United States Court of Appeals
United States of America,
Consolidated with 98-5012
A. Douglas Melamed, Deputy Assistant Attorney General, U.S. Department of Justice, argued the cause for appellee. With him on the briefs were Joel I. Klein, Assistant Attorney General, Catherine G. O'Sullivan and Mark S. Popofsky, Attorneys.
Daniel E. Lungren, Attorney General of California, James E. Ryan, Attorney General of Illinois, Carla J. Stovall, Attor-ney General of Kansas, J. Joseph Curran, Jr., Attorney General of Maryland, Scott Harshbarger, Attorney General for the Commonwealth of Massachusetts, Jeremiah W. (Jay) Nixon, Attorney General of Missouri, Joseph P. Mazurek, Attorney General of Montana, Frankie Sue Del Papa, Attorney General of Nevada, Tom Udall, Attorney General of New Mexico, and W.A. Drew Edmondson, Attorney General of Oklahoma, members of the Bar of this Court, were on the brief of certain States as amici curiae, with whom appeared the various other Attorneys General of the participating States.
Before: Wald, Williams and Randolph, Circuit Judges.
Opinion for the Court filed by Circuit Judge Williams.
Opinion concurring in part and dissenting in part filed by Circuit Judge Wald.
Williams, Circuit Judge: The district court entered a preliminary injunction prohibiting Microsoft Corporation from requiring computer manufacturers who license its operating system software to license its internet browser as well. In granting the preliminary injunction the court also referred the government's motion for a permanent injunction to a special master. Microsoft appeals the preliminary injunction and applies for a writ of mandamus revoking the reference to a master. We find that the district court erred procedurally in entering a preliminary injunction without notice to Microsoft and substantively in its implicit construction of the consent decree on which the preliminary injunction rested. We also grant the petition for mandamus and direct the district court to revoke or revise its reference.
This case arises from Microsoft's practices in marketing its Windows 95 operating system. An operating system is, so to speak, the central nervous system of the computer, control-ling the computer's interaction with peripherals such as key-boards and printers. Windows 95 is an operating system that integrates a DOS shell with a graphical user interface, i.e., a technology by which the operator performs functions not by typing at the keyboard but by clicks of his mouse. Operating systems also serve as "platforms" for application software such as word processors. As the word "platform" suggests, the operating system provides a basic support structure for an application via "application programming interfaces" ("APIs"), which provide general functions on which applications can rely. Each operating system's APIs are unique; hence applications tend to be written for particular operating systems. The primary market for operating systems consists of original equipment manufacturers ("OEMs"), which make computers, install operating systems and other software that they have licensed from vendors such as Microsoft, and sell the package to end users. These may be either individual consumers or businesses.
In an earlier opinion, also arising from litigation generated by the Justice Department's 1994 antitrust suit against Microsoft, we briefly described Microsoft's role in the software industry and some of the industry's economics. United States v. Microsoft Corp., 56 F.3d 1448, 1451-52 (D.C. Cir. 1995). Because IBM chose to install Microsoft's operating system on its personal computers, Microsoft acquired an "installed base" on millions of IBM and IBM-compatible PCs. That base constituted an exceptional advantage, and created exceptional risks of monopoly, because of two characteristics of the software industry--increasing returns to scale and network externalities. First, because most of the costs of software lie in the design, marginal production costs are negligible. Production of additional units appears likely to lower average costs indefinitely. (I.e., the average cost curve never turns upward.) Second, an increase in the number of users of a particular item of software increases the number of other people with whom any user can share work. As a result, Microsoft's large installed base increases the incentive for independent software vendors to write compatible applica-tions and thereby increases the value of its operating system to consumers.
The Department's 1994 complaint alleged a variety of anti-competitive practices, chiefly in Microsoft's licensing agreements with OEMs. Along with it, the Department filed a proposed consent decree limiting Microsoft's behavior, the product of negotiations between Microsoft, the Department and European competition authorities. Most relevant here is § IV(E) of the decree:
The Department sees a violation of § IV(E)(i) in Microsoft's marketing of Windows 95 and its web browser, Internet Explorer ("IE").
The Internet is a global network that links smaller networks of computers. The World Wide Web ("the Web") is the fastest-growing part of the Internet, composed of multimedia "pages" written in Hypertext Markup Language ("HTML") and connected to other pages by hypertext links. Browsers enable users to navigate the Web and to access information.
Most browsers are designed according to a "multi-platform" approach, with different versions for each of a variety of different operating systems. Joint Appendix ("J.A.") 81. Browsers also have the potential to serve as user interfaces and as platforms for applications (which could then be written for the APIs of a particular browser rather than of a particular operating system1), providing some of the traditional functions of an operating system. Widespread use of multi-platform browsers as user interfaces has some potential to reduce any monopoly-increasing effects of network externalities in the operating system market. Browsers can enable the user to access applications stored on the Internet or local networks, or to operate applications that are independent of the operating system. J.A. 103-05.
Microsoft has developed successive versions of IE, the first of which was initially released with Windows 95 in July 1995. Microsoft's Windows 95 license agreements have required OEMs to accept and install the software package as sent to them by Microsoft, including IE, and have prohibited OEMs from removing any features or functionality, i.e., capacity to perform functions such as browsing. J.A. 86-89.
The first three versions of IE were actually included on the Windows 95 "master" disk supplied to OEMs. Department Br. at 4; J.A. 1277-78. IE 4.0, by contrast, was initially distributed on a separate CD-ROM and OEMs were not required to install it. Microsoft intended to start requiring OEMs to preinstall IE 4.0 as part of Windows 95 in February 1998. On learning of Microsoft's plans, the Department became concerned that this practice violated § IV(E)(i) by effectively conditioning the license for Windows 95 on the license for IE 4.0, creating (in its view) what antitrust law terms a "tie-in" between the operating system and the browser.2 (It is not clear why extension of Microsoft's established IE policy to IE 4.0 aroused the Department's concern.) It filed a petition seeking to hold Microsoft in civil contempt for its practices with respect to IE 3.0, and requesting "further" that the court explicitly order Microsoft not to employ similar agreements with respect to any version of IE.
