Full opinion text
SCALIA, Circuit Judge: This is an appeal from the District Court’s grant of summary judgment in favor of the Internal Revenue Service, in a Freedom of Information Act suit brought by the Church of Scientology under 5 U.S.C. § 552(a)(4)(B) (1982). The only issue addressed by this en banc opinion is the meaning of the so-called Haskell Amendment, which excepts from the Internal Revenue Code’s definition of nondisclosable “return information” “data in a form which cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer.” 26 U.S.C. § 6103(b)(2) (1982). Specifically, we consider whether to adhere to a 1981 panel decision of this court which held that that provision removes from the defined category of protected information all material which, either in its original form or as redacted in response to a FOIA request, does not disclose the identity of the taxpayer to whom it pertains. I The facts of the present case are set forth in the panel opinion issued simultaneously with this opinion. For present purposes, it suffices to recite that the central issue in the appeal is the adequacy of the IRS’s search for requested records; that one of the principal points bearing upon that issue is whether certain files could reasonably be excluded from the search as containing only “return information”; and that the latter point depends to a considerable extent upon whether redaction (specifically, elimination of portions of documents that would disclose the taxpayer’s identity) removes the material from the protected category. After the case had been briefed and argued before the assigned panel, the . court en banc, on its own motion, requested supplemental briefing and, on December 5, 1985, heard oral argument limited to the following issue: Should the Court adhere to the interpretation of 26 U.S.C. § 6103(b)(2) adopted by the panel opinion in Neufeld v. IRS, 646 F.2d 661, 665 (D.C.Cir.1981), or should it adopt a different interpretation, in particular that announced by the Seventh Circuit in King v. IRS, 688 F.2d 488, 490-94 (7th Cir.1982)? Briefs amicus curiae were received from the American Civil Liberties Union Foundation of Washington and from Professor John L. Neufeld and the Freedom of Information Clearinghouse. II In relevant part, 26 U.S.C. § 6103(a) provides as follows: Returns and return information shall be confidential, and except as authorized by this title— (1) no officer or employee of the United States,____ shall disclose any return or return information obtained by him in any manner in connection with his service as such an officer or an employee or otherwise or under the provisions of this section____ Willful violation of this provision is a felony. 26 U.S.C. § 7213(a)(1). “Return information” is defined in the statute as follows: (A) a taxpayer’s identity, the nature, source, or amount of his income, payments, receipts, deductions, exemptions, credits, assets, liabilities, net worth, tax liability, tax withheld, deficiencies, over-assessments, or tax payments, whether the taxpayer’s return was, is being, or will be examined or subject to other investigation or processing, or any other data, received by, recorded by, prepared by, furnished to, or collected by the Secretary with respect to a return or with respect to the determination of the existence, or possible existence, of liability (or the amount thereof) of any person under this title for any tax, penalty, interest, fine, forfeiture, or other imposition, or offense, and (B) any part of any written determination or any background file document relating to such written determination (as such terms are defined in section 6110(b)) which is not open to public inspection under section 6110, but such term does not include data in a form which cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer. 26 U.S.C. § 6103(b)(2) (emphasis added). The last clause in the defining paragraph is the Haskell Amendment, so called because it was inserted into the committee-proposed bill through a floor amendment introduced by that Senator. On the basis of that clause, the Ninth Circuit held in 1979 that data that do not identify a particular taxpayer because names, identifying numbers and other similar information have been deleted are not return information. Long v. IRS, 596 F.2d 362 (9th Cir. 1979), cert. denied, 446 U.S. 917, 100 S.Ct. 1851, 64 L.Ed.2d 271 (1980). In a later case before this court in which the IRS had not briefed the question, the panel found it necessary to reach the issue and, without analysis of its own, followed what was at the time the only court of appeals precedent. Neufeld v. IRS, 646 F.2d 661, 665 (D.C.Cir.1981). In so doing, the panel observed that “[w]hile the IRS wishes to reserve the question of the proper statutory definition of return information for another day, it appears to concede, for this case only, that [no harmful error occurred] if in fact [the district court] employed the definition of return information articulated in Long.” Id. (footnote omitted). Subsequently, the Seventh Circuit reached a conclusion different from Long, holding that the statute “protects from disclosure all non-amalgamated items listed in subsection (b)(2)(A), and that the Haskell Amendment provides only for the disclosure of statistical tabulations which are not associated with or do not identify particular taxpayers.” King v. IRS, 688 F.2d 488, 493 (7th Cir.1982). The newly emerged circuit conflict has induced us to reconsider the position stated in our 1981 panel decision. The starting point of analysis, of course, is the text of the provision at issue, which, we agree with the Seventh Circuit, is ill suited to achieve the result pronounced in Long. It would be most peculiar to catalogue in such detail, in subparagraph (A) of the body of the definition, the specific items that constitute “return information” (e.g., “income, payments, receipts, deductions, exemptions, credits, assets, liabilities, net worth, tax liability, tax withheld, deficiencies, over-assessments, or tax payments, ... or any other data, received by, recorded by, prepared by, furnished to, or collected by the Secretary with respect to a return”) while leaving to an afterthought the major qualification that none of those items counts unless it identifies the taxpayer. Such an intent would more naturally have been expressed not in an exclusion (“but such term does not include ... ”) but in the body of the definition — by stating, for example, that “the term ‘return information’ means the following information that can be associated with or identify a particular taxpayer: ____” If the intended scope of the exclusion is as broad as Long holds, the structure of the provision is akin to defining mankind as “all mammals in the world, but excluding those that are not relatively hairless bipeds with the power of abstract reasoning.” While such a form of definition is conceivable, it would constitute “everyday language” (as the dissent characterizes it, Dissent at 174) only for one of Lewis Carroll’s characters, and it hardly takes “talmudic dissection!]” or “microscopic scrutiny,” id., to reject it as implausible. The Long interpretation produces a similarly mindless consequence in subparagraph (B) of the definition of return information. That subparagraph includes within the definition of return information IRS-written determinations and related background files that are not open to public