The Department of Commerce (“DOC”) appeals the district court’s summary judgment in favor of the Assembly of California (“Assembly”) requiring that computer tapes containing statistically adjusted figures from the 1990 census be released to Assembly under the Freedom of Information Act (“FOIA”). DOC argues that the tapes are protected by Exemption 5’s deliberative process privilege, which protects documents that are both “predecisional” and “deliberative.” We affirm.

FACTS

For many years, statisticians within and without the Census Bureau (an agency of the Department of Commerce) have recognized that the decennial census under-counts the actual number of persons living in the United States, and that the under-counts are most severe among urban minority populations. According to Census Bureau estimates, “Blacks appear to have been undercounted in the 1990 census by 4.8%, Hispanics by 5.2%, Asian-Pacific Islanders by 3.1%, and American Indians by 5.0%, while non-Blacks appear to have been undercounted by 1.7%.” Statement of Secretary Robert A. Mosbacher on Adjustment of the 1990 Census, 56 Fed.Reg. 33582 (July 22, 1991) (“Statement”). Fol *918 lowing the 1980 census, states with large minority populations urged the Census Bureau to make statistical adjustments to correct for the undercount. Pursuant to a settlement in City of New York v. Department of Commerce, 713 F.Supp. 48 (E.D.N.Y.1989), DOC agreed to undertake a Post-Enumeration Survey (explained below) to determine the extent of the under-count, and to reconsider de novo a 1987 decision not to undertake statistical adjustments of the 1990 census. The settlement required a decision by July 15, 1991.

The Census Bureau developed a method to adjust for the undercount. Its starting point was the actual enumeration resulting from the 1990 head count (the “unadjusted data”). The Bureau then conducted a Post-Enumeration Survey (“PES”), a sample survey of 170,000 housing units taken after the census. Each person counted in the PES was assigned to one of 1392 statistical categories, known as “post-strata,” which grouped together persons of similar characteristics. The information from the PES was then compared to information for those same blocks that was obtained from the official census head count. From this the Bureau developed an adjustment factor which estimated the extent to which a given post-stratum was incorrectly counted in the initial enumeration. By multiplying the number of persons actually counted in each post-stratum in each of the nearly 5,000,000 inhabited blocks in the country by the adjustment factor, an adjusted census was produced. The Bureau prepared computer tapes containing the adjusted census data on the national, state, city and block level for release to the states in the event the Secretary chose to adjust the census.

The Secretary appointed a special advisory panel to advise him as to whether to adopt the adjusted census as the official United States census. It split 4-4 on the issue. The Undercount Steering Committee within the Census Bureau voted 7-2 in favor of adjustment, and the Director of the Census, Barbara Bryant, recommended adjustment as well. On July 15, 1991, the Secretary announced his decision not to adopt the adjusted census, stating it had not been proved to be more accurate than the unadjusted census. His rationale was that the adjusted census is less reliable on the local level than it is on the national level, so that in some localities, unadjusted counts are more accurate than adjusted counts. Statement, 56 Fed.Reg. at 33583. Right or wrong, the Secretary’s decision is not challenged here.

As part of his July 15 announcement, the Secretary said, “The Department has tried to make the process leading to this decision as open as possible. In that spirit, we will provide the full record of the basis for our decision as soon as it is available.” Id. at 33583. The Secretary was true to his word. A detailed report was published in the Federal Register of July 22, 1991. Id. at 33582 et seq. It explained the arguments raised by all of the Secretary’s senior advisors, naming them by name. The district court found that the Federal Register report contained “essentially every item of opinion, advice and recommendation that led to the decision.” Order re Preliminary Injunction (August 20, 1991) at 15. The Administrative Record of the decision, totalling some 18,000 pages, was also made available, as was the adjusted census data on the national, state and city level. On July 29, 1991, the Secretary honored Assembly’s FOIA request for the formulas used to create the adjusted census data, and for “all analyses prepared by or for the Census Bureau or the Department of Commerce analyzing the PES (Post-Enumeration Survey) or its accuracy.”