A party seeking to hold another in contempt faces a heavy burden, needing to show by "clear and convincing evidence" that the alleged contemnor has violated a "clear and unambiguous" provision of the consent decree. Armstrong v. Executive Office of the President, 1 F.3d 1274, 1289 (D.C. Cir. 1993). Finding § IV(E)(i) ambiguous, the district court denied the Department's contempt petition. But that left open the possibility that Microsoft's practices might in fact violate the consent decree (though not so clearly as to justify contempt), and the district court continued the proceedings in order to answer that question, appointing a special master not only to oversee discovery but also to propose findings of fact and conclusions of law. For the meantime, the court entered a preliminary injunction forbidding Microsoft
A detour is necessary to explore what this injunction meant. In some of its papers before the court the Depart-ment had argued for an order barring Microsoft from "forc-ing OEMs to accept and preinstall the software code" that it separately distributes at retail as IE 3.0. J.A. 996. Micro-soft had responded that a Windows 95 operating system without IE software code simply would not function. The government characterized that assertion as "greatly over-blown." J.A. 1237. The district court, in its justifying memo-randum, referred to the injunction as barring Microsoft from "forcing OEMs to accept and preinstall the software code" separately distributed as IE 3.0, J.A. 1296-97; i.e., it employed the Department's exact words on the subject. After the injunction was issued, Microsoft and the Department had further consultations, at the end of which they entered a stipulation that Microsoft would be in compliance with the injunction if it extended to OEMs the options of (1) running the Add/Remove Programs utility with respect to IE 3.x and (2) removing the IE icon from the desktop and from the Programs list in the Start menu and marking the file IEXPLORE.EXE "hidden." J.A. 1780-81. It appears not to be disputed that these alternate modes of compliance do not remove the IE software code, which indeed continues to play a role in providing non-browser functionality for Windows. In fact, browser functionality itself persists, and can be summoned up either by entering four lines of code or by running any application (such as Quicken) that contains the code necessary to invoke the functionality. J.A. 1649-55. The agreed-upon means of compliance simply enable the OEMs to make user access to IE more difficult.3
Microsoft appealed, as authorized by 28 U.S.C. § 1292(a)(1), and also sought mandamus directing the district court to revoke the reference to the special master.
Microsoft claims at the outset that the district court, after finding no contempt, should simply have dismissed the Department's petition. But although the petition was styled simply as one for an order to show cause "Why Respondent Microsoft Corporation Should Not Be Held in Civil Contempt," its prayer for relief sought not only pure contempt remedies (such as the attention-grabbing request for $1,000,000 a day in damages), but also an order directing Microsoft to cease and desist from requiring "OEMs to license any version of Internet Explorer as an express or implied condition of licensing Windows 95." J.A. 41. This was plainly a request for clarification of the consent decree, pinning down its application to the browser issue. Such a clarification may properly take the form of an injunction. See Brewster v. Dukakis, 675 F.2d 1, 3-4 (1st Cir. 1982). Indeed, as a consent decree contains an injunction already, a clarification naturally acquires the same character. (Of course, if the supplementary language goes beyond the consent decree, it is a modification rather than a clarification, and is governed by different standards. See, e.g., United States v. Western Elec. Co., 894 F.2d 430, 435 (D.C. Cir. 1990) ("Manufacturing Appeal").) Although the framing of this request as part of a remedy for contempt may have been odd, Microsoft does not contest that the proceeding put in controversy the meaning of § IV(E)(i) as applied to its browser technology.
This government request for clarification appeared in its petition shortly after its primary request--that the court adjudge Microsoft to be in contempt--and the word "further." Following "further" are a raft of requests for orders, this being just one. Microsoft says this clearly shows that the request was contingent on a finding of contempt. It further (here, in the sense of "additionally") presses on us some lines from a colloquy between the district court and a Department lawyer during the final hearing (December 5, 1997) before the court made its decision to issue a preliminary injunction:
In fact we think this dialogue suggests either that the Department's request was always in the alternative or that it modified the request to make it such. "I think that is exactly how the Court should read it" comes in response to a suggestion that the petition be read as a request for specific enforcement, and the interest in clarification is presented as the Department's central concern. J.A. 1295. Given the district court's participation in the colloquy, we might be inclined, if necessary, to defer to its understanding of the Department's prayer for relief.
Even if we found that the Department's request was in fact contingent on a finding of contempt, however, we do not think the district court would have erred in clarifying the decree sua sponte as an incident to its denial of the contempt petition. Contempt motions are often accompanied by requests for clarification in the alternative. But they also often elicit declaratory clarifications, and sometimes even amendments, as accompaniments to denials even without (so far as appears) explicit alternative requests for clarification. See, e.g., Wilder v. Bernstein, 49 F.3d 69, 71-72 (2d Cir. 1995); Thermice Corp. v. Vistron Corp., 832 F.2d 248, 250-51 (3d Cir. 1987); Gov't of the Virgin Islands v. Sun Island Car Rentals, Inc., 819 F.2d 430, 431 (3d Cir. 1987); Movie Sys-tems, Inc. v. MAD Minneapolis Audio Distributors, Inc., 717 F.2d 427, 429-30 (8th Cir. 1983); Vertex Distributing, Inc. v. Falcon Foam Plastics, Inc., 689 F.2d 885, 888 n.2, 892 (9th Cir. 1982); Stolberg v. Board of Trustees for State Colleges, 541 F.2d 890, 892 (2d Cir. 1976); Red Ball Int. Demolition Corp. v. Palmadessa, 947 F. Supp. 116, 121 (S.D.N.Y. 1996); Johnson v. Heckler, 604 F. Supp. 1070, 1075-76 (N.D. Ill. 1985).
We are aware of no case raising doubts about the propriety of clarification incident to the denial of a contempt petition. Indeed, at oral argument Microsoft conceded "in principle" the court's authority to continue the proceeding in order to clarify the decree. Transcript at 11.4 Of course, the above cases characteristically did not explicitly affirm the district court's authority, although one did just that. See Vertex, 689 F.2d at 892 ("[T]he district court could properly clarify that ambiguous language, and this it did, requiring defendants to change their future advertising to comply with the consent judgment, as clarified.").
Microsoft points out that the question of district court authority is jurisdictional, so that mere practice may not be enough. But much repeated practice illumines the generally understood meaning of petitions for contempt citations. A court granting a clarification that the parties have not explicitly requested has at most construed the petition to contain an implicit request for declaratory relief. This construction seems altogether reasonable where, as here, the petition clearly puts the meaning of the consent decree in issue and the petition makes the standard request (in the rather typical words of this petition) for "such further orders as the nature of the case may require and as the Court may deem just and proper to compel obedience to and compliance with the orders and decrees of this Court." J.A. 43. Cf. Johnson, 604 F. Supp. at 1075-76 (denying contempt petition but clarifying decree in response to request for further relief under 28 U.S.C. s 2202).
Because it was not error for the court to address the issue of clarification, we must decide whether the preliminary injunction was correctly granted.
Microsoft argues that the district court failed to comply with Federal Rule of Civil Procedure 65(a)(1)'s command, "No preliminary injunction shall be issued without notice to the adverse party." We agree. Obviously the Department's request for a contempt citation provided no such notice, for the governing criteria are completely different. To defeat the contempt petition, all Microsoft had to do was to show that the Department failed to meet its burden of showing that the consent decree unambiguously barred its conduct. For a preliminary injunction, by contrast, traditional equitable standards would require the government to show substantial likelihood of success on the merits (here, that the decree, properly construed, barred the conduct), plus risk of irreparable injury, lack of substantial injury to the opposing party, and consistency with the public interest. See CityFed Financial Corp. v. Office of Thrift Supervision, 58 F.3d 738, 746 (D.C. Cir. 1995). The contempt petition did not alert Microsoft to contest these factors.