inspection under § 6110. The latter section excludes from the public inspection requirement not only identifying data, § 6110(c)(1), but many other matters, such as trade secrets, § 6110(c)(4), information prepared for the use of an agency regulating financial institutions, § 6110(c)(6), and (with respect to most written determinations) material relating to a taxpayer’s change of annual accounting period, § 6110(g)(5)(B)(ii). It would be absurd to incorporate these exclusions so precisely into the body of the definition of return information, and then, in the immediately following clause, to write all of them back out — except the identifying data exclusion (§ 6110(c)(1)), which is not deleted by the Haskell exclusion but merely rendered entirely redundant. We also agree with the Seventh Circuit that the formulation of the Haskell provision itself suggests something other than merely the absence of identifying information. It would be strange to express the latter thought by excluding “data in a form which cannot be associated with, or otherwise identify ... a particular taxpayer” (emphasis added). The emphasized phrase would be superfluous for that purpose, as reading the provision without it will demonstrate. A more natural formulation for the purposes which Long assigns would be similar to that contained in the provision of FOIA that “an agency may delete identifying details,” 5 U.S.C. § 552(a)(2) (emphasis added); or similar to the formulation used elsewhere in this same Subchapter of the Internal Revenue Code, that no publication shall “permit ... information ... to be associated with, or otherwise identify, directly or indirectly, a particular taxpayer,” 26 U.S.C. § 6108(c). Moreover, it is curious usage to describe an item of return information (a particular taxpayer’s tax “payments,” for example) as having one “form” when made public in a document that includes the taxpayer’s name, and taking a different “form” when made public in the very same document with only the name deleted. What is suggested by the language of the provision itself is strongly confirmed by other provisions of § 6103. Subsections 6103(f)(1) & (2) and subsection 6103(f)(4)(A) permit disclosure of return information to certain committees of Congress, and subsection 6103(f)(4)(B) to the full Senate or House; under all four provisions, however, unless the taxpayer consents in writing the disclosure must be made to the pertinent committee or house “sitting in closed executive session” when it concerns “return information which can be associated with, or otherwise identify, directly or indirectly, a particular taxpayer.” If Long's interpretation of the Haskell Amendment is adopted, the exception in these provisions completely consumes the rule. That is to say, if return information consists, as Long says, of nothing but identifying data, then whenever it is provided under these provisions the receiving committee or house must sit in executive session. Quite plainly, these provisions contemplate return information that is nonidentifying. In addition to clear textual indications, rejection of the Long interpretation is suggested by assessment of plausible legislative intent. It is of course true, as two of the amici have asserted, that there is no reason “why Congress would have wanted to forbid the disclosure of information which would not threaten the privacy of individual taxpayers.” Brief of Neufeld and Freedom of Information Clearinghouse at 5. But it is also true that the threat to privacy is not entirely eliminated by agency and (ultimately) judicial assessment that certain deletions in response to a FOIA request will suffice to conceal the taxpayer’s identity. The protection afforded by such assessment is always problematic, not only because of the risk of human error, but also because the assessment depends to a large extent upon uninformed estimations as to what data the requester possesses. Consider, for example, a FOIA request for the amounts and beneficiaries of all charitable deductions claimed by taxpayers within a particular postal ZIP code area during a particular tax year. That information would normally not identify the charitable gift of any particular taxpayer; but it would do so if the requester had been told by his neighbor that the latter made a charitable gift last year of $2,775. For most information possessed by the government, Congress has determined that the risk of occasional unknowing disclosure of facts entitled to be withheld under FOIA is outweighed by the benefits of openness. But it has not made that judgment for all information. See, e.g., 50 U.S. C.A. § 431 (West Supp.1985) (exempting Central Intelligence Agency operational files from FOIA). It is significant that FOIA’s nonidentification protection has not been considered adequate for the other major category of personal information that the government directs all its citizens to provide: Under the same Exemption 3 at issue here, 5 U.S.C. § 552(b)(3), all census data are protected from disclosure, whether or not they identify the individual to whom they pertain. See Baldridge v. Shapiro, 455 U.S. 345, 102 S.Ct. 1103, 71 L.Ed.2d 199 (1982). We think similarly heightened protection was intended with regard to tax information, in order to encourage the full, voluntary self-assessment of taxes upon which our internal revenue system largely depends. The intent to provide this increased assurance of confidentiality is conveyed by the detailed provisions of § 6108 rigidly restricting the use of tax information within the government itself, and by the severe criminal penalty (up to five years imprisonment) for unlawful disclosure. See 26 U.S.C. § 7213(a)(1). It is particularly apparent, however — and the incompatibility of the Long interpretation is particularly clear — from the provisions of § 6110, which set forth procedures for public inspection of IRS written determinations and related background files. Unlike most governmental information obtainable under the Freedom of Information Act, which one or more members of the public may be interested in for reasons that amount to no more than curiosity, there is special reason for making written determinations public, since without such a requirement agencies could develop “secret law.” Thus, FOIA requires such determinations not merely to be provided upon written request, but to be made available in the agency’s reading room, and to be reflected in a current index that is publicly distributed. 5 U.S.C. § 552(a)(2). Yet in the case of tax information, § 6110 provides greater protection against improper disclosure of this publicly essential information than FOIA provides against disclosure of data in which there is no reason to posit any public need to know. Specifically, the subject of the written determination is given a right to prior written notice of the Secretary’s intention to disclose, an administrative remedy to prevent the disclosure, a cause of action in the Tax Court if that remedy is unsuccessful, a right to intervene in any action seeking disclosure, and even a cause of action for damages in the Claims Court for improper disclosure. 