The unknown belongs inside the map.
A research system becomes dangerous when missing evidence disappears behind a confident conclusion. Evidence-gap analysis represents what is absent, inaccessible, stale, structurally uncovered or genuinely unknowable. The blank space is not empty. It is a bounded research state.
The claim graph specifies what should exist but does not.
Its value remains outside available evidence.
The range or reliability is bounded rather than absent.
Disagreement is not the same as missing evidence.
Do not fill the gap with inference. Represent it.
An evidence gap is a missing or unusable support relation required by a material claim. An unknown is the unresolved value behind that gap. The distinction matters: the gap belongs to the evidence architecture; the unknown belongs to the state being investigated.
Required support is absent.
A claim node lacks a valid incoming edge, a material attribute is unobserved or the available record cannot reach the proposition.
The value is unresolved.
The answer may exist in reality while remaining outside the current record. Unknown is a valid state, not a placeholder for a guess.
No feasible observation can recover it.
The event left no trace, the measurement is impossible or the resolution criterion cannot be satisfied. Further searching may have zero value.
Only some unknowns can change action.
Prioritize missing evidence capable of moving confidence across the declared decision threshold.
Reduce the known field. Watch the unknown expand.
Five dimensions define how completely the evidence graph can support the current claim. The scanner exposes the dominant gap and recommends the next observation with the greatest decision value.
Different voids demand different research actions.
A gap label should identify why support is missing. Otherwise every absence produces the same useless instruction: search more.
The state was never captured.
No evidence object records the required event, attribute, population or outcome under the necessary conditions.
The record may exist but cannot be inspected.
A restricted filing, deleted page, private dataset or unavailable archive blocks validation of its contents.
The relevant interval is unobserved.
Evidence exists before and after the decision window, but not at the time needed by the claim.
A material subclaim is orphaned.
The compound conclusion contains a proposition with no valid incoming evidence edge.
Only one observation channel exists.
Repeated sources reproduce the same measurement limitations and cannot expose method-specific error.
The available test cannot decide the claim.
Evidence is present, but the competing explanations predict the same observed result.
Do not collect more. Collect what can change the decision.
Select a candidate observation. The planner compares expected gap reduction, independence and acquisition burden, then reveals whether the action adds information or merely adds volume.
VALUE
FITINDEPENDENCEFRESHNESSCOST
Absence of evidence is not automatically evidence of absence.
A negative finding becomes informative only when the observation process had a high probability of detecting the thing if it were present.
“We did not find it; therefore it does not exist.”
The searched sources may be incomplete, the terminology may differ, the relevant interval may be missing or the object may not be publicly observable.
NOT FOUND + UNKNOWN DETECTION POWER → UNKNOWN“It was absent from the declared observable space.”
The search universe, time window, detection rule and expected visibility are explicit. The conclusion remains bounded to that inspected space.
COMPLETE SEARCH + HIGH DETECTION POWER → EVIDENCE OF BOUNDED ABSENCESeven passes. No invented certainty.
The protocol converts a vague request for “more research” into a finite plan with an explicit stop condition.
Fix the target
Use the exact atomic claim and decision context.
Map the known
List valid evidence edges and observed intervals.
Name each gap
Separate access, time, coverage, method and resolution.
Measure relevance
Ask whether resolving the gap can change action.
Choose next evidence
Maximize expected decision-relevant information.
Update the graph
Add the observation without erasing prior unknowns.
Declare the limit
Stop when value is exhausted or the state is unknowable.
Not detected does not mean not present.
Negative evidence becomes probative only when the search boundary, detection capability and expected trace are strong enough to make a missing observation informative.
The process found nothing.
This records a search outcome, not the state of reality. The target may fall outside the searched space, exist below the detection threshold or leave no expected trace. A search engine result, incomplete database query or short observation window can therefore support only a bounded statement about what the procedure detected.
VALID FORM → “No qualifying record was detected in corpus C, interval T, using procedure P.”A capable test failed to detect an expected trace.
The negative result bears on absence only when presence would probably have produced an observable signal inside the complete declared search space. The method must be sensitive enough, the target must be expected to leave a trace and the null result must be preserved like any positive observation.
REQUIRES → bounded universe + sensitive method + expected trace + documented null resultAn unknown becomes manageable when its location, consequence and resolution test are declared.
A useful gap register does more than list unanswered questions. It links every unknown to an exact claim, identifies why the observation is missing, records the decision it can affect and defines the evidence capable of resolving it. The matrix below prevents a vague research backlog from becoming an endless request for more sources.
No valid record observes the proposition under the required conditions.
Existing context cannot substitute for the absent state.
Material only if the observation could change the selected action.