One requested item was not released: a computer tape containing “census population data for California, by block and census tract, broken down by race and age ... after adjustment in accordance with the Post-Enumeration Survey taken by the Bureau of the Census in 1990.” Assembly wishes to have access to census figures for use in its redistricting for the California legislature. 1 When DOC refused the re *919 quest, relying on the deliberative process privilege of FOIA’s Exemption 5, Assembly filed this action in district court. Subsequently, DOC released to the House Subcommittee on Census and Population half of all of the adjusted block counts, in checkerboard fashion, for all of the states. The Subcommittee in turn made the data available to the public. 2 Now all that is at issue is disclosure of the other half of the California checkerboard.

The case first came before the district court on a motion for preliminary injunction. It ordered that the tapes be released. Order re Preliminary Injunction (August 20, 1991). A divided panel of this court denied DOC’s motion to stay the preliminary injunction. No. 91-16266 (August 30, 1991). On September 10, the Supreme Court stayed the injunction pending final disposition. - U.S. -, 112 S.Ct. 19 , 115 L.Ed.2d 1103 (1991). The case proceeded in district court to cross-motions for summary judgment. In January, prior to the decision, DOC released the checkerboard block data to Congress. The district court granted Assembly’s motion for summary judgment and ordered the data released. Order 1992 WL 164141 (February 10, 1992). DOC timely-appealed.

JURISDICTION

The district court had jurisdiction under FOIA, 5 U.S.C. § 552 (a)(4)(B). We have jurisdiction under 28 U.S.C. § 1291 .

STANDARD OF REVIEW

Ordinarily, summary judgments are reviewed de novo. Intel Corp. v. Hartford Accident and Indemnity Co., 952 F.2d 1551, 1556 (9th Cir.1991). However,

National Wildlife Fed’n v. U.S. Forest Service, 861 F.2d 1114, 1116 (9th Cir.1988) (review of summary judgment). See also Van Bourg, Allen, Weinberg & Roger v. NLRB, 751 F.2d 982, 984 (9th Cir.1985) (on review of summary judgment in a FOIA case, “this court will reverse the district court’s findings that a particular document is exempt from mandatory disclosure only if the finding is clearly erroneous”).

At first glance this standard seems anomalous. It can best be explained by reflecting upon the task confronting the district court in a FOIA ease. It must examine the requested document (usually in camera, to avoid the risk of premature disclosure) to determine whether it falls within any of FOIA’s statutory exemptions from disclosure. Because there will rarely be any genuine issues of material fact — the document says whatever it says — the case may usually be decided on summary judgment. Even so, the proceeding might better be described as a trial on a hidden record, where the district court’s characterization of the requested document more closely resembles a finding of fact than a conclusion of law. Of course, we grant substantial deference to a district court’s fact finding.

Our precedents make abundantly clear that a district court’s decision on summary judgment that a given document does or does not fall within one of FOIA’s exceptions will not be reversed lightly. Essentially, the present case hinges on whether disclosure of the requested information would reveal anything about the agency’s decisional process. This is a fact-based inquiry where deference to the district court’s findings is appropriate. See United States v. McConney, 728 F.2d 1195, 1204 (9th Cir.) (en banc) (mixed questions of law and fact are not reviewed de novo when “the applicable legal standard provides for a strictly factual test”), cert. denied, 469 *920 U.S. 824 , 105 S.Ct. 101 , 83 L.Ed.2d 46 (1984).

DISCUSSION

FOIA requires that government agencies disclose to the public any requested documents. 5 U.S.C. § 552 (a). The agency may avoid disclosure only if it proves that the documents fall within one of nine enumerated exemptions. 5 U.S.C. § 552 (b)(1)-(9). FOIA’s purpose is to encourage disclosure, and to that end, its exemptions are to be interpreted narrowly. Department of Justice v. Julian, 486 U.S. 1, 8 , 108 S.Ct. 1606, 1611 , 100 L.Ed.2d 1 (1988); Department of the Air Force v. Rose, 425 U.S. 352, 360-61 , 96 S.Ct. 1592, 1599 , 48 L.Ed.2d 11 (1976). The government has the burden to prove that a requested document falls within one of FOIA’s exemptions. 5 U.S.C. § 552 (a)(3).

