United States Court of Appeals
         FOR THE DISTRICT OF COLUMBIA CIRCUIT



Argued September 12, 2019         Decided November 1, 2019

                        No. 18-5316

                  INSTITUTE FOR JUSTICE,
                        APPELLANT

                             v.

                INTERNAL REVENUE SERVICE,
                        APPELLEE


        Appeal from the United States District Court
                for the District of Columbia
                    (No. 1:16-cv-02406)


     Andrew D. Prins argued the cause for appellant. With him
on the briefs was Gregory in den Berken.

    Douglas C. Rennie, Attorney, U.S. Department of Justice,
argued the cause for appellee. With him on the brief were
Richard E. Zuckerman, Principal Deputy Assistant Attorney
General, Travis A. Greaves, Deputy Assistant Attorney
General, Gilbert S. Rothenberg, and Michael J. Haungs,
Attorneys.
                                 2

    Before: PILLARD and RAO, Circuit Judges, and WILLIAMS,
Senior Circuit Judge.

   Opinion for the Court filed by Senior Circuit Judge
WILLIAMS.

     WILLIAMS, Senior Circuit Judge: The non-profit, public
interest law firm Institute for Justice seeks information kept by
the Internal Revenue Service about asset forfeitures. In its
request, the Institute followed a lead supplied by the IRS. The
latter’s own manual repeatedly refers to the Asset Forfeiture
Tracking and Retrieval System (“AFTRAK”) as the “database”
in which the agency compiles information about asset
forfeitures.    See Internal Revenue Manual, §§ 9.7.5.2,
9.7.5.4.1(1). Accordingly, the Institute submitted a Freedom of
Information Act (“FOIA”) request for “all records contained
in” the AFTRAK database.

     The IRS’s legal response, both in the district court and
here, is that the Institute’s FOIA request fails from the start
because AFTRAK “is not a ‘database’” and therefore its
“‘contents’ do not qualify as ‘records’ under the FOIA.” Joint
Meet & Confer Statement 2, ECF No. 10. The modifier “legal”
is critical in the sentence above, as the factual declaration the
IRS submitted conspicuously includes no assertion that
AFTRAK is not a database. More affirmatively, the IRS says
that AFTRAK is (legally and factually) “a web-based
application that aggregates information from various other
sources within the [IRS] into a single user interface.” Def.’s
Mem. Supp. Summ. J. 3, ECF No. 14-1. According to the IRS,
this distinction renders the Institute’s request not just imprecise,
but unintelligible: “[b]ecause AFTRAK only generates reports,
there was [] no traditional search to perform in AFTRAK.”
Def.’s Reply Br. Supp. Mot. Summ. J. 7, ECF No. 26.
                               3

Nonetheless, after the Institute filed suit, the IRS created what
it describes as the “most comprehensive standard report from
the AFTRAK system, the Open/Closed Asset Report,” Joint
Status Report 5, ECF No. 12, saved the Report in PDF format,
heavily redacted it, and provided it to the Institute. Pl.’s Br.
Supp. Mot. Discovery 9–10, ECF No. 15-1. By the IRS’s
telling, creating the Report was “arguably” an act of
administrative grace; FOIA imposes no duty on agencies to
create new records in response to FOIA requests, but here the
agency created the Open/Closed Report. Appellee’s Br. 44.

     Unhappy with that result, the Institute filed a cross-motion
for summary judgment, but the district court awarded summary
judgment in large part in favor of the IRS. The court reasoned
that “there is no need to resolve the technical question of
whether AFTRAK is or is not a database,” because regardless
of that determination, “[t]he IRS generated a comprehensive
report that revealed every possible data point on seized assets
in the AFTRAK system during the relevant timeframe.”
Institute for Justice v. IRS, 

, 41 (D.D.C.
2018). The court also granted summary judgment on the IRS’s
application of FOIA Exemptions 7(A) and 7(C), 5 U.S.C.
§ 552(b)(7)(A), (C), and deemed the Institute to have forfeited
its objection to the IRS’s redaction of Asset Description column
entries under Exemption 7(F), 

Institute for

–45.

     We reverse the district court and remand the case for
further proceedings. First, whether the Open/Closed Report
covers all records “contained in” AFTRAK is itself a material,
genuinely disputed question of fact, and the answer in turn
depends on other disputed and material facts. Second, whether
AFTRAK is correctly classified as a database, a matter on
which the IRS’s Manual and other official documents
                              4

contradict its legal denial here, appears to be an intermediate
fact with potential consequences for resolving the parties’
claims. The disputes on these matters preclude summary
judgment.

