A guide to reading federal records responses without converting uncertainty into an accusation. All invented requests, agencies, correspondence and numerical datasets below are hypothetical. This is general information, not legal advice or a prediction about a particular appeal. Sources checked October 29, 2026.
1. A response about disclosure is not an admission about the world
The Department of Justice's 2026 Chief FOIA Officer Report identifies a limitation that deserves attention before anyone calculates a secrecy rate. Nineteen of the department's thirty-two components either did not track their use of Glomar responses or reported that they had issued none during the reporting period. Those two situations appear in the same description, but they mean different things. An unmeasured count is not an observed zero. The report's discussion concerns administrative reporting, not a tally of secret operations. DOJ's April 22, 2026 report, Glomar discussion.
That distinction mirrors the challenge faced by someone opening a records-response letter. A document saying that an agency will neither confirm nor deny the existence of responsive records can be frustrating precisely because it withholds a fact that an ordinary response would reveal. But frustration does not supply the missing fact. The letter is evidence of the agency's response and stated legal position. It is not, simply by existing, evidence that the requested event occurred.
Imagine a fictional requester asking whether a fictional agency has records concerning a person called Subject Q. The response declines to acknowledge whether responsive records exist. An online summary announces that the government has “admitted it has a file.” A second summary says the response proves that there is no file. Both summaries supply an answer the letter withheld. The opposite conclusions do not become more reasonable because they are emotionally opposite.
There is a third possibility: read the letter as a claim about what the agency believes it may disclose, identify the cited exemption and the precise scope, then evaluate the response through the available administrative or judicial process. That approach does not assume the agency is correct. It identifies the dispute that can actually be examined.
The practical challenge is therefore twofold. Requesters need to understand what sort of decision they received. Readers and publishers need to avoid turning that decision into an unsupported story about its subject. This article follows both tasks through original worked records, distinguishing a refusal to acknowledge existence from a search result, a withholding decision and a statement about an underlying event.
2. Why a salvage ship became a legal shorthand
The CIA's historical account describes Project AZORIAN's attempt to recover a sunken Soviet submarine using the Hughes Glomar Explorer in 1974. The operation's public cover story involved deep-ocean mining; later reporting exposed the covert purpose. That historical account supplies the connection between the vessel and the familiar response. It does not establish that every modern use of the term concerns an intelligence operation, much less an operation resembling the salvage mission. CIA's history of the exposure of Project AZORIAN.
The National Archives' Office of Government Information Services explains the legal distinction: an ordinary withholding protects information within records, while a neither-confirm-nor-deny response protects the fact of whether responsive records exist. The agency cites the exemption underlying that position. The office also explains that the response can apply to part of a request while other parts receive ordinary processing. OGIS's March 2024 Glomar explainer.
The ship's name is memorable, but the administrative distinction is more useful. “Glomar” is not a tenth exemption that can be added to the nine in the statute. Nor is it a synonym for every redacted page. A blacked-out paragraph and a refusal to acknowledge a category of records have different informational effects.
Consider an invented publication containing three images: a historical vessel, a modern response letter and a heavily redacted document from an unrelated matter. Placed together without careful captions, the images could suggest that the modern requester uncovered a covert operation and that the redacted document is its evidence. Nothing in the mere juxtaposition establishes either link. The vessel explains the name; the response explains the agency's position; the unrelated page explains nothing about the request.
A responsible visual or written account would identify each item's relationship to the claim. The historical photograph would be labeled as background. The letter would be tied to the exact request it answers. Any released record would be identified by provenance and scope. This article does not reproduce such images because the analysis does not require them.
That separation is not pedantry. A colorful origin story can make a dry response feel like confirmation of a dramatic secret. Keeping the history brief preserves its explanatory value without letting it replace the evidence required for a contemporary claim.
3. Four response categories that should not share one label
FOIA.gov explains that the federal process concerns agency records; it does not generally require an agency to create new records, conduct research, analyze data or answer questions on a requester's behalf. A request framed as “explain why this happened” may therefore differ from a request for existing communications, policies or reports about the subject. Understanding the requested record is the first step toward understanding the response. FOIA.gov's frequently asked questions.
For an original reading exercise, suppose a newsroom has four fictional letters. Letter A says the office searched identified locations and found no responsive records. Letter B says it located responsive records but withheld them in full under stated exemptions. Letter C releases records with certain passages removed. Letter D neither confirms nor denies the existence of responsive records within a specified category. None of these fictional letters is an actual agency determination.
