POSNER, Circuit Judge.

This appeal requires us to consider both the standard for granting (and reviewing grants of) preliminary injunctions, and substantive issues of exclusive dealing under section 3 of the Clayton Act, 15 U.S.C. § 14, which makes it unlawful to sell goods "on the condition, agreement, or understanding that the . . . purchaser . . . shall not use or deal in the goods . . . of a competitor or competitors of the . . . seller, where the effect of such . . . condition, agreement, or understanding may be to substantially lessen competition or tend to create a monopoly in any line of commerce."

Roland Machinery Company, a substantial dealer (its gross revenues exceed $10 million a year) in construction equipment and related items, serving a 45-county area in central Illinois, was for many years the area's exclusive distributor of International Harvester's line of construction equipment. International Harvester got into serious financial trouble and in 1982 sold its construction-equipment division to Dresser Industries. Dresser promptly signed a dealership agreement with Roland. The agreement provided that it could be terminated by either party, without cause, on 90 days' notice. It did not contain an exclusive-dealing clause (that is, a clause forbidding the dealer to sell any competing manufacturer's construction equipment). Eight months after signing the agreement Roland signed a similar agreement with Komatsu, a Japanese manufacturer of construction equipment. Several months after discovering that Roland had done this, Dresser gave notice that it would exercise its contract right to terminate its dealership agreement with Roland without cause. Roland brought this suit shortly before the end of the 90-day notice period, charging that Dresser had violated section 3 of the Clayton Act and other provisions of federal and state law. The district judge granted Roland a preliminary injunction based solely on the section 3 charges, and Dresser has appealed under 28 U.S.C. § 1292(a)(1). None of Roland's other charges is before us on this appeal.

At the hearing on Roland's motion for preliminary injunction, Dresser presented evidence that it had cut off Roland because it was afraid that Roland intended to phase out the Dresser line and become an exclusive Komatsu dealer, and because it believed that as long as Roland (a well-established firm) remained a Dresser dealer, no other dealer in the area would be willing to handle Dresser equipment, as this would mean competing with Roland. The usual practice in the industry is for dealers not to carry competing lines, and Dresser presented evidence that this makes for more aggressive promotion of each line. Roland, however, presented evidence that it had no intention of phasing out Dresser equipment, that it was terminated because the dealership contract contained what Roland at argument called a "secret" term requiring Roland to deal exclusively in Dresser equipment, and that the sudden termination would bankrupt it or at least cause it serious loss. But it seems that only about 50 percent of Roland's revenues are derived directly or indirectly from Dresser equipment, and only about 10 percent from selling new Dresser equipment (the other 40 percent coming from renting and servicing equipment, and from selling parts); and Dresser argues that Roland could survive simply by promoting Komatsu equipment aggressively — which it intended to do anyway.

After a two and a half day hearing, the district judge concluded that if Dresser were allowed to cut off Roland pending the trial of the case Roland would probably go out of business — an injury to Roland greater than the harm to Dresser from being forced to continue dealing with Roland in the interim. But because comments by Roland's general manager "raise some doubts as to the sincerity of Roland's current claims that it plans to aggressively distribute Dresser products both now and in the future," the judge conditioned the preliminary injunction on Roland's "maintain[ing], within normal economic fluctuations, the approximate market share which it now has obtained for Dresser products."

On the probable merits of Roland's section 3 claim, the judge found that while Dresser's contract with Roland contained no exclusive-dealing requirement, Roland "has adequately shown that an implied exclusive dealing arrangement existed between itself and Dresser. The Defendant's own evidence showed that it considered a distributor which carried a competing line as inimical to its own interest. In the mind of Dresser, a distributor must either live or die with the manufacturer's product." Having thus found an agreement (which is a prerequisite to liability under section 3 of the Clayton Act, see, e.g., Ron Tonkin Gran Turismo, Inc. v. Fiat Distributors, Inc., 637 F.2d 1376, 1389 (9th Cir. 1981), and cases cited there), the judge considered whether Roland had raised a "substantial question" as to whether the agreement was likely to lessen competition substantially. Dresser manufactures 16 or 17 percent of the construction equipment sold in central Illinois (defined as Illinois south of Chicago and north of the latitude of St. Louis), and Komatsu only 1 percent. Roland accounts for all these sales. The judge found that Komatsu (which has no exclusive dealerships anywhere in the midwest) could not have gotten into this market except by persuading a dealer for another manufacturer, such as Roland, to carry Komatsu equipment as a second line.

The first bone of contention between the parties is the proper standard for granting a preliminary injunction and for appellate review of such a grant. Each party is able to cite numerous decisions in support of its view of the proper standard, simply because the relevant case law is in disarray in both this and other circuits. Many of our cases say that to get a preliminary injunction a plaintiff must show each of four things: that he has no adequate remedy at law or will suffer irreparable harm if the injunction is denied; that this harm will be greater than the harm the defendant will suffer if the injunction is granted; that the plaintiff has a reasonable likelihood of success on the merits; and that the injunction will not harm the public interest. See, e.g., Technical Publishing Co. v. Lebhar-Friedman, Inc., 729 F.2d 1136, 1138-39 (7th Cir. 1984); Alexander v. Chicago Park District, 709 F.2d 463, 467 (7th Cir. 1983); O'Connor v. Board of Education, 645 F.2d 578, 580 (7th Cir. 1981). Although described in the cases as a four-factor test, the test actually involves five factors, unless "no adequate remedy at law" and "irreparable injury" mean the same thing. In ordinary equity parlance they do not. See 11 Wright Miller, Federal Practice and Procedure § 2944, at p. 401 (1973). But Fox Valley Harvestore, Inc. v. A.O. Smith Harvestore Products, Inc., 545 F.2d 1096, 1097 n. 1 (7th Cir. 1976), suggests that they may mean the same thing in the preliminary-injunction setting. Further adding to the uncertainty is the fact that we usually say no adequate remedy at law or irreparable harm (as in Technical Publishing, Alexander, and O'Connor), but sometimes no adequate remedy at law and irreparable harm, as in Fox Valley, supra, 545 F.2d at 1097. And our recent decision in American Can Co. v. Mansukhani, 742 F.2d 314, 325 (7th Cir. 1984), states that the plaintiff must establish "the threat of irreparable harm for which there is no adequate remedy at law."

Maybe there is a sixth factor. Some cases, by stating that the purpose of a preliminary injunction is to preserve the status quo, see, e.g., EEOC v. City of Janesville, 630 F.2d 1254, 1259 (7th Cir. 1980), imply that this is another thing the plaintiff must prove; and it is true that if the plaintiff asks for more than a return to the status quo he is apt to be turned down on that ground. See, e.g., SCM Corp. v. Xerox Corp., 507 F.2d 358, 361 (2d Cir. 1974). But "status quo" is ambiguous. The preliminary injunction in this case maintained the status quo in one sense: it continued Roland as a Dresser dealer. But it changed the status quo in another: it made the dealership agreement no longer terminable by either party on 90 days' notice.

