POSNER, Circuit Judge.
We vacated the panel decision, and took the case en banc to decide whether a court of appeals is empowered to dismiss a party in order to retain federal jurisdiction. Previous decisions by this court had answered the question “no,” but the panel disagreed and held that we may [918]*918do this, 832 F.2d 417, 419-20 (7th Cir.1987), and it went on to reverse the district court on the merits. The full court adheres to our previous decisions. “Where the record reveals no jurisdiction, we are powerless to do anything but recognize the defect.” Alderman v. Elgin, Joliet & Eastern Ry., 125 F.2d 971, 973 (7th Cir.1942); see also Carson v. Allied News Co., 511 F.2d 22 (7th Cir.1975). The record reveals no jurisdiction, so we remand the case to the district court — of course without expressing any view on the merits — for such further proceedings as may be consistent with our jurisdictional ruling.
The suit began in 1982 when the plaintiff, Newman-Green, Inc., filed a complaint in the federal district court against five individuals who had guaranteed a debt owed to the plaintiff. The district court had no jurisdiction over the case. The naming of Bettison — a citizen of the U.S. but not of any state — as a defendant had destroyed complete diversity, Sadat v. Mertes, 615 F.2d 1176, 1180 (7th Cir.1980) (per curiam); see 28 U.S.C. § 1332(a); Strawbridge v. Curtiss, 7 U.S. (3 Cranch) 267 (1806); Fidelity & Deposit Co. v. City of Sheboygan Falls, 713 F.2d 1261, 1264 (7th Cir.1983); 13B Wright, Miller & Cooper, Federal Practice and Procedure § 3621 (2d ed. 1984); cf. Currie, The Federal Courts and the American Law Institute, 36 U.Chi.L.Rev. 1, 9-10 (1968), and there was no possible basis of federal jurisdiction except diversity, since the complaint was based exclusively on state law.
Years passed. Extensive discovery, virtually all of it directed against Bettison, was conducted, and the district judge made numerous rulings, climaxed by the grant of summary judgment for the defendants — all this in a suit over which the court had no jurisdiction. Still acting without jurisdiction, the district court entered a final judgment for the defendants under Rule 54(b) of the Federal Rules of Civil Procedure. On October 28, 1987, almost five years after the complaint had been filed, the panel before which the plaintiffs appeal had been argued — having itself raised the jurisdictional issue at oral argument, and having invited supplemental memoranda in which the plaintiff asked the panel to order Bettison dismissed and the defendants asked the panel to order the case dismissed — dismissed Bettison and proceeded to the merits.
Protracted proceedings have unfolded in a case over which the district court never acquired jurisdiction. For us now to confer jurisdiction on the district court retroactively to the date of the complaint— to pretend that Bettison was never a party — would violate the principle that the existence of federal jurisdiction depends on the facts when the complaint is filed, not on later facts. See, e.g., New Orleans & Bayou Sara Mail Co. v. Fernandez, 79 U.S. (12 Wall.) 130, 134, 20 L.Ed. 249 (1870); St. Paul Mercury Indemnity Co. v. Red Cab Co., 303 U.S. 283, 289-90, 58 S.Ct. 586, 590-91, 82 L.Ed. 845 (1938); Fidelity & Deposit Co. v. City of Sheboygan Falls, supra, 713 F.2d at 1266; Field v. Volkswagenwerk AG, 626 F.2d 293, 304-05 (3d Cir.1980). “Jurisdiction cannot be obtained retroactively.” Denberg v. United States Railroad Retirement Bd., 696 F.2d 1193, 1197 (7th Cir.1983). To the same effect see Mansfield, Coldwater & Lake Michigan Ry. v. Swan, 111 U.S. 379, 381-82, 4 S.Ct. 510, 511-12, 28 L.Ed. 462 (1884) (a case, like the present one, in which complete diversity was lacking when the case was brought); American Nat’l Bank & Trust Co. v. Bailey, 750 F.2d 577, 585 (7th Cir.1984); Illinois v. General Electric Co., 683 F.2d 206, 209 (7th Cir.1982).
