POSNER, Circuit Judge.

The district judge, in this suit by participants in an ERISA pension plan, has asked us, and we have agreed, to entertain an interlocutory appeal from a ruling in which she answered in favor of the plaintiffs a potentially controlling question of law that had arisen in the course of the litigation. 28 U.S.C. § 1292(b). (The plaintiffs have cross-appealed, but since they are defending rather than attacking the judge's ruling, the cross-appeal is improper, Rose Acre Farms, Inc. v. Madigan, 956 F.2d 670, 672 (7th Cir. 1992), and is hereby dismissed.) The question is the correct approach to deciding whether an ERISA pension plan — in this case a defined-contribution plan in which the employer matched contributions that its employees made by means of payroll deductions to individual retirement accounts — has been partially terminated.

Provided that certain requirements are met, the interest or other earnings in an individual retirement account are not taxed as they accrue. John D. Colombo, "Paying for the Sins of the Master: An Analysis of the Tax Effects of Pension Plan Disqualification and a Proposal for Reform," 34 Ariz. L.Rev. 53, 56-58 (1992); see 26 U.S.C. §§ 402(a), 501(a). Suppose the employer terminates the plan. Were it not for the special rule on terminations that is the focus of this case, an employee whose pension entitlement had not yet fully vested would receive (i.e., would be deemed fully vested as to), over and above the vested portion of the employer's contribution, only the contributions he had made to his retirement account, plus the earnings on them. The portion of the employer's contributions that had not yet vested would revert to the employer, compare 29 U.S.C. § 1053(a)(2) with id. § 1053(a)(1), yielding it a tax benefit because the amount by which its contributions had grown as a result of the pension plan's investing them would have escaped being taxed. But this is where the special rule clicks in: in the event of termination the rights of all the participants in the terminated plan vest in full, so that none of the money that the employer contributed is returned to it. 26 U.S.C. § 411(d)(3).

The purpose of the rule is to prevent plan terminations motivated by the prospect of a tax windfall. Matz v. Household International Tax Reduction Investment Plan, 227 F.3d 971, 975 (7th Cir. 2000), vacated on other grounds, 533 U.S. 925, 121 S.Ct. 2545, 150 L.Ed.2d 713 (2001) (per curiam); Bruch v. Firestone Tire Rubber Co., 828 F.2d 134, 151 (3d Cir. 1987), reversed in part on other grounds, 489 U.S. 101, 109 S.Ct. 948, 103 L.Ed.2d 80 (1989); Vincent Amoroso et al., "A Policy Premise Approach to Partial Terminations," New York University Review of Employees Benefits and Executive Compensation § 8.01[3], pp. 8-9 (2002); E. Thomas Veal Edward R. Mackiewicz, Pension Plan Terminations 364-65 (2d ed. 1998). We are unconvinced by an alternative rationale sometimes suggested for the rule — to protect nonvested employees' expectations of receiving pension benefits. Tipton Kalmbach, Inc. v. Commissioner, 83 T.C. 154, 160-61, 1984 WL 15598 (1984); see also Matz v. Household International Tax Reduction Investment Plan, supra, 227 F.3d at 975; Halliburton Co. v. Commissioner, 100 T.C. 216, 227-28, 1993 WL 83409 (1993). Until his pension benefits have vested, an employee at will, lacking as he does any job tenure, has no reasonable expectations of receiving benefits. The point of vesting is to create such an expectation.

To prevent evasion, the rule requiring immediate vesting of all participants' pension benefits applies to "partial" terminations as well as to complete ones. The statute does not define "partial termination," however, although a Treasury Regulation tells the IRS to base the determination on "all the facts and circumstances of a particular case." 26 C.F.R. § 1.1411(d)-2(b)(1). Despite the phrasing of the regulation, and "despite their origin in tax law, disputes as to whether a partial termination has occurred rarely involve the IRS, which has been a party to only a small minority of the reported cases and rulings." Veal Mackiewicz, supra, at 363. This case is not one of the small minority. (Actually not such a small minority, as the table later in this opinion reveals.) The case law has assumed that the regulation is intended to guide adjudicators as well as the IRS, and we shall indulge the assumption.