Nor could the Department's request for a permanent injunction serve as notice--even putting aside Microsoft's argument that the request was contingent on a situation that never arose (see section II). The request for a permanent injunction amounted to no more than a request for a clarification, and thus would require only a showing that the Department's reading of the consent decree was correct. It did not put into play the equitable factors of interim irreparable injury to the requester, harm to the party to be enjoined, and effects on the public interest. But resolution of these is essential in granting a preliminary injunction; the proponent must make a showing about the interim risks precisely in order to counterbalance the lack of any final ruling in its favor on the merits.
At oral argument the Department claimed that when the government seeks a preliminary injunction under a statute, its showing of a likelihood of success on the merits supplants the normal need for assessment of the interim risks. This is true under some circumstances. First, it is clear that if a statute confers a right to an injunction once a certain showing is made, no plaintiff--neither governmental agencies nor private parties--need show more than the statute specifies. See, e.g., Illinois Bell Telephone Co. v. Illinois Commerce Comm'n, 740 F.2d 566, 571 (7th Cir. 1984) (irreparable injury not required for preliminary injunction under 47 U.S.C. § 401(b), which provides that court "shall enforce" FCC order upon showing of disobedience). The statutory specification displaces the traditional equitable standards, but this displacement is "not lightly assume[d]." Weinberger v. Romero-Barcelo, 456 U.S. 305, 312 (1982). Plaintiffs must show that Congress intended to "intervene and guide or control the exercise of the courts' discretion." Id. at 313. See also Amoco Production Co. v. Village of Gambell, 480 U.S. 531, 541-45 (1987).
Second, and not entirely distinct in the cases, lurks the proposition that when a governmental entity sues to enforce a statute, irreparable injury is presumed to flow from the violation itself. See, e.g., United States v. Diapulse, 457 F.2d 25, 27-28 (2d Cir. 1972) ("The passage of the statute is, in a sense, an implied finding that violations will harm the public and ought, if necessary, [to] be restrained."). It is unclear, however, whether such decisions actually turn on the identity of the plaintiff, or whether they are simply instances where the court read the statute as providing for injunction on a reduced showing and mentioned the enforcing agency because it happened to be the plaintiff before the court. As the cases did not purport to apply the doctrine to all statutes under which a government agency might seek relief, but limited it, for example, to "remedial" ones,5 see, e.g., Commodity Futures Trading Comm'n v. Muller, 570 F.2d 1296, 1300 (5th Cir. 1978), the latter seems more probable.
Some cases antedating Romero-Barcelo took the view that mere statutory authorization of injunctive relief displaced equitable standards, see, e.g., Atchison, Topeka & Santa Fe Railway Co. v. Lennen, 640 F.2d 255, 259 (10th Cir. 1981) (citing cases). Under this view § 4 of the Sherman Act, 15 U.S.C. s 4, authorizing such temporary injunctive relief "as shall be deemed just," but setting no standards for its award, would evidently be enough. But such cases seem outmoded by Romero-Barcelo's view that displacement of the usual equitable standards requires a "clear and valid legislative command." 456 U.S. at 313, quoting Porter v. Warner Holding Co., 328 U.S. 395, 398 (1946). In fact, even before Romero-Barcelo the Court construed s 4's grant as con-trolled by the ordinary principles of equity courts. De Beers Consol. Mines v. United States, 325 U.S. 212, 218-19 (1945).
In any event, the Department's suggestion is irrelevant. The preliminary injunction was entered in a suit to enforce a consent decree, not a statute. As the settlement of a litigation, the decree may require less than the statute under which the suit was brought, or more, United States v. Armour & Co., 402 U.S. 673, 681-82 (1971); Ass'n for Retarded Citizens v. Thorne, 30 F.3d 367, 369 (2d Cir. 1994), so the violation of one is not necessarily a violation of the other. Thus a finding of probable violation of the consent decree could not support a presumption of irreparable harm even under the most extravagant version of the doctrine the gov-ernment invokes.
So the district court did need to find irreparable injury before granting the preliminary injunction. The absence of notice had the effect of precluding the introduction of evidence and argument on this requirement, and the omission was far from trivial. At oral argument the Department conceded that it had offered no evidence on the subject, and Microsoft suggested that it could have forcefully contested anything the Department might have offered, alluding to information--obviously not in the record--that no OEM had, since entry of the preliminary injunction, sought to take advantage of its terms. Transcript at 10, 61.6
The purpose of Rule 65(a)(1)'s notice requirement is to allow the opposing party a fair opportunity to oppose the preliminary injunction, see Weitzman v. Stein, 897 F.2d 653, 657 (2d Cir. 1990), and compliance is mandatory, Parker v. Ryan, 960 F.2d 543, 544 (5th Cir. 1992). Preliminary injunctions entered without notice to the opposing party are generally dissolved. See, e.g., Williams v. McKeithen, 939 F.2d 1100, 1105-06 (5th Cir. 1991); Weitzman, 897 F.2d at 657-68; Phillips v. Chas. Schreiner Bank, 894 F.2d 127, 130-31 (5th Cir. 1990). Appellate courts have, however, on occasion allowed a procedurally flawed injunction to remain in place pending a proper hearing on remand if the equities support such a disposition. See, e.g., Rosen v. Siegel, 106 F.3d 28, 33 (2d Cir. 1997).
The Department urges us to do so here. Evaluating such a request requires the court to consider the traditional equitable factors as apparent on the existing record. See, e.g., Inverness Corp. v. Whitehall Laboratories, 819 F.2d 48, 51 (2d Cir. 1987). We do not believe that a reviewing court must entertain such a request; if the record were so deficient as to make effective evaluation of the equities impossible, a court might do better simply to vacate the injunction as a matter of course--especially where, as here, the injunction was sought only rather obliquely. As later sections will show, however, the record here is enough for us at least to make a reasonable appraisal of the Department's eventual likelihood of success on the merits, and this factor proves dispositive. Silence at this stage would risk considerable waste of litigative resources. "When the district court's estimate of the probability of success depends on an incorrect or mistakenly applied legal premise, 'the appellate court furthers the interest of justice by providing a ruling on the merits to the extent that the matter is ripe, though technically the case is only at the stage of application for preliminary injunction.' " Air Line Pilots Ass'n Int'l v. Eastern Air Lines, Inc., 863 F.2d 891, 895 (D.C. Cir. 1988) (quoting Natural Resources Defense Council, Inc. v. Morton, 458 F.2d 827, 832 (D.C. Cir. 1972). When reviewing preliminary injunctions we have generally not been hesitant to offer interpretation and guidance on the substantive legal issues. See, e.g., Defenders of Wildlife v. Andrus, 627 F.2d 1238, 1243 (D.C. Cir. 1980); Energy Action Educational Found'n v. Andrus, 631 F.2d 751, 761 (D.C. Cir. 1979); Maryland-National Capital Park & Planning Comm'n v. U.S. Postal Service, 487 F.2d 1029, 1041 (D.C. Cir. 1973). We thus turn to the interpretation of § IV(E)(i), on which the merits of the Department's case depend. Our review of a district court's interpretation of a consent decree is de novo.7 See, e.g., Richardson v. Edwards, 127 F.3d 97, 101 (D.C. Cir. 1997); United States v. Western Elec. Co., 12 F.3d 225, 229 (D.C. Cir. 1993); United States v. Western Elec. Co., 900 F.2d 283, 293 (D.C. Cir. 1990) ("Triennial Review ").
