Dissenting opinion filed by Circuit Judge RANDOLPH.
STEPHEN F. WILLIAMS, Circuit Judge:This case appears before us for the second time, with the defendants — the mayor and other officials of the District of Columbia— asking for relief from a consent decree pursuant to which the district court is exercising broad supervisory authority over the District’s child welfare system. Because the district court has not re-examined the validity of the federal claims underlying its jurisdiction since the Supreme Court issued a decision that seems to undermine the statutory support for federal jurisdiction (though not the constitutional basis of certain claims), we remand the ease to the district court to perform that re-examination. See United Mine Workers v. Gibbs, 383 U.S. 715, 726-27, 86 S.Ct. 1130, 1139, 16 L.Ed.2d 218 (1966).
* * *
Plaintiffs, a class of children who either are in foster care under the supervision of the D.C. Department of Human Services (“DHS”), or have been reported to be abused or neglected but are not yet in DHS care, sought injunctive relief based on alleged violations of federal statutory and constitutional law, as well as of local law.
The district court agreed for the most part with plaintiffs. LaShawn A. v. Dixon, 762 F.Supp. 959 (D.D.C.1991). It found explicitly that the Adoption Assistance and Child Welfare Act of 1980, 42 U.S.C. §§ 620-27 and §§ 670-79, afforded plaintiffs rights actionable under 42 U.S.C. § 1983 (1988), and held implicitly that the Child Abuse Prevention and Treatment Act, 42 U.S.C. §§ 5101-06, did so as well. See 762 F.Supp. at 988-89. It detailed widespread violations of those statutes. Id. at 968-87. The court proceeded to find that children in foster care under the supervision of the District enjoyed a liberty interest under the U.S. Constitution, id. at 991-92, with a concomitant right to such services as were “essential to preventing harm” to the children, id. at 993. Compliance with this norm was to be measured by the “professional judgment standard”, drawn from Youngberg v. Romeo, 457 U.S. 307, 102 S.Ct. 2452, 73 L.Ed.2d 28 (1982). Under that standard, “liability may be imposed only when the decision by the professional is such a substantial departure from accepted professional judgment, practice, or standards as to demonstrate that the person responsible actually did not base the decision on such a judgment.” Id. at 323, 102 S.Ct. at 2462 (quoted at 762 F.Supp. at 994). Finally, the court held that various practices of the District violated that standard, as well as the two pertinent federal statutes and various provisions of local law. 762 F.Supp. at 996-98. It did not find a private right of action for the children under D.C. law, although its discussion of possible entitlements protected under the due process clause may have assumed such rights. Id. at 993-94.
In response to this merits determination, the parties negotiated a broad consent decree, which in its current form occupies 90 single-spaced pages and constitutes a rather comprehensive manual for the conduct of the *559District’s child welfare activities. The agreement reserved defendants’ right to appeal the district court’s liability ruling and addressed the possibility that the court’s merits opinion might be “vacated” in whole or in part. Section XXII(C) provides:
In the event that the court’s Memorandum Opinion of April 18, 1991, is vacated on appeal in its entirety, this Order and any subsequent implementation plan or plans shall be null and void. In the event that the court’s Memorandum Opinion of April 18, 1991, is vacated on appeal in part, the portions of this Order or any subsequent implementation plan or plans that are directly based on that part of the Memorandum Opinion that is vacated shall be null and void.
In their first appeal, defendants challenged the findings of constitutional violations. They also claimed that, in light of the intervening Supreme Court decision in Suter v. Artist M., 503 U.S. 347, 112 S.Ct. 1360, 118 L.Ed.2d 1 (1992), holding that §§ 671(a)(9) & (a)(15) of the Adoption Assistance Act are not enforceable in a § 1983 action, the federal statutes on which the district court had relied were not enforceable by private plaintiffs. This court found it unnecessary “to confront these constitutional and federal statutory issues, for the district court judgment is completely supportable on the grounds of local law”, which, we said, creates a private cause of action both “for children in foster care and for children reported to have been abused or neglected but not yet in the District’s custody.” LaShawn A. ex rel. Moore v. Kelly, 990 F.2d 1319, 1322, 1325 (D.C.Cir.1993). Because the consent decree had been drafted to conform to federal as well as local law, we remanded to the district court “with instructions to fashion an equally comprehensive order based entirely on District of Columbia law, if possible. If there are any portions of the consent decree that depend entirely on a federal statute, the district court should consider the impact of [Suter ] on those provisions before it includes them in the revised consent decree.” Id. at 1326.
On remand, the district court declared, despite our having explicitly sidestepped the issue of liability under federal law, that we had “clearly affirmed this Court’s Memorandum Opinion.” LaShawn A v. Kelly, Civ. No. 89-1754, Order at 1 (D.D.C. Nov. 12, 1993) (citing our decision, 990 F.2d at 1326 (“Because the district court’s judgment is independently supportable by District of Columbia law, we affirm the court’s decision in favor of the children in this case.”)). On that basis, it rejected defendants’ argument that § XXII(C) of the consent decree required modification of the decree. Id. at 2. It did not in any way address whether other circumstances — such as the Suter decision itself — might require modification of the decree or re-examination of the court’s jurisdiction under United Mine Workers v. Gibbs. After modifying certain provisions of the consent decree to remove terms that the parties evidently agreed were in violation of District law, it rejected defendants’ contention that some provisions still violated local law. It acknowledged that the decree “exceeds the specific mandates of local law”, but found the excess “a necessary and appropriate use of [the court’s] equitable authority” in light of defendants’ “widespread local law violations”. LaShawn A. v. Kelly, Civ. No. 89-1754, Order at 1 (D.D.C. Jan. 27, 1994). See also Order of Nov. 12, 1993 at 2 (“To the extent that portions of the remedial order exceed the terms of local law, the Court invokes its equitable authority in approving the consent decree.”).
* * *
Defendants now return to us, questioning whether the district court has jurisdiction to enforce a wide-ranging institutional reform order against the government of the District of Columbia based entirely on District of Columbia law. We regard the defendants’ claim as basically posing a question under step 2 of Gibbs, namely, whether, given power at the outset in a generalized sense to decide the plaintiffs’ state law claims, the district court should, despite Suter, exercise that power in the form of continuing to enforce a massive institutional reform decree of the sort it had originally adopted.