DOC relies on Exemption 5, which permits nondisclosure of “inter-agency or in-tra-agency memorandums or letters which would not be available by law to a party other than an agency in litigation with the agency.” 5 U.S.C. § 552 (b)(5). This “somewhat Delphic provision,” Julian, 486 U.S. at 11 , 108 S.Ct. at 1613 , shields “those documents, and only those documents, normally privileged in the civil discovery context.” NLRB v. Sears, Roebuck & Co., 421 U.S. 132, 149 , 95 S.Ct. 1504, 1515 , 44 L.Ed.2d 29 (1975). DOC invokes the “deliberative process” privilege. Its purpose is to allow agencies freely to explore possibilities, engage in internal debates, or play devil’s advocate without fear of public scrutiny. Congress expressed concern that if agencies were forced to “operate in a fishbowl,” S.Rep. No. 813, 89th Cong., 1st Sess. 9 (1965), candid exchange of ideas within an agency would cease and the quality of decisions would suffer. See generally NLRB v. Sears, 421 U.S. at 150-54 , 95 S.Ct. at 1516-18 ; EPA v. Mink, 410 U.S. 73, 85-94 , 93 S.Ct. 827, 835-39 , 35 L.Ed.2d 119 (1972).

We give the government the benefit of the doubt and assume for purposes of our review that the tapes are “inter-agency or intra-agency memorandums” within the meaning of Exemption 5. In order to be protected by the deliberative process privilege, such a document must be both “pre-decisional” and “deliberative.” National Wildlife Fed’n, 861 F.2d at 1117 . The D.C. Circuit has succinctly defined these terms of art:

Formaldehyde Inst. v. Department of Health and Human Services, 889 F.2d 1118, 1122 (D.C.Cir.1989). In practice, there is some overlap between the categories.

A. The “Predecisional” Prong

Exemption 5 cases contrast agency documents leading to a decision with documents explaining or interpreting the decision after the fact. Because an agency’s interpretations of its decisions often become the “working law” of the agency, documents deemed “postdecisional” do not enjoy the protection of the deliberative process privilege. NLRB v. Sears, 421 U.S. at 153, 95 S.Ct. at 1517 . This insures that the agency does not operate on the basis of “secret law.” Coastal States, 617 F.2d at 866 .

“[T]he line between predecisional documents and postdecisional documents may not always be a bright one.” NLRB v. Sears, 421 U.S. at 152 n. 19, 95 S.Ct. at 1517 n. 19. This is surely the case here. Chronologically, the adjusted census tapes were created before the Secretary an *921 nounced his decision not to adjust the census. But the tapes were prepared solely for the purpose of post-decision dissemination.

A document may be considered pre-decisional if it was “prepared in order to assist an agency decisionmaker in arriving at his decision.” Grumman Aircraft, 421 U.S. at 184 , 95 S.Ct. at 1500 . In this case, the district court found that “the record does not support the inference that the adjusted block counts played a meaningful role in that process.” - This finding is not clearly erroneous. The only evidence DOC produced that the adjusted block counts, rather than the formulas that generated them, contributed to thé decision not to adjust the census is the declaration of DOC official Mark Plant. . Plant declared that he examined a random sample of 500 adjusted block counts (not all necessarily from California), but “was unable to complete the analysis, which [he] did not regard as essential because of other analyses of this issue.” In deposition, Plant explained that the sample he consulted was not included in the administrative record “because it did not enter in any way direct or indirect into the Secretary’s decision.” Material which predates a decision chronologically, but did not contribute to that decision, is not prede-cisional in any meaningful sense.

DOC makes a further argument that the adjusted census tapes are predeci-sional because they may be used in the future by the Census Bureau in calculating its intercensal population estimates. This argument proves far too much. Any memorandum always will be “predecisional” if referenced to a decision that possibly may be made at some undisclosed time in the future. The district court did not clearly err by limiting its consideration to whether the tapes were “predecisional” to the Secretary’s July 15, 1991 decision not to adjust the census. “Characterizing these documents as ‘predecisional’ simply because they play into an ongoing audit process would be a serious warping of the meaning of the word.” Coastal States, 617 F.2d at 868 .

B. The “Deliberative” Prong

The underlying purpose of the deliberative process privilege is to ensure that agencies are not forced “to operate in a fishbowl.” To that end, courts focus “on the effect of the materials’ release: the key question in Exemption 5 cases [is] whether the disclosure of materials would expose an agency’s decision-making process in such a way as to discourage candid discussion within the agency and thereby undermine the agency’s ability to perform its functions.” Dudman Communications, 815 F.2d at 1568 . “We should focus on whether the document in question is a part of the deliberative process.” National Wildlife Fed’n, 861 F.2d at 1118 (emphasis in original). Predecisional materials are privileged “to the extent that they reveal the mental processes of decision-makers.” Id. at 1119 . The district court found that disclosure of the adjusted census tapes would reveal nothing about the deliberative process. We agree.