Section IV(E) arose from a 1993 complaint filed with the Directorate General IV of the European Union ("DG IV") (the principal competition authority in Europe). Novell, a rival software vendor, alleged that Microsoft was tying its MS-DOS operating system to the graphical user interface provided by Windows 3.11. Before the introduction of Windows 95, which integrated the two, Microsoft marketed the DOS component and the Windows component of the operating system separately, and Windows 3.11 could be operated with other DOS products. But Novell, which marketed a competing DOS product, DR-DOS, complained that by means of specific marketing practices--particularly "per processor and per system licenses," J.A. 754--Microsoft was creating economic incentives for OEMs to preinstall MS-DOS as well as Windows 3.11, thereby using its power in the market for DOS-compatible graphical user interfaces (where it commanded a near 100% market share) to affect OEM choice in the DOS market.8 J.A. 839-48.
During June 1994 negotiations with the Department, Microsoft proposed the possibility of a joint settlement, and representatives of DG IV participated in meetings in Brussels and later in Washington, D.C. On July 15, 1994, the three sides reached agreement and Microsoft and the Department signed a stipulation agreeing to entry of the consent decree, including § IV(E). Both Microsoft and the Department characterize § IV(E) as an "anti-tying" provision.
Microsoft and the Department engage in a brief battle over the extent to which antitrust law may be relevant to this dispute. Without wasting time on the parties' somewhat exaggerated positions, we can simply say that Microsoft is clearly right that the decree does not embody either the entirety of the Sherman Act or even all "tying" law under the Act, and the Department is equally right to point out that the consent decree emerged from antitrust claims, unresolved though they were, so that we must keep procompetitive goals in mind in the interpretive task.
As Armour makes clear, however, an antitrust consent decree cannot be read as though its animating spirit were solely the antitrust laws. "[T]he decree itself cannot be said to have a purpose; rather the parties have purposes, generally opposed to each other, and the resultant decree embodies as much of those opposing purposes as the respective parties have the bargaining power and skill to achieve." 402 U.S. at 681-82.
The court's task, then, is to discern the bargain that the parties struck; this is the sense behind the proposition that consent decrees are to be interpreted as contracts. See, e.g., ITT Continental Baking Co., 420 U.S. at 236-37; Richardson, 127 F.3d at 101; Manufacturing Appeal, 894 F.2d at 1390. To find the meaning of an ambiguous provision we look for the intent of the parties, just as we would with a contract. See Western Elec. Co., 12 F.3d at 231-32 (reading ambiguous provision of consent decree "in light of the parties' jointly intended purpose" (internal quotation omitted)); NRM Corp. v. Hercules, Inc., 758 F.2d 676, 681-82 (D.C. Cir. 1985) (contract interpretation). In that quest we may rely on the same aids to construction as we would when interpreting an ambiguous contract, including "the circumstances surrounding the formation of the consent order." See ITT, 420 U.S. at 238.
Section IV(E)(i) represented the parties' agreed "solution" to the problem posed by the Novell complaint. The practices complained of there, coupled with the decree's explicit acceptance of Windows 95, establish the competing models that guide our resolution of the present dispute. Whatever else s IV(E)(i) does, it must forbid a tie-in between Windows 3.11 and MS-DOS, and it must permit Windows 95. Thus if the relation between Windows 95 and IE is similar to the relation between Windows 3.11 and MS-DOS, the link is presumably barred by § IV(E)(i). On the other hand, a counter-analogy is Windows 95 itself, which the decree explicitly recognizes as a single "product" (it defines it as a "Covered Product," § II(1)(v)), even though, as we have said, Windows 95 com-bines the functionalities of a graphical interface and an oper-ating system. If the Windows 95/IE combination is like the MS-DOS/graphical interface combination that comprises Windows 95 itself, then it must be permissible.
The parties offer us little help in picking the correct analogy. Both propose readings of § IV(E)(i) that fail to reconcile its language with the facts of the Novell complaint and the later permissible release of Windows 95. The Department claims that § IV(E)(i) prohibits Microsoft from bundling together a Covered Product and anything that "Microsoft simultaneously treats" and "antitrust law regards" as "a distinct commercial product." Department Br. at 37-38. It says that the browser-Windows pair is caught in the first filter (Microsoft's treatment of IE as a separate product) because Microsoft provides it separately to end users, sells versions of IE 4 for different operating systems, advertises IE 4, tracks its performance in a "browser market," and distributes it on a separate CD-ROM. J.A. 32-37. For antitrust criteria, the Department draws on Jefferson Parish Hosp. District No. 2 v. Hyde, 466 U.S. 2 (1984), for the proposition that products are distinct for tying purposes if consumer demand exists for each separately. (The Department notes correctly that this does not require demand for one product without the other but simply demand for the two products from different sellers. See id. at 19 & n.30.)
We are not convinced that these indicia necessarily point to separateness, especially those that depend on Microsoft's treatment. Microsoft plausibly characterizes the IE that it provides to end users as an operating system upgrade, as does its rival Netscape, J.A. 589, and the Department offers no means of distinguishing an upgrade from a separate product. Versions developed for different operating systems may be better understood as different products altogether; hence, their relevance to separateness is obscure. Distribution of software code on a separate CD-ROM shows nothing at all about whether the code is integrated into an operating system (software for an operating system that is clearly a single product may take up many disks).
The Department's interpretation of the "integrated products" proviso does nothing to remedy its reading of the body of § IV(e)(i). On the Department's account, the proviso allows Microsoft to incorporate new features into an operating system and offer the package to OEMs--as long as it and antitrust law do not simultaneously treat those features as "a distinct commercial product." Department Br. at 37-38. But these are just the criteria deployed to argue that IE is an "other product"; if the proviso merely reiterates them (to say that what is not an "other product" is "integrated"), it does nothing. And while the Department says that the proviso would protect Microsoft from a charge that it had violated the decree by adopting a technology incompatible with other firms' products, Department Br. at 37 n.17, it is not apparent how § IV(E)(i)'s ban might prohibit such conduct nor how, if it did, the proviso on integration would help it. In short, the Department effectively reads the proviso out of § IV(e)(i).