So vague a regulation is no help to anyone. But some years ago the IRS, in an amicus curiae brief filed in Weil v. Retirement Plan Administrative Committee, 933 F.2d 106 (2d Cir. 1991), suggested that, with an important qualification that we'll take up at the end of this opinion, a pension plan should be deemed partially terminated if at least 20 percent of the plan's participants lose coverage. The IRS was putting welcome flesh on a skeletal regulation, and the court in Weil deferred to the IRS's position on the basis of the Chevron principle. 933 F.2d at 110. So did we the first time this protracted litigation came before us. Matz v. Household International Tax Reduction Investment Plan, supra. The Supreme Court, however, vacated our decision, 533 U.S. 925, 121 S.Ct. 2545, 150 L.Ed.2d 713 (2001) (per curiam), in light of the just-decided United States v. Mead Corp., 533 U.S. 218, 121 S.Ct. 2164, 150 L.Ed.2d 292 (2001), where the Court had ruled that informal agency actions are not to receive Chevron deference. A position stated in an amicus curiae brief has seemed to us a good example of what the Court had in mind. Keys v. Barnhart, 347 F.3d 990, 993-94 (7th Cir. 2003); see also Matz v. Household International Tax Reduction Investment Plan, 265 F.3d 572, 574-75 (7th Cir. 2001); Doe v. Mutual of Omaha Ins. Co., 179 F.3d 557, 563 (7th Cir. 1999); cf. In re New Times Securities Services, Inc., 371 F.3d 68, 80-82 (2d Cir. 2004).

On remand from the Supreme Court, freed from the IRS incubus we "adopt[ed] the rule that only non-vested participants should be counted in determining whether partial termination of a pension plan has occurred." 265 F.3d at 576. By "non-vested" we meant not fully vested. ERISA requires that at least 20 percent of the employee's benefits vest at the end of the third year, another 20 percent at the end of the fourth, and so on, so that by the end of seven years the employee is fully vested. 29 U.S.C. § 1053(a)(2)(B). (The plan can provide for more rapid vesting, § 1053(d), and this one did; vesting was complete in five years rather than seven.) The case went back down to the district court, where the question arose whether we had meant that vested participants should be excluded only from the numerator or also from the denominator in deciding how "partial" the termination had been. The 20 percent figure in the IRS's amicus brief in Weil was the percentage of the plan's participants who were terminated, irrespective of how many either of them or of the remaining participants were fully vested.

From a favorable reference in our first opinion, see 227 F.3d at 975-76, to In re Gulf Pension Litigation, 764 F.Supp. 1149 (S.D.Tex. 1991), affirmed under the name Borst v. Chevron Corp., 36 F.3d 1308 (5th Cir. 1994), the district judge inferred that we would exclude from both numerator and denominator all fully vested participants, although we hadn't said that. So if the plan had 200 participants, 50 lost their plan coverage, 10 of those were not fully vested, and the total number of not fully vested participants was 20, the relevant percentage would be not 25 percent (50 ÷ 200) but 50 percent (10 ÷ 20). The judge reasoned that since the effect of termination, either from the participants' standpoint or from the tax standpoint, is limited to those who aren't fully vested, they are the only participants who should be considered in deciding whether a partial termination has occurred. The plan agrees that only participants who were not fully vested should be in the numerator, but it argues that all the participants should be in the denominator, which would change the percentage in our hypothetical example from 50 percent to 5 percent (10 ÷ 200).

Which approach is right? And what difference does it make in this case, where, as a result of a series of reorganizations of subsidiaries of Household, a total of 2,396 of the 11,955 participants in Household's plan ceased to be participants? We can and shortly will select what we believe to be the correct approach. But because of unresolved issues in the district court (remember that the case is here on interlocutory appeal), we can't use the approach to generate a definite percentage in this case.

One of the unresolved issues is whether the terminations should be treated as a single termination. They were closely related in time (all occurred between the end of August 1994 and the end of June 1996) and appear to have had the same motive, unlike the two partial terminations in Administrative Committee of Sea Ray Employees' Stock Ownership Profit Sharing Plan v. Robinson, 164 F.3d 981, 987-88 (6th Cir. 1999), which had unrelated causes. See Matz v. Household International Tax Reduction Investment Plan, supra, 227 F.3d at 976-77 and 265 F.3d at 576; Weil v. Retirement Plan Administrative Committee, 750 F.2d 10, 13 (2d Cir. 1984). But there is no actual finding by the district court.