                            * * *

     We review de novo the district court’s grant of summary
judgment, Nation Magazine v. U.S. Customs Serv., 

,
889 (D.C. Cir. 1995), and “must draw ‘all justifiable
inferences’ in favor of the non-movant,” Aguiar v. Drug
Enforcement Administration, 

, 735 (D.C. Cir.
2017) (quoting Anderson v. Liberty Lobby, Inc., 

,
255 (1986)).

     An agency “fulfills its obligations under FOIA ‘if it can
demonstrate beyond material doubt that its search was
reasonably calculated to uncover all relevant documents.’”
Ancient Coin Collectors Guild v. U.S. Dep’t of State, 

, 514 (D.C. Cir. 2011) (quoting Valencia-Lucena v. U.S.
Coast Guard, 

, 325 (D.C. Cir. 1999)). In the
context of a request for a database, “FOIA requires agencies to
disclose all non-exempt data points,” National Security
Counselors v. CIA, 

, 272 (D.D.C. 2012),
subject, as always, to limits aimed at protecting agencies from
undue burdens.

     Viewing the case in this light, we are not convinced
beyond a material doubt that the Open/Closed Report contains
all AFTRAK records or that its delivery to the Institute could
serve as a substitute for the search required by FOIA.

    The key passage from the IRS’s sole declaration on the
matter explains that the Open/Closed Report is “the most
                                5

utilized and complete standard report available from the
AFTRAK system” and “contains comprehensive data about
every asset seized by [the Criminal Investigation section of the
IRS] within a specified time period.” Dean Martin Decl. ¶ 10
(July 10, 2017), ECF No. 14-3. But this terse statement and the
Institute’s evidence to the contrary leave many unanswered
questions.

     First, does “comprehensive data about every asset seized”
mean all the records “contained in” AFTRAK, which the
Institute requested? “Comprehensive” can mean “including or
dealing with all or nearly all elements or aspects of something,”
New Oxford American Dictionary (2d ed. 2005) (emphasis
added), or being “of large content or scope,” 

“comprehensive” does not always mean “all.” Plus, this
comprehensiveness pertains to “every asset seized,” but what
about data not specific to individual assets or their seizure, such
as statistical and reporting data? Cf. Martin Decl. ¶ 7
(“AFTRAK is also used for monitoring assets in inventory and
for law enforcement statistical and reporting purposes
exclusively by CI personnel.”).

    Second, does AFTRAK include data about assets seized by
IRS sections other than Criminal Investigation? If so, why
should such data not be disclosed (unless for some reason it
doesn’t constitute a “record”)?

     Third, even if the Open/Closed Report is “the most utilized
and complete standard report,” that seems no reason for the IRS
to withhold data absent from the Open/Closed Report but
included in a non-standard report or other record.

    Fourth, and perhaps most devastating to the IRS’s position:
In 2014 the IRS furnished the Institute a limited dataset in
                               6

response to a FOIA request; that dataset included three columns
of information—“State Seized,” “Forfeiture Actual Amount,”
and “Primary Statute Violated”—that aren’t found in the IRS’s
“comprehensive” Open/Closed Report. See Appellant’s Br.
19–20 & n.10 (citing Joint Appendix (“J.A.”) 112–89, 647–57).

     Fifth and finally, the IRS’s counsel confirmed in an email
to the Institute’s counsel, in reference to a “sample” of the
Open/Closed Report, that either the Report itself, or perhaps
only its sample—that remains ambiguous—does not include all
“fields” of data from AFTRAK. Email from Joycelyn Peyton,
Trial Attorney, U.S. Dep’t of Justice, to Andrew Prins, Partner,
Latham & Watkins LLP (Mar. 21, 2017, 13:37 EDT), ECF No.
23-2; Andrew Prins Decl. ¶ 2 (Apr. 20, 2018), ECF No. 23-1.
Were the missing fields simply rows of the sample to be added
before delivery of the full report, or whole columns of
information to be left out even from the final production?