To the principle that federal courts cannot obtain jurisdiction retroactively, as to virtually every legal generalization, there are exceptions. One is where a court of appeals assumes jurisdiction over a partial final judgment entered under Fed.R.Civ.P. 54(b) after the notice of appeal had been filed. See, e.g., Sutter v. Groen, 687 F.2d 197, 199 (7th Cir.1982); Local P-171, Amalgamated Meat Cutters & Butcher Workmen v. Thompson Farms Co., 642 F.2d 1065, 1068 (7th Cir.1981). That exception, however, involves form rather than substance, and in any event is within the specific curative provision of [919]*919Rule 4(a)(2) of the Federal Rules of Appellate Procedure: “a notice of appeal filed after the announcement of a decision or order but before the entry of the judgment or order shall be treated as filed after such entry and on the day thereof.” Where the exercise of jurisdiction is discretionary (as with pendent jurisdiction), events occurring after the filing of the suit may result in or even compel a decision to relinquish that jurisdiction. See United Mine Workers v. Gibbs, 383 U.S. 715, 726, 86 S.Ct. 1130, 1139, 16 L.Ed.2d 218 (1966). But that is not a case of obtaining jurisdiction retroactively. More important, a decision not to exercise jurisdiction presupposes the presence of jurisdiction, not its absence. “The question of power will ordinarily be resolved on the pleadings. But the issue whether pendent jurisdiction has been properly assumed is one which remains open throughout the litigation.” Id. at 727, 86 S.Ct. at 1139. And when pendent jurisdiction is relinquished, the district court need not dismiss the case (as it would have to do if it lacked jurisdiction); it may in appropriate cases remand it to a state court. Carnegie-Mellon University v. Cohill, — U.S. -, 108 S.Ct. 614, 619, 98 L.Ed.2d 720 (1988). A genuine and significant exception to the principle that jurisdiction cannot be conferred retroactively is discussed later in this opinion.
We can find no source for a power in a court of appeals to preserve diversity jurisdiction by dismissing a nondiverse party retroactively to the date the suit was filed. The source cannot be 28 U.S.C. § 1653, which authorizes us merely to amend “defective allegations of jurisdiction.” (Emphasis added.) The predecessor statute (enacted in 1915) expressly limited such amendments to cases in which diversity jurisdiction “in fact existed at the time the suit was brought or removed, though defectively alleged.” Act of March 3,1915, ch. 90, 38 Stat. 956, 28 U.S.C. § 399 (1940 ed.); see Dollar S.S. Lines, Inc. v. Merz, 68 F.2d 594, 595 and n. 1 (9th Cir.1934); Carson v. Allied News Co., supra, 511 F.2d at 24 n. 6; Thomas v. Anderson, 223 Fed. 41, 43 (8th Cir.1915). The Revision Note to section 1653 says that, besides the extension of the power to amend from diversity cases to all cases, “changes were made in phraseology.” A change from permitting a defective allegation to be amended in order to reflect more accurately the facts that create federal jurisdiction to permitting federal jurisdiction to be conferred retroactively would be more than a change in phraseology. There is a difference between a case in which the complaint alleges that a defendant is a “resident” of a state different from the plaintiffs, thus leaving open the possibility that he is a “citizen” of the plaintiffs state — but in fact the defendant is a citizen of a state other than the plaintiffs — and a case in which the defendant is a citizen of the plaintiffs state. In the first case section 1653 allows us to correct the complaint in order to show that the case is, in fact though not in pleading, within the diversity jurisdiction. (For a collection of such cases see 13B Wright, Miller & Cooper, supra, § 3611, at p. 518 n. 30.) In the second case the defect is not in the allegations — the “defect” is that the case is not within federal jurisdiction — and section 1653 is inapplicable.
Like any other piece of legislative history, a reviser’s note is not conclusive. If the language of section 1653, or the events that led to its passage, showed that Congress wanted to empower the courts of appeals to preserve jurisdiction by adding or dropping parties, the Revision Note would have to give way. But the contrary is true. The language of the statute suggests that its only purpose was to extend the power to cure defective allegations from diversity cases to all cases. The previous statute had distinguished between the existence of jurisdiction and its being “defectively alleged,” and the reference to allegations was carried forward into section 1653. There is no hint that the authors of the new statute (who were also, of course, the authors of the Revision Note) wanted to authorize courts of appeals to create jurisdiction in district courts retroactively.
The 1915 statute, the predecessor of section 1653, had, as its wording shows, been intended to overrule cases like Denny v. [920]*920Pironi & Slatri, 141 U.S. 121, 124, 11 S.Ct. 966, 967, 35 L.Ed. 657 (1891), a case of defective allegation of jurisdiction where the Court had held that “a case cannot be amended here so as to show [show, not create ] jurisdiction, but the court below, in its discretion, may allow it to be done,” and Continental Life Ins. Co. v. Rhoads, 119 U.S. 237, 7 S.Ct. 193, 30 L.Ed. 380 (1886), a similar case. The statute had not been intended to erase the distinction between “an existing diversity of citizenship defectively alleged,” Thomas v. Anderson, supra, 223 Fed. at 43, and an absence of diversity jurisdiction. The original statute had made a remand to the district court to correct a defective jurisdictional allegation no longer necessary in diversity cases. Now, thanks to the amended statute, the court of appeals can correct such allegations in other types of cases as well.