Law of the Case Considerations
Plaintiffs respond that the jurisdictional question has already been decided, by the panel in the first appeal, so that defen*560dants’ claim is barred by law-of-the-case doctrine. That doctrine embodies the general rule that
a court involved in later phases of a lawsuit should not re-open questions decided (i.e., established as the law of the ease) by that court or a higher one in earlier phases. When there are multiple appeals taken in the course of a single piece of litigation, law-of-the-case doctrine holds that decisions rendered on the first appeal should not be revisited on later trips to the appellate court.
Crocker v. Piedmont Aviation, Inc., 49 F.3d 735, 739 (D.C.Cir.1995). The doctrine normally applies to issues decided explicitly or “by necessary implication”. Id.
These conventional requirements have certainly been fulfilled in this case. The prior panel’s decision necessarily implied that a remedial order supported entirely by local law could properly be entered. The court explicitly recognized that its affirmance rested “entirely on the basis of local law”, LaShawn A. v. Kelly, 990 F.2d at 1326, and implicitly held that an order devoid of federal support would present no jurisdictional problem. We said that our “authority to decide the case entirely on pendent state grounds is incontrovertible”, id., citing United Mine Workers v. Gibbs, 383 U.S. 715, 722, 86 S.Ct. 1130, 1136, 16 L.Ed.2d 218 (1966), for the proposition that a federal court presented with federal and local law claims may, “even though the federal ground be not established ... nevertheless retain and dispose of the case upon the non-federal ground”, id. (emphasis deleted). As to the second step required by Gibbs, the decision whether under the conditions prevailing the district court should exercise power over the pendent local-law claim by continuing enforcement of a 90-page child welfare code, the prior panel passed sub silentio. Its remand to the district court for entry of an order based entirely on local law necessarily implied that the answer to this question was “yes”.
Law of the case, however, is a prudential doctrine that “merely expresses the practice of courts generally to refuse to reopen what has been decided, not a limit to their power.” Messinger v. Anderson, 225 U.S. 436, 444, 32 S.Ct. 739, 740, 56 L.Ed. 1152 (1912) (Holmes, J.) (citations omitted). “A court has the power to revisit prior decisions of its own or of a coordinate court in any circumstance”. Christianson v. Colt Industries Operating Corp., 486 U.S. 800, 817, 108 S.Ct. 2166, 2178, 100 L.Ed.2d 811 (1988) (emphasis added). The Court in Christian-son noted, however, that “as a rule courts should be loathe to do so in the absence of extraordinary circumstances such as where the initial decision was ‘clearly erroneous and would work a manifest injustice.’ ” Id. (citation omitted).
In Christianson itself, the Court rejected defendant’s theory that the Seventh Circuit’s determination that it lacked jurisdiction bound the Federal Circuit to that view, on transfer from the Seventh. Once the Federal Circuit decided that the Seventh’s view was “ ‘clearly wrong’ it [the Federal Circuit] was obliged to decline jurisdiction,” id. at 817, 108 S.Ct. at 2178, thus applying the conventional exception for clear error. In any event, law of the case in the lower courts could not constrain the Supreme Court’s own review. “Just as a district court’s adherence to law of the case cannot insulate an issue from appellate review, a court of appeals’ adherence to the law of the case cannot insulate an issue from this Court’s review.” Id. In a footnote dictum, however, the Court said,
There is no reason to apply law-of-the-case principles less rigorously to transfer decisions that implicate the transferee’s jurisdiction. Perpetual litigation of any issue— jurisdictional or nonjurisdictional—delays, and therefore threatens to deny, justice. But cf. Potomac Passengers Assn. v. Chesapeake & Ohio R. Co., ... 520 F.2d 91, 95 n. 22 [(D.C.Cir.1975)].
486 U.S. at 816 n. 5, 108 S.Ct. at 2178 n. 5.
As the “But cf.” cite suggests, our holding in Potomac Passengers directly contradicted the Christianson dictum. In Potomac Passengers this court found that it had erroneously decided a jurisdictional issue in a prior appeal and proceeded to correct the mistake. Though apologizing for the earlier mistake, the court insisted on the need for correctness as to issues of subject matter jurisdiction:
*561“When an appellate court makes so fundamental an error as that of sustaining federal subject matter jurisdiction where none exists, we think the court must exercise its discretion to correct that mistake.” 520 F.2d at 95 n. 22. Potomac Passengers has been widely cited for the proposition that jurisdictional questions are relatively unrestrained by law of the ease.1
Courts have applied Christianson’s discussion of law of the case doctrine relating to jurisdictional questions only in cases involving the propriety of transfers.2 See, e.g., Ukiah Adventist Hospital v. FTC, 981 F.2d 543, 546 n. 4 (D.C.Cir.1992) (“Review of a transfer order [including jurisdictional grounds] in a transferee court is exceedingly limited.”) (dictum citing Christianson); Wang Laboratories v. Applied Computer Sciences, 958 F.2d 355, 358 (Fed.Cir.1992) (accepting case transferred by First Circuit, the court reasoned that “if the transferee court can find the transfer decision plausible, its jurisdictional inquiry is at an end”); Moses v. Business Card Express, 929 F.2d 1131, 1137 (6th Cir.1991) (upholding, under Christian-son, district court’s decision to apply law-of-the-ease principles to transfer decision made pursuant to forum selection clause; technically venue rather than jurisdictional issue). While courts have sometimes applied law of the case to jurisdictional questions, in doing so they have neither relied on the Christian-son dictum nor even expressly considered whether jurisdictional issues called for special treatment. See McKesson Corp. v. Islamic Republic of Iran, 52 F.3d 346, 351 (D.C.Cir.1995); In the Matter of Memorial Estates, 950 F.2d 1364, 1367 (7th Cir.1991); Oneida Indian Nation of New York v. State of New York, 860 F.2d 1145, 1151 (2d Cir.1988); Gould v. Mutual Life Ins. Co., 790 F.2d 769 (9th Cir.1986) (pre-Christianson). And two of the circuits that once applied law of the case to a jurisdictional issue without discussion, the Ninth and the Second, later refused to do so when they zeroed in on the special character of jurisdictional matters. In Matek v. Murat, 862 F.2d 720, 724 n. 1 (9th Cir.1988), decided several months after Christianson, the Ninth Circuit held that a district court was not barred by law of the case from reconsidering its own non-final orders. The court cited Potomac Passengers for the proposition that “[s]ubject matter jurisdiction, because of its intrinsic importance to the judicial power of the federal courts, is particularly suitable for reconsideration.” Id. And the Second Circuit applied the same principle in DiLaura v. Power Authority, 982 F.2d 73, 77 (2d Cir.1992), noting that “subject matter jurisdiction is particularly suited for reconsideration”, although, to be sure, as Judge Winter noted in a separate opinion, the issue was in fact moot because what might have been law of the case for the district court could not have prevented review by the court of appeals, see id. at 81. Wright & Miller have noted that “[questions of subject matter jurisdiction are particularly apt to be free of law of the case principles”, citing Potomac Passengers. Wright & Miller, Federal Practice and Procedure § 4478 at 799 n. 32 (1981) (under the heading “Suitable to Reconsider”). “In addition to the great importance that is generally attributed to jurisdictional limits, such questions may at times involve matters of discretion that inherently require reexamination as a case progresses”, id. (emphasis added), the latter phrase being apparently an allusion to step 2 of Gibbs. The 1995 Supplement to Wright & Miller does not modify that passage, 1995 Supplement at 739-40 n. 32, citing Christian-son under the separate category of “Propriety of Transfer: Transferor reconsideration”, id. at 730 n. 26.