     As to the significance of AFTRAK’s constituting a
“database,” nearly forty years ago we treated databases as
containing the modern equivalent of old-fashioned records,
saying, “Although accessing information from computers may
involve a somewhat different process than locating and
retrieving manually-stored records, these differences may not
be used to circumvent the full disclosure policies of the FOIA.”
Yeager v. Drug Enforcement Administration, 

, 321
(D.C. Cir. 1982). See also Elliott v. Dep’t of Agriculture, 

, 852 (D.C. Cir. 2010) (using the terms “filing cabinet”
and “electronic database” interchangeably when addressing the
adequacy of a search); Multi Ag Media LLC v. Dep’t of
Agriculture, 

, 1226 (D.C. Cir. 2008) (applying
standard FOIA rules to “a massive database” in agency hands);
Petroleum Information Corp. v. Dep’t of Interior, 

, 1431–32 (D.C. Cir. 1992) (applying standard rules to
                                7

“data elements—i.e., categories of information” in “records
from a computer data bank”). The essence of our approach—
treating electronic records the same as analog records—was
later codified by the Electronic Freedom of Information Act
Amendments of 1996, which newly defined a “record” as “any
information that would be an agency record . . . when
maintained by an agency in any format, including an electronic
format,” Pub. L. No. 104-231, § 3, 110 Stat. 3048, 3049 (1996)
(codified as amended at 5 U.S.C. § 552(f)(2)(A) (emphasis
added)). See also H.R. Rep. 104-795, at 22 (1996), 1996
U.S.C.C.A.N. 3448, 3465 (analogizing “[c]omputer records
found in a database” to records in a “file cabinet”).

     We follow the parties’ and the district court’s lead in using
“database”—a term that has no independent legal significance
under FOIA—to mean a system that stores or contains records
subject to FOIA. Absent further evidence to the contrary, it
seems safe to say that if AFTRAK is a database, the Institute is
entitled to more than has been delivered, very possibly much
more. And even if it should prove that AFTRAK isn’t a
database, we have no showing at all why AFTRAK’s (as yet
undisclosed) non-database characteristics in any way imply that
a search for records “contained in” it would not yield material
more closely approximating the IRS’s expansive descriptions
of AFTRAK in the IRS Manual and other publications than
does the Open/Closed Report.

     In sum, the IRS’s declaration, on its own terms, does not
contain the “reasonable specificity of detail” necessary to
sustain summary judgement, 

, and
worse, its claims have been “called into question by
contradictory evidence in the record,” 

require explanation. Given the existing record, the district
court erred in concluding that the IRS’s production of the
                                8

Open/Closed Report satisfied the IRS’s ordinary obligation to
search AFTRAK for “all records.”

                             * * *

    On remand, the district court will have to ascertain the
nature of the AFTRAK system—i.e., whether and to what
extent AFTRAK itself contains records and/or provides access
to records stored elsewhere. To the extent that AFTRAK
provides access to records stored elsewhere, the district will
need to “ascertain the scope of the [Institute’s] request,” Nation

, specifically, whether it requests such
records, and then determine whether the IRS’s search was
adequate in light of that scope.

      As to scope, the Institute argued before the district court
that the IRS should have construed its request for all records
“contained in” AFTRAK as also including a request for records
“accessible by” AFTRAK, in the event that technical facts
about how AFTRAK works, unknown to the Institute, would
make a request for records “accessible by” AFTRAK better
targeted and more encompassing than a request for records
“contained in” it. Pl.’s Br. Supp. Mot for Discovery 10, 16,
ECF No. 15-1. We leave it to the district court to decide in the
first instance the proper scope of the Institute’s request. We do
not require technical precision in FOIA requests, and a request
certainly should not fail where the agency knew or should have
known what the requester was seeking all along. For example,
a FOIA request for emails would not fail because the request
was for emails “in” an Outlook inbox rather than “accessible
through” Outlook. Nor do we require a requester to know
anything about where or how the agency stores those emails.
Instead, the statutory standard for a FOIA request is that it
                               9

“reasonably describes” the records sought.            5 U.S.C.
§ 552(a)(3)(A).

     Congress added the “reasonably describes” standard for a
FOIA request in 1974, replacing the requirement that a request
name “identifiable records.” See Act of Nov. 21, 1974, Pub. L.
No. 93-502, § 1(b)(1), 88 Stat. 1561, 1561. According to the
Senate Judiciary Committee Report accompanying the
amendment, the change “makes explicit the liberal standard for
identification that Congress intended.” S. Rep. No. 93-854, at
10 (1974). Accordingly, our case law has often repeated that
an agency “has a duty to construe a FOIA request liberally.”
Nation 

(citing Truitt v. Dep’t of State,

, 544–45 (D.C. Cir. 1990), and Founding Church
of Scientology v. NSA, 

, 836–37 (D.C. Cir. 1979)).
Here, the IRS’s narrow interpretation would be lawful only if
the agency were truly not “able to determine ‘precisely what
records [were] being requested.’” 

(quoting S. Rep. No. 93-854, at 10).