But that is as far as section 1653 goes; to interpret it as going further would be to give an innocuous “change ... in phraseology” far-reaching substantive effect. The mistake in the present case was not in the allegation; it was in the assertion of federal jurisdiction over a case not within that jurisdiction. So clear is the inapplicability of section 1653 — from its language and history, from the Revision Note, and from the background principle that jurisdiction cannot attach retroactively — that all seven circuits to have discussed the scope of section 1653 have said that it is intended only to enable the courts of appeals to remedy inadequate jurisdictional allegations, not defective jurisdiction itself. Besides Carson, and our later decision in Sarnoff v. American Home Products Corp., 798 F.2d 1075, 1079 (7th Cir.1986) (“section 1653 of the Judicial Code, which allows defective allegations of jurisdiction to be corrected even on appeal, can thus be of no help to Fletcher [a plaintiff who had failed to demonstrate that his diversity claim had been worth more than $10,000 when filed]. His problem is not just pleading but also proof; we simply do not know whether his claim met the statutory minimum at the relevant time”), see Laborers Local 938 Joint Health & Welfare Trust Fund v. B.R. Starnes Co., 827 F.2d 1454, 1457 n. 1 (11th Cir.1987); Rockwell International Credit Corp. v. United States Aircraft Ins. Group, 823 F.2d 302, 304 (9th Cir.1987); Boelens v. Redman Homes, Inc., 759 F.2d 504, 512 (5th Cir.1985); Aetna Casualty & Surety Co. v. Hillman, 796 F.2d 770, 775-76 (5th Cir.1986); Pressroom Unions, Etc. v. Continental Assurance Co., 700 F.2d 889, 893 (2d Cir.1983); Brennan v. University of Kansas, 451 F.2d 1287, 1289 (10th Cir.1971); Field v. Volkswagenwerk AG, supra, 626 F.2d at 306.
No statute or rule authorizes us, by the expedient of dropping — retroactively to the date the complaint was filed — an inconvenient party whose presence destroyed jurisdiction, to enter a judgment on the merits in a case that has never been within the jurisdiction of the federal courts. If the party is a nominal party, whose presence or absence does not affect jurisdiction, Wormley v. Wormley, 21 U.S. (8 Wheat.) 421, 451, 5 L.Ed. 651 (1823); Wood v. Davis, 59 U.S. (18 How.) 467, 15 L.Ed. 460 (1856); Bacon v. Rives, 106 U.S. (16 Otto) 99, 27 L.Ed. 69 (1882); Navarro Savings Ass’n v. Lee, 446 U.S. 458, 461, 100 S.Ct. 1779, 1782, 64 L.Ed.2d 425 (1980); Matchett v. Wold, 818 F.2d 574, 576 (7th Cir.1987); Iowa Public Service Co. v. Medicine Bow Coal Co., 556 F.2d 400, 404 (8th Cir.1977), there is no harm in dismissing him, though no point either, cf. Chathas v. Smith, 848 F.2d 93 (7th Cir.1988). The dismissal of such a party is a mere gesture, and the issue of power does not arise. “[I]f the ‘nondi-verse’ plaintiff is not a real party in interest, and is purely a formal or nominal party, his or its presence in the case may be ignored in determining jurisdiction. And such a party may be dropped from the case.” Iowa Public Service Co. v. Medicine Bow Coal Co., supra, 556 F.2d at 404 (citations omitted).
A similar type of case is where the real party in interest is substituted for a nominal party. Mullaney v. Anderson, 342 U.S. 415, 72 S.Ct. 428, 96 L.Ed. 458 (1952), was a suit by a fishermen’s union and its secretary-treasurer on behalf of nonresident union members, challenging the constitutionality of a statute that im[921]*921posed a higher tax on nonresident fishermen than on local fishermen. In the Supreme Court the defendant for the first time questioned the plaintiffs’ standing to maintain the suit on behalf of the nonresident union members. (It is now plain that the union did have standing. See Sierra Club v. Morton, 405 U.S. 727, 739, 92 S.Ct. 1361, 1368, 31 L.Ed.2d 636 (1972).) In response the union moved to add two of those members as plaintiffs, and the Supreme Court granted the motion. “The original plaintiffs alleged without contradiction that they were authorized by the nonresident union members to bring this action in their behalf. This claim of authority is now confirmed in the petition supporting the motion to add the member-fishermen as plaintiffs. To grant the motion merely puts the principal, the real party in interest, in the position of his avowed agent.” Id. 342 U.S. at 417, 72 S.Ct. at 430. It was as if the complaint had mistakenly named the lawyer rather than the client as the plaintiff. Mistakes of a clerical or otherwise purely nominal character can generally be corrected at any time, cf. Fed.R.Civ.P. 60(a); but cf. Torres v. Oakland Scavenger Co., — U.S. -, 108 S.Ct. 2405, 101 L.Ed.2d 285 (1988), but no one suggests that bringing Bettison into this suit as a defendant was a mistake of that character.