Likewise unresolved are whether the figure for participants whose coverage is canceled includes any of the employees who Household asserts left voluntarily (the plaintiffs claim on the contrary that those employees were constructively discharged) and therefore shouldn't count in determining whether a partial termination has occurred, Sage v. Automation, Inc. Pension Plan Trust, 845 F.2d 885, 891-92 (10th Cir. 1988); whether only participants who were employed by the reorganized entities, as opposed to Household itself, should be counted; and how to treat participants who became fully vested during or at the end, rather than at the beginning, of the reorganizations. By our calculations (the district court can redo them if we've made a mistake), the percentage reduction in coverage under the IRS's approach ranges from 15.4 percent, if all three issues are resolved in the plan's favor, to 35.8 percent if all three issues are resolved in favor of the plaintiffs. The corresponding percentages under the district court's (and the plaintiffs') approach are 13.5 percent and 79.8 percent, and under the plan's approach they are 7.2 percent and 16.4 percent. We do not know what further adjustments would be necessary if the terminations were treated separately rather than as one.

Although our decision in the last appeal of this case gestured toward the Gulf Pension approach (not fully vested over not fully vested, the district court's and the plaintiffs' preferred approach), it did not adopt that approach explicitly. This has opened the way for the plan to argue as we noted that while the numerator should be limited to the not fully vested, the denominator should not be. It is true that if only a very small percentage of plan participants lose benefits, the policy of the statute is not strongly engaged. But that does not justify the plan's approach, which has bizarre consequences. For suppose, in a variant of our hypothetical case, that the plan was terminated as to all but 10 of the 200 participants and of those 10, five were not fully vested. On the plan's view, this would be only a 2.5 percent termination (5/200), and hence not a partial termination on anyone's view. But how can a reduction in the coverage of a plan from 200 to 10 employees — a reduction of 95 percent — not be considered even a partial termination? To say it is not would do extreme violence to the language of the statute. But so does the district court's approach. Suppose that only one of the 200 participants is nonvested, and now the plan is modified so that he loses coverage. Is this a partial termination? Under the district court's approach, absolutely — the termination percentage is 100 percent.

Where both approaches go astray is in confusing the purpose of a statute with its terms, a common error. Brogan v. United States, 522 U.S. 398, 402-04, 118 S.Ct. 805, 139 L.Ed.2d 830 (1998); Board of Governors of Federal Reserve System v. Dimension Financial Corp., 474 U.S. 361, 373-75, 106 S.Ct. 681, 88 L.Ed.2d 691 (1986); Wood v. Thompson, 246 F.3d 1026, 1035 (7th Cir. 2001); United States v. Medico Industries, Inc., 784 F.2d 840, 844 (7th Cir. 1986). (A further stumble was to assume that the 20 percent figure, which the IRS adopted with reference to all participants, fully vested and not, could have any application when all participants are replaced, in either numerator or denominator or both, with not fully vested participants. Such an alteration would change the premises of the IRS's choice of the 20 percent benchmark.) The purpose of section 411(d)(3) of the Internal Revenue Code is to prevent an employer from obtaining a tax windfall at the expense of the not fully vested participants in his plan. But the statute does not provide that plan alterations which result in a tax windfall at the expense of such participants shall be deemed terminations and precipitate full vesting. It provides that terminations precipitate full vesting, and to prevent evasion adds that terminations include partial terminations.

The natural way to decide whether a partial termination has occurred is to see how close it is to a complete termination. On the one hand, clearly an employer shouldn't be able to get away with ejecting 99 percent of the plan's participants. On the other hand, no one is arguing that an employer should be forbidden to alter his plan in the slightest degree without forfeiting tax benefits. Such a rule would go far toward erasing the distinction between fully vested and not fully vested employees. For every time the plan was altered to remove even a small handful of not fully vested participants, there would be a case for treating the alteration as a partial termination, requiring immediate full vesting of all not yet fully vested participants.

So where to draw the line? The IRS, which is not famous for encouraging tax windfalls, draws it at 20 percent. As we look back upon the course of this litigation, now in its ninth year and its third interlocutory appeal to this court, we find our-selves drawn back to the IRS's position. Not because it is entitled to Chevron deference — we adhere to our holding that it is not — but because, having toyed with the alternatives, we think it is the best available and we respect the IRS's experience in formulating tax rules.