The dissent attempts to elevate law of the circuit doctrine (which supposedly has no *562exceptions) above law of the case doctrine (which concededly does). Dissent at 571-572. Plaintiffs, although they argued law-of-the-case vigorously, never suggested this distinction, perhaps because they thought there was none—that at the court of appeals level exceptions to law of the case were necessarily the same as exceptions to law of the circuit. Nor did our own precedent on the subject, Potomac Passengers, perceive such a distinction.
Among jurisdictional issues, the ones most strongly inviting a relaxed application of law of the case are those that the prior decision never explicitly confronted. In Pennhurst State School & Hospital v. Halderman, 465 U.S. 89, 104 S.Ct. 900, 79 L.Ed.2d 67 (1984) (“Pennhurst II”), the Court certainly acted as if such an exception to ordinary law of the case doctrine existed. In Pennhurst State School & Hospital v. Halderman, 451 U.S. 1, 101 S.Ct. 1531, 67 L.Ed.2d 694 (1981) (“Pennhurst I”), the Court itself cast doubt on the federal basis for a consent decree enforced against state officials, see id. at 18-27, 101 S.Ct. at 1540-45, and remanded the case to the Court of Appeals to determine the validity of the federal claims. The Court also remanded state law claims to determine “whether state law provides an independent and adequate ground which can support the court’s remedial order.” Id. at 31, 101 S.Ct. at 1547. The remand order thus necessarily implied that a federal court would have jurisdiction to enter the order on state law grounds alone.
On remand, the Court of Appeals acted on the implied permission Pennhurst I had given and found that the Pennsylvania statute provided “adequate support for [the order] independent of federal law”. Halderman v. Pennhurst State School & Hospital, 673 F.2d 647, 656 (3d Cir.1982). Again the Supreme Court reversed, holding that the Eleventh Amendment denies federal courts jurisdiction to order state officials to conform their conduct to state law. 465 U.S. at 117-21, 104 S.Ct. at 917-19. It brushed aside several earlier implicit rulings to the contrary in other cases, announcing that stare decisis prineiples did not bind its reconsideration of the jurisdictional issue. (“When questions of jurisdiction have been passed on in prior decisions sub silentio, this Court has never considered itself bound when a subsequent case finally brings the jurisdictional issue before us.” Id. at 119, 104 S.Ct. at 918.)
Pennhurst II is not, of course, identical to the current issue. First, the direct bar of the Eleventh Amendment is plainly of greater force than the discretionary jurisdictional limit of Gibbs step 2. Second, though the Court alluded to implicit holdings to the contrary in other cases, it did not acknowledge the necessary implication of Pennhurst I that there was federal jurisdiction for an injunction resting solely on state law, and thus an obvious ground for applying law-of-the-case. Especially given the merely “prudential” character of law-of-the-case restrictions, and the doctrine’s grounding in concerns of judicial economy, see, e.g., Crocker, 49 F.3d at 739-40, however, the Court’s action strongly suggests that law-of-the-case considerations give way more readily to jurisdictional concerns, at least where the initial decision has failed to address the issue explicitly.
We thus turn to consider whether a Gibbs step 2 inquiry is jurisdictional. When a federal court is presented with a combination of federal and state claims, it must pursue the two-step analysis outlined in Gibbs. 383 U.S. at 725-27, 86 S.Ct. at 1138-40.3 The first step requires the court to evaluate the substantiality of the federal claim and whether the state and federal claims “derive from a common nucleus of operative fact”. Id. at 725, 86 S.Ct. at 1138. If these requirements are met, then “there is power in the federal courts to hear the whole”. In our first pass at this case, we got that far and stopped. Finding that there had been (prior to Suter) a substantial federal claim, we asserted “incontrovertible” authority “to decide the case entirely on pendent state grounds”. 990 F.2d at 1326.
But Gibbs makes clear that that should not be the end of the inquiry. The Supreme Court has spoken of Gibbs step 2 in terms *563that mark it as a jurisdictional inquiry, in the sense that it requires continuous re-examination as the litigation and surrounding legal context develop. As Gibbs itself says:
That power need not be exercised in every ease in which it is found to exist.... Needless decisions of state law should be avoided both as a matter of comity and to promote justice between the parties, by procuring for them a surer-footed reading of applicable law.... The question of power will ordinarily be resolved on the pleadings. But the issue whether pendent jurisdiction has properly been assumed is one which remains open throughout the litigation. Pretrial procedures or even the trial itself may reveal a substantial hegemony of state law claims, or likelihood of jury confusion, which could not have been anticipated at the pleading stage. Although it will of course be appropriate to take account in this circumstance of the already completed course of the litigation, dismissal of the state claim might even then be merited.