Some of our cases imply that although each of the four factors must be considered, the plaintiff need not prevail on all four. See, e.g., Reinders Bros. v. Rain Bird Eastern Sales Corp., 627 F.2d 44, 49 (7th Cir. 1980); Fox Valley, supra, 545 F.2d at 1098. But other cases say that "a preliminary injunction is an extraordinary remedy which is not available unless plaintiffs carry their burden of persuasion as to all of the [four] prerequisites." Shaffer v. Globe Protection, Inc., 721 F.2d 1121, 1123 (7th Cir. 1983). To the same effect see, e.g., Technical Publishing Co., supra, 729 F.2d at 1139; Singer Co. v. P.R. Mallory Co., 671 F.2d 232, 234 (7th Cir. 1982) ("All of these conditions must be satisfied before the drastic remedy of an injunction will be ordered."); but cf. Illinois Bell Tel. Co. v. Illinois Commerce Comm'n, 740 F.2d 566, 571 (7th Cir. 1984) ("where the plaintiff seeks an injunction [preliminary or permanent] to prevent the violation of a federal statute that specifically provides for injunctive relief, it need not show irreparable harm"). Still other cases do not mention the four factors, see, e.g., Valley Liquors, Inc. v. Renfield Importers, Ltd., 678 F.2d 742 (7th Cir. 1982), or propose a different standard, see Omega Satellite Products Co. v. City of Indianapolis, 694 F.2d 119, 123 (7th Cir. 1982).

With respect to the individual factors, some cases imply that the first (first two, if no adequate remedy at law and irreparable harm are meant to be separate factors) is presumptively satisfied in any dealer-termination case, presumably because the dealer's loss of profits would be difficult to quantify. See Menominee Rubber Co. v. Gould, Inc., 657 F.2d 164, 167 (7th Cir. 1981); Reinders Bros. v. Rain Bird Eastern Sales Corp., supra, 627 F.2d at 53. But these are alternative holdings, and the strongest statement of this position, in Milsen Co. v. Southland Corp., 454 F.2d 363, 366 (7th Cir. 1971), appears to be a dictum. Fox Valley can be read to require that the plaintiff go much further and show that the termination has "crippled if not destroyed [its] ability to carry on [its] antitrust case." 545 F.2d at 1098. See generally Note, The Irreparable Harm Requirement for Preliminary Injunction Relief in Antitrust Distributor Termination Cases, 61 B.U.L.Rev. 507, 516-21 (1981).

It is in the dealer-termination cases, too, that we find an alternative formulation to "reasonable likelihood of success," another of the traditional four (or five, or six) factors. Fox Valley alternates "reasonable likelihood of success," with "some likelihood of success," 545 F.2d at 1098, while Milsen just says "success on the merits," 454 F.2d at 366, leaving unclear just how fair a prospect of success is necessary. The Supreme Court has said, "likely to prevail on the merits." Doran v. Salem Inn, Inc., 422 U.S. 922, 931, 95 S.Ct. 2561, 2567, 45 L.Ed.2d 648 (1975). "Some likelihood of success" seems to be an abbreviated version of the formula in Hamilton Watch Co. v. Benrus Watch Co., 206 F.2d 738, 740 (2d Cir. 1953) (Frank, J.), that the plaintiff's contentions on the merits must be "so serious, substantial, difficult and doubtful, as to make them a fair ground for litigation and thus for more deliberate investigation." We adopted this formula in Mytinger Casselberry, Inc. v. Numanna Laboratories Corp., 215 F.2d 382, 385 (7th Cir. 1954), but later references are more gingerly. See Milsen Co. v. Southland Corp., supra, 454 F.2d at 366 n. 3; Mullis v. Arco Petroleum Corp., 502 F.2d 290, 293 (7th Cir. 1974); Fox Valley, supra, 545 F.2d at 1097-98; W.A. Mack, Inc. v. General Motors Corp., 260 F.2d 886, 890 (7th Cir. 1958); see also Benson Hotel Corp. v. Woods, 168 F.2d 694, 697 (8th Cir. 1948). Most of our cases continue to speak of "reasonable likelihood," but as the appearance of both "reasonable" and "some" likelihood in Fox Valley suggests, it is unclear how much practical difference there is between these formulations.

With respect to the scope of appellate review, the picture is especially blurred, as pointed out by Judge Friendly in a dictum in Buffalo Courier-Express, Inc. v. Buffalo Evening News, Inc., 601 F.2d 48, 59 (2d Cir. 1979), that Judge Eschbach of our court has called cogent. Shango v. Jurich, 681 F.2d 1091, 1097 (7th Cir. 1982). Most cases in this and other circuits denote the standard of review either as abuse of discretion, see, e.g., SEC v. Suter, 732 F.2d 1294, 1301 (7th Cir. 1984), or as "clear" abuse of discretion. See quotation in next paragraph from Alexander, United States Steel Corp. v. Fraternal Ass'n of Steelhaulers, 431 F.2d 1046, 1048 (3d Cir. 1970). The older cases, such as Mytinger Casselberry, supra, 215 F.2d at 384-85, are especially emphatic about the limited scope of appellate review. Although the Supreme Court routinely employs the abuse of discretion formulation, see Doran v. Salem Inn, Inc., supra, 422 U.S. at 932, 95 S.Ct. at 2568; Brown v. Chote, 411 U.S. 452, 457, 93 S.Ct. 1732, 1735, 36 L.Ed.2d 420 (1973), and cases cited, it is not clear whether the standard of review was actually at issue in any of the Court's cases; the abuse of discretion standard seems to have been assumed rather than examined. One of our cases implies that there is broader appellate review of an order granting than of an order denying a preliminary injunction, Milsen Co. v. Southland Corp., supra, 454 F.2d at 369 — but how much broader is unclear. Several of our cases do not specify a standard of review, but imply that it is the same standard as is used to review district court judgments involving the application of a substantive rule to the facts found by the district judge. Mullis, Valley Liquors, and Omega Satellite Products are all such cases.