But what about that Treasury Regulation we quoted earlier, which tells the IRS and presumably us as well to base the determination of whether a partial termination has occurred on "all the facts and circumstances of a particular case"? This language has given rise to judicial formulations like the following that are distressingly vague: "The regulation's plain language clearly directs us to consider all the facts and circumstances, of which the percentage of excluded plan participants is but one, albeit generally the most persuasive one. Of course, in a particular case the percentage may be so high or so low as to be determinative standing alone, but as a general matter we must look beyond the mere percentages unless and until Congress or the Treasury Department provides otherwise." Kreis v. Charles O. Townley, M.D. Associates, P.C., 833 F.2d 74, 79-80 (6th Cir. 1987) (citations omitted); see also Administrative Committee of Sea Ray Employees' Stock Ownership Profit Sharing Plan v. Robinson, supra, 164 F.3d at 987; Freeman v. Central States, Southeast Southwest Areas Pension Fund, 32 F.3d 90, 92 n. 5 (4th Cir. 1994); cf. Gluck v. Unisys Corp., 960 F.2d 1168, 1183 (3d Cir. 1992).

We acknowledge that even the 20 percent rule proposed in the IRS's amicus brief in Weil was not intended to be hard and fast despite its appearance of mathematical exactitude. See Brief for the United States as Amicus Curiae in Weil v. Retirement Plan Administrative Committee, pp. 7-8; Veal Mackiewicz, supra, at 368-69; Halliburton Co. v. Commissioner, supra, 100 T.C. at 237; Internal Revenue Manual § 7.12.1.2.7, http://www.irs.gov/irm/index.html. And yet the following table (adapted from Veal Mackiewicz, supra, at 367), which summarizes the cases (treating cases involving distinct plans as multiple cases) and rulings on partial termination, reveals the surprising robustness of the 20 percent benchmark: PARTIAL TERMINATION CASES AND RULINGS Total Number Who Percentage Who Partial Case or Ruling Paricipants Lost Coverage Lost Coverage Termination? Collignon v. Reporting Services Co., 796 F.Supp. 1136 Peter M. Boruta M.D., P.C. v. Commissioner, 55 T.C.M. 670 Tipton Kalmbach, Inc. v. Commissioner, supra In re Gulf Pension Litigation, supra In re Gulf Pension Litigation, supra Weil v. Retirement Plan Administrative Committee, Tipton Kalmbach, Inc. v. Commissioner, supra Administrative Committee of Sea Ray Employees' Stock Ownership Profit Sharing Plan v. Robinson, Halliburton Co. v. Commissioner, supra Administrative Committee of the Sea Ray Employees' Stock Ownership Profit Sharing Plan v. Robinson, supra Morales v. Pan American Life Ins. Co., 718 F.Supp. 1297 1303 914 F.2d 83 Kreis v. Townley, supra Babb v. Olney Paint Co., 764 F.2d 240 Kreis v. Townley, supra Wishner v. St. Luke's Hospital Center, 550 F.Supp. 1016 Ehm v. Phillips Petroleum Co., 583 F.Supp. 1113 Bruch v. Firestone Tire Rubber Co., 640 F.Supp. 519 530

In an effort to make the law as certain as possible without opening up gaping loopholes, we shall generalize from the cases and the rulings a rebuttable presumption that a 20 percent or greater reduction in plan participants is a partial termination and that a smaller reduction is not. How rebuttable? One can imagine cases in which a somewhat smaller reduction in the percentage of plan participants would be tax-driven and might on that account be thought a "partial" termination, and other cases, like Sea Ray, in which the reduction is perhaps not so far above 20 percent that further inquiry is inappropriate. We assume in other words that there is a band around 20 percent in which consideration of tax motives or consequences can be used to rebut the presumption created by that percentage. A generous band would run from 10 percent to 40 percent. Below 10 percent, the reduction in coverage should be conclusively presumed not to be a partial termination; above 40 percent, it should be conclusively presumed to be a partial termination.

We have considered whether we should invite the IRS to submit an amicus curiae brief advising us of its current view of the proper approach to determining partial termination. We have decided not to do so because of the great age of the case. Obviously should the IRS decide on its own to revisit the issue, we would give its views significant weight and therefore the rule we have just formulated for deciding such cases as this should be considered tentative.

The range of possible reduction-of-coverage percentages in the present case — 15.4 percent to 35.8 percent (treating the series of terminations as a single event — if on remand the district court rejects that treatment this will affect the range, perhaps decisively) — lies within the band, and so a remand will be necessary in which the district court will have to consider additional "facts and circumstances." But given that the statutory purpose is to prevent tax windfalls, it seems to us that the only relevant facts and circumstances should be the tax motives and tax consequences involved in the reduction in plan coverage.

VACATED AND REMANDED.