The newsroom initially enters all four into a spreadsheet under “request denied.” That may be a convenient broad administrative field, but it cannot support a detailed inference about existence. Letter B expressly acknowledges responsive records. Letter C provides some content. Letter A reports the result of a stated search. Letter D withholds the existence answer for its specified category. The same broad label has erased the distinctions the journalist needs.
A better local classification retains the response wording and adds separate fields for acknowledged existence, released content, stated search, cited legal basis and unresolved scope. “Not stated” should be available as a value. So should “applies only to category two.” A forced yes-or-no field may create false certainty even when every source letter has been copied accurately.
The response categories also have different implications for follow-up research. A dispute about which locations were searched is not identical to a dispute about whether acknowledging a record would reveal protected information. A disagreement over a redaction is not resolved merely by proving that the document exists, because its existence has already been acknowledged.
The point of the classification is not to decide whether each letter is lawful. That requires the applicable law and facts. It is to stop the preliminary data entry from deciding a different question by accident. Before a reader evaluates an agency's explanation, the summary must preserve what the agency actually said.
4. Read the exemption and the protected fact together
The FOIA statute contains distinct grounds for withholding. Exemption 1 concerns properly classified national-security information; Exemption 3 concerns qualifying statutory nondisclosure provisions. Exemption 6 addresses certain personal-privacy interests, while Exemption 7(C) addresses personal privacy in records compiled for law-enforcement purposes. The statute also includes the foreseeable-harm or legal-prohibition framework and requirements concerning partial disclosure and segregable nonexempt information. These provisions should not be compressed into a single claim that any reference to “security” ends the inquiry. FOIA's statutory text.
For classified information, section 3.6(a) of Executive Order 13526 expressly addresses refusing to confirm or deny when the existence or nonexistence of requested records is itself classified. This is a statement about the protected existence fact, not an automatic conclusion that every record associated with a controversial topic is classified. Executive Order 13526, section 3.6.
Imagine a fictional letter citing an exemption without explaining whether the relevant category is personnel information, a law-enforcement record, or a classified relationship. A reader should not fill the gap with the most sensational option. The useful annotation is “basis cited, connection to requested category unclear in this letter.” That identifies a communication problem without pretending to resolve the legal merits.
Now imagine another letter that identifies three numbered request items and applies different explanations to each. The newsroom's database needs three scope entries or an equivalent nested structure. If the editor copies only the broadest phrase into the headline, the published summary may falsely imply that every requested record received the same treatment.
The protected fact should also be distinguished from the fact the requester wants to prove. A hypothetical request might concern records that mention a person. Even an acknowledgment that such records exist would not, on its own, establish why the person was mentioned, whether the information was accurate or whether the person did anything wrong. The inferential gap remains important before and after any existence question is answered.
This is the central reading discipline: match the exemption claim to the category and then distinguish that category from the broader story. Neither the agency's citation nor the requester's suspicion is a substitute for that matching work.
5. “No records found” is not the same dispute
DOJ's procedural guide describes the adequacy of a FOIA search in terms of whether its methods were reasonably designed to locate responsive records, rather than simply whether the requester received the records expected. That discussion is about ordinary search obligations and their review. It should not be silently substituted for the separate question presented by a properly invoked Glomar response. The linked chapter is the version identified by DOJ's guide index as updated in August 2025. DOJ's procedural guide, search discussion.
Consider a fictional request for procurement correspondence from January through June. The response says the office searched an email archive covering that period and found no responsive records. The requester possesses an authentic purchasing memo referring to a related contract. The memo may provide a concrete lead about likely custodians, terminology or record systems. It does not automatically prove that the specified email archive contains the exact requested correspondence.
A useful research worksheet would record the request's dates, subject terms, offices and record types next to what the response says was searched. If the response identifies only January through March, the mismatch is visible without speculation about motive. If the dates match but the requester used a former project name and the office searched only a later name, the wording discrepancy becomes another concrete question. The invented example illustrates analysis of the record, not a prediction that a court would find the search inadequate.
A Glomar response creates a different situation. The D.C. Circuit's June 2023 decision in Schaerr v. Department of Justice affirmed that an agency properly invoking that response need not first search for responsive records. The court evaluated the agencies' explanations under the exemptions at issue; it did not treat a demand for a search as automatically defeating the existence-based protection. The court's 2023 opinion.