But the most immediate problem with this reading is that it produces the wrong result on the Novell allegations. In its attempts to define the "product" IE, the Department consistently invokes the concept of "browser functionality." Department Br. at 10; Department Motion for Contempt, J.A. 1317-19; Department Reply Memorandum in Support of Motion for Contempt, J.A. 1424, 1429. But if functionality is the criterion of identity (which the Department asserts so as to claim that the "browser functionality" in Windows 95 is the same product as IE 4 for other operating systems), Windows 95 looks like a tie-in of two products (MS-DOS and Windows 3.11) that were sold separately in the market: it contains the functionalities of both. On the Department's reading, it should thus be prohibited unless Microsoft refrains from marketing MS-DOS separately. There is some suggestion that Microsoft has in fact continued to license MS-DOS separately, at least to end users. Microsoft Reply Br. at 15-16. More significantly, the consent decree does not condition its approval of Windows 95 on Microsoft's marketing behavior with respect to MS-DOS. The failure to produce the right result when applied to Windows 95, one of the situations clearly resolved by the decree, is a fatal flaw.9
The interpretation that Microsoft advances most strongly suffers mirror-image defects: it lacks much logical sense and it fails to fit the decree's setting, the disposition of the Novell complaint. Microsoft stresses § IV(E)(i)'s "integrated products" proviso, saying that the addition of any feature to an operating system, as by simply putting the disk containing a compatible application in the same box with the operating system disk and requiring an OEM to install both, creates an integrated product--unless Microsoft also licenses the feature on a stand-alone basis "in the OEM channel."
This interpretation neatly matches the failure of the Department's theory to account for the permissibility of Windows 95: Microsoft's reading would provide zero relief to Novell, for it would allow Microsoft to bundle MS-DOS with Windows 3.11 as long as it did not license MS-DOS separately to OEMs. In short, the Department's reading does not permit Windows 95, and Microsoft's does not prohibit a bundle of Windows 3.11 and MS-DOS. Neither can be the correct interpretation of a provision that was intended to do both.
Curiously, in both parties' readings Microsoft's behavior determines the permissibility of conditioned licensing. This would be no defect if the behavior were in some way relevant to the economic principles of tie-ins. But it is not. The Department offers no theory as to how a seller's abstaining from separate marketing of the tied good might blunt the possible anticompetitive effects of bundling.10 It seems especially beside the point where the goods are complements used in fixed proportions. A monopolist who ties two such goods has no obvious reason to market the tied good separately: since all buyers of the tying good will also take the tied good, the residual market for the tied good will be minimal. If the concern is that the tie-in makes it more difficult for competitors to enter the market for the tying good (because they must also offer the tied good), see Grappone, Inc. v. Subaru of New England, 858 F.2d 792, 795-96 (1st Cir. 1988) (Breyer, J.), separate marketing of the tied good actually mitigates the posited harm by facilitating new entry into the market for the tying good. Thus both readings allow legitimation by behavior that is either irrelevant or actively harmful.
We think it quite possible, however, to find a construction of § IV(E)(i) that is consistent with the antitrust laws and accomplishes the parties' evident desires on entering the decree. The Department and DG IV were concerned with the alleged anticompetitive effects of tie-ins. Microsoft's goal was to preserve its freedom to design products that consumers would like. Antitrust scholars have long recognized the undesirability of having courts oversee product design, and any dampening of technological innovation would be at cross-purposes with antitrust law. Thus, a simple way to harmonize the parties' desires is to read the integration proviso of § IV(E)(i) as permitting any genuine technological integration, regardless of whether elements of the integrated package are marketed separately.
This reading requires us, of course, to give substantive content to the concept of integration. We think that an "integrated product" is most reasonably understood as a product that combines functionalities (which may also be marketed separately and operated together) in a way that offers advantages unavailable if the functionalities are bought separately and combined by the purchaser.
The point of the test is twofold and may be illustrated by its application to the paradigm case of the Novell complaint and the subsequent release of Windows 95. First, "integration" suggests a degree of unity, something beyond merely placing disks in the same box. If an OEM or end user (referred to generally as "the purchaser") could buy separate products and combine them himself to produce the "integrated product," then the integration looks like a sham. If Microsoft had simply placed the disks for Windows 3.11 and MS-DOS in one package and covered it with a single license agreement, it would have offered purchasers nothing they could not get by buying the separate products and combining them on their own.11
Windows 95, by contrast, unites the two functionalities in a way that purchasers could not; it is not simply a graphical user interface running on top of MS-DOS. Windows 95 is integrated in the sense that the two functionalities--DOS and graphical interface--do not exist separately: the code that is required to produce one also produces the other. Of course one can imagine that code being sold on two different disks, one containing all the code necessary for an operating system, the other with all the code necessary for a graphical interface. But as the code in the two would largely overlap, it would be odd to speak of either containing a discrete functionality. Rather, each would represent a disabled version of Windows 95. The customer could then "repair" each by installing them both on a single computer, but in such a case it would not be meaningful to speak of the customer "combining" two products. Windows 95 is an example of what Professor Areeda calls "physical or technological interlinkage that the customer cannot perform." X Areeda, Antitrust Law s 1746b at 227, 228 (1996).
So the combination offered by the manufacturer must be different from what the purchaser could create from the separate products on his own. The second point is that it must also be better in some respect; there should be some technological value to integration. Manufacturers can stick products together in ways that purchasers cannot without the link serving any purpose but an anticompetitive one. The concept of integration should exclude a case where the manufacturer has done nothing more than to metaphorically "bolt" two products together, as would be true if Windows 95 were artificially rigged to crash if IEXPLORE.EXE were deleted. Cf. ILC Peripherals Leasing Corp. v. International Business Machines Corp., 448 F. Supp. 228, 233 (N.D. Cal. 1978) ("If IBM had simply bolted a disk pack or data module into a drive and sold the two items as a unit for a single price, the 'aggregation' would clearly have been an illegal tying arrangement.") aff'd per curiam sub nom. Memorex Corp. v. International Business Machines Corp., 636 F.2d 1188 (9th Cir. 1980); X Areeda, Antitrust Law p 1746 at 227 (discussing literal bolting). Thus if there is no suggestion that the product is superior to the purchaser's combination in some respect, it cannot be deemed integrated.12
It might seem difficult to put the two elements discussed above together. If purchasers cannot combine the two functionalities to make Windows 95, it might seem that there is nothing to test Windows 95 against in search of the required superiority. But purchasers can combine the functionalities in their stand-alone incarnations. They can install MS-DOS and Windows 3.11. The test for the integration of Windows 95 then comes down to the question of whether its integrated design offers benefits when compared to a purchaser's combination of corresponding stand-alone functionalities. The decree's evident embrace of Windows 95 as a permissible single product can be taken as manifesting the parties' agreement that it met this test.