Willingham v. Morgan, 395 U.S. 402, 407 n. 3, 89 S.Ct. 1813, 1816 n. 3, 23 L.Ed.2d 396 (1969), and Rogers v. Paul, 382 U.S. 198, 86 S.Ct. 358, 15 L.Ed.2d 265 (1965), likewise involved the correction of purely technical deficiencies. Carneal v. Banks, 23 U.S. (10 Wheat.) 181, 188, 6 L.Ed. 297 (1825), may or may not have been such a case. In one place the Court described the joinder of the nondiverse parties as a “mistake” (id.), but in another (on the same page) the Court treated the suit as if it were two suits, one satisfying the requirement of complete diversity, the other dismissable and dismissed. In any event, Carneal involved the power of a district court, not of an appellate court, to save jurisdiction by dismissing a party; and that is an important distinction, as we shall see. Horn v. Lockhart, 84 U.S. (17 Wall.) 570, 579, 21 L.Ed. 657 (1873), likewise involved dismissal of the nondiverse parties by the trial court. The distinction is express in some cases, such as Levering & Garrigues Co. v. Morrin, 61 F.2d 115, 121 (2d Cir.1932), aff’d, 289 U.S. 103, 53 S.Ct. 549, 77 L.Ed. 1062 (1933), and there is nothing in either Carneal or Horn (decisions whose authority we accept absolutely) to suggest that the Court was opining on the powers of appellate courts. Parenthetically, we note that in Horn there had in fact been complete diversity from the start. The heirs of John Horn had sued the executor of his estate. One of the heirs was a citizen of the same state as the executor. For reasons entirely unclear— for there appears to have been no dispute among the heirs — the remaining heirs wanted to make this heir (Mrs. McPhail) and her husband defendants, even though the suit was on her behalf as well as theirs and even though they were seeking no relief against the McPhails. To this end the plaintiffs asked “that McPhail and his wife be made defendants, if they come within the jurisdiction of the court.” The McPhails answered the complaint, admitting the truth of all of its allegations. The Court held that the district court’s eventual dismissal of these two nondiverse defendants had restored complete diversity. Actually they had been joined in the suit only conditionally, and they were real parties plaintiff, not defendant, and should have been deemed realigned as such. See City of Dawson v. Columbia Avenue Saving Fund, Etc., 197 U.S. 178, 180, 25 S.Ct. 420, 421, 49 L.Ed. 713 (1905); Fidelity & Deposit Co. v. City of Sheboygan Falls, supra, 713 F.2d at 1266-68.
Although section 1653 does not apply to this case, our inquiry is not ended. Federal courts possess not only the powers conferred on them by statute but also inherent powers, powers necessary to the courts’ effective functioning as courts — for example, the power to punish for contempt, the power to sanction persons who file frivolous pleadings, the power to determine whether there is jurisdiction, the power, in short, to preserve the integrity of the judi[922]*922cial process. See, e.g., United States v. Hastings, 461 U.S. 499, 505-06, 103 S.Ct. 1974, 1978-79, 76 L.Ed.2d 96 (1983); In re Martin-Trigona, 737 F.2d 1254, 1261 (2nd Cir.1984); Dunn v. United States Dept. of Agriculture, 654 F.2d 64, 68 (Ct.Cl.1981); United States v. Reagan, 453 F.2d 165, 172 (6th Cir.1971). But “ ‘inherent authority' is not a substitute for a good reason,” Soo Line R.R. v. Escanaba & Lake Superior R.R., 840 F.2d 546, 551 (7th Cir.1988), and the fact that courts have inherent authority to do some things does not empower them to disregard the limitations on their jurisdiction. Since judges are sometimes careless about jurisdiction (and sometimes make mistakes about jurisdiction without being careless), and naturally are reluctant to set at naught what may have been protracted and expensive proceedings (as in this case), one can find cases in which courts, straining at their jurisdictional leashes, have acted in excess of their jurisdiction and have tried to cure the usurpation by amending the pleadings. See, e.g., Long v. District of Columbia, 820 F.2d 409, 416-17 (D.C.Cir.1987); Continental Airlines, Inc. v. Goodyear Tire & Rubber Co., 819 F.2d 1519, 1523 n. 3 (9th Cir.1987); Othman v. Globe Indemnity Co., 759 F.2d 1458, 1463 (9th Cir.1985); Caspary v. Louisiana Land & Exploration Co., 725 F.2d 189 (2d Cir.1984) (per curiam). (There are other cases, but all from the Ninth Circuit; Long and Caspary are the only cases from other circuits.) Judge Sneed spoke with refreshing candor about these cases when he said in Continental Airlines that “conferring jurisdiction retroactively is a curious idea,” and “knowledgeable judges have said it cannot be done,” but “they underestimated, it appears, judicial — or at least appellate — resolve.” 819 F.2d at 1523-24 n. 3. None of the cases in this line relies on section 1653. No court has ever held that the statute empowers an appellate court to add or drop parties. The cases that have exercised an “inherent” power to preserve diversity jurisdiction by dropping a nondiverse party appear to have proceeded on the theory that courts should do what they think needs doing without inquiring too closely into their authority to do it.