26 U.S.C. § 6110(f), (i). In the judicial proceedings to restrain disclosure or to require further disclosure, there is no requirement similar to the provision of FOIA that “the burden is on the agency to sustain” the withholding. See 5 U.S.C. § 552(a)(4)(B). It would be absurd to provide such guarantees against disclosure of identifying information in the context of written determinations while relying upon no more than the FOIA protections (through Long’s interpretation of the Haskell Amendment) when a request for less publicly important return information is received. The dissent criticizes our use of standard textual analysis on the ground that, while it may be appropriate where Congress “labored arduously over each choice of word and each comma,” it is improper “when the legislative history shows that a provision was injected into the bill at the tail end of the process.” Dissent at 174. We need not pause to consider the theoretical deficiencies of such an approach to statutory construction, since it is in any case not properly applicable here. The (ill-considered) Haskell Amendment was not adopted separately and distinctly from the other provisions that we seek to reconcile with it. As we noted earlier, it was not an amendment to a preexisting law, but an amendment to the bill as originally presented on the floor. Congress did not pass into law the Haskell amendment by itself, but as part and parcel of an exceedingly detailed and complex legislative scheme, on which it had “labored arduously over each choice of word and each comma.” Since all the provisions were enacted simultaneously, there is no plausible justification for focusing on the hastily considered nature of one of them and ignoring the carefully crafted character of the remainder. In fact, far from militating in favor of the broad Long interpretation, the last-minute and cursory manner in which the Haskell Amendment was proposed and adopted greatly augments the implausibility of that interpretation. The massive effect of the amendment, if Long is correct, was to change the scope of protection from all “return information,” as carefully and expansively described in § 6103(b)(2), to merely all such information which would identify the taxpayer. That change would not only make superficially nonidentifying information available to FOIA requesters, and thereby frustrate the carefully drawn protections against such disclosure contained in § 6110, but it would also allow such information to be circulated freely within the government (since the defined term “return information” is central to both those provisions governing public disclosure and those governing inter-agency dissemination). We are asked to believe that this fundamental change in the committee proposal that the members of the Senate had (presumably) studied was made by this brief proviso at the last minute, without any statement by its sponsor that it had an effect upon anything except statistical studies and compilations of data, see infra at 161, without a floor vote (it was adopted by consent), without dissent from even a single member of the Senate, and indeed without any comment by any members of the body who might have been present except Senator Long’s remark: “Mr. President, I will be happy to take this amendment to conference. It might not be entirely necessary, but it might serve a good purpose.” 122 Cong.Rec. 24,012 (1976). Rather than embrace this implausibility it would make more sense — if one were to favor the dissent’s approach of using supposed inadequacy of consideration as a basis to ignore, rather than seeking to reconcile, textual conflicts — to endorse the position urged by the government, to wit, that all the befuddled Congress meant to do (never mind that the text will not bear it, see infra at 161-62) was to add the tax model to the disclosure exceptions of § 6108. Ill It is much easier to discern what the Haskell Amendment does not mean (viz., what Long suggests) than what it does. If, as we have concluded, it does not exclude from the definition of return information all nonidentifying data, what particular nonidentifying data does it exclude? Again we think the key is the crucial phrase “in a form.” It is significant that this phrase is not contained in the provisions discussed earlier which seek — in language otherwise almost identical to the Haskell Amendment — to describe all identifying data. See §§ 6103(f)(1) & (2); 6103(f)(4)(A) & (B). But it is contained in two other portions of § 6103. Subsection (j), entitled “Statistical use,” permits disclosure of return information (1) to the Secretary of Commerce “for the purpose of, but only to the extent necessary in, the structuring of censuses and national economic accounts and conducting related statistical activities authorized by law”; (2) to the Chairman of the Federal Trade Commission, “for the purpose of ... administration ... of legally authorized economic surveys of corporations”; and (3) to the Department of the Treasury, “for the purpose of ... preparing economic or financial forecasts, projections, analyses, and statistical studies and conducting related activities.” The last paragraph of the subsection, entitled “Anonymous form,” concludes: No person who receives ... return information under this subsection shall disclose such ... return information to any person other than the taxpayer to whom it relates except in a form which cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer. § 6103(j)(4) (emphasis added). In this context, the meaning of the emphasized phrase seems clear: It is evidently meant to permit the publication and distribution of the statistical studies, forecasts and surveys that are the purpose of the permitted disclosures to Commerce, the FTC and Treasury. The phrase envisions, in other words, not merely the deletion of an identifying name or symbol on a document that contains return information, but agency reformulation of the return information into a statistical study or some other composite product — presumably on the theory that such reformulation gives added assurance that a taxpayer’s identity will in fact not be disclosed. The same meaning fits the other instance in which the phrase “in a form” appears as a disclosure limitation in § 6103. Subsection (i)(7)(A) provides that return information “shall be open to inspection by, or disclosure to, officers and employees of the General Accounting Office” for the purpose of conducting legally required audits. Such officers and employees are prohibited, however, from disclosing to others “return information ... in a form which can be associated with, or otherwise identify, directly or indirectly, a particular taxpayer.” This seems designed to assure that the reformulations of raw return information in the audit reports prepared by GAO officers and employees, if they are to be made public, be carefully devised to avoid the disclosure of identifying data. Once again the phrase is associated with a reformulation of the return information. Significantly, the phrase is not used where the subject of the provision is not “return information” but material that has already been reformulated. The concluding provision of § 6108 prescribes that no publication or disclosure of the statistical studies and compilations authorized by that section “shall in any manner permit the statistics, study, or any information so published, furnished, or otherwise disclosed to be associated with, or otherwise identify, directly or indirectly, a particular taxpayer.” 