The short answer is thus that integration may be considered genuine if it is beneficial when compared to a purchaser combination. But we do not propose that in making this inquiry the court should embark on product design assessment. In antitrust law, from which this whole proceeding springs, the courts have recognized the limits of their institutional competence and have on that ground rejected theories of "technological tying." A court's evaluation of a claim of integration must be narrow and deferential.13 As the Fifth Circuit put it, "[S]uch a violation must be limited to those instances where the technological factor tying the hardware to the software has been designed for the purpose of tying the products, rather than to achieve some technologically beneficial result. Any other conclusion would enmesh the courts in a technical inquiry into the justifiability of product innovations." Response of Carolina, Inc. v. Leasco Response, Inc., 537 F.2d 1307, 1330 (5th Cir. 1976).
In fact, Microsoft did, in negotiations, suggest such an understanding of "integrated." In response to the Department and DG IV's statement of concern about tying, it asserted its right to "continue to develop integrated products like [Windows 95] that provide technological benefits to end users." J.A. 756 (emphasis added). Microsoft later withdrew this qualifying phrase, J.A. 760, in order, it claims, to avoid the application of "vague or subjective criteria"--though why the absence of criteria should cure a vagueness problem is unclear. But we do not think that removing the phrase can drain the word "integrated" of all meaning, and we do not accept the suggestion that the Department and DG IV bargained for an "integrated products" proviso so boundless as to swallow § IV(E)(i). Significantly, Microsoft assured the Department and DG IV that the elimination of the qualifying phrase "did not represent a substantive change." J.A. 761.
We believe this understanding is consistent with tying law. The Court in Eastman Kodak Co. v. Image Tech. Servs., 504 U.S. 451 (1992), for example, found parts and service separate products because sufficient consumer demand existed to make separate provision efficient. See id. at 462. But we doubt that it would have subjected a self-repairing copier to the same analysis; i.e., the separate markets for parts and service would not suggest that such an innovation was really a tie-in. (The separate opinion, we take it, makes roughly the same point by its observation about digital cameras. See Sep. Op. at 3-4.) Similarly, Professor Areeda argues that new products integrating functionalities in a useful way should be considered single products regardless of market structure. See X Areeda, Antitrust Law p 1746b.14
We emphasize that this analysis does not require a court to find that an integrated product is superior to its stand-alone rivals. See ILC Peripherals Leasing Corp. v. International Business Machines Corp., 458 F. Supp. 423, 439 (N.D. Cal. 1978) ("Where there is a difference of opinion as to the advantages of two alternatives which can both be defended from an engineering standpoint, the court will not allow itself to be enmeshed 'in a technical inquiry into the justifiability of product innovations.' ") (quoting Leasco, 537 F.2d at 1330), aff'd per curiam sub nom. Memorex Corp. v. IBM Corp., 636 F.2d 1188 (9th Cir. 1980). We do not read § IV(E)(i) to "put[ ] judges and juries in the unwelcome position of designing computers." IX Areeda, Antitrust Law p 1700j at 15. The question is not whether the integration is a net plus but merely whether there is a plausible claim that it brings some advantage. Whether or not this is the appropriate test for antitrust law generally, we believe it is the only sensible reading of § IV(E)(i). On the facts before us, Microsoft has clearly met the burden of ascribing facially plausible benefits to its integrated design as compared to an operating system combined with a stand-alone browser such as Netscape's Navigator.15 Incorporating browsing functionality into the operating system allows applications to avail themselves of that functionality without starting up a separate browser application. J.A. 944, 965.16 Further, components of IE 3.0 and even more IE 4--especially the HTML reader--provide system services not directly related to Web browsing, enhancing the functionality of a wide variety of applications. J.A. 607-22, 1646-48. Finally, IE 4 technologies are used to upgrade some aspects of the operating system unrelated to Web browsing. For example, they are used to let users customize their "Start" menus, making favored applications more readily available. J.A. 490-95; 1662-64. They also make possible "thumbnail" previews of files on the computer's hard drive, using the HTML reader to display a richer view of the files' contents. J.A. 1664-69. Even the Department apparently concedes that integration of functionality into the operating system can bring benefits; responding to a comment on the proposed 1994 consent decree (which the Department published in the Federal Register as required by the Tunney Act), it stated that "a broad injunction against such behavior generally would not be consistent with the public interest." 59 Fed. Reg. 59426, 59428 (Nov. 17, 1994).
The conclusion that integration brings benefits does not end the inquiry we have traced out. It is also necessary that there be some reason Microsoft, rather than the OEMs or end users, must bring the functionalities together. See X Areeda, Antitrust Law p 1746b at 227; p 1747 at 229. Some more subtleties emerge at this stage, parallel to those encountered in determining the integrated status of Windows 95. Microsoft provides OEMs with IE 4 on a separate CD-ROM (a fact to which the Department attaches great significance). It might seem, superficially, that the OEM is just as capable as Microsoft of combining the browser and the operating system.
But the issue is not which firm's employees should run particular disks or CD-ROMs. A program may be provided on three disks--Windows 95 certainly could be--but it is not therefore three programs which the user combines. Software code by its nature is susceptible to division and combination in a way that physical products are not; if the feasibility of installation from multiple disks meant that the customer was doing the combination, no software product could ever count as integrated. The idea that in installing IE 4 an OEM is combining two stand-alone products is defective in the same way that it would be nonsensical to say that an OEM install-ing Windows 95 is itself "combining" DOS functionality and a graphical interface. As the discussion above indicates, IE 3 and IE 4 add to the operating system features that cannot be included without also including browsing functionality. See J.A. 1661-68. Thus, as was the case with Windows 95, the products--the full functionality of the operating system when upgraded by IE 4 and the "browser functionality" of IE 4--do not exist separately.17 This strikes us as an essential point. If the products have no separate existence, it is incorrect to speak of the purchaser combining them. Pur-chasers who end up with the Windows 95/IE package may have installed code from more than one disk; they may have taken the browser out of hiding;18 they may have upgraded their operating system--indeed, Netscape characterizes the installation of IE 4 as "really an OS [operating system] upgrade." J.A. 589. But they have not combined two distinct products.
What, then, counts as the combination that brings together the two functionalities? Since neither fully exists separately, we think the only sensible answer is that the act of combination is the creation of the design that knits the two together. OEMs cannot do this: if Microsoft presented them with an operating system and a stand-alone browser application, rather than with the interpenetrating design of Windows 95 and IE 4, the OEMs could not combine them in the way in which Microsoft has integrated IE 4 into Windows 95. They could not, for example, make the operating system use the browser's HTML reader to provide a richer view of information on the computer's hard drive, J.A. 1665--not without changing the code to create an integrated browser. This reprogramming would be absurdly inefficient. Consequently, it seems clear that there is a reason why the integration must take place at Microsoft's level. This analysis essentially replays our comparison of Windows 95 to a bundle of MS-DOS and Windows 3.11 and concludes that the Windows 95/IE package more closely resembles Windows 95 than it does the bundle. The factual conclusion is, of course, subject to reexamination on a more complete record. On the facts before us, however, we are inclined to conclude that the Windows 95/IE package is a genuine integration; consequently, § IV(E)(i) does not bar Microsoft from offering it as one product.