The temptation to seize for reasons of trivial expediency a jurisdiction that has not been granted should be resisted, as it was by another panel of the Ninth Circuit in Rockwell International Credit Group v. United States Aircraft Ins. Group, supra, decided shortly after Continental Airlines: “the proposed amendment is not merely technical like the dismissal of a nonessential nondiverse party. The proposed amendment seeks more than the correction of ‘defective allegations of jurisdiction,’ a correction permissible under 28 U.S. C. § 1653. The proposed amendment attempts to create jurisdiction where none existed. The proposed amendment is not acceptable.” 823 F.2d at 304. The word “nonessential” makes interpretation of this passage difficult. If the word is intended to be synonymous with “nominal,” the opinion is consistent with our approach; if it means “nonindispensable,” it is inconsistent but makes the passage we have quoted internally contradictory — so the former reading is preferable, though either way the reference to “dismissal of a nonessential nondiverse party” is dictum, for that was not what was being sought.
The importance of resisting temptation is especially great in cases that are within federal jurisdiction if at all only because the parties are of diverse citizenship. This is a jurisdiction that the Supreme Court has told us should be construed strictly, see City of Indianapolis v. Chase National Bank, 314 U.S. 63, 76-77, 62 S.Ct. 15, 20-21, 86 L.Ed. 47 (1941), in recognition of the sensitive issues of comity and federalism that its exercise involves. The Court has also told us that “the age-old rule that a court may not in any case, even in the interest of justice, extend its jurisdiction where none exists has always worked injustice in particular cases. Parties often spend years litigating claims only to learn that their efforts and expense were wasted in a court that lacked jurisdiction.” Christianson v. Colt Industries Operating Corp., — U.S. -, 108 S.Ct. 2166, 2178, 100 L.Ed.2d 811 (1988).
[923]*923It would be remarkable if the federal courts of appeals had a comprehensive power, somehow inherent in their being courts, to confer jurisdiction on district courts retroactively. Suppose an Illinois citizen sues five other Illinois citizens and one Iowan. Could we five years later, after a trial on the merits, grant the plaintiffs motion to dismiss the five nondiverse defendants, in order to preserve its victory on the merits? If during the district court proceedings in this case Mr. Bettison had become a citizen of Texas, could we have backdated his change of address to the date of the complaint, to preserve jurisdiction? If the suit had been for $1,000 but judgment had been entered for $12,000, would the plaintiff have been heard to argue that if it had thought harder when it filed its complaint it would have realized that its claim satisfied the amount in controversy requirement?
The members of this court have tried to impress on each other and on the district judges of this circuit the importance of scrupulous adherence to the jurisdictional limitations of the federal courts. See, e.g., Kanzelberger v. Kanzelberger, 782 F.2d 774 (7th Cir.1986). If we attempt to patch up the jurisdictional deficiencies in a case when the case comes up to us on appeal, the district courts will have less incentive to police their jurisdiction themselves. They are busy courts, and if they know we will clean up jurisdictional messes, we will see little attention to detail by them on questions of jurisdiction. It will seem to them easier to get on with other pressing matters and leave the esoteric jurisdictional points to the court of appeals. That would not be the disciplined regard for the limits of authority that we want to encourage in the courts of this circuit; and by subjecting litigants to proceedings before a judicial body that had no authority to act, it would impose a heavy expense on the public. It is not our office to create a (federal) lawsuit where none exists and in the process create our own authority by overruling our previous decisions.