One case says that an order denying a preliminary injunction will not be reversed "unless there is demonstrated to be a clear abuse of the trial court's discretion or clear error in the trial court's findings." Alexander v. Chicago Park District, supra, 709 F.2d at 467. A variant of this test, found in several cases, is that a grant or denial of a preliminary injunction will not be reversed unless there has been a clear abuse of discretion, "a certain mistake of law," or clearly erroneous factfindings. E.g., Atari, Inc. v. North American Philips Consumer Electronics Corp., 672 F.2d 607, 613 (7th Cir. 1982). Now it might seem that if the trial judge's findings are not clearly erroneous and he has made no legal errors, there would be no basis for reversal, and hence no place for the abuse of discretion standard — unless the judge wasn't applying a standard; but we have seen that preliminary injunctions are supposed to be granted or denied in accordance with a standard, and not as a matter of judicial grace. Determinations that involve applying a legal standard to a state of facts are commonly treated as factfindings for purposes of the clearly erroneous standard, see 9 Wright Miller, Federal Practice and Procedure §§ 2589-2590 (1971), but in any event are not considered discretionary determinations. The reference to "certain" mistake of law adds little, for any purely legal determination made by the trial judge in granting or denying a preliminary injunction is subject to plenary appellate review, see e.g., Pratte v. NLRB, 683 F.2d 1038, 1044 (7th Cir. 1982); Delaware Hudson Ry. v. United Transport. Union, 450 F.2d 603, 620 (D.C. Cir. 1971), just as any purely legal determination underlying a final judgment is. Maybe, however, the formulation in Atari is intended to distinguish between ordinary factfindings (including findings resulting from the application of a substantive standard, such as that of negligence, to the who-did-what-to-whom facts), as to which the clearly-erroneous standard clearly applies, and the balancing of harms that is required in a preliminary injunction decision. The striking of the balance often requires a comparison of imponderables that invites an exercise of judgment by the district judge to which the court of appeals will defer even more broadly than where the judge has applied a substantive standard such as negligence — provided the judge has not committed a clear error of fact, or an error of law.

Several other decisions in this circuit, however, suggest that the term abuse of discretion when used in the context of appellate review of orders granting or denying preliminary injunctions means just that there are no errors of law and no clearly erroneous findings of fact. Support for this position in this circuit is found in Hillhaven Corp. v. Wisconsin Dep't of Health Social Services, 733 F.2d 1224, 1226 (7th Cir. 1984) (per curiam) ("we can only reverse the district court's action [granting a preliminary injunction] if it is clearly erroneous or represents a mistake of law. The totality of the factors must point to a clear departure from the proper exercise of the district judge's discretion."); Machlett Laboratories, Inc. v. Techny Industries, Inc., 665 F.2d 795, 798 (7th Cir. 1981) ("The latter injury . . . is plainly outweighed by the certain injury to Techny of going out of business, and therefore it must be deemed an abuse of discretion to find the opposite."), and Charles v. Carey, 627 F.2d 772, 776 (7th Cir. 1980) ("misapplication of the law to particular facts is an abuse of discretion"). Other circuits have suggested that "abuse of discretion" means error, period: "What we mean, when we say that a court abused its discretion, is merely that we think that it made a mistake." Pearson v. Dennison, 353 F.2d 24, 28 n. 6 (9th Cir. 1965); see also In re Josephson, 218 F.2d 174, 182 (1st Cir. 1954) (Magruder, J.). Although neither of these was an injunction case, Clemons v. Board of Education, 228 F.2d 853, 857 (6th Cir. 1956), was, and it says that "misapplication of the law to the facts is in itself an abuse of discretion" requiring reversal. See also Omega Importing Corp. v. Petri-Kine Camera Co., 451 F.2d 1190, 1197 (2d Cir. 1971) (Friendly, J.). Finally, many opinions that say "abuse of discretion" appear to review the district court's order as searchingly as they would any nondiscretionary determination. See, e.g., Godinez v. Lane, 733 F.2d 1250, 1262 (7th Cir. 1984); United Church of the Medical Center v. Medical Center Comm'n, 689 F.2d 693, 698-701 (7th Cir. 1982); Dos Santos v. Columbus-Cuneo-Cabrini Medical Center, 684 F.2d 1346, 1349-52 (7th Cir. 1982).

Our discussion of the standard for ruling on requests for preliminary injunctions, and of the standard for appellate review of such rulings, should have made clear that it is not possible to reconcile all the precedents, or even just all the ones in this circuit. But the apparent discord is mostly verbal. Beneath the welter of apparently conflicting precedents we sense agreement on the following principles:

1. In every case in which the plaintiff wants a preliminary injunction he must show that he has "no adequate remedy at law," and (unless the statute under which he is suing excuses a showing of irreparable harm, as in Illinois Bell Tel. Co. v. Illinois Commerce Comm'n, supra, 740 F.2d at 571) that he will suffer "irreparable harm" if the preliminary injunction is not granted. The absence of an adequate remedy at law is a precondition to any form of equitable relief. The requirement of irreparable harm is needed to take care of the case where although the ultimate relief that the plaintiff is seeking is equitable, implying that he has no adequate remedy at law, he can easily wait till the end of trial to get that relief. (On the distinction between "no adequate remedy at law" and "irreparable harm" see Fiss Rendleman, Injunctions 59 (2d ed. 1984).) Only if he will suffer irreparable harm in the interim — that is, harm that cannot be prevented or fully rectified by the final judgment after trial — can he get a preliminary injunction. Where the only remedy sought at trial is damages, the two requirements — irreparable harm, and no adequate remedy at law — merge. The question is then whether the plaintiff will be made whole if he prevails on the merits and is awarded damages.

2. In saying that the plaintiff must show that an award of damages at the end of trial will be inadequate, we do not mean wholly ineffectual; we mean seriously deficient as a remedy for the harm suffered. See Semmes Motors, Inc. v. Ford Motor Co., 429 F.2d 1197, 1205 (2d Cir. 1970) (Friendly, J.); 11 Wright Miller, supra, § 2944, at p. 396. A damages remedy can be inadequate for any of four reasons:

(a) The damage award may come too late to save the plaintiff's business. He may go broke while waiting, or may have to shut down his business but without declaring bankruptcy. Of course, even if the plaintiff declares bankruptcy, his trustee in bankruptcy can get a damage award; but probably it will not cover all the losses incident to the bankruptcy. The award may be inadequate even if the plaintiff leaves the business without becoming insolvent. As Judge Friendly noted in a case involving the termination of an automobile dealer, "the right to continue a business in which William Semmes had engaged for twenty years and into which his son had recently entered is not measurable entirely in monetary terms; the Semmes want to sell automobiles, not to live on the income from a damages award." Semmes Motors, Inc. v. Ford Motor Co., supra, 429 F.2d at 1205.

(b) The plaintiff may not be able to finance his lawsuit against the defendant without the revenues from his business that the defendant is threatening to destroy. But in an age of contingent-fee contracts this will rarely be a decisive consideration.

(c) Damages may be unobtainable from the defendant because he may become insolvent before a final judgment can be entered and collected. See discussion in Signode Corp. v. Weld-Loc Systems, Inc., 700 F.2d 1108, 1111 (7th Cir. 1983).