For the newsroom, the practical consequence is to maintain distinct questions in the file. “What did the agency search?” belongs beside an ordinary search result. “Why does the agency contend that even acknowledging existence is protected?” belongs beside the Glomar determination. The questions can interact in a complex case, but they are not interchangeable.
A conclusion about the underlying event requires still more evidence. An unsuccessful search is not a universal proof that nothing happened anywhere. A refusal to acknowledge existence is not a successful search secretly translated into coded language. The actual scope of the response controls what can responsibly be inferred from it.
6. A split response is not necessarily a contradiction
Suppose a fictional request has three parts. Part One seeks the agency's publicly issued records-request manual. Part Two seeks correspondence about a named person's alleged confidential relationship. Part Three seeks published aggregate processing statistics. The response releases the manual, links to the statistics and declines to confirm or deny records within Part Two. This is an invented exercise, not a claim about a real person or agency.
An impatient summary might say the agency contradicted itself by both finding records and refusing to say whether records exist. The apparent contradiction disappears once the referents are restored. The released manual answers Part One. The existence-based response concerns Part Two. The statistics answer Part Three. “Records” is not a single indivisible object shared by all three sentences.
For this exercise, create a three-row crosswalk. Each row contains the request item's exact scope, the response paragraph that addresses it, any released material, and the remaining question. Part One has a located manual. Part Two has an unresolved existence question and the agency's stated legal basis. Part Three has an aggregate table whose time period must still be checked. The crosswalk does not decide the merits; it prevents one response paragraph from being applied to a different item.
Now alter the fictional letter so that its numbered references are missing. It says only that “some records” are released and that the agency neither confirms nor denies “other records.” The right response in an editorial file is not to invent the mapping. Mark the scope as ambiguous and retain the full letter. Clarification of which part received which disposition would materially improve the account.
There is a similar problem when a requester revises a request during processing. If the published letter answers the narrowed version but a social-media post displays the original broader request, readers may compare documents that no longer form a matched pair. Preserving the clarification correspondence can establish what the agency was actually asked to decide.
This is one place where document management directly affects public understanding. A response can be accurately quoted and still be inaccurately represented if the associated request is the wrong version. A sound account identifies the request, its accepted revisions and the scope of each disposition before drawing conclusions about inconsistency.
7. Official acknowledgment is a matching problem, not a popularity contest
DOJ's current Exemption 1 guide summarizes the official-acknowledgment doctrine through specificity, matching information and an official documented disclosure. Its Glomar discussion distinguishes the existence of records from their contents and discusses limits on relying on broad public statements. The linked chapter was posted in January 2025; it is a guide to authorities, not a guarantee about a future case or a substitute for reading the relevant opinion. DOJ's Exemption 1 guide.
For an original evidence-organizing exercise, imagine four public items concerning a fictional project. Item A is an anonymous online claim. Item B is a newspaper story quoting an unidentified source. Item C is an agency-issued report acknowledging a broad program. Item D is an authenticated official release identifying a particular memorandum. These items should not be assigned the same provenance simply because they discuss the same subject.
The worksheet for each item would preserve its publisher, date, speaker or author, exact passage, record identifier if any, and the proposition it actually establishes. The last field is crucial. “The program existed” is different from “this agency holds this memorandum,” which is different from “every alleged activity occurred.” A document can support one proposition without supporting the others.
A useful editorial technique is to write the claimed match in two short columns. The first says what the request seeks. The second says what the official item acknowledges. If the columns differ in agency, period, record category or specificity, the difference remains visible. This is not an automated legal test. It is a way to keep an argument's premises available for qualified review.
Suppose the fictional official report discusses a program in 2018, while the request concerns a specific 2022 operation. An editor cannot erase the date difference by calling both items “the same project.” Conversely, a directly relevant official release deserves careful examination rather than being dismissed merely because the agency later uses a standard response phrase.
The value of the matching worksheet is that it can improve either side's account. It may reveal that a claimed acknowledgment is much narrower than advertised. It may also reveal a concrete prior disclosure that the response has not addressed. The aim is to make that comparison inspectable, not to turn popularity, repetition or confidence into legal authority.