26 U.S.C. § 6108(c). The United States has argued in this appeal that the only type of reformulation that the Haskell Amendment exempts is that envisioned by the last mentioned section. The consequence of this interpretation, . of course, is that the Haskell Amendment becomes substantively superfluous, amounting to no more than a reminder in the definition section that § 6108 permits disclosure of statistical data. That alone may not be fatal since, as far as we can discern, any interpretation of the amendment, including the one we adopt, creates some redundancy. The insuperable problem, however, is that there is absolutely no textual basis for limiting the phrase “in a form” to the precise types of reformulation set forth in § 6108. It is not likely that such a surgically exact result would be described by that vague term, rather than by the simple and precise statement that return data “does not include statistical studies and compilations prepared under authority of § 6108.” To support the suggested limitation, the government resorts to what the Seventh Circuit in King called the “scant legislative history” of the Haskell Amendment, 688 F.2d at 492, consisting principally of the following statement by Senator Haskell on the Senate floor: [T]he purpose of this amendment is to insure that statistical studies and other compilations of data now prepared by the Internal Revenue Service and disclosed by it to outside parties will continue to be subject to disclosure to the extent allowed under present law. Thus the Internal Revenue Service can continue to release for research purposes statistical studies and compilations of data, such as the tax model, which do not identify individual taxpayers. The definition of “return information” was intended to neither enhance nor diminish access now obtainable under the Freedom of Information Act to statistical studies and compilations of data by the Internal Revenue Service. Thus, the addition by the Internal Revenue Service of easily deletable identifying information to the type of statistical study or compilation of data which, under its current practice, has been subject to disclosure, will not prevent disclosure of such study or compilation under the newly amended section 6103. In such an instance, the identifying information would be deleted and disclosure of the statistical study or compilation of data be made. 688 F.2d at 493 (quoting 122 Cong. Rec. 24,012 (1976)). As King noted, this statement was made in response to a question whether the IRS could avoid disclosing statistical studies simply by adding identifying information, and thus was not necessarily intended as a comprehensive expression of the purpose of the amendment (though it assuredly adds no support to the textually implausible Long interpretation). Even disregarding that limitation, however, the statement simply does not support the government’s narrow construction. It refers not only to “statistical studies” but also to “compilations of data, such as the tax model.” The latter is not a statistical tabulation. At the time the Haskell Amendment was adopted, it appears to have been an actual return with identifying details eliminated. Several years later, perhaps out of recognition that the 1976 legislation no longer permitted such redacted material to be made public, it was altered to consist of a reformulated return, with substitution of new figures for certain items — a partly factual, partly fictional return, so to speak. There is no way that the tax model can be brought within the publication exemptions of § 6108. Even if the provision of § 6108(b) which permits preparation and disclosure of “special statistical studies and compilations” is interpreted in such fashion that the. adjective “statistical” does not modify “compilations” (which seems to us strained), the tax model — which the Secretary had prepared and made public for years before the Haskell Amendment and has continued to prepare and make public since — could not possibly come within that provision, since it is by no stretch of the imagination a “special” compilation prepared “upon written request by any party or parties,” as § 6108(b) requires. We may add that, for similar reasons, we find no support in text or legislative history for the Seventh Circuit’s statement in King that “the Haskell Amendment provides only for the disclosure of statistical tabulations which are not associated with or do not identify particular taxpayers.” 688 F.2d at 493 (emphasis added). We hold, more broadly than King, that as used in § 6103(b)(2) the phrase “data in a form which cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer” requires — in addition to the fact of nonidentification — some alteration by the government of the form in which the return information was originally recorded. That reformulation will typically consist of statistical tabulation or of some other form of combination with other data so as to produce a unitary product that disguises the origin of its components (as in the tax model). We need not define, for purposes of the present appeal, all other manners of reformulation that may be included. It suffices to say that the mere deletion of the taxpayer’s name or other identifying data is not enough, since that would render the reformulation requirement entirely duplicative of the nonidentification requirement. We do not pretend that the interpretation we have given the Haskell Amendment causes it to fit with perfect consistency into the body of Chapter 61 or even, less ambitiously, § 6103. It causes the provisions of § 61030(4) and § 6103(i)(7)(A), discussed above, to be superfluous. But that superfluity is nothing beside the textual and policy absurdities produced by the interpretation in Long — and exceeds the superfluity produced by the government’s interpretation only because the latter inexplicably limits its “statistical” restriction to the statistics referred to in § 6108. As we have construed the Haskell Amendment, it creates no more disruption of the carefully drawn statutory scheme than is commonly produced by the dread genre of floor amendment; indeed, with a scheme as detailed as this it is remarkable that the dislocations are not greater. We are persuaded, in any case, that the meaning we have assigned is the meaning most faithful to the text, most compatible with the remainder of the legislation, and most supportable by a plausible legislative intent. * * * * * * Application of our holding to the facts of the present case, and the other issues presented by the instant appeal, are left to the disposition of the panel, whose opinion is issued simultaneously with this. So ordered. . Judge Wald's dissent expresses concern over "the court’s recent practice of issuing en banc opinions on legal issues, as opposed to concrete factual scenarios, see also Foster v. United States, 783 F.2d 1082 (D.C.Cir.1986) (en banc).” Dissent at 172 n. 1. That concern must logically extend, of course — and should indeed have heightened application — to the court’s common practice of rendering en banc decisions on isolated legal issues without en banc rehearing, by means of a so-called “Irons footnote” added to the panel opinion, reflecting the fact that departure from prior law of the circuit has been approved by the full court. See, e.g., Telecommunications Research & Action Center v. FCC, 750 F.2d 70, 75 n. 24 (D.C.Cir.1984); In re: Commodity Futures Trading Comm'n, 738 F.2d 487, 496 n. 19 (D.C.Cir.1984); Irons v. Diamond, 670 F.2d 265, 268 n. 11 (D.C.Cir.1981). Indeed, it must logically extend to the even more venerable practice of en banc