The earliest cases to examine the deliberative process privilege contrasted “factual” and “deliberative” materials. The idea behind this distinction was that agencies had no legitimate interest in keeping the public ignorant of the facts the agencies worked from, while they did have a legitimate interest in shielding their preliminary opinions and explorations. Soon, however, the Supreme Court warned against a “wooden” application of this distinction that would make the amount of material deemed factual within a document the deciding factor. EPA v. Mink, 410 U.S. at 91 , 93 S.Ct. at 838 . The key to the inquiry is whether revealing the information exposes the deliberative process. National Wildlife Fed’n, 861 F.2d at 1119 . The factual/deliberative distinction survives, but simply as a useful rule-of-thumb favoring disclosure of factual documents, or the factual portions of deliberative documents where such a separation is feasible. Julian v. Department of Justice, 806 F.2d 1411 (9th Cir.1986) (“communications containing purely factual material are not typically within the purview of Exemption 5”), aff'd, 486 U.S. 1 , 108 S.Ct. 1606 , 100 L.Ed.2d 1 (1988); Wolfe v. Department of Health and Human Ser *922 vices, 839 F.2d 768 (D.C.Cir.1988) (en banc) (“The fact/opinion distinction 'offers a quick, clear and predictable rule of decision’ for most cases”). 3

The district court found that the requested tapes were factual. “The numerical data contained on these tapes, like the official census data, provides a list of numbers of people, broken down by race, associated with census tracts. The court finds the material is purely factual and in no way divulges the reasoning process through which the data was derived or in any way explains any recommendation or decision not to adjust the census.” Order at 27 .

DOC challenges the district court’s finding, citing to Quarles v. Department of the Navy, 893 F.2d 390 (D.C.Cir.1990), for the proposition that ideas expressed in the form of numbers are not necessarily facts. True enough. Some numbers, like the military cost estimates at issue in Quarles , may “derive from a complex set of judgments” and “partake of that elasticity that has persuaded courts to provide shelter for opinions generally.” Id. at 392-93 . However, the district court did not say that the adjusted census data was factual simply because it was expressed in numbers; nor does Quarles say that numbers never represent facts. The district court acknowledged that both the unadjusted and adjusted census figures were estimates of the actual population, derived through different means. “In the court’s view, the two processes are not so fundamentally different as to render the product of one ‘classic, objective, non-deliberative material that must be disclosed under FOIA’ and the product of the other deliberative material that must not be disclosed.” Order at 30 (quoting DOC’s brief). In sum, the district court analyzed the requested tapes to see where they fell along the continuum of deliberation and fact, and found that they, like the unadjusted census data from which they were derived, fell closer to fact and would not reveal the agency’s protectable thought processes. This finding was not clearly erroneous.

As is typical for factual matter, release of the adjusted block-level data would not enable the public to reconstruct any of the protected deliberative process. First, as explained at oral argument, it is not possible to recreate the adjusted data solely from the raw data and the adjustment formulas. 4 For that same reason, it would not be possible to derive the formulas, or the process that created the formulas, from the adjusted data. Second, the adjusted data would not reveal anything about the most sensitive decision the Secretary had to make, namely which set of figures (adjusted or unadjusted) should be adopted as the official United States census. That decision involved the Secretary’s judgment as to which figures best estimated the actual population of the United States. The bare numbers reveal nothing about the process informing that judgment.

We note that DOC has already revealed most of its decisional process by revealing the method used to generate the adjusted census data. 5 The raw census figures are *923 a matter of public record. So are the formulas and procedures used to adjust the census. So are the adjusted census figures for the national, state and city levels, and, at the block level, for every other block. Only half of the block level adjusted data is currently withheld.

Order at 30-31 (citations omitted).

We agree. The complete deliberative process has been revealed. The adjustment formulas have been revealed. Almost all of the product of the process has been revealed. The portion of the product that has not been disclosed reveals no remnants of the deliberative process.