(d) The nature of the plaintiff's loss may make damages very difficult to calculate. Consider a loss, but not a crippling loss (that would be case (a)), of business profits. in principle, any profits lost by Roland as a result of being terminated for breach of an implied exclusive-dealing contract can be monetized, and awarded as damages; but in practice it may be very difficult to distinguish the effect of the termination from the effect of other things happening at the same time, and to project that effect into the distant future. On the difficulties encountered in trying to calculate damages for lost profits, see, e.g., Taylor v. Meirick, 712 F.2d 1112, 1119-22 (7th Cir. 1983); Hayes v. Solomon, 597 F.2d 958, 976-77 (5th Cir. 1979); Note, Private Treble Damage Antitrust Suits: Measure of Damages for Destruction of All or Part of a Business, 80 Harv.L.Rev. 1566, 1577-86 (1967).

3. In deciding whether to grant a preliminary injunction, the court must also consider any irreparable harm that the defendant might suffer from the injunction — harm that would not be either cured by the defendant's ultimately prevailing in the trial on the merits or fully compensated by the injunction bond that Rule 65(c) of the Federal Rules of Civil Procedure requires the district court to make the plaintiff post. The cases do not usually speak of the defendant's irreparable harm, but the qualification is implicit; if the defendant will not be irreversibly injured by the injunction because a final judgment in his favor would make him whole, the injunction will not really harm him. But since the defendant may suffer irreparable harm from the entry of a preliminary injunction, the court must not only determine that the plaintiff will suffer irreparable harm if the preliminary injunction is denied — a threshold requirement for granting a preliminary injunction — but also weight that harm against any irreparable harm that the defendant can show he will suffer if the injunction is granted.

4. Besides showing that he has no adequate remedy at law and that he will suffer irreparable harm unless the preliminary injunction is granted, the plaintiff has another threshold to cross: that of showing some likelihood of succeeding on the merits. It is not enough that the failure to get the injunction will be a disaster for him whereas the injunction would be only a minor inconvenience to the defendant. Equity jurisdiction exists only to remedy legal wrongs; without some showing of a probable right there is no basis for invoking it. But although unwilling therefore to go as far as the Ninth Circuit in Costandi v. AAMCO Automatic Transmissions, Inc., 456 F.2d 941, 943 (9th Cir. 1972) (per curiam), which upheld the grant of a preliminary injunction without any consideration at all of the probable outcome of the trial, we agree that the threshold is low. It is enough that "the plaintiff's chances are better than negligible. . . ." Omega Satellite Products Co. v. City of Indianapolis, supra, 694 F.2d at 123. See also Washington Metropolitan Area Transit Comm'n v. Holiday Tours, Inc., 559 F.2d 841, 842 n. 1, 842-44 (D.C. Cir. 1977) ("substantial case on the merits" good enough for a stay pending appeal (analogous to a preliminary injunction), even though "ultimate success by the movant is [not] a mathematical probability," and the court's "own approach may be contrary to movant's view of the merits"); West Virginia Highlands Conservancy v. Island Creek Coal Co., 441 F.2d 232, 235 (4th Cir. 1971) ("probable right," used interchangeably with "substantial issues"). And see the excellent discussion in Comment, Probability of Ultimate Success Held Unnecessary for Grant of Interlocutory Injunction, 71 Colum.L.Rev. 165 (1971).

5. If the plaintiff does show some likelihood of success, the court must then determine how likely that success is, because this affects the balance of relative harms (point 3 above). The more likely the plaintiff is to win, the less heavily need the balance of harms weigh in his favor; the less likely he is to win, the more need it weigh in his favor. This is a most important principle, and one well supported by cases in this and other circuits, and by scholarly commentary. See Hyatt Corp. v. Hyatt Legal Services, 736 F.2d 1153, 1159 (7th Cir. 1984); Omega Satellite Products Co. v. City of Indianapolis, supra, 694 F.2d at 123; American Hospital Ass'n v. Harris, 625 F.2d 1328, 1331 (7th Cir. 1980) ("The required showing of probability of success on the merits `varies with the quality and quantum of harm that [the moving party] will suffer from the denial of an injunction,'" quoting District 50, United Mine Workers of America v. International Union, United Mine Workers of America, 412 F.2d 165, 168 (D.C. Cir. 1969)); Dan River, Inc. v. Icahn, 701 F.2d 278, 283 (4th Cir. 1983), and cases cited there; Roth v. Bank of the Commonwealth, 583 F.2d 527, 538 (6th Cir. 1978); Washington Metropolitan Area Transit Comm'n v. Holiday Tours, Inc., supra, 559 F.2d at 843-44; Canal Authority v. Callaway, 489 F.2d 567, 576-77 (5th Cir. 1974); Virginia Petroleum Jobbers Ass'n v. FPC, 259 F.2d 921, 925 (D.C. Cir. 1958) (per curiam) (stay pending appeal) ("injury held insufficient to justify a stay in one case may well be sufficient to justify it in another, where the applicant has demonstrated a higher probability of success on the merits"); 11 Wright Miller, supra, § 2948, at pp. 453-55; Developments in the Law, Injunctions, 78 Harv.L.Rev. 994, 1056 (1965) ("Clear evidence of irreparable injury should result in a less stringent requirement of certainty of victory; greater certainty of victory should result in a less stringent requirement of proof of irreparable injury") (footnote omitted).

A variant of this "sliding scale" approach is illustrated by Charlie's Girls, Inc. v. Revlon, Inc., 483 F.2d 953, 954 (2d Cir. 1973) (per curiam). "One moving for a preliminary injunction assumes the burden of demonstrating either a combination of probable success and the possibility of irreparable injury or that serious questions are raised and the balance of hardships tips sharply in his favor." See also William Inglis Sons Baking Co. v. ITT Continental Baking Co., 526 F.2d 86, 88 (9th Cir. 1975). This approach and the "sliding scale" approach are treated as identical in American Hospital Ass'n v. Harris, supra, 625 F.2d at 1331.

The idea underlying these equivalent approaches is that the task for the district judge in deciding whether to grant or deny a motion for preliminary injunction is to minimize errors: the error of denying an injunction to one who will in fact (though no one can know this for sure) go on to win the case on the merits, and the error of granting an injunction to one who will go on to lose. The judge must try to avoid the error that is more costly in the circumstances. That cost is a function of the gravity of the error if it occurs and the probability that it will occur. The error of denying an injunction to someone whose legal rights have in fact been infringed is thus more costly the greater the magnitude of the harm that the plaintiff will incur from the denial and the greater the probability that his legal rights really have been infringed. And similarly the error of granting an injunction to someone whose legal rights will turn out not to have been infringed is more costly the greater the magnitude of the harm to the defendant from the injunction and the smaller the likelihood that the plaintiff's rights really have been infringed.