It may seem, however, that the course of this litigation could not have been affected by the jurisdictional flaw first discovered when the case was appealed. If the plaintiff had known that Bettison’s presence as a defendant precluded federal jurisdiction, would it not have refrained from naming him as a defendant? But this is uncertain. The plaintiffs pretrial discovery was directed primarily at Bettison, and it is easier to get certain forms of discovery from a party than from a nonparty. See Fed.R. Civ.P. 33-35. This is not a purely theoretical distinction in the present case. Bet-tison, and only Bettison, responded to interrogatories and to requests for production of documents — forms of discovery available only against parties. See Fed.R.Civ.P. 33(a), 34(a). Forced to choose between a state court suit with Bettison as a defendant and a federal court suit without Betti-son, the plaintiff might have chosen the state court — and who knows what view a state court would have taken of the merits of the case? But all this is beside the point, since the issue is not how severely (if at all) we should punish the plaintiff for having filed a suit that was not within the jurisdiction of the district court, but whether we have the power to determine the appropriate sanction.
Having determined that the district court lacked jurisdiction of this case and that we are powerless to cure this lack, we must decide whether to vacate the judgment with instructions to dismiss the suit or vacate the judgment and remand the case to the district court. We choose the latter course even though neither party has explicitly asked us to. The defendants asked us to dismiss the case; the plaintiff asked us to retain the case by dismissing Bettison. But the plaintiff also said that it would “present a similar motion as necessary in the District Court,” and implicit in this wording is a request that if this court should decide not to drop Bettison the district court should be given a shot at the question. There was not even an implicit request in Kanzelberger v. Kanzelberger, supra, where, rather than remand to the district court, we directed that court to dismiss the case.
Rule 21 of the Federal Rules of Civil Procedure provides that “parties may be [924]*924dropped or added by order of the court ... at any stage of the action and on such terms as are just.” The principle expressed in this rule is a response to the harsh common law rule that required automatic dismissal in the event of misjoinder. See Clark, Handbook of the Law of Code Pleading 352-68 (2d ed. 1947). The relevance of the principle to the question whether a misjoinder affecting jurisdiction is ever curable is apparent, and together with the broad wording of the rule may explain the large number of decisions which hold that Rule 21 (or its predecessors) empowers a district court to dismiss, at any stage of the proceedings, a party whose presence destroys complete diversity, if such dismissal is necessary to preserve diversity jurisdiction and would not harm other parties. Horn v. Lockhart, as noted, was one such case (and Carneal v. Banks may have been another). Other cases in this line include — and this is a small sample—Conolly v. Taylor, 27 U.S. (2 Pet.) 556, 565, 7 L.Ed. 518 (1829); American Fire & Casualty Co. v. Finn, 341 U.S. 6, 19 n. 18, 71 S.Ct. 534, 542 n. 18, 95 L.Ed. 702 (1951); Fritz v. American Home Shield Corp., 751 F.2d 1152, 1155 (11th Cir.1985); Park v. Didden, 695 F.2d 626, 631 (D.C.Cir.1982); Publicker Industries, Inc. v. Roman Ceramics Corp., 603 F.2d 1065, 1069 (3d Cir.1979); Caperton v. Beatrice Pocahontas Coal Co., 585 F.2d 683, 692 (4th Cir.1978); Dollar S.S. Lines, Inc. v. Merz, supra. This court first embraced the principle in 1897, see Mason v. Dullagham, 82 Fed. 689 (7th Cir.1897), and it is mentioned with approval in both Alderman v. Elgin, Joliet & Eastern Ry., supra, 125 F.2d at 973, and Carson v. Allied News Co., supra, 511 F.2d at 24.
Judge Mack, whose opinion in Dollar S.S. Lines remains the fullest statement of the principle, pointed out that where a case is remanded to enable the district court to exercise its powers under Rule 21, the court of appeals should not express a view of the merits. See 68 F.2d at 597. For by doing so it may be handing the plaintiff an unearned tactical advantage when the plaintiff comes to decide whether to refile the case in a state court or to ask the district court to retain the case. This point provides a practical reason for the lodging of the power in question in the district courts rather than in the courts of appeals: we are spared the temptation that would be created if we had to decide whether to exercise our discretion to retain a case in the knowledge that if we did so this would allow us to decide the merits, ordinarily with finality.
Most important, the district court is better placed than the court of appeals to determine whether any party was harmed by the presence of the party that destroyed jurisdiction. In their petition for rehearing the defendants took sharp issue with the confident observation in the panel opinion that Bettison’s presence in the district court had not affected the conduct of the litigation. They said that Bettison had “had a substantial impact on the posture of the case,” that “virtually the sole basis for the district court’s determination of liability was from the testimony of Mr. Bettison and from documents that Mr. Bettison produced,” and that “none of the other guarantors had access to such materials and no other person was deposed in this litigation. In short, without Bettison, NGI [the plaintiff] had little prospect of proving its case. Guarantors were therefore clearly affected to their detriment by Mr. Bettison’s presence and NGI profited enormously.” These allegations may or may not be true, a question on which we express no opinion; the district court is in a better position than this court to evaluate them — indeed, that is our point.