     The IRS’s argument to the contrary is unpersuasive. The
agency mistakenly cites Kowalczyk v. U.S. Dep’t of Justice, 

, 389 (D.C. Cir. 1996), for the proposition that it may
narrowly construe the scope of the Institute’s request. See
Appellee’s Br. 30. But in Kowalczyk the request never
mentioned, and contained no clues suggesting, a location that
the requester later claimed was a likely site for documents he
sought (the FBI’s New York 

. The court
rejected this attempted demand on FBI’s intuition or second
sight. Once again it said that a “request reasonably describes
records if ‘the agency is able to determine precisely what
records are being requested,’” 

(quoting 

), but an agency need not “speculate about potential
leads” to responsive documents, 

In the same breath,
                               10

the court clarified that “[t]his is not to say that the agency may
ignore what it cannot help but know.” 

Valencia-

(“It is well-settled that if an agency
has reason to know that certain places may contain responsive
documents, it is obligated under FOIA to search barring an
undue burden.”).

     The IRS hasn’t revealed what limits it may believe inhere
in the words “contained in.” A request for records contained in
the “Asset Forfeiture Tracking and Retrieval System” certainly
sounds at first blush as if it seeks records that can be tracked
and retrieved through that system. Is the IRS arguing that the
words exclude a record that is retained elsewhere by the agency
and which AFTRAK accesses and displays? If so, then in what
sense would such a document not be “contained in” AFTRAK?
Is such a link a “lead”?

     The second inquiry for the district court on remand is
whether the IRS’s search was adequate, considering the scope
of the request. The IRS has a tidy two-step argument: the scope
of the FOIA request includes only records “contained in”
AFTRAK, and the only “records” contained in AFTRAK are
the reports it generates, because it is not a database. Appellee’s
Br. 35. See also Def.’s Reply Br. Supp. Mot. Summ. J. 6–7,
ECF No. 26. We have already expressed our doubts as to the
plausibility of the IRS’s scoping, and now review the IRS’s
legal claim that AFTRAK is not a database.

     The IRS’s declaration states that AFTRAK is a “web-
based application,” but does not state that AFTRAK is not also
a database—as IRS’s counsel asserts. Compare Martin Decl.
¶ 6 (“AFTRAK is a web-based application system specifically
designed to track assets seized for forfeiture, evidence, and
abandonment.”) with Def.’s Mem. Supp. Mot. Summ. J. 3, ECF
                               11

No. 14-1 (“[T]hough [the Institute] resolutely asserts that
AFTRAK is a ‘database’ . . . it is not.”). Nor does the IRS give
the court reason to infer that there’s any inherent conflict in
classifying the AFTRAK “system” as both a “web-based
application” and a database. Given that the IRS’s legal
argument turns on AFTRAK’s not being a database, the failure
of the IRS’s sole declaration even to assert (much less explain
or justify) mutual exclusivity between the two is bewildering.

     By contrast, the Institute’s claim to the contrary, that
AFTRAK is “almost certainly” a database, finds considerable
support in the record. The Institute presented three separate
documents: the IRS Manual, IRS “Information Quality”
guidelines, and a report by the Treasury Department’s Inspector
General for Tax Administration, which all independently refer
to AFTRAK as a “database.” The IRS’s attempt to belittle
these sources as merely “documents [the Institute] appears to
have found on the internet,” Appellee’s Br. 25, does nothing to
undercut their force as official agency descriptions of
AFTRAK. The Institute also entered into the record a
declaration submitted by the IRS in other federal litigation
which describes AFTRAK as “an asset inventory system” in
which the declarant, an “Asset Forfeiture Coordinator,” is said
to “enter” information about seized assets. Teri Schultz Decl.
¶ 4, Stott v. Internal Revenue Service, No. 14-cv-176 (W.D.
Wis. Sept. 10, 2014), Dkt. 11. If a person can “enter”
information into AFTRAK, why can’t the information so
entered be “reasonably described” as a record “contained in”
it?

     We thus think it clear that viewing the evidence in the light
most favorable to the Institute, the material fact of AFTRAK’s
status as a database remains in dispute, unless other facts and
analysis render it possible to conclude that material by then
                               12

supplied by the IRS satisfies the Institute’s request. Of course
we don’t foreclose either the possibility that despite
AFTRAK’s classification as a database some of the records the
Institute seeks are somehow beyond its request, or the
possibility that despite AFTRAK’s classification as something
other than a database the Institute is fully entitled to those
records.