* * *
A few words with respect to our colleague's separate opinion may clarify our position. Judge Wald suggests that "the prohibition and the proviso could reasonably be construed to state that Microsoft may offer an 'integrated' product to OEMs under one license only if the integrated product achieves synergies great enough to justify Microsoft's extension of its monopoly to an otherwise distinct market." Sep. Op. at 3. We are at a loss to understand how a section that (1) articulates a prohibition and (2) sets a limit on the reach of the prohibition 19 can be read to state a balancing test. Apart from the lack of textual support, we think that a balancing test that requires courts to weigh the "synergies" of an integrated product against the "evidence of distinct markets," Sep. Op. at 5, is not feasible in any predictable or useful way. Courts are ill equipped to evaluate the benefits of high-tech product design,20 and even could they place such an evaluation on one side of the balance, the strength of the "evidence of distinct markets," proposed for the other side of the scale, seems quite incommensurable. Both Jefferson Parish and Eastman Kodak use their "distinct markets" analysis in a binary fashion: markets are distinct or they are not. See 466 U.S. at 21-22; 504 U.S. at 462. If, as the record suggests, Microsoft proposed modification of the integration proviso because of concern about "vague or subjective criteria," J.A. 760, an interpretation requiring courts to weigh evidence that establishes distinctness (or does not) against a sliding scale of net synergistic value looks like the most total transvaluation one can imagine.
Institutional competence may not have been foremost in the parties' minds in drafting the consent decree, and judicial inability to apply a test does not ipso facto mean that the parties did not intend it. But if they did intend a balancing test, they kept that intent well hidden. Nothing in their contemporaneous conduct (or in the conduct of anyone, at any time) suggests that they contemplated a balancing inquiry. Windows 95 was not subjected to any such analysis, though the markets it unified were substantial and obviously distinct. J.A. 790-96. Indeed, one might think that an especially compelling case would be required to "justify Microsoft's extension of its monopoly," Sep. Op. at 3, via Windows 95. Windows 95 leveraged Microsoft's Windows 3.11 market power into the operating system market, where network external-ities are most apparent. See Microsoft, 56 F.3d at 1451. According to the separate opinion, Windows 95 should have required the utmost justification. But nothing suggests that it was analyzed that way, and it seems more likely that it passed muster not because the synergies of its integration outweighed the evidence of distinct markets but because it was, simply, integrated.
The view expressed in the separate opinion seems sure to thwart Microsoft's legitimate desire to continue to integrate products that had been separate--and hence necessarily would have been provided in distinct markets. By its very nature "integration" represents a change from a state of affairs in which products were separate, to one in which they are no longer. By focusing on the historical fact of separate provision, the separate opinion puts a thumb on the scale and requires Microsoft to counterbalance with evidence courts are not equipped to evaluate. We do not think that this makes sense in terms of the text of the consent decree, the evidence of the parties' intents, the values the decree was presumably intended to promote, or the competence of the judiciary.
* * *
At this stage, then, the Department has not shown a reasonable probability of success on the merits. Given this failure, there is no reason to allow the preliminary injunction to remain in effect pending a proper hearing, and we reverse the district court's grant.
Though the proceedings below may continue, they must do so without the preliminary injunction. With respect to the continuation, Microsoft argues that the "blanket" reference to a special master violates Article III, and, in the alternative, that the case does not present the exceptional circumstances required under Federal Rule of Civil Procedure 53(b)21 and La Buy v. Howes Leather Co., 352 U.S. 249 (1957), to justify a nonconsensual reference. Future developments may moot this problem. In view of our interpretation of the consent decree, tentative though it be, see University of Texas v. Camenisch, 451 U.S. 390, 395 (1981); Taylor v. FDIC, 132 F.3d 753, 766 (D.C. Cir. 1997), the Department may well regard further pursuit of the case as unpromising, especially given the alternate avenues developing in its recently launched separate attacks on Microsoft's practices, Nos. 98-1232 and 98-1233. But no such mooting has yet occurred. Accordingly, we address the questions of whether mandamus is available to correct the asserted illegality in the reference and, if so, whether it should issue in this case. We answer yes to both.
The Department claims that mandamus jurisdiction is lack-ing. Its primary theory is evidently that an unlawful refer-ence can be reached by mandamus only if it is so outrageous as to fail to "comport[ ] with Article III." Supplemental Br. of United States 9. But La Buy, the Supreme Court's authoritative pass at the use of mandamus to correct illegal references, rests simply on the presence of "an abuse of [the judge's] power under Rule 53(b)," 352 U.S. at 256, and on that abuse being "clear," id. at 257. To be sure, La Buy observed that the court of appeals there confronted a district court "practice" of freewheeling references to masters: "[T]here is an end of patience." Id. at 258. But we have explicitly rejected the idea that mandamus is available only to correct "persistent disregard" of the limits on references. In re Bituminous Coal Operators' Ass'n, 949 F.2d 1165, 1168 n.4 (D.C. Cir. 1991).
Of course it may be that the boundaries of Rule 53(b) and Article III are close, so that the La Buy Court, in addressing the Rule 53 "exceptional condition[s]" criterion, was actually finding unconstitutional conduct. If true, this helps the De-partment not at all. If mandamus is available only to correct unconstitutional abdications of Article III power, but a clear abuse of discretion under Rule 53 amounts to such an abdication, we may properly exercise mandamus jurisdiction if we find the same sort of clear abuse of discretion that the La Buy Court found.
It is objected that a challenge at the end of the proceeding provides ample remedy, and black-letter mandamus law tells us that such adequacy of remedy defeats mandamus. In re Minister Papandreou, 1998 WL 163561 at *1 (D.C. Cir. 1998) (discussing mandamus criteria). See also Stauble v. Warrob, Inc., 977 F.2d 690, 693 & n.4 (1st Cir. 1992) (dictum to effect that "improvident" reference presents no danger of irrepara-ble harm); In re Thornburgh, 869 F.2d 1503, 1508 (D.C. Cir. 1989) (reference to a master for purposes of determining relief is not an "abdication of the total judicial power" such as to justify mandamus). But the Court's action in La Buy appears necessarily to depend on the view that, at least at some point, even the temporary subjection of a party to a Potemkin jurisdiction so mocks the party's rights as to render end-of-the-line correction inadequate.