383 U.S. at 726-27, 86 S.Ct. at 1139 (emphasis added). And more recently the Court stated in Carnegie-Mellon University v. Cohill, 484 U.S. 343, 108 S.Ct. 614, 98 L.Ed.2d 720 (1988):
Under Gibbs, a federal court should consider and weigh in each case, and at every stage of the litigation, the values of judicial economy, convenience, fairness, and comity in order to decide whether to exercise jurisdiction over a case brought in that court involving pendent state-law claims. When the balance of these factors indicates that a ease properly belongs in state court, as when the federal-law claims have dropped out of the lawsuit in its early stages and only state-law claims remain, the federal court should decline the exercise of jurisdiction by dismissing the case without prejudice.
Id. at 360, 108 S.Ct. at 619 (emphasis added).4
Lower federal courts of appeals, including this circuit, have followed the Supreme Court’s lead and treated the Gibbs step 2 inquiry as jurisdictional — and accordingly have raised the issue on their own. The Seventh Circuit raised a Gibbs step 2 issue sua sponte in Maguire v. Marquette University, 814 F.2d 1213, 1218 & n. 4 (7th Cir.1987). After dismissing plaintiffs federal claim before trial, the district court had proceeded to review a pendent state law claim on the merits. Although defendant did not raise a Gibbs step 2 issue on appeal, the Seventh Circuit declared that “because the rule is jurisdictional, we are obligated to raise it ourselves.” Id. at 1218 n. 4 (emphasis added). The court then vacated the district court’s dismissal of the pendent claim on the merits. This court has likewise treated Gibbs step 2 analysis as jurisdictional and raised the issue sua sponte. In Minker v. Baltimore Annual Conference of United Methodist Church, 894 F.2d 1354, 1361 (D.C.Cir.1990), the district court had previously dismissed both federal statutory and state contract claims under the Free Exercise Clause of the First Amendment. This court held that the district court had correctly dismissed the federal statutory claims, but had incorrectly dismissed the state contract claims on First Amendment grounds. Without prompting, the court proceeded to hold that because the district court had properly dismissed the federal claim before trial, under Gibbs step 2 it should also dismiss the pendent state law claim unless it found diversity of citizenship. Id.
The dissent dispatches Maguire and Minker by characterizing them as applications of a newly coined Gibbs step one-and-a-half — eases where the federal claim survives step 1 but is dismissed before trial. Dissent at 575. Thus, in its view, the Seventh Circuit’s and our treatment of that situation as jurisdictional can’t possibly mean that Gibbs step 2 is jurisdictional. But until the dissent, no one has perceived this circumstance as *564involving a special kind of Gibbs step; it is simply an instance of Gibbs step 2 so clear that any continued assertion of jurisdiction by the district court would be an abuse of discretion. See Gibbs, 383 U.S. at 726, 86 S.Ct. at 1139.
In sum, the key issue here—the application of Gibbs step 2—(1) is jurisdictional, and thus is subject to continuous reexamination throughout the enforcement of a consent decree (even without a party’s raising it) and (2) was never explicitly addressed in our prior decision. Our non-discussion of the finding of continued jurisdiction suggests that it consumed few or no judicial resources. Taking our cue from the Supreme Court in Pennhurst II, therefore, we conclude that the interest in limiting federal courts to the exercise of their proper jurisdiction outweighs the marginal contribution to stability and judicial economy that could be achieved by giving finality to our prior sub silentio treatment of the Gibbs step 2 issue.
Gibbs Step 2 Analysis
As we said before, the first part of the Gibbs test is satisfied once a federal court is determined to have power over state claims because of a substantial related federal claim. That part is indeed satisfied here. Step 2 of the Gibbs test requires a weighing of factors to determine whether, although power exists, it should be exercised in favor of jurisdiction. These factors include judicial economy, convenience and fairness to litigants, avoidance of needless decisions of state law (in order to promote comity and obtain a “surer-footed reading of applicable law”), the timing of the dismissal of the federal claims, the predominance of state versus federal issues, and the potential for jury confusion. United Mine Workers v. Gibbs, 383 U.S. 715, 726-27, 86 S.Ct. 1130, 1139, 16 L.Ed.2d 218 (1966). The initial decision lies in the discretion of the district court, subject to review for abuse of discretion in the court of appeals. Edmondson & Gallo-gher v. Alban Towers Tenants Ass’n, 48 F.3d 1260, 1265-66 (D.C.Cir.1995).
Two circumstances weigh heavily against allowing pendent jurisdiction: federal claims have either dropped out or are greatly weakened, and the local claims lead to extensive entanglement in local government operations.
First, the elimination or weakening of federal claims obviously weighs against exercise of pendent jurisdiction.5 Gibbs itself said: “Certainly, if the federal claims are dismissed before trial, even though not insubstantial in a jurisdictional sense, the state claims should be dismissed as well.” 888 U.S. at 726, 86 S.Ct. at 1139. Courts have held that pendent state claims should be dismissed if the federal claim drops out quite late, even after trial. See, e.g., Kidder, Peabody & Co. v. Maxus Energy Corp., 925 F.2d 556, 564 (2d Cir.1991) (“The judicial economy factor should not be the controlling factor, and it may be appropriate for a court to relinquish jurisdiction over pendent claims even where the court has invested considerable time in their resolution.”); Powell v. Gardner, 891 F.2d 1039, 1047 (2d Cir.1989) (holding that even after a trial on the merits, district court properly dismissed pendent state law claims after directing a verdict in favor of defendant on federal claim).
