When Can a Derivative Classifier Determine Combining Two Unclassified Facts Yields a Classified Fact?

A derivative classifier can make that determination when the combined facts reveal a classified association, aggregate, or pattern that no single input discloses and that the derivative classifier is authorized to analyze at the resulting level. The decision hinges on three tests applied jointly: whether the combination produces new classified information, whether classification guidance maps the combination to a specific level and category, and whether the classifier holds the clearance and need-to-know for the output. If any test fails, the classifier must stop and refer the question upward rather than infer a level.

That answer matters in practice because derivative classification is the everyday engine of the U.S. classification system. Most classified documents are not marked by the original classifying authority. They are marked by analysts who restate, paraphrase, incorporate, or synthesize existing classified information under authority that someone else created.

The hard question is the edge of that authority. Restating one secret is clearly derivative classification. Slicing it into two unclassified pieces, then reassembling them, is not obviously derivative classification at all. This article separates the mechanism, the thresholds, the edge cases, and the practical failure modes.

The mechanism, step by step

Derivative classification rests on a legal and procedural chain. Executive Order 13526 defines derivative classification as incorporating, paraphrasing, restating, or generating in new form information that is already classified, and marking the new material consistently with the source. That definition is quoted in the Information Security Oversight Office's marking guide (ISOO 32 CFR Part 2001). Unclassified source material is not part of the chain. So the first mechanism question is whether each input truly lacks a classification, or whether the classifier simply lacks access to the sources that would reveal one.

The second mechanism is aggregation. Aggregation occurs when individually unclassified or lower-classified items combine to create a new sensitivity, often because the relationship itself is sensitive. A single refinery's output is public. The same refinery's rail loading schedule is public. The correlation of the two across many sites can reveal national distribution patterns that are themselves classified. The combination, not the components, carries the secret.

The third mechanism is the classification guide. Original classifiers prepare guides that tell derivative classifiers how to mark derived work. Guides typically operate at the paragraph level, listing what to classify at what level and for how long. A guide that addresses combinations will say so explicitly, for example that combining program A with country B requires a higher marking. A guide that is silent on combinations does not authorize a derivative classifier to invent one.

The fourth mechanism is authority. Derivative classifiers must have the clearance and access appropriate to the output, and must be trained and designated in writing. Even where a combination is logically classifiable, an analyst without the required access cannot legally perform the synthesis, because performing it would itself require exposure to the higher-level result.

Quantified context: how classification systems scale

The scale of the system explains why combination analysis is a standing risk rather than an occasional puzzle. The U.S. government classifies roughly 50 million to 70 million new documents per year in recent ISOO annual reports, the vast majority through derivative classification. Original classification decisions number in the thousands. In other words, the ratio runs about 100 to 1 or higher in favor of derivative over original decisions.

Clearance volume sets the universe of possible combiners. Roughly 4 million to 5 million Americans hold security clearances, and about 1.2 million to 1.3 million of those hold top secret eligibility, per government reporting summarized in ISOO and ODNI materials. Each cleared person is a potential derivative classifier for some subset of material, which makes the combination question a population-scale governance problem.

Declassification timelines add a time dimension. Most classified records carry a declassification date or event, commonly 10, 25, or 50 years depending on the category. Combination-created facts often carry the longest duration of their inputs, because the aggregate can remain sensitive after each component lapses.

Training requirements quantify the human control point. Derivative classifiers must complete training before marking, and refresher training is typically required annually. The mandated content includes the aggregation concept, which is the institutional acknowledgment that combination decisions are expected to happen routinely.

Edge cases that break simple rules

The open-source-plus-classified case is the most common trap. An unclassified news report plus a classified cable may combine into a classified assessment of a foreign government's intentions. The derivative classifier may legitimately produce that assessment only if an authorized classification guide covers it or an original classifier has classified the assessment itself. Absent that, the analyst has generated new analysis, not derived a marking.

The mosaic-in-reverse case is subtler. If two unclassified facts combine into a classified conclusion, the person who first performs the combination is arguably an original classifier for that conclusion, because nothing in the derivative chain classified it. That person then needs original classification authority or must refer the conclusion to someone who has it. Very few analysts hold that authority.

Compartmented access creates asymmetric cases. Two facts may each be unclassified at the collateral level yet jointly reveal a special access program's existence. Here the sensitivity is in the relationship, and the only correct move for a classifier without program access is to stop and escalate.

Foreign disclosure adds a second layer. A combination unclassified in U.S. channels may be classified under a bilateral agreement or controlled under export rules. The derivative classifier must apply the more restrictive governing guidance, not the more convenient one.