reconsideration of issued panel opinions limited to specific issues — a practice we indulged in only a few weeks ago. See Northern Natural Gas Co. v. FERC, 780 F.2d 59, 64 (D.C.Cir.1986) (order granting rehearing en banc "for the limited purpose of deciding whether the Court should reconsider its holding in Panhandle Eastern Pipe Line Co. v. FERC, 613 F.2d 1120 (D.C.Cir.1979)’’). The practice of segregating legal issues requiring the attention of the full court from the remainder of the case reflects the fact that appellate review serves a dual purpose: the correction of legal error and the establishment of legal rules for future guidance. Only the latter is ordinarily worthy of the attention of the full court. The dissent’s perception that judges "typically" dispose of all the issues in a case, see Dissent at 172 n. 1, is simply not true at the second appellate level, where the law-clarifying function predominates. The Supreme Court often, if not usually, grants certiorari only on one or more discrete points of law, and issues its opinions (in that sense) on "legal issues, as opposed to concrete factual scenarios.” En banc consideration (or Irons footnote disposition) effectively constitutes such second-level appellate review — at least where, as is the case here, the full court has before it the full text of a proposed panel opinion. It would be especially perverse to abandon our efficient practice of limited en banc disposition just as our caseload has steeply increased, and as the size of the court has been expanded, magnifying the number of judge-hours devoted to each issue handled en banc. The proposal to restrict "piecemeal” consideration would have the effect of making it inordinately difficult to alter prior law of the circuit or to correct panel decisions that adopt erroneous new rules. . The dissent refers to all the strange textual consequences of the Long interpretation as “stylistic superfluity,” which it equates in character with textual imperfections that remain under our interpretation of the amendment. Dissent at 176-. The latter, however, consist of nothing more than repetition, in later sections, of the exemption which (under our interpretation) the Haskell Amendment has already provided. This cannot reasonably be compared with the Alice-in-Wonderland definitional structure, see supra at 157, the pointless incorporation in the definition of exceptions that have no application, see supra at 157, and the provisions for open meetings that can never occur, see supra at 157-158, that are the consequences of Long. The dissent’s indiscriminate totaling of textual imperfections also happens to be inaccurate. One of the "superfluities” which it attributes to our interpretation is not that, since the section in question, § 6108(c), is not specifically tied to the term "return information.” Moreover, both of the minor imperfections in our interpretation subsist under the Long interpretation as well, and the dissent seems to have miscounted the distinctive problems of Long we have discussed. The simplistic "score” is not 5-3, as the dissent asserts, but 9-2. . The three sentences preceding this footnote sign did not appear in our original opinion, as issued in slip form. There the corresponding passage read as follows: The latter is not a statistical tabulation but a sample return, derived from an actual return but reformulated to substitute new figures for certain items — a partly actual, partly fictional return, so to speak. The revision was made to correct a factual inaccuracy brought to our attention by a post-decision motion of amicus American Civil Liberties Union of Washington, which noted that the description of the tax model presented to us by the government in oral argument and reflected in the foregoing passage was accurate for the period beginning about 1980, but was not accurate as of 1976, when the model was an actual return with identifying details eliminated. The dissent’s suggestion, Dissent at 176 n. 7, that our decision turned upon this mistaken factual assumption is demonstrably wrong. Our rejection of Long, set forth in Part II of this opinion, was made and continues to be made without any reference to this snippet of legislative history. And the only reason for raising it in this Part III of our opinion is to refute the government’s position, which relies upon it. The factual inaccuracy in the case as originally presented to us shows the wisdom of relying upon the text and structure of the statute rather than this statement by a single senator as a means of ascertaining the Congress’ intent. We have no way of knowing whether Senator Haskell’s understanding of the tax model — much less that of his colleagues, if any of them relied upon his remark — comported with our original understanding or rather with what we now know. The mere term “Tax Model” assuredly does not suggest a redacted actual return. . The absence of any textual basis in § 6108 for publication of the tax model seems to us a complete response to the concurring opinion’s contention that Chevron, U.S.A., Inc. v. Natural Resources Defense Council, 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984), requires us to defer to the agency’s interpretation of the stat- ute. There is some question, to begin with, whether an interpretive theory put forth only by agency counsel in litigation, which explains agency action that could be explained on different theories, constitutes an "agency position” for purposes of Chevron. Even granting that principle, however, it cannot possibly have application where counsel’s interpretation in fact does not explain agency action but is, to the contrary, incompatible with the agency’s settled course of conduct. That is the situation here, since the IRS has regularly released, and plans to continue to release, the tax model. Nor does it suffice to appeal to Senator Haskell’s explicit reference to the tax model as the agency’s justification for this singular departure from its (supposed) § 6108 theory. Legislative history is used to clarify the meaning of a text, not to create extra-statutory law. If it can ever be the basis for plainly departing from the text, it assuredly cannot be so when an interpretation that honors both the text and the history is available. The concurrence claims that agency counsel did not take the position that the Haskell Amendment referred exclusively to § 6108, but rather maintained that the phrase “data in a form” in § 6103 referred exclusively to the statistical studies covered by § 6108 and to the tax model. Concurrence at 171. The oral argument contained the following exchange: [COUNSEL]: ... [I]t is certainly arguable that the Haskell Amendment is redundant in light of 6108. QUESTION: ... Your interpretation of the Haskell Amendment is § 6108? [COUNSEL]: That’s right and that is why Senator Long didn’t think it was all that necessary____ Corrected Tr. of Oral Argument at 28 (Dec. 5, 1985). This concession was not, as the concurrence contends, a matter of government counsel’s ”hav[ing] been momentarily caught off guard by the court’s vigorous questioning.” Concurrence at 171 n. 10. To the contrary, what prompted the concession (and what prompted the questioning) was the carefully considered and frequently repeated assertion in the IRS’s brief to the full court that the Haskell Amendment covered only "information ... amalgamated into statistical data,” Supp.Brief for Appellees at 