Similarly, predominance of local law issues weighs against exercise of pendent jurisdiction. In Gibbs, the Supreme Court noted “if it appears that the state issues substantially predominate, whether in terms of proof, of the scope of the issues raised, or of the comprehensiveness of the remedy sought, the state law claims may be dismissed without prejudice”. 383 U.S. at 726, 86 S.Ct. at 1139 (emphasis added). As Gibbs suggests, considerations of comity weigh especially heavily in cases involving massive consent decrees controlling government bodies. Even before Pennhurst, federal courts *565were concerned about controlling state agencies through exercise of pendent jurisdiction under Gibbs step 2. In Evans v. Buchanan, 468 F.Supp. 944, 956 (D.Del.1979), for example, the court refused to exercise pendent jurisdiction under Gibbs to redesign a state school district tax, citing “the intricacies of the state-law issue” and “questions of comity raised by the remedies proposed.”6
Although injunctions against District of Columbia officials based on District law are not outright banned under the Eleventh Amendment, which applies only to states,7 Pennhurst II does suggest that, in general, injunctions against nonfederal officials based on nonfederal law should be disfavored. Indeed, we have consistently stressed the respect due the courts of the District of Columbia in applying District law. In applying jurisdictional aspects of exhaustion doctrine, for example, we have noted that “the federal courts owe to the District of Columbia the comity extended to a state”. Committee of Blind Vendors of the District of Columbia v. District of Columbia, 28 F.3d 130, 134 (D.C.Cir.1994). Moreover, we have followed this principle in applying Gibbs step 2. In Financial General Bankshares, Inc. v. Metzger, 680 F.2d 768, 772 (D.C.Cir.1982), we held that the district court had abused its discretion in exercising pendent jurisdiction over novel and unsettled questions of D.C. law (fiduciary duty issues). And in Grano v. Barry, 733 F.2d 164 (D.C.Cir.1984), we underlined the importance of avoiding unnecessary federal court invasions of local self-government. One part of an injunction had become moot and the other rested solely on a mistaken belief that local officials’ violation of local law ipso facto violated the federal Constitution. After explaining that local law violations were not automatically federal violations and that rights created by local law should normally be vindicated in local courts, we turned to the possibility of pendent jurisdiction and rejected it:
In general, “principles of comity and the desirability of a ‘surer-footed reading of applicable law1 support the determination of state claims in state court.” [Citing Gibbs and a D.C. Circuit case.]3 Determination by the state court is especially important where the case involves “novel and unsettled” issues of state law.... Here, the law in question is new, its meaning ambiguous and sharply disputed. Moreover, the district court should not retain jurisdiction because this case directly implicates the processes by which a locality governs itself. Appellees’ remedy, if any, lies in the courts of the District of Columbia.
Id. at 169 (emphasis added). Similarly, in Angela R. by Hesselbein v. Clinton, 999 F.2d 320 (8th Cir.1993), while the court found the first step of Gibbs satisfied in a case under the Fourteenth Amendment, the Adoption Assistance and Child Welfare Act, and the Child Abuse Prevention and Treatment Act, and while it for unexpressed reasons did not regard the Pennhurst II bar as insurmountable, id. at 325, it cautioned the district court about devoting federal judicial resources to enforcement of a decree under a pendent state law claim — even where the state legislature had explicitly imposed the precise terms of the decree on the state welfare agency. Noting the “decidedly minor” interest of federal courts in remedying violations of state law, id. at 325-26, the court vacated the decree and remanded the case to the district court, with instructions to weigh the “constitutional, statutory, and institutional *566factors that bear upon the task of crafting a suitable equitable decree”. Id.8
It is true that we have upheld an injunction against District officials based solely on pendent District law grounds. See District of Columbia Common Cause v. District of Columbia, 858 F.2d 1, 10 (D.C.Cir.1988). But there our application of Gibbs step 2 found only that the district court had not abused its discretion by ordering an injunction against the District of Columbia prohibiting it from spending public funds to oppose citizens’ initiatives. Such a limited, prohibitory injunction offends the principle of comity to a far lesser extent than the wide-ranging, positive institutional reform injunction here.
So far, we have considered the two factors of a vanishing federal claim and an intrusive remedy based on local law grounds separately. But the two factors are closely related. Gibbs sets out a multi-factor balancing test, and presumably a factor pointing in one direction could offset one pointing in the other, and a strong factor pointing in one direction would reinforce a weak factor pointing in the same direction. So, for instance, the greater the intrusion on local government, the more substantial the federal claim should be. It seems at least doubtful that an intrusion such as the consent decree here could be supported where the Supreme Court has found two of the federal statutory provisions inadequate to ground a § 1983 claim — unless the remaining statutory clauses can be successfully distinguished.
Because the district court failed to consider the impact of Gibbs step 2 factors — particularly considerations of comity — on its exercise of jurisdiction, we remand the case with instructions on how to proceed. Cf. Webb v. Bladen, 480 F.2d 306, 309-10 (4th Cir.1973) (remanding case for Gibbs step 2 analysis where the district court failed to recognize that it had discretion under Gibbs to hear pendent claims and erroneously thought it was required to dismiss state claims for lack of jurisdiction).
Further Considerations on Remand
Any specific provision of the consent decree that can be grounded in what the court concludes is a winning federal claim should of course be preserved.9 The parties here have *567not briefed the precise impact of Suter on each of the various sections of the federal statutes at issue. Accordingly, we remand the case to the district court to make that inquiry in the first instance, in order to determine which, if any, federal grounds survive that could support an injunction. We note that at a minimum, the “reasonable efforts” clause of the Adoption Assistance Act, § 671(a)(15), relied on heavily by the district court, see 762 F.Supp. at 961, 970, 980, 986, and § 671(a)(9), related to its findings at id. at 986, are no longer enforceable under the express holding of Suter. See 503 U.S. at 359 n. 10, 364, 112 S.Ct. at 1368 n. 10, 1370.
Ancillary mandates may be appropriate under the court’s equitable jurisdiction, but only to the extent shown necessary to remedy violations that any decree properly addresses under the above standards. See Missouri v. Jenkins, — U.S. —, —, 115 S.Ct. 2038, 2056, 132 L.Ed.2d 63 (1995) (“[T]he remedial quality education program should be tailored to remedy the injuries suffered by the victims of prior de jure segregation. ... On remand, the District Court must bear in mind that its end purpose is not only ‘to remedy the violation’ to the extent practicable, but also ‘to restore state and local authorities to the control of a school system that is operating in compliance with the Constitution.’ ”) (citations omitted); Milliken v. Bradley, 433 U.S. 267, 280-81, 97 S.Ct. 2749, 2757, 53 L.Ed.2d 745 (1977) (“[T]he nature of the desegregation remedy is to be determined by the nature and scope of the constitutional violation_ [T]he federal courts in devising a remedy must take into account the interests of state and local authorities in managing their own affairs, consistent with the Constitution.”) (citations omitted). Specific provisions that cannot be supported on any of the above bases should be excised from the decree.