C. DOC’s Other Arguments

DOC opposes Assembly’s use of the adjusted block-level census because it might be less accurate in discrete local situations than the unadjusted census data, leading to public confusion. Although the Secretary chose not to adjust the census because he was not convinced that the adjusted numbers were more accurate than the unadjusted numbers, inaccuracy is not a basis for FOIA exemption. If the tapes are not protected by Exemption 5 because they are predecisional and deliberative, they must be disclosed. DOC has warned the public that both the unadjusted and adjusted data are inaccurate. DOC, perhaps predictably, worries that the adjusted data will be misused. However, it is not among FOIA’s functions to control the use of disclosed information. The public, not DOC or the courts, should decide whether to heed DOC’s warnings.

DOC also expresses fear that the Census Bureau’s reputation for reliability would be diminished if it released information known to be inaccurate. The district court rejected this theory as a factual matter. “Any harm to the Department’s reputation has already occurred; because of the Department’s candor, the public already knows that the block level data are inaccurate, and has a good idea of the extent of any inaccuracies.” Order at 33 (citation omitted). Even if DOC could prove that its reputation would be harmed by disclosure, we note that the core purpose of FOIA is to allow the public to know “what their government is up to.” Department of Justice v. Reporters Committee for Freedom of the Press, 489 U.S. 749, 773 , 109 S.Ct. 1468, 1481 , 103 L.Ed.2d 774 (1989). Concealing information from the public in order to protect agency reputations is precisely the sort of behavior FOIA was enacted to eliminate.

CONCLUSION

We agree with the district court that “production of the contested document would [not] be ‘injurious to the consultative functions of government that the privilege of nondisclosure protects’.” Mink, 410 U.S. at 87 , 93 S.Ct. at 836 (quoting Kaiser Aluminum & Chem. Corp. v. United *924 States, 141 Ct.Cl. 38 , 157 F.Supp. 939, 946 (1958)).

We AFFIRM.

. The states are not obliged to use official census data when drawing their state legislative districts, Burns v. Richardson, 384 U.S. 73, 91 , 86 S.Ct. 1286 , 16 L.Ed.2d 376 (1966), or their congressional districts, Young v. Klutznik, 652 F.2d 617 , 624 (6th Cir.1981), cert. denied, 455 U.S. 939 , 102 S.Ct. 1430 , 71 L.Ed.2d 650 (1982).

. Unlike the adjusted block-level data released to plaintiffs as part of discovery in the City of New York litigation, the figures released to Congress were not subject to a protective order. Subcommittee Chairperson Rep. Thomas Sawyer indicated that the checkerboard data would be released to any interested member of the public on request. Letter from Rep. Thomas Sawyer to Barbara Brenner, January 16, 1992.

. In a case raising identical issues brought by the Florida Legislature, the Eleventh Circuit rejected the functional approach and held instead that the fact/opinion distinction provides a bright-line rule of decision in all cases. See Florida House of Representatives v. Department of Commerce, 961 F.2d 941, 947-49 (11th Cir.1992). In doing so, the Eleventh Circuit expressly disapproved our decision in National Wildlife Fed’n. Id. at 947. We believe that National Wildlife Fed’n states the better rule and more closely follows Supreme Court precedent. At any rate, we are bound to follow our own circuit's law.

. The formulas used to generate the adjusted census data require as input an Internal Detail File (“IDF"). The IDF links individuals counted in the PES to the appropriate demographic post-stratum. This file necessarily contains personal information: age, sex, race, income, and so on. Assembly does not seek access to the IDF, and would not be entitled to it if it did; the Census Bureau is forbidden by law to reveal personal information collected by the census. 13 U.S.C. § 9 . This is why Assembly seeks the adjusted block data through this FOIA action.

. When considering the effect of substantial disclosure by the agency, every court to examine the issue has expressly avoided basing its decision on waiver. Order at 32 n. 15 (“[T]he extent of prior disclosure does not waive the privilege but merely is a factor that the court considers”); *923 Florida House of Representatives v. Department of Commerce, No. TCA 91-40387-WS (N.D.Fla. January 9, 1992) at 24 ("This court does not hold that the Department waived its deliberative process privilege”), rev'd on other grounds, 961 F.2d 941, 947 (11th Cir.1992) ("The district court did not err in finding no waiver"). DOC devotes considerable briefing to the waiver argument. We agree that this case should not turn on waiver. Agencies should not be penalized for openness. We consider prior disclosures only to determine whether the disclosure of these tapes would expose the decision-making process any more than it has already been disclosed.