4, 6, and information “likely rendered anonymous by amalgamation into general statistics,” id. at 9; and that the purpose of the Amendment was "to permit the Internal Revenue Service to continue its collection and release of general statistics,” id. at 11, and "to permit the release of historically provided statistical data compiled from return information,” id. at 20. These expressions are subject to no interpretation other than that the Haskell Amendment excludes from the definition of "return information" only the statistical studies and compilations already covered by § 6108. The problem counsel faced at oral argument was reconciling this theory with the embarrassing fact (first brought to the IRS’s attention, evidently, by the Appellant's Supplemental Brief, filed simultaneously with its own) that the tax model is not a statistical study or compilation. We thought, in light of the above-quoted concession at oral argument, that counsel was seeking to mend his hold by adhering to the IRS’s original theory of equivalence between the Haskell Amendment and § 6108, but explaining the tax model as a legislative-history-prescribed expansion of what the word "statistics” includes. The concurrence chooses to interpret the discussion as utterly abandoning the position that "data in a form” means statistics only (§ 6108) and embracing instead the view that it means statistics plus (exclusively) the tax model. This merely substitutes for the vice of nontextual amendment to § 6108 the vice of nontextual amendment to § 6103, since in no way can the crucial words “data in a form" be categorically limited to statistical studies, the tax model, and nothing else. (Nor is there any apparent reason why Congress should assume that, of all conceivable nonstatistical amalgamations, only the tax model would be sufficiently helpful and would sufficiently secure anonymity to justify publication.) Whatever position counsel was taking, one thing is clear: it is impossible to find here the sort of clear and consistent agency view (even as purportedly expressed by counsel) that must be given deference. . Given the purpose of the reformulation requirement, however — to wit, the increased assurance of anonymity — we can readily opine that it does not include the dissent’s example of copying the same data “onto a fresh piece of paper, perhaps in narrative style." Dissent at 175.
SILBERMAN, Circuit Judge, concurring: The court confronts in this case a difficult issue of statutory interpretation. The puzzle begins with 26 U.S.C. § 6103 (1982), which provides generally for nondisclosure of “return information” in the interest of taxpayer confidentiality. All sides agree that the heart of this controversy is the proper interpretation of the Haskell amendment, which authorizes the agency to release material that otherwise could not be released because it would be (but for the Haskell amendment) “return information.” In other words, what does the phrase “data in a form which cannot be associated with or otherwise identify, directly or indirectly, a particular taxpayer” mean? But I believe there is another question presented of equal importance: Did Congress intend that the IRS’s interpretation of what constitutes such “data in a form ...” be given deference by the judiciary? I think it must be conceded that Section 6103 is, at the very least, ambiguous as to the issue now before the court. Nothing in the language of the statute itself yields a decisive understanding as to whether Neufeld or King is a more faithful rendering of the statute’s purpose. The legislative history is sparse and inconclusive on the Neufeld/King question. Against the backdrop of this ambiguity the parties argue for divergent interpretations of this provision. The agency argues that the Haskell amendment means that data must be amalgamated or combined in a different form (Appellee’s Supp.Br. at 5 n. 4) and specifically points to data “rendered anonymous by amalgamation into general statistics” as satisfying that test (Id. at 9). The agency relies heavily on the adoption of this interpretation by the Seventh Circuit in King and further argues that a broader definition — such as that adopted by the majority — apart from creating certain anomalies and inconsistencies in the statutory scheme, presents troublesome administrative problems. The Secretary, we are told, has grave difficulty in determining the ability of one who requests information to “correlate facially nonidentifying data with specific taxpayers.” Appellees’ Supp.Br. at 13. The Church, on the other hand, argues that any document in the IRS files that contains return information must be disclosed provided that identifying information is first redacted. Through the majority opinion, this court rejects as excessively narrow the agency’s construction of the statute and offers yet another construction. It concludes that the Haskell amendment exempts from the general rule of nondisclosure nonidentifying return information that has also undergone “agency reformulation ... into a statistical study or some other composite product____” Maj.Op. at 160 (emphasis in original). Each of these interpretations is, in my view, a reasonable construction of a difficult statute, but I disagree with the majority’s decision to treat its own construction as authoritative. I believe the majority's approach oversteps the limitations on the court’s proper role as defined in Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). The message of Chevron is emphatic. A court’s duty in matters of statutory construction is to give effect to congressional intent. If that intent is not precisely apparent, however, the court does not simply impose its own construction on the statute, as would be necessary in the absence of an administrative interpretation. Rather, if the statute is silent or ambiguous with respect to the specific issue, the question for the court is whether the agency’s answer is based oh a permissible construction of the statute. Id. at 843 (footnote omitted). Beyond this we cannot go. I believe Chevron’s instruction applies to the issue of statutory construction now before the court. Section 6103(b)(2), set out in full in the majority opinion at page 156, defines return information by enumerating several different categories of information, but qualifying the enumeration with the words of the Haskell amendment “but such term does not include data in a form which cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer.” 26 U.S.C. § 6103(b)(2) (1982). The words of the Haskell amendment seem to me to cry out for application of the doctrine of deference expounded in Chevron. No one, it would seem, is better qualified than the Secretary to decide whether certain classes of information may be released in compliance with the Haskell amendment. I Before proceeding, I must address a question that the majority raises without answering — “whether an interpretive theory put forth only by agency counsel in litigation, which explains agency action that could be explained on different theories, constitutes an ‘agency position’ for purposes of Chevron.” Maj.Op. at 162 n. 4. I think it is much too late to question whether the construction of Section 6103(b)(2) urged by agency counsel in this case really represents the IRS’s position. We know that the IRS has been advancing its interpretation in courts throughout the country at least since 1982, when it persuaded