Plaintiffs cite Local No. 93 v. City of Cleveland, 478 U.S. 501, 106 S.Ct. 3063, 92 L.Ed.2d 405 (1986), for the proposition that defendants may and should be held to the provisions of the consent decree, even though those provisions exceed what the law requires. But as we explain in more detail below, the District, like defendants in Rufo v. Inmates of Suffolk County Jail, 502 U.S. 367, 112 S.Ct. 748, 116 L.Ed.2d 867 (1992), consented to the provisions of the decree only to the extent that they were necessary to remedy legal violations and has consistently objected to being subjected to more demanding standards. In Local No. 93, by contrast, the defendant city voluntarily accepted and continued to embrace extra obligations, to which only third parties objected. Compare Rufo, 502 U.S. at 389, 112 S.Ct. at 762-63 (“Federal courts may not order States or local governments, over their objection, to undertake a course of conduct not tailored to curing a constitutional violation that has been adjudicated.... [MJisunderstanding of the law could form a basis for modification. In this connection, we note again that the decree itself recited that it ‘sets forth a program which is both constitutionally adequate and constitutionally required.’ ”) (citations omitted).
* * *
The district court will confront a number of other issues on remand that the parties have adequately briefed and on which we offer the following guidance.
First, plaintiffs suggest that § XXII(C) forecloses termination of the consent decree, because it calls for termination only on the occurrence of an event — vacation of the district court’s original merits opinion in its entirety — that has not taken place. There are a number of flaws in this contention. First, § XXII(C) clearly contemplates partial relief calibrated to partial failure of the grounds on which the district court acted. The consent decree thus does not represent a “compromise [on] genuine uncertainties”, Evans v. City of Chicago, 10 F.3d 474, 478-79 (7th Cir.1993) (en banc), but rather a knuckling under to the district court’s legal conclusions, contingent on those conclusions being upheld on appeal. The decree specifies that “portions of this Order ... that are directly based on [any] part of the Memorandum Opinion that is vacated shall be null and void.” In a literal sense, of course, even the *568partial condition never had any serious possibility of being fulfilled, for appellate courts rarely vacate district court opinions as opposed to judgments. But the parties plainly did not bargain for so hyperteehnical a reading. Rather, since the provisions of the consent decree were negotiated as responses to specific findings of violations, the parties clearly anticipated that any undermining of those findings would require that the District be freed of the affected provisions. In effect, § XXII(C) is congruent with the lessons of Pennhurst II — that major changes in federal law may have the effect of undermining the jurisdictional basis of a continuing decree that is based only on local law. Accordingly, § XXII(C) cannot be read to preclude a consideration of SutePs impact on the decree’s legal foundation.
Resting heavily on their contrary reading of § XXII(C), plaintiffs object that defendants failed to move below for modification of the consent decree. Although the defendants did not specifically invoke Fed. R.Civ.P. 60(b)(5), authorizing relief when “it is no longer equitable that the judgment should have prospective application”, they indisputably sought relief on the grounds that Suter had eroded the basis for the decree, see Defendants’ Memorandum on Remand at 4 & n. 1 (Oct. 20, 1993), and the district court’s orders recognize as much, see, e.g., Order of Nov. 12, 1993, at 1 (referring to defendants’ requests that it “vacate or substantially modify the remedial order”).10 Even apart from § XXII(C), modification of a consent decree under Rule 60(b)(5) would proceed under “a flexible modification standard in institutional reform litigation because such decrees ‘reach beyond the parties involved directly in the suit and impact on the public’s right to the sound and efficient operation of its institutions.’ ” Rufo, 502 U.S. at 381,112 S.Ct. at 759 (citations omitted). See also Evans, 10 F.3d at 480, 477 (finding that any continued enforcement of a consent decree entered into by the City of Chicago had become “ ‘inequitable’ within the meaning of Rule 60(b)(5)” where the federal grounds had all been found wanting and continuing supervision would burden the district court and entangle it in the operations of “another sovereign”); Sweeton v. Brown, 27 F.3d 1162 (6th Cir.1994) (en banc) (similar).
Second, plaintiffs contend that insofar as Suter may draw in question the district court’s conclusion that the two federal statutes create rights enforceable under § 1983, Congress has reversed it. They point to the following 1994 enactment:
In an action brought to enforce a provision of the Social Security Act, such provision is not to be deemed unenforceable because of its inclusion in a section of the Act requiring a State plan or specifying the required contents of a State plan. This section is not intended to limit or expand the grounds for determining the availability of private actions to enforce State plan requirements other than by overturning any such grounds applied in Suter v. Artist M. [503 U.S. 347], 112 S.Ct. 1360 [118 L.Ed.2d 1] (1992), but not applied in prior Supreme Court decisions respecting such enforceability; Provided, however, that this section is not intended to alter the holding in Suter v. Artist M. that section [671(a)(15) ] of the Act is not enforceable in a private right of action.
Social Security Act Amendments of 1994, Pub.L. No. 103^32, § 211, 108 Stat. 4398, 4460 (Oct. 31, 1994) (to be codified at 42 U.S.C. § 1320a-2); see also Improving America’s Schools Act of 1994, Pub.L. No. 103-382, § 555, 108 Stat. 3518, 4057-58 (Oct. 20, 1994) (enacting the same language). Defendants respond that this instruction about “how to go about construing legislation enacted by other Congresses ... impermissibly intrudes on the judicial power.” Reply Br. at 22. We see no need to evaluate defendants’ constitutional argument, because we cannot see that the “instruction” bears on the interpretive issues before the district court.
Contrary to the assumptions of plaintiffs’ (and the dissent’s) arguments, Su-*569ter did not find provisions of the Adoption Assistance Act unenforceable “because of ... inclusion in a section of [the Act] requiring a State plan or specifying the required contents of a State plan.” 42 U.S.C. § 1320a-2. Rather, as we read Suter, its holding was based on the combination of (1) the Perm-hurst I requirement of a clear statement for any conditions on state receipt of federal grants, see 503 U.S. at 356,112 S.Ct. at 1366, (2) the absence of “statutory guidance ... as to how ‘reasonable efforts’ are to be measured”, id. at 360, 112 S.Ct. at 1368, and (3) the establishment of an alternative enforcement mechanism, see id. at 360-61, 112 S.Ct. at 1368-69. The discussion of the section’s being embedded in the requirements for a state plan, id. at 358, 112 S.Ct. at 1367, seems largely a way station to the point that alternative means of enforcement are provided. The Suter Court’s principal concern was the Act’s vagueness. In discussing the “reasonable efforts” clause, § 671(a)(15) (state must make “reasonable efforts” to prevent need for removing child from his home and to return child to home after removal), the Court noted: “How the State was to comply with this directive, and with the other provisions of the Act, was, within broad limits, left up to the State.” 503 U.S. at 360, 112 S.Ct. at 1368 (emphasis added).