It could be argued that the principle which allows the district court to preserve jurisdiction by dropping a party at any stage in the lawsuit is in tension with the principle that the existence of federal jurisdiction depends on the facts as they exist when the complaint is filed, and not on later events. And it could be pointed out that Rule 21 must be read in light of Rule 82 (“These rules shall not be construed to extend or limit the jurisdiction of the United States district courts”). See Kerr v. Compagnie de Ultramar, 250 F.2d 860, 864 (2d Cir.1958); Hill v. Western Electric [925]*925Co., 672 F.2d 381, 398 n. 10 (4th Cir.1982) (dissenting opinion); cf. Audi Vision Inc. v. RCA Mfg. Co., 136 F.2d 621, 624 (2d Cir.1943). But as law is an instrument of governance rather than a hymn to intellectual beauty, some consideration must be given to practicalities. If an easily curable jurisdictional defect is discovered shortly after a case is filed, the district court should have the power to decide whether the plaintiff must be put to the bother of filing a fresh suit rather than allowed simply to amend the complaint. See Hackner v. Guaranty Trust Co., 117 F.2d 95, 98 (2d Cir.1941) (Clark, J.); cf. American Law Institute, Study of the Division of Jurisdiction Between State and Federal Courts 366 (1969). And if this point is conceded it becomes a matter of judgment when the curative power of the district court runs out. Rule 21 specifies no deadline.
None of the parties has asked us to reconsider the interpretation of Rule 21 under which the district court could have done — and, more to the point, still can do— what the plaintiff has asked us to do. This is not surprising. So many cases have adopted that interpretation that a court at our level probably can no longer reexamine it. Stare decisis has its claims, as much in interpreting Rule 21 as in interpreting section 1653. Stability is an important value in law, and is promoted by declining to revisit issues that have long been settled. Modesty alone would counsel hesitation in setting one’s personal view against a view that had over a period of many years commanded the unanimous support of judges of diverse background.
And certainly this appeal is not the right time and place in which to reconsider the scope of Rule 21. The district court has not yet been asked to retain jurisdiction by dismissing Bettison, and if upon being asked the district court decides in a proper exercise of its discretion to do nothing the question of its power will be moot.
We would need a crystal ball to forecast the fate of this litigation on remand to the district court. It may turn out that, rather than do without Bettison as a defendant, the plaintiff would prefer to start over in state court, as it can do without encountering the bar of the statute of limitations. See Ill.Rev.Stat. ch. 110, ¶ 13-217. If the plaintiff decides that it would prefer to stay in federal court (its current preference), it may ask the district court to dismiss Bettison, but the district court may decide that six years of federal litigation in a suit outside its jurisdiction are enough. Conversely, that court may decide that the plaintiffs jurisdictional error was forgivable; that the plaintiff has been sufficiently punished already (for it has seen the victory it won before the panel snatched from its jaws); that when one considers that the plaintiff’s error involves both an esoteric and nonintuitive point of federal jurisdiction and was an oversight by the defendants as well as by the plaintiff, it is not so grave an error as to warrant the sanction of dismissal. Excessive sanctions are as inappropriate as inadequate ones. See, e.g., Brown v. Federation of State Medical Bds., 830 F.2d 1429, 1437 (7th Cir.1987). So the district court may decide to dismiss Bettison after all, though perhaps on terms different from those proposed in the panel opinion. We express no view on these questions; they are for the district court in the first instance.
Long v. District of Columbia, supra, 820 F.2d at 417, holds that if the district court has the power under Rule 21 to preserve diversity jurisdiction by dropping a nondiverse party, we should have the same power. Long gave no reason for this conclusion, however, but merely cited two cases. One, Reed v. Robilio, 376 F.2d 392, 394 (6th Cir.1967), makes no reference to Rule 21, and was in any event a case, like Mullaney v. Anderson, supra, that involved identifying the real parties in interest — not dropping a real party in interest. The other case cited in Long was Mullaney itself, which as have noted involved the joining of the real party in interest as an additional named plaintiff. Mullaney cited Rule 21 but without discussion of its application at the appellate level, beyond the mysterious statement (342 U.S. at 417, 72 S.Ct. at 430) that “Rule 21 will rarely come into play at this [i.e., the Supreme Court] stage of litigation” — mysterious because [926]*926“the Federal Rules of Civil Procedure, of course, apply only in the federal district courts.” International Union, UAW, Local 283 v. Scofield, 382 U.S. 205, 217 n. 10, 86 S.Ct. 373, 381 n. 10, 15 L.Ed.2d 272 (1965). To this generalization, too, there are exceptions, not only where appellate courts adopt the civil rules to govern proceedings in the appellate court (see, e.g., 9th Cir.R. 1-1), but also where one of those rules clearly constrains appellate action, notably in the case of Rule 52(a), which provides that findings of fact by the district court shall not be set aside unless clearly erroneous. These exceptions have no application to this case. Rule 21 has not been incorporated by reference or otherwise in the Federal Rules of Appellate Procedure or in our circuit rules. Cf. Hill v. Norfolk & Western Ry., 814 F.2d 1192, 1200 (7th Cir.1987) (“this court has not incorporated Rule 11 into its own rules, and therefore the rule does not apply directly to proceedings in this court”); Hays v. Sony Corp., 847 F.2d 412, 429 (7th Cir.1988) (same). The result may be an irksome gap in our authority, but there it is.