6. Sometimes an order granting or denying a preliminary injunction will have consequences beyond the immediate parties. If so, those interests — the "public interest" if you will — must be reckoned into the weighing process just described. See Yakus v. United States, 321 U.S. 414, 440-41, 64 S.Ct. 660, 674-75, 88 L.Ed. 834 (1944); Delaware River Port Authority v. Transamerican Trailer Transport, Inc., 501 F.2d 917, 924 (3d Cir. 1974).

7. The application of the above principles requires the seasoned judgment of the district court — what is usually called the court's "discretion," although the use of the term in the present context has been criticized. "Perhaps the most important area where parroting the discretion phrase is likely to lead to wrong decision is the review of the grant of denial of preliminary injunctions." Friendly, Indiscretion About Discretion, 31 Emory L.J. 747, 773 (1982). As so often in law the trouble comes from using the same word in different senses. When we speak of an issue as being committed to the district court's discretion, intending to imply by this that the scope of appellate review should be extremely limited, we generally have reference to one of three different (but closely related, and sometimes identical) types of issue: one not decided according to a standard; one not susceptible of uniform treatment across cases; one not decided on the basis of evidence reasonably accessible to the appellate court through the record of the proceedings in the trial court. See Mosey Mfg. Co. v. NLRB, 701 F.2d 610, 615 (7th Cir. 1983) (en banc); United States v. McCoy, 517 F.2d 41, 44 (7th Cir. 1975); United States v. Criden, 648 F.2d 814, 817-18 (3d Cir. 1981); Noonan v. Cunard S.S. Co., 375 F.2d 69, 71 (2d Cir. 1967) (Friendly, J.). Criminal sentences, rulings on most discovery questions, and rulings on evidentiary questions where the issue is prejudicial impact on the jury illustrate the three types of discretionary judgment. Their common characteristic is that either the need for or the feasibility of appellate review is very limited. The trial judge is supposed to tailor a criminal sentence to the particular characteristics of the defendant (including his demeanor) and may weight the various characteristics as he pleases; a proper assessment of jury prejudice requires first-hand observation of the jurors; discovery issues frequently call for managerial rather than juristic judgments. All are determinations where appellate review is limited to deciding whether the district judge abused his discretion — or even, as in the review of a criminal sentence, whether he exercised discretion; if he did, and the sentence is within statutory limits, the court of appeals cannot reverse. See United States v. Ely, 719 F.2d 902, 906-07 (7th Cir. 1983), and cases cited there.

Rulings on requests for preliminary injunctions, it may be argued, do not have the relevant characteristics of these examples. As in this case, usually they are made on a trial-type record. Many preliminary-injunction hearings take as long as the average federal trial, which lasts only three days. (Estimated from Tables C8 and C9 in the 1983 Annual Report of the Director of the Administrative Office of the U.S. Courts.) The hearing here lasted two and a half days. Moreover, the analytical procedure for deciding whether to grant or deny a motion for preliminary injunction, sketched earlier, constitutes a true legal standard. It does not call for a discretionary — in the sense of standardless or intuitive — judgment, or for a consideration of numerous factors none enjoying any particular weight, or for an evaluation of factors inaccessible to a reviewing court (there may of course be credibility issues, but this is no more likely than in the mine run of federal civil cases). The factors to be considered are few and definite; they are as we have said spread out on a record; and they are to be compared in a particular sequence and in accordance with a specific formula which requires first deciding whether the plaintiff has crossed specified thresholds and then weighting the parties' likely harms from the grant and denial of the preliminary judgment, respectively, by the strength of the plaintiff's case.

Although there is a sense in which equitable relief is inherently discretionary because historically, and still to a large extent, there is no absolute right to an equitable remedy, see Weinberger v. Romero-Barcelo, 456 U.S. 305, 311-13, 102 S.Ct. 1798, 1802-03, 72 L.Ed.2d 91 (1982); Yakus v. United States, supra, 321 U.S. at 440, 64 S.Ct. at 674; Meredith v. City of Winter Haven, 320 U.S. 228, 235, 64 S.Ct. 7, 11, 88 L.Ed. 9 (1943); Wisconsin v. Weinberger, 745 F.2d 412, 425 (7th Cir. 1984), this is a different meaning of discretion, one that actually cuts against a highly deferential standard of review when as in this case the district court grants the preliminary injunction. "[T]he granting of a preliminary injunction is an exercise of a very far-reaching power, never to be indulged in except in a case clearly demanding it." Warner Bros. Pictures, Inc. v. Gittone, 110 F.2d 292, 293 (3d Cir. 1940) (per curiam), quoted in Buffalo Courier-Express v. Buffalo Evening News, supra, 601 F.2d at 59; see also 11 Wright Miller, supra, § 2942, at p. 368, and cases cited there; and the Milsen, Shaffer, and Singer cases cited earlier in this opinion. The exercise of a power so far-reaching ought to be subject to effective, and not merely perfunctory, appellate review. "Congress would scarcely have made orders granting or refusing temporary injunctions an exception to the general requirement of finality . . . if it intended appellate courts to be mere rubber-stamps save for the rare cases when a district judge has misunderstood the law or transcended the bounds of reason." Omega Importing Corp. v. Petri-Kine Camera Co., supra, 451 F.2d at 1197.

These considerations may explain why, as we pointed out earlier, many cases that recite "abuse of discretion" as the standard of review actually review the district judge's order granting or denying a preliminary injunction as if the standard were clear error (both for pure factfindings and for the district judge's application to those findings of the principles for granting or denying preliminary injunctions) or mere error (for pure legal determinations). In these cases, the words "abuse of discretion" are not being used literally, to refer only to cases where the ruling being reviewed is a discretionary one. It seems, therefore, that the words describe a range of standards (see, e.g., United States v. Criden, supra, 648 F.2d at 817), in which the amount of deference given the trial judge's rulings varies with the nature of the ruling — a range that overlaps clear error, and even error, period. See Friendly, supra, at 764, 784; Rosenberg, Judicial Discretion of the Trial Court, Viewed from Above, 22 Syracuse L.Rev. 635, 650-53 (1971); the cases cited in these two articles; and the Pearson, Clemons, and Omega Importing cases cited earlier in this opinion. The other end of the range is described in Delno v. Market St. Ry., 124 F.2d 965, 967 (9th Cir. 1942): "Discretion . . . is abused when the judicial action is arbitrary, fanciful or unreasonable, which is another way of saying that discretion is abused only where no reasonable man would take the view adopted by the trial court." See also Deitchman v. E.R. Squibb Sons, Inc., 740 F.2d 556, 563 (7th Cir. 1984); Harrington v. DeVito, 656 F.2d 264, 269 (7th Cir. 1981). This may be the appropriate formula when reviewing a determination as to whether a discovery order should have been granted, or whether an objection to a question asked a witness should have been sustained as unduly prejudicing a party in the eyes of the jury. But it does not follow that it is the appropriate formula for reviewing orders granting or denying preliminary injunctions.