The counterintuitive professional caveat: the fact that a combination is obvious to an adversary does not make it unclassified. Classification depends on the damage an unauthorized disclosure could cause, not on whether outsiders have guessed. Many classifiers wrongly downgrade on the logic that the conclusion is already apparent.

Misconceptions worth retiring

The first misconception is that unclassified plus unclassified equals unclassified. That holds only when no guide, category, or aggregation principle covers the combination. The arithmetic of classification is not closed under addition.

The second is that a derivative classifier can always reclassify downward. Derivative classifiers apply existing decisions. They do not change levels, remove markings, or create new classifications. Any perceived error goes to the original classifier or the classifying authority.

The third is that aggregation requires secrecy to be intentional. Aggregation is a property of the information, not of anyone's intent. Accidental synthesis still produces a classified fact if the underlying guidance covers it.

The fourth is that paraphrasing away the sensitive detail declassifies the derived product. Paraphrase changes form, not substance. If the derived sentence still reveals the classified fact or its combination, the marking must follow.

The fifth is that automated tools can decide combinations. Current classification assistance tools flag keywords and known patterns. They cannot weigh damage, apply need-to-know, or exercise classification judgment, so they generate candidates for human review, not determinations.

Comparison of decision paths

The table below summarizes how a derivative classifier should route a proposed combination under common conditions. Each row reflects a distinct authority situation, and the last column names the required action.

The controlling logic: a guide must positively authorize the combination, access must match the output, and any ambiguity routes upward.

Timeline of a combination decision

The sequence below shows how a well-run review of a proposed combination should unfold, with typical durations drawn from common agency practice rather than a single mandated schedule.

Two thresholds in this timeline deserve explicit reconciliation. The training threshold says a derivative classifier may mark only within published guidance. The access threshold says a classifier may analyze only at or below cleared levels. These are different gates, and passing one does not imply the other. A fully trained analyst without program access still cannot perform a program-revealing combination, and an analyst with broad access but no applicable guide cannot invent a marking. When the gates conflict, the stricter gate controls.

The red flags below mark situations where the safest institutional answer is to pause and refer.

The combination produces a conclusion not stated in any single source. No guide paragraph describes the combination or its level. The proposed output would be marked lower than one of its inputs. The synthesis requires access the classifier does not hold. An automated tool flagged the combination, but no human authority has reviewed it. The classifier feels pressure to publish quickly and skip guide verification.

Practical actions for classifiers and reviewers

The first step is to inventory inputs before synthesis. Record each fact's source, marking, guide reference, and declassification date. If any input lacks a traceable authority, treat it as unclassified only provisionally and verify before combining.

The second step is to search guides for combination clauses, not just topic headings. Guides frequently address aggregation in a general section rather than under each topic. Document the search so reviewers can see what was checked.

The third step is to apply an explicit three-part test and write down the result: new information created, guide authority identified, access confirmed for the output. If any answer is no, the combination is not derivative classification and must be routed to an original classifier or the classifying authority.

The fourth step is to set an escalation threshold. A practical rule used in many offices: any combination that would raise the level of an input, reveal a relationship across programs, or touch a compartmented area goes to the classifying authority before drafting. This threshold catches most aggregation risk with modest delay.

The fifth step is to preserve the derivation trail. Record which sources were combined, which guide language applied, and who approved the marking. This trail is what allows later reviewers, and declassification reviewers decades on, to reconstruct the decision.

The sixth step is to refresh training annually and treat aggregation as a tested skill, not a footnote. Quick scenario exercises, where analysts must decide whether a combination is derivable, measurably improve consistency.

Limitations and what this analysis cannot settle

This article describes stable mechanisms and governance logic. It cannot resolve a specific combination in a specific agency, because the controlling guidance is program-specific and often itself classified. Nothing here substitutes for the classification guide that actually governs your material, your agency's derivative classification training, or a determination by your classifying authority.

A further limit is evidentiary. Public reporting on clearance populations and classification volumes varies by year and methodology, and the figures cited here are ranges rather than audited counts. Treat them as order-of-magnitude context, not precise statistics.

The deepest limitation is conceptual. Classification is a damage judgment, and damage judgments are inherently contextual, depend on what adversaries already know, and change over time. No rule set fully automates them. The derivative classifier's real skill is recognizing when a combination has crossed from restating to revealing, and escalating at that boundary.

This article is general educational information about classification governance. It is not a legal opinion, an agency determination, or guidance for handling any specific classified or controlled record. For decisions about real material, follow your agency's classification guide and consult your classifying authority or security office.