8. Winning the existence argument does not release every page
In American Civil Liberties Union v. CIA, the D.C. Circuit rejected the agency's broad Glomar response concerning drone-strike records in March 2013. Its analysis considered official statements and the particular explanation offered for refusing to acknowledge any responsive documents. The court also expressly distinguished that result from a decision requiring disclosure of all contents: exemption questions remained for further proceedings. The decision is an example of a successful challenge with a bounded consequence, not a universal shortcut around withholding. The court's 2013 opinion.
A fictional newsroom can represent that distinction with a status timeline. At Stage One, existence is neither confirmed nor denied. At Stage Two, an authorized decision requires the agency to proceed beyond that position. At Stage Three, the agency processes responsive material and provides a disposition. At Stage Four, some content may be released while other content remains disputed. The invented timeline is a tracking proposal, not a prediction of deadlines or outcomes.
If the newsroom marks Stage Two as “documents released,” its database is ahead of the evidence. If it marks every later withholding as proof that the earlier challenge failed, it erases the narrower result actually obtained. Different stages can produce different kinds of success: recognition of existence, a better explanation, a new search, partial disclosure or a final merits decision. The labels should say which occurred.
This becomes especially important when a headline uses the word “ordered.” A court may order further proceedings, a search or a new explanation without ordering immediate public delivery of every requested record. The operative language and procedural posture matter more than a celebratory press release. A reader should be able to locate the order and see what action it requires.
In the fictional timeline, the editor records the date and source for each status change and leaves later fields blank until the corresponding evidence arrives. A blank release field is not a failure to celebrate transparency. It is a refusal to confuse one achieved step with a later one.
The same discipline applies when a response changes voluntarily rather than through litigation. A new acknowledgment may materially improve what is known. It does not retroactively validate every interpretation someone previously attached to the older letter. Keep the old response, the new response and the actual difference between them visible.
9. Privacy protection is not a finding of misconduct
A useful dated example is the D.C. Circuit's 2014 decision in People for the Ethical Treatment of Animals v. National Institutes of Health. The court upheld Glomar treatment for a category that would reveal whether particular people had been investigated, but concluded that the request also encompassed another category for which that rationale did not justify the same treatment. The partial outcome shows why the scope of a request and the protected interest require careful matching. It did not turn the response into a finding that the individuals committed misconduct. The court's 2014 opinion.
For a fictional editorial exercise, suppose a request asks for complaints concerning Employee R. A response cites privacy and declines to acknowledge responsive records. The draft headline says “Agency refuses to release Employee R's misconduct file.” That headline has introduced at least two unsupported propositions: that the relevant file exists and that its proper description is misconduct rather than an allegation, inquiry or other record.
A narrower description would identify the request and the response without adopting the allegation as a fact. The editorial file should keep separate fields for what was alleged by the requester, what the agency acknowledged, what an actual adjudicative finding established, and what remains unknown. A field labeled “allegation” should never automatically populate a field labeled “finding.”
The potential harm is not merely semantic. A person may appear in a record as a witness, complainant, recipient, background reference or subject of an allegation that was not substantiated. In this invented example, none of those roles is known. Choosing the most damaging one would be an editorial invention rather than a disclosure obtained through FOIA.
Public accountability can still be investigated through questions about institutional process: what published standards apply, how aggregate cases are counted, what oversight reports say, or whether an agency has explained its handling of a defined category. The suitability and availability of particular records depend on their circumstances. The point is not to guarantee that a broader request avoids privacy issues. It is to separate an institutional question from an unsupported accusation about a person.
This distinction protects the integrity of criticism as well as the subject's reputation. An account grounded in what the records actually establish is easier to examine and correct than one that treats every missing answer as evidence of guilt.
10. Exemptions and exclusions are different mechanisms
DOJ's exclusions chapter distinguishes a Glomar response, which expressly invokes an exemption-based refusal to acknowledge records, from the statute's three narrow record-exclusion provisions. The latter permit qualifying records to be treated as outside FOIA's requirements in specifically defined circumstances. The guidance also describes general notice language used by relevant components regardless of whether an exclusion was invoked in an individual case. Its stated case-law coverage is principally through May 2024, so it should not be mistaken for an exhaustive account of later litigation. DOJ's exclusions guide.
Imagine a fictional response letter that includes a standard paragraph explaining that certain records can be excluded from FOIA. An online reader circles that paragraph and declares that the agency has admitted using an exclusion in this case. The inference does not follow from a notice expressly designed to describe a general legal mechanism. The reader would need case-specific evidence, not merely the existence of standardized explanatory text.