Where in this range should appellate review of a decision granting (or denying) a preliminary injunction be located? We have mentioned the factors pushing review toward the upper end of the range, where abuse of discretion merges with clear error (or even mere) error. Other factors push in the other direction. Judge Friendly himself, in his Omega Importing opinion from which we quoted earlier, stated that the court of appeals in reviewing such a decision not only must accept the district court's credibility findings, and give due regard to its (other) factfindings, but also must "give proper regard to . . . its `feel' of the case." 451 F.2d at 1197. The processing of an application for a preliminary injunction often requires the district judge to negotiate informally but intensively with the parties over such matters as the scheduling of discovery and the scope and time of the hearing, and these negotiations may give the judge a sense for the proper disposition of the application for a preliminary injunction that the appellate court will not be able to obtain from the appeal papers. And we have already noted the presence of imponderable factors when the judge is balancing the harms to the parties from granting or denying the preliminary injunction. Although, considering the nature of the judge's determination and the stakes to the parties, we do not think the term abuse of discretion can be limited to cases where the judge can be said to have acted irrationally or fancifully ( Delno), we also do not think, considering the imponderable character of the balancing process and the judge's superior feel for the issues which a cold transcript may not fully communicate to the reviewing court, that we are entitled to substitute our judgment for the district judge's. See, e.g. Wesley-Jessen Division v. Bausch Lomb Inc., 698 F.2d 862, 864 (7th Cir. 1983). The question for us is whether the judge exceeded the bounds of permissible choice in the circumstances, not what we would have done if we had been in his shoes.

We need not be any more precise than this about the meaning of "abuse of discretion" in the context of appellate review of orders granting or denying preliminary injunctions in order to decide this case. (Nor need we decide whether the denial of a preliminary injunction should be reviewed more deferentially than a grant, on the theory that all equitable relief is in some — but perhaps largely historical — sense discretionary with the trial judge; nor whether the standard of review should be different when the district judge grants or denies a preliminary injunction without having held an evidentiary hearing.) For as we are about to see, the district judge committed a clear factual error and a legal error that require us to set aside his conclusion on the balance of harms, and another legal error with respect to the probable success of Roland on the merits of its antitrust claim.

Applying to the present case the general approach to the analysis of preliminary-injunction cases that we have outlined, we begin by noting that Roland has satisfied the threshold requirements of proving that it has no adequate remedy at law and that it will suffer irreparable harm if preliminary injective relief is denied it. Of course it can if it wins at trial get a damage award that will, in principle, compensate it for any interim losses. But the nature of the losses that it fears is such as to make the calculation of these damages a difficult and not very satisfactory process — better than nothing, but not so sure a method of rectifying the wrong to Roland (if there was a wrong) that we can be confident that none of the harm to it in the interim will be irreparable harm. But we think the district judge clearly erred in finding that without the preliminary injunction Roland probably would go out of business. So far as we can tell, although half of Roland's revenues are derived directly or indirectly from the sale of Dresser construction equipment only 10 percent come from sales of new equipment, which would be cut off if the dealership were cancelled. Roland's revenues from renting Dresser equipment should not be affected by the cancellation of the franchise, at least until its present stock wears out, and that should not be till after the trial is over. Roland's parts business (and therefore its maintenance business as well) will, however, be affected, by the difference between the cost of parts when bought directly from the manufacturer (as Roland now does) and the cost of parts bought from another dealer. The record indicates that the second cost is 20 percent higher than the first. Although Roland sells parts to its customers at an average mark-up of 27 or 28 percent over the cost of buying parts from Dresser, the difference between that mark-up and the 20 percent mark-up that it will have to pay to get the same parts from a dealer might well be too little to cover Roland's selling expenses. But Roland is selling service as well as parts, and its labor costs will not be affected by its not being able to buy parts directly from Dresser.

The greatest weakness in Roland's proof of business losses is that Roland projected its sales of Komatsu equipment to continue at the same, naturally very low, level at which it sold that equipment in the first year of its Komatsu dealership. This is unrealistic. Surely Roland expects to do better in its second and subsequent years. In sum, the loss of the Dresser distributorship will be painful, but it will not be fatal. And bear in mind that we are concerned only with the period between now and the entry of a final judgment; it is only these interim losses that are relevant in deciding whether Roland is entitled to a preliminary injunction.

Turning to the harm to Dresser from the grant of the preliminary injunction, we think the district judge was too confident of his ability to ward off that harm by requiring as a condition of the injunction that Roland maintain Dresser's market share. The judge was right that Dresser faces the prospect of a substantial loss of goodwill if, while this litigation is wending its way to completion, Roland curries favor with Komatsu (with whom Roland's long-range future as a dealer in construction equipment probably lies) by selling Komatsu equipment in preference to Dresser equipment. This is the danger that called forth the restriction that the judge imposed. But a federal judge, however able and experienced, is in no position to supervise a dealership in construction equipment, as the judge undertook to do by requiring that Roland maintain, within "normal economic fluctuations," Dresser's "approximate" market share in central Illinois. The decree may protect Dresser to some extent — but in doing so it will harm competition between Dresser and Komatsu. And it cannot protect Dresser fully. Dresser "will be frozen into an intimate and continuous relationship with a dealer it no longer wishes to be associated with. . . . One of the reasons the contract was of so short duration and subject to such short cancellation notice may be assumed to be that one of the undoubted rights a manufacturer still has is to use its own judgment with respect to those whom it wishes to appoint as its dealers. To bear this burden for an indefinite number of years is a great hardship." Jack Kahn Music Co. v. Baldwin Piano Organ Co., 604 F.2d 755, 764 (2d Cir. 1979).

Ordinarily the trial judge's balancing of the harms to the parties would be entitled to substantial deference by the reviewing court, in accordance with our earlier discussion of the scope of appellate review — but not if the balancing is infected by a clear error of fact, or an error of law. The judge in this case made a clearly erroneous finding of fact when he found that Roland probably would go out of business if the injunction was not issued; and he committed an error of law when he failed to consider the possible impact on competition of the provision of the injunction that freezes Dresser's market share until the end of the lawsuit.