This interpretation of Suter corresponds with the ordinary inquiry as to whether to infer a private right of action from a statutory scheme, see, e.g., Cort v. Ash, 422 U.S. 66, 95 S.Ct. 2080, 45 L.Ed.2d 26 (1975), and has been endorsed by other circuits as well. See, e.g., Marshall v. Switzer, 10 F.3d 925, 929 (2d Cir.1993) (“[T]he significant point in Su-ter was not that the statute in question only required a state to submit a plan to the federal agency but that the statute provided no guidance for measuring ‘reasonable efforts.’ ”); Wood v. Tompkins, 33 F.3d 600, 605-06 n. 14 (6th Cir.1994) (“[T]he [Suter ] Court noted that other sections of the Adoption Act, in contrast to § 671(a)(15), did indeed impose precise requirements on the states in addition to requiring the mere submission of a plan.... [T]he holding is simply that the one particular provision that was at issue is too amorphous to confer enforceable rights.”). Likewise, in Doe v. District of Columbia, Civ. No. 93-0092, slip op. at 17, 1994 WL 71547 (D.D.C. Feb. 25, 1994), the court found that a provision that had originally been relied on in this case, 42 U.S.C. § 5106a(b)(2) (a provision of the Child Abuse Prevention and Treatment Act), “invests only ‘generalized dut[ies]’ upon the states and does not ‘unambiguously confer’ an enforceable right to bring a private action against the government.” The court at no point relied on Congress’s having specified that the requirement should be embodied in a State plan.11 Of course neither the Social Security Act of 1994 nor the identical language of the Improving America’s Schools Act of 1994 did anything to supply more precise standards for the Adoption Assistance Act, or to alter the clear statement requirement or the Adoption Assistance Act’s non-judicial enforcement provisions; it thus changed none of the factors on which the Suter Court’s reasoning depended.
Congress may have been misled by the language of Suter1 s footnote 10, which says, after setting forth the language of § 671(a)(9), “As this subsection is merely another feature which the state plan must include to be approved by the Secretary, it does not afford a cause of action to the respondents anymore than does the ‘reasonable efforts’ clause of § 671(a)(15).” 503 U.S. at 359 n. 10, 112 S.Ct. at 1368 n. 10. In conjunction with the opinion’s text, discussed above, we read the footnote to state essentially that the combination of factors noted above applies with equal force to § 671(a)(9).
The legislative history of the identical 1994 provisions suggests that Congress may also have been confused by SutePs statement that “the Act does place a requirement on the States, but that requirement only goes so far as to ensure that the State have a plan approved by the Secretary which contains the 16 listed features.” 503 U.S. at 358, 112 *570S.Ct. at 1367. See H.R.Conf.Rep. No. 213, 103d Cong., 1st Sess. 863 (1993), reprinted in 1993 U.S.C.C.A.N. 1088, 1552 (quoting above sentence). The point of the sentence, however, is simply to distinguish what the statute specifically requires of states (adoption of a plan approved by the Secretary) from what — for the reasons given elsewhere in the opinion — it does not require. It certainly does not announce a general rule that mandatory inclusion in a plan demonstrates intent not to create an enforceable right.
The legislative history has another formulation of congressional purpose, but it is no more helpful than the reference to state plans. The Conference report stated that
The intent of this provision is to assure that individuals who have been injured by a State’s failure to comply with the Federal mandates of the State plan titles of the Social Security Act are able to seek redress in the federal courts to the extent that they were able to prior to the decision in Suter v. Artist M., while also making clear that there is no intent to overturn or reject the determination in Suter that the reasonable efforts clause to Title IV-E does not provide a basis for a private right of action.
H.R.Conf.Rep. No. 761, 103d Cong., 2d Sess. 926 (1994), reprinted in 1994 U.S.C.C.A.N. 2901, 3257 (emphasis added). But the Court decided Suter by applying previously established judicial criteria to a specific instance (§ 671(a)(15)) of a general question (whether Congress intended to create an enforceable right). The dissent in Suter, cited by the dissent here, was not correct in claiming that the Suter majority had “changed the rules of the game.” 503 U.S. at 377, 112 S.Ct. at 1377 (Blackmun, J., dissenting). Even if Congress could wipe Suter from the books, we think that lower courts addressing the same issue would have no basis for assuming that, if once again confronted with a Suter-like issue, the Supreme Court would not do just what it did in Suter. Thus, unless § 211 actually changed part of the test that led to the outcome in Suter (which, as we have said, it did not), courts should find equally vague provisions of similar acts equally unenforceable for the reasons that the Court found convincing in Suter.
Nor does the second sentence of the 1994 amendments assist plaintiffs. It reads:
This section is not intended to limit or expand the grounds for determining the availability of private actions to enforce State plan requirements other than by overturning any such grounds applied in Suter v. Artist M. [503 U.S. 347], 112 S.Ct. 1360 [118 L.Ed.2d 1] (1992), but not applied in prior Supreme Court decisions respecting such enforceability; Provided, however, that this section is not intended to alter the holding in Suter v. Artist M. that section [671(a)(15) ] of [the Act] is not enforceable in a private right of action.
(Emphasis added.) But Congress identified no “grounds” applied in Suter that were “not applied in prior Supreme Court decisions respecting ... enforceability.” Thus the sentence has no effect on the interpretive issues raised by the Adoption Assistance Act.
After what may seem a long journey, the core of our conclusion is no more than that we take seriously the Supreme Court’s statement in Gibbs that “the issue whether pendent jurisdiction has properly been assumed is one which remains open throughout the litigation.” 383 U.S. at 727, 86 S.Ct. at 1139. When a Supreme Court decision such as Suter flatly removes some of the supports of pendent jurisdiction, and at a minimum enfeebles most of the others, the “open” issue must be reexamined. To that end, we remand the case to the district court.
So ordered.