This is a familiar document-reading problem beyond FOIA. A form can describe several possible actions without recording which action occurred. A privacy notice can explain data practices without establishing that a particular person's information was used in every listed way. An appeal instruction can explain a right without showing that an appeal was filed. In each instance, the document's function matters.
For the fictional letter, a useful annotation separates “general notice” from “case-specific determination.” The paragraph's placement, wording and relationship to the rest of the response should be preserved. If the distinction is genuinely unclear, the file should say so rather than promote a possibility into an admission.
The difference also matters for aggregate reporting. Counting every occurrence of the word “exclusion” in downloaded letters would count boilerplate as well as any case-specific discussion. That would be a word-frequency statistic, not a validated count of exclusion use. An automated extraction can be useful for finding passages that need review, but it should not be presented as an administrative total without checking what the passages mean.
The practical lesson is to classify the function of a statement before extracting a factual claim from it. Legal vocabulary appearing in a response is evidence that the vocabulary appears there. What it establishes about the individual request depends on the actual statement being made.
11. Build a denominator before drawing a trend
DOJ's guidelines for 2026 Chief FOIA Officer Reports ask agencies that track Glomar responses to report full and partial uses during fiscal year 2025, separating them if possible, and to report use by exemption. The guidelines distinguish the specific fiscal-year question from the broader reporting period for other narrative activities. A report year, a fiscal year and a publication date are not automatically the same interval. DOJ's 2026 reporting guidelines.
Consider an original synthetic dataset with four components. Component A reports six full and four partial Glomar responses. Component B reports twelve such responses without separating full and partial. Component C expressly reports zero. Component D says it does not track the category. The observed total among reporting components is twenty-two. The whole organization's total is not established by assigning zero to Component D.
Suppose A processed 100 requests, B processed 200 and C processed 50 during the same stipulated period. Among those three reporting components, twenty-two divided by 350 is approximately 6.29 percent. If D processed another 150 requests, dividing the same numerator by all 500 produces 4.4 percent. That second calculation has added an unobserved component to the denominator without adding its unknown numerator. It can be labeled as a ratio of reported responses to all processed requests, but not as a complete measured organization-wide incidence rate.
Even the first ratio needs its unit identified. Are the numerator's “responses” one per closed request, or can a request receive several tracked determinations? Does a partial response count once, or separately for multiple request items? This synthetic exercise stipulates one tracked response per request for the first calculation. A real dataset would need documentation supporting an equivalent assumption.
Exemption totals introduce another trap. Suppose the twenty-two synthetic responses cite Exemption X eighteen times and Exemption Y twelve times, with eight responses citing both. Adding the exemption counts gives thirty, but the union is twenty-two: eighteen plus twelve minus eight. A table of cited grounds is not necessarily a table of mutually exclusive cases.
Now imagine a second year in which D begins tracking and reports seven uses. A raw increase in the published total may partly reflect improved measurement coverage. That possibility does not prove the underlying practice stayed constant. It means the trend cannot be interpreted responsibly without the coverage change. Keeping a field for “not tracked” is therefore substantive transparency, not a minor spreadsheet preference.
These fictional calculations do not estimate DOJ's actual use or any other agency's rate. They show why the reporting limitation noted at the start of this article matters. A credible trend begins with comparable definitions, populations and periods, not with whichever two totals are easiest to divide.
12. An appeal, a clarification and a new request do different work
DOJ's response-letter guidance explains that adverse determinations must include notice of administrative appeal rights and access to dispute-resolution services, and that requesters must be afforded at least ninety days to appeal. The relevant agency's current rules and actual determination letter govern the particular process. This article does not calculate a deadline for anyone, and seeking informal assistance should not be assumed to suspend a filing requirement. DOJ's guidance on response letters.
Imagine a fictional requester with a confusing split response. The requester has three possible objectives: understand which request item received which disposition, challenge a stated adverse determination, or seek a different category of records. Those objectives may overlap, but they should be named separately. A clarification seeks a clearer understanding of the existing response. An appeal contests a decision through the specified process. A new request changes what records are being sought.
A simple correspondence log can prevent those objectives from becoming entangled. Record the original request identifier, the determination date, the issue raised in each message and whether the agency treats that message as clarification, appeal or new request. Preserve any confirmation of receipt. Do not assume that a conversational email has been accepted as a formal appeal merely because it expresses disagreement.