Once we set aside the erroneous aspects of the district judge's analysis of the balance of harms, rejecting his finding that Roland probably would go out of business, and his attempt to freeze Dresser's share of Roland's business, we are left with no certain sense of where the balance of harms lies. We cannot say that without the restriction we have disapproved, Dresser will be harmed more if the preliminary injunction is kept in force until the case is tried (no trial date has been set) and judgment entered than Roland will be harmed if the injunction is dissolved. But as there is no clear balance of hardships in favor of the injunction, Roland cannot prevail merely by showing that it has raised a substantial question regarding the legality of Dresser's action in terminating it. It must show that it is more likely than not to win. For only then would the error of denying Roland a preliminary injunction should it later win on the merits be more costly than the error of granting it the injunction should it later lose on the merits.

In order to prevail on its section 3 claim, Roland will have to show both that there was an agreement, though not necessarily an explicit agreement (see Tire Sales Corp. v. Cities Service Oil Co., 637 F.2d 467, 474 (7th Cir. 1980); McElhenney Co. v. Western Auto Supply Co., 269 F.2d 332, 338 (4th Cir. 1959)), between it and Dresser that it not carry a line of construction equipment competitive with Dresser's, and that the agreement was likely to have a substantial though not necessarily an immediate anticompetitive effect. Regarding the first of these required showings, the record of the preliminary-injunction proceeding contains no evidence that either Roland or any other Dresser dealer agreed with Dresser not to carry a competing manufacturer's line. Nothing in the dealership agreement even hints at a requirement of exclusive dealing, and the fact that after signing the agreement with Dresser, Roland applied for a Komatsu dealership is evidence that Roland itself did not think it had made an implied commitment to exclusive dealing. True, Dresser prefers exclusive dealers — so much so as to be willing to terminate its only dealer in a large marketing area. The district judge believed that evidence of this preference, coupled with the absence of any reason for Dresser's having terminated Roland other than Roland's having taken on an additional line of construction equipment, established a prima facie case of agreement. But an agreement requires a meeting of minds, and there is no evidence that Roland ever thought itself bound to carry only the Dresser line. Indeed, at argument Roland disclaimed any knowledge of what it describes as the implied exclusive-dealing term in the contract; it called it a "secret" term, echoing the district judge's description of exclusive dealing as something "in the mind of" Dresser. One mind is not enough for a meeting of minds. The fact that Dresser was hostile to dealers who would not live and die by its product (as the district judge put it), and acted on its hostility by canceling a dealer who did the thing to which it was hostile, does not establish an agreement, but if anything the opposite: a failure to agree on a point critical to one of the parties. See Turner, The Definition of Agreement Under the Sherman Act: Conscious Parallelism and Refusals to Deal, 75 Harv.L. Rev. 655, 686 (1962); Filco v. Amana Refrigeration, Inc., 709 F.2d 1257, 1266-67 (9th Cir. 1983).

Actually, it is not important whether Dresser's antipathy to nonexclusive dealing was secret. Assume that Dresser made clear to Roland and its other dealers that it wanted only exclusive dealers and would exercise its contract right to terminate, immediately and without cause, any dealer who took on a competing line. The mere announcement of such a policy, and the carrying out of it by canceling Roland or any other noncomplying dealer, would not establish an agreement. See, e.g., Dillon Materials Handling, Inc. v. Albion Industries, 567 F.2d 1299, 1306 (5th Cir. 1978). It would be a classic example of the conduct permitted by United States v. Colgate Co., 250 U.S. 300, 305-06, 39 S.Ct. 465, 467, 63 L.Ed. 992 (1919), a decision whose continued vitality despite much criticism is attested to by the Supreme Court's approving reference to it in Monsanto Co. v. Spray-Rite Service Corp., ___ U.S. ___, 104 S.Ct. 1464, 1469, 79 L.Ed.2d 775 (1984). See also Russell Stover Candies, Inc. v. FTC, 718 F.2d 256, 260 (8th Cir. 1983). True, there is some evidence that Dresser, fearing the competition of Komatsu, made efforts to discover which of its dealers were considering dealing with Komatsu. Surveillance could in some circumstances intimidate — could be the stick that forced dealers into a tacit understanding not to handle a competitor's goods. See Yentsch v. Texaco, Inc., 630 F.2d 46, 54 (2d Cir. 1980). But there is no evidence that it was such here, when Roland within months after signing a dealership agreement with Dresser went and signed another one with Komatsu. The district judge did not rely on evidence of surveillance in concluding that there was an implicit exclusive-dealing agreement.

Dresser's preference for exclusive dealers, its efforts to find out whether its dealers were exclusive dealers, and its terminating Roland when it found out that Roland no longer was its exclusive dealer do not support an inference "both that the distributor communicated its acquiescence or agreement" to exclusive dealing "and that this was sought by the manufacturer." Monsanto Co. v. Spray-Rite Service Corp., supra, 104 S.Ct. at 1471 n. 9. But even if Roland can prove at trial that there was an exclusive-dealing agreement, it will have grave difficulty — we infer from this record — in proving that the agreement is anticompetitive. The objection to exclusive-dealing agreements is that they deny outlets to a competitor during the term of the agreement. At one time it was thought that this effect alone would condemn exclusive-dealing agreements under section 3 of the Clayton Act, provided that the agreements covered a large fraction of the market. See Standard Oil Co. v. United States, 337 U.S. 293, 314, 69 S.Ct. 1051, 1062, 93 L.Ed. 1371 (1949) (" Standard Stations"). Although the Supreme Court has not decided an exclusive-dealing case in many years, it now appears most unlikely that such agreements, whether challenged under section 3 of the Clayton Act or section 1 of the Sherman Act, will be judged by the simple and strict test of Standard Stations. They will be judged under the Rule of Reason, and thus condemned only if found to restrain trade unreasonably. See, e.g., Jefferson Parish Hospital District No. 2 v. Hyde, ___ U.S. ___, 104 S.Ct. 1551, 1575-76, 80 L.Ed.2d 2 (1984) (concurring opinion); Tampa Elec. Co. v. Nashville Coal Co., 365 U.S. 320, 333-35, 81 S.Ct. 623, 631-32, 5 L.Ed.2d 580 (1961); American Key Corp. v. Cumberland Associates, 579 F. Supp. 1245, 1256-57 (N.D. Ga. 1983).

The exclusion of one or even several competitors, for a short time or even a long time, is not ipso facto unreasonable. The welfare of a particular competitor who may be hurt as the result of some trade practice is the concern not of the federal antitrust laws, see Copperweld Corp. v. Independence Tube Corp., ___ U.S. ___, 104 S.Ct. 2731, 2740 n. 14, 81 L.Ed.2d 628 (1984), but of state unfair competition law, see Sutliff, Inc. v. Donovan Cos., 727 F.2d 648, 655 (7th Cir. 1984), which is not involved in this appeal. The competitor in question, Komatsu, is not even a party to this case.