the Seventh Circuit in King to adopt its position, and perhaps even before then. To suggest in these circumstances that the King analysis is not an “agency position” is to imply that IRS counsel are mavericks, disembodied from the agency that they represent. I reject that supposition. See Ashland Oil, Inc. v. FTC, 548 F.2d 977, 984-85 (D.C.Cir.1976) (MacKinnon, J., dissenting). The precept that the agency’s rationale must be stated by the agency itself stems from proper respect for the separation of powers among the branches of government. In the seminal case of SEC v. Chenery Corp., 318 U.S. 80, 63 S.Ct. 454, 87 L.Ed. 626 (1943), the Court explained that “a judicial judgment cannot be made to do service for an administrative judgment.” Id. at 88, 63 S.Ct. at 459. Similarly, in Investment Co. Inst. v. Camp, 401 U.S. 617, 91 S.Ct. 1091, 28 L.Ed.2d 367 (1971), the Court declined to defer to agency counsel’s interpretation, noting that “Congress has delegated to the administrative official and not to appellate counsel the responsibility for elaborating and enforcing statutory commands.” Id. at 628, 91 S.Ct. at 1097. In these cases, the Supreme Court declined “to substitute [its own] or counsel’s discretion for that of the [agency]” because to do so would be “incompatible with the orderly functioning of the process of judicial review. This is not to deprecate, but to vindicate ... the administrative process, for the purpose of the rule is to avoid ‘propel[ling] the court into the domain which Congress has set aside exclusively for the administrative agency.’ ” Burlington Truck Lines, Inc. v. United States, 371 U.S. 156, 169, 83 S.Ct. 239, 246, 9 L.Ed.2d 207 (1962) (quoting SEC v. Chenery Corp., 332 U.S. 194, 196, 67 S.Ct. 1575, 1577, 91 L.Ed. 1995 (1947) (citation omitted) (Chenery II)). In other words, the doctrine developed as a manifestation of judicial restraint. If courts were to accept an agency counsel’s position that significantly differed from the agency’s position, they would in effect substitute a judicial interpretation for the agency’s. The doctrine has been applied in a variety of cases. Courts have rejected as inadequate agency counsel’s articulation of a statutory interpretation when that interpretation has been inconsistent with a prior administrative construction, see Securities Indus. Ass’n v. Board of Governors of the Fed. Reserve Sys., 468 U.S. 137, 104 S.Ct. 2979, 2983-84, 82 L.Ed.2d 107 (1984); Pitzak v. OPM, 710 F.2d 1476, 1479 n. 2 (10th Cir.1983); when the record evidence before the court demonstrates no link between counsel’s interpretation and administrative practice, Alaniz v. OPM, 728 F.2d 1460, 1465 (Fed.Cir.1984); or when agency counsel’s interpretation is revealed as no more than a “current litigating position.” Ames v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 567 F.2d 1174, 1177 n. 3 (2d Cir.1977); see McMillan & Peterson, “The Permissible Scope of Hearings, Discovery, and Additional Factfinding During Judicial Review of Informal Agency Action,” 1982 Duke L.J. 333, 363 n. 163. And in Investment Co. Inst. v. Camp, the Supreme Court dedined to credit agency counsel’s statutory interpretation, offered in support of an agency regulation, where the Comptroller had “adopted no expressly articulated position at the administrative level as to the meaning and impact” of the relevant statutory provisions. 401 U.S. at 627, 91 S.Ct. at 1097. Those circumstances are not present in this case, however. The construction of the Haskell amendment was not even a central issue in this case in its administrative phase before the agency. There was no need then for the agency to set forth its position on that issue. It was not until the case came before the district court that the issue arose in any form. There the Church argued that in light of Neufeld, which had been decided during the district court litigation, the IRS could not tenably maintain that it need not even search files known to contain return information. The IRS, bound, of course, by the Neufeld court’s construction of the Haskell amendment, argued to the district court that since the Church’s request identified the taxpayer, “no information could be released by the Service that would be anonymous.” Defendants’ Memorandum in Reply to Plaintiff’s Statement of Points and Authorities in Opposition to Defendants’ Motion for Summary Judgment at 3. On appeal before the D.C. Circuit panel, the IRS acknowledged that it was bound by Neufeld, but urged the court to reconsider Neufeld in light of King. Thus, at the first moment in the case when it was appropriate and relevant for the IRS to articulate its construction of the statute, it did so. En banc consideration of the issue ensued. What is clear from all this is that the King analysis is in no way inconsistent with the basis for the agency’s decision in the administrative appeal. See Appellee’s Panel Br. at 24 n. 11. This is not a case like Investment Co. Inst. v. Camp. In that case, the Comptroller of the Currency, to whom Congress had only recently reassigned regulatory responsibility for national banks’ trust activities, adopted, three decades after enactment of the Glass-Steagall Act, a regulation that departed from a long-settled interpretation of the statute. See 401 U.S. at 621-22, 91 S.Ct. 1094-95. The Comptroller plainly neglected to furnish an appropriate rationale for his actions at a stage in the proceedings where there was a compelling need to articulate a link between the statutory provisions and the new regulation. The time to provide a statutory interpretation that underpins a newly promulgated rule that sweeps away earlier interpretations of the statute is obviously at the time of the rulemaking. In informal administrative adjudicative proceedings that evolve into federal court litigation, by contrast, the issues may well be shaped and sharpened throughout the proceedings. I think it is enough that the agency, through its counsel, set forth its interpretation of the statute at the first moment when it was appropriate and relevant to do so. Were the rule to be otherwise — were the courts to withhold deference unless an agency asserted its interpretation of a statute in a formal adjudication or agency rule-making — we would be creating a strong incentive for government agencies and departments to undertake their business strictly through formal procedures. Although judicial review of administrative action has seemed in recent times to push in that direction, I doubt that much good can come of this trend or, more importantly, that it is justified by congressional direction. See Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc., 435 U.S. 519, 98 S.Ct. 1197, 55 L.Ed.2d 460 (1978). We should take care that a doctrine developed to restrain the judiciary not be transformed to serve as a justification for excessive judicial intervention. In any event, if this court rejects King on the unwarranted assumption that King is not the IRS’s “true” interpretation of the Haskell amendment, the agency presumably could undercut the court’s holding merely by taking some more formal step to adopt King as the Commissioner’s interpretation of the statute. This strikes me as an unseemly institutional pas de deux. II FOIA’s general policy favors disclosure, but the statute also recognizes nine categories of exemptions from the general rule of disclosure. One of the exemptions, found at Section 