The principle underlying a particular rule may apply to the appellate court independently of the rule itself. The principle behind Fed.R.Civ.P. 12(h)(3) — “Whenever it appears by suggestion of the parties or otherwise that the court lacks jurisdiction of the subject matter, the court shall dismiss the case” — requires this court, subject to the district court’s curative powers under Rule 21, to order a case dismissed when it notices a lack of subject-matter jurisdiction. See, e.g., Rice v. Rice Foundation, 610 F.2d 471, 474 (7th Cir.1979). Rule 12(h)(3) is an instance of the fundamental principle, which is as applicable to appellate as to trial courts, that objections to subject-matter jurisdiction cannot be waived. See, e.g., Mansfield, Coldwater & Lake Michigan Ry. v. Swan, supra. There is no similar overarching principle applicable alike to trial and to appellate courts that authorizes courts to repair jurisdictional defects by changing parties.
Appellate courts are not automatically invested with all the powers of district courts, as the Supreme Court has been at pains to emphasize in discussing factfinding. See Anderson v. City of Bessemer City, 470 U.S. 564, 573-75, 105 S.Ct. 1504, 1511-12, 84 L.Ed.2d 518 (1985). For us to assume the powers conferred by Rule 21 would be to cross the line that separates our authority from that of the district courts. Our power to act in this suit — to decide the merits, to mete out sanctions, and so forth — depends on a valid final judgment in the district court. If that court was without jurisdiction, it is not for us to create a federal ease upon which to exercise our powers.
The line that the Supreme Court has with the consent of Congress drawn in Rule 21 between the powers of the trial and appellate courts of the federal system is not an irrational one. Jurisdictional problems in the district court should be resolved in the first instance by that court, rather than by this court’s speculating on how that court might respond to a motion to add or drop parties. Since, moreover, that court has the power, there is no urgent need for this court to confer the same power on itself (even if we could do such a thing), however convenient this might be in a case in which we were utterly confident that a party should be added or dropped to preserve jurisdiction and enable us to reach the merits without waiting for the district court to decide a motion under Rule 21. And there is such a thing as overconfidence.
Federal judges spend lots of time telling other officials to stay within constitutional and statutory bounds, however those bounds may chafe in particular cases. If we exceed the limits of our own jurisdiction we shall be setting a bad example for litigants in the federal courts and for other persons subject to federal law. We shall be enforcing the lesson that rules on jurisdiction are inconveniences that can be got round by interpreting statutes such as 28 U.S.C. § 1653 expansively and by stretching the elastic concept of inherent judicial authority to the breaking point. We shall also be ignoring the precept (repeated just the other day, in Kennedy v. Wright, 851 F.2d 963, 967 (7th Cir.1988)) that “jurisdictional rules should be as simple as possi[927]*927ble.” Coté v. Wadel, 796 F.2d 981, 983 (7th Cir.1986). “The first characteristic of a good jurisdictional rule is predictability and uniform application.” Exchange National Bank of Chicago v. Daniels, 763 F.2d 286, 292, modified on other grounds, 768 F.2d 140 (7th Cir.1985). Cf. Budinich v. Becton Dickinson & Co., — U.S. -, 108 S.Ct. 1717, 1722, 100 L.Ed.2d 178 (1988). The simple rule, which is also the sensible rule and the one established by the weight of precedent, is that, subject to well-defined exceptions that do not include a free-floating discretionary power in this court to cure defects in the jurisdiction of the district courts, jurisdiction is determined on the pleadings as originally filed. If the jurisdictional allegations are defective but there was jurisdiction when the complaint was filed, we can cure the defect under section 1653. But if there was no jurisdiction when the complaint was filed the case must be dismissed, subject to the curative powers of the district court under Fed.R. Civ.P. 21.
The judgment is vacated and the case returned to the district court (where it will remain with Judge Shadur pursuant to Circuit Rule 36) for further consideration consistent with this opinion. No costs in this court.