. See, e.g., Green v. Dept., of Commerce, 618 F.2d 836, 839 & n. 9 (D.C.Cir.1980); Amusement & Music Operators Ass'n v. Copyright Royalty Tribunal, 636 F.2d 531, 532-33 & nn. 2-4 (D.C.Cir.1980); Acton Corp. v. Borden, 670 F.2d 377, 380 n. 2 (1st Cir.1982); Crane Co. v. American Standard, 603 F.2d 244, 247 (2d Cir.1979); EEOC v. Neches Butane Products Co., 704 F.2d 144, 147 n. 2 (5th Cir.1983); Amen v. City of Dearborn, 718 F.2d 789, 794 (6th Cir.1983); Christianson v. Colt Industries Operating Corp., 798 F.2d 1051, 1056 (7th Cir.1986).
. The lone exception seems to be a case involving personal, not subject matter, jurisdiction. In the Matter of Oil Spill by the Amoco Cadiz Off the Coast of France on March 16, 1978, 954 F.2d 1279, 1292 (7th Cir.1992).
. Because this litigation was commenced before December 1, 1990, the new supplemental jurisdiction statute, 28 U.S.C. § 1367 (Supp. V 1993), does not apply.
. Cf. Evans v. City of Chicago, 10 F.3d 474, 479 (7th Cir.1993) (en banc) (holding that mere consent of local officials, without viable federal claim, could not justify continued enforcement of decree against city) ("the court must ensure that there is a substantial federal claim, not only when the decree is entered hut also when it is enforced, and that the obligations imposed by the decree rest on this rule of federal law rather than the bare consent of the officeholder") (emphasis added).
. We do not hold that the disappearance of federal claims from a case automatically deprives a federal court of jurisdiction over all types of pendent claims, which would be contrary to Rosado v. Wyman, 397 U.S. 397, 404-05, 90 S.Ct. 1207, 1213-14, 25 L.Ed.2d 442 (1970). The disappearance or weakening of federal claims are simply factors to be considered in step 2 of a Gibbs analysis. We do not understand why the dissent flogs us with "conflict[ing] directly with Rosado." Dissent at 576-577.
. See also Catalano v. Dept. of Hospitals, 299 F.Supp. 166, 175 (S.D.N.Y.1969) (refusing to exercise pendent jurisdiction under Gibbs to determine whether income exemption regulation conformed to state law, since issue "presents a question of state legislative intent and delegation of powers to an administrative body, which more appropriately lies within the ambit of the state court’s expertise with respect to its own laws”).
Although the District of Columbia is constitutionally distinct from the states, it may nevertheless be treated as a state for present purposes ....
. We note that the First Circuit has interpreted the term “state”, as used in the Eleventh Amendment, to include more than just the 50 obvious ones, ruling that Puerto Rico is covered by the Eleventh Amendment "in the same manner, and to the same extent, as if [it] were a State.” See De Leon Lopez v. Corporacion Insular de Seguros, 931 F.2d 116, 121 (1st Cir.1991). But we decline to follow the First Circuit’s approach here.
. See also Evans v. City of Chicago, 10 F.3d 474 (7th Cir.1993) (en banc). Although Evans involved a pure "municipal corporation”, the city of Chicago, the court recognized the force of comity principles even as to political entities not covered by the Eleventh Amendment. When making inquiries into jurisdictional support for a consent decree, "courts are bound by principles of federalism (and by the fundamental differences between judicial and political branches of government) to preserve the maximum leeway for democratic governance.” Id. at 479.
The dissent appears to suppose that we rely on the Eleventh Amendment. Dissent at 578-579. If for some reason the paragraphs above leave any doubt on the subject, No, we do not; we rely primarily on circuit precedent affording a similar (and doubtless weaker) comity to the District of Columbia, with support from other circuits recognizing the role of comity where it was not (or was perceived as not) mandated by Pennhurst II.
. We express no opinion as to the merits of the district court's original finding that the post-custody children have valid constitutional claims under the dictum of DeShaney v. Winnebago County Dept. of Social Services, 489 U.S. 189, 201 n. 9, 109 S.Ct. 998, 1006 n. 9, 103 L.Ed.2d 249 (1989) (“Had the State by the affirmative exercise of its power removed Joshua from free society and placed him in a foster home operated by its agents, we might have a situation sufficiently analogous to incarceration or institutionalization to give rise to an affirmative duty to protect.”). If the district court does reach the issue on remand and adheres to its ruling, it should consider the possibility of bifurcating the class, since in-custody plaintiffs raise issues distinct from those who rely entirely on federal and local statutes.
We note in this connection that although courts ordinarily refrain from deciding constitutional questions when a case can be resolved on the basis of state law, see Ashwander v. TVA, 297 U.S. 288, 346-47, 56 S.Ct. 466, 482-83, 80 L.Ed. 688 (1936) (Brandeis, J., concurring); Siler v. Louisville & Nashville R.R. Co., 213 U.S. 175, 191, 29 S.Ct. 451, 454, 53 L.Ed. 753 (1909), Pennhurst II "had the effect of overruling the entire history of Siler/Ashwander doctrine in cases where injunctive relief is requested against the state.” Cuesnongle v. Ramos, 835 F.2d 1486, 1497 (1st Cir.1987) (citing Pennhurst II, 465 U.S. at 123, 104 S.Ct. at 920 (majority opinion); 465 U.S. at 159-63, 104 S.Ct. at 939-42 (Stevens, J., dissenting)). The comity afforded the District under Grano v. Barry in the application of Gibbs step 2, paralleling the rule of Pennhurst II, similarly implies that the district court may not exercise jurisdiction over the pendent claims merely to avoid confronting the reverse-DeShaney issue. Of course Ashwander would under no circum*567stances suggest any need to refrain from addressing the issues of federal statutory law.
. Even if the defendants had not sought relief on these grounds, we would still be required to investigate the basis of the district court’s jurisdiction on our own initiative. See Minker v. Baltimore Annual Conference of United Methodist Church, 894 F.2d 1354, 1361 (D.C.Cir.1990); Maguire v. Marquette University, 814 F.2d 1213, 1218 & n. 4 (7th Cir.1987).
. We note that two district courts interpreting § 211 have deferred to the congressional view that inclusion in a plan was central to Suter's analysis — see Jeanine B. v. Thompson, 877 F.Supp. 1268, 1283 (E.D.Wis.1995); Harris v. James, 883 F.Supp. 1511, 1520 (M.D.Ala.1995)— but we disagree.