For this invented case, the requester drafts a one-page issue summary with three columns: disputed statement, supporting document and requested clarification or reconsideration. The first row identifies ambiguous scope. The second identifies a particular official disclosure the requester believes is relevant. The third notes a mismatch between the cited request dates and the accepted revision. These are concrete documentary issues rather than a general assertion that secrecy proves wrongdoing.
Such an issue summary is not a universal appeal form or legal strategy. It is a way to make the record understandable to someone qualified to assess the situation. Actual disputes can involve jurisdiction-specific precedent, administrative-exhaustion questions, costs and deadlines. A requester facing a consequential dispute should consult the relevant official instructions and, where appropriate, a qualified lawyer.
The editorial benefit of the log continues regardless of the outcome. It lets a later reader distinguish a pending challenge from a completed one, a new scope from an old one, and an agency's response from a requester's interpretation of it.
13. Mediation is assistance, not a secret appellate court
OGIS describes its role as a neutral, nonbinding, voluntary avenue for resolving federal FOIA disputes. Its FAQ expressly says it cannot compel agencies to release records, reverse administrative decisions or immediately answer requests, and that its services do not replace administrative appeals. These limits are important both for setting expectations and for describing an OGIS communication accurately. OGIS's mediation-program FAQ.
Suppose an invented OGIS exchange helps a requester learn that an agency's response covered only two of three numbered items. The parties then clarify the remaining item and agree on next steps. That can be a meaningful improvement. It should not be described as OGIS ordering disclosure, because the hypothetical improvement occurred through clarification and agreement rather than a binding adjudication.
Conversely, suppose OGIS explains the agency's position but the requester remains dissatisfied. An editor should not write that OGIS has judicially upheld the withholding. The office's explanation and the agency's determination have different institutional roles. Describing them as interchangeable would give the communication an authority it does not possess.
For a proposed newsroom status log, distinguish “assistance requested,” “issue clarified,” “agreement reached,” “no resolution reported” and “formal appeal outcome.” These are not official OGIS categories offered as a substitute for its records. They are suggested editorial labels that prevent different events from collapsing into a single “case closed” field.
A further practical caution concerns the substance of correspondence. A requester may share more personal information in a dispute-resolution exchange than is necessary for a public article. The fact that a journalist possesses the exchange does not mean every detail should be republished. An evidence packet can establish the relevant procedural event while minimizing unrelated personal information and preserving confidentiality obligations where applicable.
The goal is not to diminish mediation. It is to describe its genuine value without promising powers it lacks. A clearer scope, a workable communication channel or a corrected misunderstanding can materially improve access even when the office cannot command a particular release.
14. Publish the uncertainty without manufacturing mystery
Return to the fictional newsroom's file. It now contains an original request, an accepted scope revision, a split response, a potentially relevant official statement, an administrative appeal and a later clarification. The story can be much more informative than a screenshot captioned “what are they hiding?” It can show exactly which category remains unacknowledged, what the agency says protects that category and what the requester has supplied in response.
A proposed publication packet would include a short chronology, a scope crosswalk, links to the public documents that can safely be shared, and a clear account of pending issues. The chronology should distinguish the date a document was written from the date it was released or received. The crosswalk should identify which version of the request the response addresses. The public links should not expose unrelated personal details merely to make the package look complete.
For the synthetic example, a careful summary might say that the agency released a manual and aggregate statistics but declined to acknowledge records within a separately specified personal category; the requester challenged that determination using an identified official document, and the appeal remained pending as of the stated date. That sentence is less dramatic than claiming an admitted secret file. It is also substantially more useful because another reader can check its parts.
A later development should change only the claims it actually changes. If the agency subsequently acknowledges a document, update the existence status. If it releases content, describe the content with its limitations. If an appeal is denied, identify the decision and reasoning without treating it as a finding about every underlying allegation. If litigation begins, distinguish a complaint's assertions from a court's conclusions.
The article's central rule is therefore not “trust the agency” or “assume concealment.” It is to keep the evidence levels separate: the request, the response, the legal justification, any official acknowledgment, the released content and the underlying event. Each level can be challenged, but one should not be substituted for another without support.
A Glomar response leaves a particular question unanswered. That unanswered question can justify further scrutiny, careful records work and, where appropriate, a legal challenge. It does not license inventing the answer. The most valuable public account is the one that tells readers what remains unknown, why it remains unknown according to the record, and which concrete evidence could change that assessment.