980) (adopting a four-factor discretionary standard) with McCord v. Maguire, 873 F.2d 1271, 1273-74 (9th Cir.) (limiting Traver to general verdicts encompassing multiple legal theories and applying a new-trial rule for general verdicts based on a single legal theory encompassing multiple factual theories), amended by 885 F.2d 650 (9th Cir.1989).

7 The general remand rule in civil cases, derived fromMaryland v. Baldwin, 112 U.S. 490 , 493, 5 S.Ct. 278, 28 L.Ed. 822 (1884), has been reaffirmed in more recent cases, e.g., Sunkist Growers, Inc. v. Winckler & Smith Citrus Prods. Co., 370 U.S. 19, 29-30, 82 S.Ct. 1130, 8 L.Ed.2d 305 (1962). The contrary rule in criminal cases involving insufficient evidence, see Griffin v. United States, 502 U.S. 46 , 49-51, 112 S.Ct. 466, 116 L.Ed.2d 371 (1991), makes sense when one looks at the cases in which the rule is often applied, e.g., id. (no evidence on one theory and enough on another).

8 See, e.g., Gen. Indus. Corp. v. Hartz Mountain Corp., 810 F.2d 795, 801 (8th Cir.1987) (applying waiver rule); Ryko Mfg. Co. v. Eden Servs., 823 F.2d 1215, 1236-37 (8th Cir.1987) (not applying rule); Anixter v. Home-Stake Prod. Co., 77 F.3d 1215, 1231-32 (10th Cir.1996) (stating a waiver rule but exercising discretion not to apply it because an objection would have been futile at the time of trial); Counts v. Burlington N. R.Co., 952 F.2d 1136, 1140 (9th Cir.1991) (not applying rule for defects other than sufficiency of the evidence); McCord, 873 F.2d at 1274 (applying rule).

9 Compare, e.g., Barber v. Whirlpool Corp., 34 F.3d 1268, 1278 (4th Cir.1994) (refusing to apply the state's "two-issue" rule in a diversity case because "federal law controls verdicts"), and Royal Typewriter Co. v. Xerographic Supplies Corp., 719 F.2d 1092, 1098-99 (11th Cir.1983) (applying the federal rule); with Mich. Abrasive Co. v. Poole, 805 F.2d 1001, 1005 (11th Cir.1986) (applying without discussion the state two-issue rule), and Rimer v. Rockwell Int'l Corp., 739 F.2d 1125, 1130 (8th Cir.1984) (same).

10 The 1993 version of Rule 26 had once mandated automatic discovery of everything that might help the other side,see Fed.R.Civ.P. 26(a)(1) (1993), but this rule was supplanted by a 2000 amendment limiting automatic discovery to material that the possessing party might intend to use to support its claims or defenses. See Fed.R.Civ.P. 26(a)(1) (2000) & 2000 advisory committee notes to Rule 26(a)(1). The district judge initially assumed that the old rule was in force or had been preserved through a standing order, which does not appear to have been authorized by the rule. See Fed.R.Civ.P. 26, 2000 advisory committee notes to Rule 26(a)(1). On the other hand, from discussions with the judge, both sides should arguably have understood that he expected all relevant evidence to be provided.

11 Gillespie's further assertion that the event record should have been produced because of a document request accompanying a notice of a Rule 30(b)(6) deposition was not relied upon by the district judge and further appears to be a weak claim given the district judge's orders limiting document requests and interrogatories