The exclusion of competitors is cause for antitrust concern only if it impairs the health of the competitive process itself. See Products Liability Ins. Agency, Inc. v. Crum Forster Ins. Cos., 682 F.2d 660, 663-65 (7th Cir. 1982). Hence a plaintiff must prove two things to show that an exclusive-dealing agreement is unreasonable. First, he must prove that it is likely to keep at least one significant competitor of the defendant from doing business in a relevant market. If there is no exclusion of a significant competitor, the agreement cannot possibly harm competition. Second, he must prove that the probable (not certain) effect of the exclusion will be to raise prices above (and therefore reduce output below) the competitive level, or otherwise injure competition; he must show in other words that the anticompetitive effects (if any) of the exclusion outweigh any benefits to competition from it.

Roland has as yet made very little effort to establish either of these two things. On the present record it appears that Komatsu cannot be kept out of the central Illinois market even if every manufacturer of construction equipment prefers exclusive dealers and will cancel any dealer who switches to the Komatsu line. Komatsu is the second largest manufacturer of construction equipment in the world. Its total sales of such equipment are four times as great as Dresser's. Already it is a major factor in the U.S. construction-equipment market; in some items it outsells Dresser. The nationwide practice of exclusive dealing has not kept Komatsu from becoming a major factor in the U.S. market, apparently in a short period of time. The reason is evident. Since dealership agreements in this industry are terminable by either party on short notice, Komatsu, to obtain its own exclusive dealer in some area, has only to offer a better deal to some other manufacturer's dealer in the area. It need not fear being sued for interference with contract; Roland would not be breaking its contract with Dresser if it gave Dresser the heave-ho, provided it gave 90 days' notice. Maybe if Roland had known that it would be cut off by Dresser as soon as it was, it would have demanded some guarantees from Komatsu to tide it over the period of transition when Komatsu was not yet as well established a name in central Illinois as Dresser (though Dresser itself was in a sense new to the market); and probably Komatsu would have given Roland these guarantees to get a foothold in the central Illinois market. The likeliest consequence of our dissolving the preliminary injunction would be to accelerate Komatsu's efforts to promote its brand through the Roland dealership.

Admittedly this analysis may exaggerate the smoothness with which the competitive process operates. Knowing that it cannot move gradually into central Illinois by persuading dealers to carry its line as a second line, Komatsu may expand more slowly than it would otherwise have done, and at somewhat higher cost. And since the national market in construction equipment appears to be highly concentrated (although the record is scanty in this regard, particularly in its omission of any data on foreign sales, which may conceivably be part of the U.S. market, properly defined), any impediments to new competition may harm consumers by keeping prices at noncompetitive levels — though whether the industry at present is or is not highly competitive must be a matter of conjecture on this record. But with all this conceded we still cannot agree that Roland has shown a substantial anticompetitive effect, actual or potential, from the alleged exclusive-dealing agreement, when we reflect on Komatsu's strength and on the fact that neither Dresser nor, it appears, any other manufacturer has long-term exclusive-dealing contracts. Exclusive-dealing contracts terminable in less than a year are presumptively lawful under section 3. See ABA Antitrust Section, Antitrust Law Developments 98 (2d ed. 1984); Sullivan, Handbook of the Law of Antitrust 485-86 (1977); cf. Marvel, Exclusive Dealing, 25 J. Law Econ. 1, 6 (1982). This one was terminable in 90 days. Finally, Komatsu undoubtedly has the resources to establish its own dealership in central Illinois, if it cannot lure away someone else's dealer despite the lack of long-term contracts binding dealers to their existing suppliers.

The calculus of competitive effect must also include some consideration, of the possible competitive benefits of exclusive dealing in this industry. Competition is the allocation of resources in which economic welfare (consumer welfare, to oversimplify slightly) is maximized; it is not rivalry per se, or a particular form of rivalry, or some minimum number of competitors. See Products Liability Ins. Agency, Inc. v. Crum Forster Ins. Cos., supra, 682 F.2d at 663-65. If, as Dresser argues, exclusive dealing leads dealers to promote each manufacturer's brand more vigorously than would be the case under nonexclusive dealing, the quality-adjusted price to the consumer (where quality includes the information and other services that dealers render to their customers) may be lower with exclusive dealing than without, even though a collateral effect of exclusive dealing is to slow the pace at which new brands, such as Komatsu, are introduced. Cf. Beltone Electronics Corp., FTC Complaints and Orders ¶ 21,934 at p. 22,394 (FTC July 6, 1982). The evidence on this point is slim. But it is at least plausible that Dresser, having if we may judge from its operation in central Illinois only one dealer in a large territory, would want that dealer to devote his efforts entirely to selling Dresser's brand. A dealer who expresses his willingness to carry only one manufacturer's brand of a particular product indicates his commitment to pushing that brand; he doesn't have divided loyalties. See Sulmeyer v. Coca-Cola Co., 515 F.2d 835, 840 n. 2 (5th Cir. 1975). If the dealer carries several brands, his stake in the success of each is reduced. Suppose, though there is contrary evidence in the record, that Roland intended to promote Dresser and Komatsu products with equal vigor. It is still the case that if Roland failed to promote Dresser vigorously, it would have Komatsu to fall back on — but Dresser might suffer a drastic decline in the central Illinois market, all of its eggs being in the Roland basket. Exclusive dealing may also enable a manufacturer to prevent dealers from taking a free ride on his efforts (for example, efforts in the form of national advertising) to promote his brand. The dealer who carried competing brands as well might switch customers to a lower-priced substitute on which he got a higher margin, thus defeating the manufacturer's effort to recover the costs of his promotional expenditures by charging the dealer a higher price. See Marvel, supra, at 6-25.

Therefore, even if, in signing on with Komatsu, Roland did not intend to discontinue its sales of the Dresser line eventually and in the meantime to begin phasing Dresser out, Dresser still has a plausible argument that an exclusive dealer would promote its line more effectively than a nonexclusive dealer, and by doing so would increase competition in the market for construction equipment. The argument is no more than plausible; it is supported by very little evidence; it may be wrong. But when we consider how tenuous is the evidence that exclusive dealing in this market will exclude or even significantly retard Komatsu — how tenuous even is the inference that there was an exclusive-dealing agreement — even weak evidence of competitive gains from exclusive dealing must reinforce our conclusion that Roland has failed to show that it is more likely than not to prevail at the trial on the merits.

It should go without saying that although we have concluded that the district judge should not have granted Roland's motion for a preliminary injunction, our discussion of the probable merits of Roland's antitrust claim is tentative. We do not exclude the possibility that on the fuller record made in the trial on the merits Roland will succeed in establishing its claim.

REVERSED.