552(b)(3), provides that (b) This section does not apply to matters that are ... (3) specifically exempted from disclosure by statute (other than section 552b of this title) provided that such statute (A) requires that the matters be withheld from the public in such a manner as to leave no discretion on the issue, or (B) establishes particular criteria for withholding or refers to particular types of matters to be withheld; 5 U.S.C. § 552(b)(3) (1982). Thus, the language of Section 552 exempts statutes such as Section 6103 from “[t]his section,” which I take to mean the entire Section 552 in the absence of any other plausible reference. But Section 552 contains not only the disclosure requirements; it also contains the provision for de novo judicial review. While it is obvious that materials covered by exemption 3 are exempted from the disclosure requirement of Section 552, mere assertion that requested documents are covered by an exemption 3 statute will not serve to immunize items from disclosure. The de novo review section of FOIA requires courts to examine withheld documents to determine whether they are in fact covered by an exemption 3 statute. But that cannot possibly mean that the law applied in those proceedings is other than that embodied in the exemption 3 statute. That is to say, the definition or scope of “matters” covered by the exemption 3 statute must derive from the exemption 3 statute. It could come from nowhere else. I have found nothing in the legislative history of FOIA or its subsequent amendments that contradicts this analysis. The House Report accompanying S. 1160, the bill whose provisions were eventually codified as amended at 5 U.S.C. § 552, explained that “[t]he proceedings are to be de novo so that the court can consider the propriety of the withholding instead of being restricted to judicial sanctioning of agency discretion.” H.R.Rep. No. 1497, 89th Cong., 2d Sess. 9 (1966), U.S.Code Cong. & Admin.News 1966, p. 2418. The Senate Report stated: “That the proceeding must be do novo is essential in order that the ultimate decision as to the propriety of the agency’s action is made by the court and prevent [sic] it from becoming meaningless judicial sanctioning of agency discretion.” S.Rep. No. 813, 89th Cong., 1st Sess. 8 (1965). But does de novo review mean more than an independent examination of the facts in light of applicable law; is it inconsistent with deference to an agency’s reasonable construction of a statute it administers? In other words, must Chevron give way whenever a court is charged with de novo review? I have concluded that Chevron does apply in such circumstances. The rule of Chevron is not a rule of judicial administration for courts to apply in reviewing administrative decisionmaking. It is an articulation of the fundamental principle that when Congress intends to delegate authority to an agency, that purpose demands recognition by the courts. I reason as follows. In entrusting administration of a statute to an agency, Congress typically delegates to the agency concomitant authority to “fill any gap” that Congress has left. Morton v. Ruiz, 415 U.S. 199, 231, 94 S.Ct. 1055, 1072, 39 L.Ed.2d 270 (1974). Part of the legislative purpose with respect to a particular statute, then, is Congress’ intent that the agency will exercise its delegated authority in fulfillment of the statute’s purpose. Courts appropriately defer to an agency’s reasonable interpretation of a statute that the agency administers, then, not simply out of recognition of the agency’s expertise, but primarily out of respect for congressional intent. In Sims v. CIA, 642 F.2d 562 (D.C.Cir. 1980), this court took an approach contrary to my analysis and treated the issue of statutory construction of an exemption 3 statute as a question of law “reserved ultimately to our determination.” Id. at 568. In that case, the CIA declined to disclose documents that revealed the names of individuals and institutions that had participated in a controversial research program. The agency argued that the documents were “specifically exempted by statute” from disclosure within the meaning of Section 552(b)(3). The exemption 3 statute upon which the CIA relied was 50 U.S.C. § 403(d)(3) (1982), which authorizes the Director to protect “intelligence sources and methods from unauthorized disclosure.” Appellants argued that the institutions and individuals did not constitute “intelligence sources” within the meaning of the statute. Thus, the issue in the Sims case, as in this case, turned on construction of a term in the exemption 3 statute. On appeal, this court devised a definition of the term “intelligence sources” after reviewing the statute and its legislative history. The court stated that FOIA authorized the judiciary “to undertake de novo review” not only of “agency determinations that particular records fall within exemption classifications,” but also of “agency constructions of applicable statutes.” 642 F.2d at 566 (citation and footnote omitted). The court acknowledged that “[bjecause the term ‘intelligence ... sources’ appears in the text of the National Security Act, it is appropriate for us to begin our analysis with the construction proposed by the CIA, an agency chartered by that statute and charged with major responsibility for its administration.” 642 F.2d at 568 (citations omitted). The court continued, “But we must not shrink from the responsibility vested in us by Congress. The question presented is one of law reserved ultimately to our determination.” Id. The Sims court’s observations as to the state of the law on deference to administrative agencies’ determinations in 1980 were, in my view, overbroad and erroneous. It has always been true that statutory interpretation is a question of law but it is equally true that an agency’s construction of its governing statute traditionally has been viewed as entitled to deference in certain contexts. As this court recognized in a case decided after Sims, although the “APA appears to require de novo review of all questions of law ... courts almost always accord some deference to an agency’s statutory construction.” Office of Communication of the United Church of Christ v. FCC, 707 F.2d 1413, 1422-23 n. 12 (D.C.Cir.1983); see Red Lion Broadcasting Co. v. FCC, 395 U.S. 367, 381, 89 S.Ct. 1794, 1801-02, 23 L.Ed.2d 371 (1969); Unemployment Comp. Comm’n v. Aragon, 329 U.S. 143, 153-54, 67 S.Ct. 245, 250-51, 91 L.Ed. 136 (1946); McLaren v. Fleischer, 256 U.S. 477, 480-81, 41 S.Ct. 577, 577-78, 65 L.Ed. 1052 (1921); Webster v. Luther, 163 U.S. 331, 342, 16 S.Ct. 963, 967, 41 L.Ed. 179 (1896). Chevron has put a new gloss on these old truisms. Chevron tells us that if congressional intent respecting a statutory provision is not clear, we must not treat the issue of statutory construction as we would treat any other garden variety question of law. We must defer to another institution’s determination — that of the agency charged to administer the law— provided, of course, that the agency’s determination is reasonable. In this sense, I believe that Chevron supersedes the Sims court’s overbroad analysis. My view is further reinforced by the Supreme Court’s treatment of the definitional issue when Sims came before that Court. CIA v. Sims, 471 U.S. 159, 105 S.Ct. 1881, 85 L.Ed.2d 173 (1985). Dismissing the court of appeals’ definition of “intelligence sources” as too narrow, the Court stated: The plain mea