     We note that the Institute also appealed the district court’s
denial of additional discovery under Rule 56(d). “Because we
conclude that appellant carried [its] burden of opposing the
motion for summary judgment” on the search’s adequacy,
however, “we do not reach the issue whether the denial of
additional discovery was appropriate” under Rule 56(d).
Greenberg v. FDA, 

, 1217 n.4 (D.C. Cir. 1986).
Instead, we leave open to the district court on remand to
consider the extent to which discovery may be appropriate in
light of our opinion.

                             * * *

     We also vacate the district court’s grant of summary
judgment insofar as it sustains the IRS’s application of
Exemptions 7(A) to entire rows of the Open/Closed Report
without showing that non-exempt information cannot be
segregated. On de novo review, we conclude that the IRS failed
to tailor the categories of information withheld to what
Exemption 7(A) protects: law enforcement records that “could
reasonably be expected to interfere with enforcement
proceedings.” Citizens for Responsibility & Ethics in Wash. v.
U.S. Dep’t of Justice, 

, 1091, 1096 (D.C. Cir.
2014) (quoting 5 U.S.C. § 552(b)(7)(A)). In so doing, the
agency does not appear to have accounted for its “obligat[ion]
to disclose ‘[a]ny reasonably segregable portion of a record’
                               13

after removing the exempt material,” Bartko v. U.S. Dep’t of
Justice, 

, 62 (D.C. Cir. 2018) (quoting 5 U.S.C.
§ 552(b)), pursuant to this exemption, or to “note the ‘amount
of information deleted,’” 

of Exemption 7(A) to redact all rows
relating to open investigations using the justification that
“[d]isclosure of any information concerning an open criminal
investigation could reasonably be expected to compromise or
interfere with ongoing law enforcement,” Martin Decl. ¶ 32, is
insufficient on the facts of this case. See Crooker v. Bureau of
Alcohol, Tobacco & Firearms, 

, 66–67 (D.C. Cir.
1986) (holding agency cannot apply a “‘blanket exemption’ for
‘all records relating to an ongoing investigation’” (quoting
Campbell v. U.S. Dep’t of Health and Human Servs., 

, 259 (D.C. Cir. 1982))). In its declaration, the IRS does not
even mention 14 of the 22 columns, much less explain why
disclosure of such columns as, for example, the “Primary
Seizure Statute” or the “Seizure Type” (civil, administrative, or
criminal), could possibly harm ongoing investigations. See
Prison Legal News v. Samuels, 

, 1150 (D.C. Cir.
2015) (reiterating that under the categorical approach to
applying exemptions, the “[t]he range of circumstances
included in the category must ‘characteristically support an
inference that the statutory requirements for exemption are
satisfied.’” (quoting Nation 

)). On
remand, we expect the agency to provide a more robust
explanation of its use of Exemption 7(A) to redact all open
investigations as a category.

     As for the IRS’s application of Exemption 7(F) to the Asset
Description column, the district court abused its discretion in
holding that the Institute’s objection was forfeited. See Trudel
v. SunTrust Bank, 

, 1286 (D.C. Cir. 2019). The
                              14

IRS originally applied Exemptions 6, 7(C), and 7(F) to
withhold the entire Asset Description column. Only after the
IRS complied with the district court’s June 5, 2018 order—
requiring production of data in the Asset Description column it
had withheld under Exemptions 6 and 7(C), Institute for

—did the types of data in that
column become apparent. Only then could the Institute
recognize that the IRS appeared to be applying Exemption 7(F)
to redact information beyond that which “could reasonably be
expected to endanger the life or physical safety of any
individual.” 5 U.S.C. § 552(b)(7)(F).

     This context shows that the Institute challenged the IRS’s
reliance on Exemption 7(F) at its first opportunity—after the
IRS submitted its newly redacted production to the court in
response to the court’s June 5, 2018 Order. Compare IRS’s
“Notice of Compliance” (Sept. 4, 2018), J.A. 1061–62, with the
Institute’s Response (Sept. 18, 2018), J.A. 1064–66, asserting
insufficiency of the former. The fact that the Institute did not
object to the IRS’s use of Exemption 7(F) to withhold a
“target’s street address”—the only information the IRS
previously claimed a right to withhold from the Asset
Description column under Exemption 7(F), Martin Decl.
¶ 49—does not mean that the Institute forfeited the right to
object to the agency’s actual—and evidently far broader—use
of Exemption 7(F).

     We leave in place only the district court’s judgment in
favor of the IRS on the applications of Exemption 7(C), which
the Institute has not challenged—specifically, redaction of the
Lead Agent, Program Area, Investigation Name, Investigation
Number, Storage Location, and SEACATS Number columns.
                              15

                            * * *

     The judgment of the district court is reversed and the case
is remanded.

                                    So ordered.