On the merits, the Department defends the reference on three grounds. First, it asserts that it falls within the well-recognized category of references for purely remedial deter-minations, such as the accountings and computations of damages, which Rule 53(b) expressly permits. Second, it claims that the technological complexity of the case and need for speed constitute "exceptional condition[s]" under the Rule. And third, it says that the order contains an implicit reservation by the district court of a power of de novo review, and that that unstated reservation saves the order.
In saying that the reference was merely for purposes of supervising remediation the Department invokes the well-established tradition allowing use of special masters to over-see compliance. See generally Apex Fountain Sales, Inc. v. Kleinfeld, 818 F.2d 1089, 1097 (3d Cir. 1987) (citing cases). But this is not such a situation. The issue here is interpretation, not compliance; the parties' rights must be determined, not merely enforced. The matters referred to the master are no more "remedial" than would be those of any total referral of a contract case. The concern about nonconsensual refer- ences turns on the determination of rights, not on a formalistic division of the juridical universe into pre-trial, trial and post-trial. It is for this reason that special masters may not decide dispositive pretrial motions. See In re United States, 816 F.2d 1083, 1090 (6th Cir. 1987).
The Department also proposes that this is a case of such technological complexity as to be "exception[al]." So far as the meaning of the consent decree is concerned, that is clearly false; the words are in plain English, and if their meaning is not clear it is not because of some deep technological issue but because of uncertainties in the purposes of the parties that drafted the decree, the sort of uncertainties that pervade contract actions. As for application of that meaning, we have just stated our interpretation, an interpretation that we regard as grounded, as are the related antitrust doctrines, in a realistic assessment of the institutional limits of the judiciary. Application of our reading of s IV(E)(i) does not require a software expert; it precludes Microsoft from any purely artificial "bolting" of functionalities but is otherwise deferential to entrepreneurs' product design choices.22
More broadly, if La Buy was not complex, with its six defendants and nearly 100 plaintiffs (87 in one suit, six in another), 27 pages of docket entries in one of the cases for preliminary pleas and motions, over 50 depositions, and intricate charges of monopolization and Robinson-Patman violations, 352 U.S. at 251-52, we see no reason to think of this case as especially complicated. In fact, it is very doubtful that complexity tends to legitimate references to a master at all. La Buy noted that the complexity of an antitrust case, rather than justifying a reference, "is an impelling reason for trial before a regular, experienced trial judge rather than before a temporary substitute appointed on an ad hoc basis and ordinarily not experienced in judicial work." Id. at 259.
Thus the only remaining question is whether some implicit district court reservation of a power of de novo review of the special master's report, as to both fact and law, could save the reference. First, we find no such implied reservation. The order itself recites that "the Special Master shall receive evidence and legal authority presented by the parties at such times and places, and in such manner as he shall prescribe, and shall propose findings of fact and conclusions of law for consideration by the Court on the issues raised by this case." J.A. 1301 (emphasis added). The Department suggests that the use of "propose" is somehow helpful, and distinguishes the reference from that in Bituminous Coal, which invited "recommended findings of fact and conclusions of law." 949 F.2d at 1166 (emphasis added). The difference in wording is immaterial, and in any event a special master's findings and conclusions are always advisory. Thus, even if we thought that a reservation of de novo review could save the reference we would have to vacate it, subject to possible re-issuance with an unequivocal commitment to non-deferential review. But we think, in fact, that no such rescue mechanism is available.
Microsoft notes Rule 53(e)(2)'s instruction that in non-jury trials, "the court shall accept the master's findings of fact unless clearly erroneous." (emphasis added). It reads this, as have courts and commentators, as making deferential review mandatory. See, e.g., Williams v. Lane, 851 F.2d 867, 884 (7th Cir. 1988); Apex Fountain, 818 F.2d at 1097; In re Crystal Palace Gambling Hall, 817 F.2d 1361, 1364 (9th Cir. 1987); In re United States, 816 F.2d at 1087, 1088; 9A Charles A. Wright & Arthur R. Miller, Federal Practice & Procedure s 2605 at 670 (1995).
The Department counters with a footnote from Stauble to the effect that a district court may refer fundamental issues of liability if he affords de novo review. 977 F.2d at 698 n.13. But compare id. at 698 n.12 (explaining that, because the parties have agreed that if the reference were constitutionally deficient the remedy would be remand for a full trial before the district court, the court will leave for "another day" the possibility that a master's findings on liability might "perhaps be salvaged, even after appeal, by having the district court conduct a deeper, more participatory sort of review."). At most, however, the thought reflected in the Stauble footnote might save a reference from invalidation under the Constitution; that possibility would do little to make it any less the sort of clear abuse of discretion that is fatal under La Buy. The same is equally true of the observation in United States v. Raddatz, 447 U.S. 667, 683 n.11 (1980), saying that the Court itself, in the exercise of its original Article III jurisdic-tion, acts on the basis of special masters' reports. Clearly that practice tells us nothing about the presence of a clear abuse of discretion under Rule 53(b).
We note some ambiguity on this point in our decision in Bituminous Coal. Our concluding directive told the judge to "revise his order of reference ... and to decide de novo, without deferring to a special master, all potentially dispositive questions of fact or law." 949 F.2d at 1169. The Department thinks that this implicitly endorsed the possibility of a non-deferential reference. We think not. In Bituminous Coal we rejected the coal operators' suggestion that "any unconsented-to reference would constitute an abuse of judicial power," id. (emphasis added), but our rejection of that broad claim was based not on variations in scope of district court review but on our endorsement of the established and legitimate practice of using unconsented-to references for pretrial preparation or for implementation of remedies, id. And in the introductory summary of our disposition we said that the judge must "revise the order of reference to reserve for himself, and not delegate to the special master, the core function of making dispositive rulings, including findings of fact and conclusions of law on issues of liability." Id. at 1166. In short, when we said in Bituminous Coal that we were granting the writ "not because the district judge simply abused his discretion, but because he has no discretion to impose on parties against their will 'a surrogate judge,' " id. at 1168 (citations omitted), we effectively ruled out nonconsensual references in nonjury cases except as to peripheral issues such as discovery and remedy. Compare In re Armco, Inc., 770 F.2d 103, 105 (8th Cir. 1985) ("Beyond matters of account, difficult computation of damages, and unusual discov-ery, it is difficult to conceive of a reference of a nonjury case that will meet the rigid standards of the La Buy decision.") (internal quotations omitted).
In short, finding the case devoid of anything remotely "exceptional" within the meaning of Rule 53(b), we believe the reference to a master must be vacated. We do not, accordingly, reach Microsoft's claims that the specific referee is biased or has conducted the proceedings on referral improperly.
The preliminary injunction was issued without adequate notice and on an erroneous reading of § IV(E)(i) of the consent decree. We accordingly reverse and remand. The reference to the master was in effect the imposition on the parties of a surrogate judge and either a clear abuse of discretion or an exercise of wholly non-existent discretion. We grant mandamus to vacate the reference.