Observe the state with a declared acquisition method.
Close when valid evidence exists or recovery is impossible.
Its existence may be referenced, but its content is unverified.
Secondary descriptions remain derivative evidence.
The inaccessible record may contain the deciding fact.
Prefer direct inspection; otherwise use independent evidence.
Declare the limitation when no lawful route remains.
Available records sit before or after the claimed state.
Do not project the nearest snapshot into the missing interval.
The gap can reverse sequence, duration or trend interpretation.
Match configuration and preserve version relations.
Do not imply historical recovery from a new capture.
One or more material propositions lack incoming evidence.
Do not average a missing subclaim into the supported whole.
A peripheral gap may not affect the core conclusion.
Support, attribute, qualify or remove the orphaned proposition.
Every material proposition has a valid status.
Many sources may repeat one measurement limitation.
Repetition within one evidence family adds limited independence.
A second method may reveal systematic distortion.
Use a channel with different assumptions and failure modes.
Stop when remaining method error cannot change action.
The same observed result is compatible with multiple states.
Occurrence evidence does not automatically prove mechanism.
Different explanations may require different interventions.
Seek an outcome that the rival explanations predict differently.
Retain plurality when no feasible test separates them.
The same blank space means something different in each evidence environment.
These cases show how to name the unknown, narrow the claim and choose the next observation. Each example separates what the current record establishes from what remains outside it. That separation is the difference between a useful limitation and an unsupported disclaimer.
A page disappears from one result capture
A current search result does not show a previously observed page. The record establishes non-detection for one query, location, language, device and time. It does not prove deindexing, a permanent ranking loss or global absence. The gap concerns other environments and intervals. Recollect matched queries, inspect direct indexability signals and preserve both result states before naming the change.
Three sources disagree about a person’s role
A company page shows a current executive role, an older profile uses a previous title and a directory omits the person. The conflict may be temporal rather than factual. The gap is the effective date and source lineage. Preserve versions, identify first-party authority and seek a dated appointment or departure record. Until then, state the role only within the source’s declared time boundary.
A percentage has no visible denominator
Two reports repeat an 18 percent market-share figure, but neither exposes the population definition, period or transaction basis. Agreement does not resolve the measurement gap if both derive from the same panel. The number can be attributed as reported, but comparison with another market remains unsafe. The next evidence is methodology or independent transaction data with a compatible denominator.
Disruption is visible, but root cause is not
Status notices, user reports and telemetry independently indicate a service disruption. This supports occurrence and a bounded duration. It does not by itself establish intrusion, responsible actor or technical cause. The resolution gap sits between event evidence and causal attribution. Preserve the event claim, keep cause unknown and seek forensic artifacts or an authoritative post-incident analysis.
An index reports a link that direct capture cannot find
The observations can coexist: the index may be stale, the referring page may vary by access state or the link may have been removed between captures. The current evidence supports discovery by the index at its recorded time and non-detection in the later direct page state. The missing interval prevents a precise removal date. Repeated matched captures can narrow, but not retroactively fill, that interval.
One generated answer cites a source once
The preserved run proves that the citation appeared under the recorded prompt and context. It does not establish stable visibility across models, accounts, sessions or future runs. The gap is behavioral distribution. A repeated-run design with controlled prompt variants can estimate recurrence, while direct source validation separately tests whether the cited page supports the generated claim.
A topic is mentioned but never resolved
A site references an entity across many pages, yet none defines the entity, answers its main question or connects it to the relevant parent topic. Mention count creates apparent breadth while the claim-level coverage remains empty. Map required questions to exact URLs, distinguish partial from complete answers and prioritize the missing page that closes the most important user path.
Search demand exists, but buyer intent is unclear
Query volume can show repeated interest without proving commercial readiness, category ownership or willingness to pay. The decision gap concerns intent quality and market behavior. Segment query modifiers, inspect result composition, observe conversion actions and compare direct inquiries. Do not convert search frequency into revenue expectation without an evidence bridge.
More evidence has value only while it can change the bounded conclusion.
Research does not end when every imaginable question is answered. It ends when material claims have adequate support, decisive alternatives have been tested and the expected value of another observation falls below its burden. A stopping rule protects the investigation from both premature certainty and endless collection.
Can the unknown move the action threshold?
Estimate the plausible values of the missing state. If every plausible value leads to the same action, the gap may be real but operationally non-decisive. If a small change reverses the choice, the gap remains high priority even when the broader evidence base looks strong.
Can the proposed observation actually close it?
A relevant source can still have low resolution value when access is doubtful, the method repeats an existing limitation or the expected output cannot discriminate between rival explanations. Score the observation itself, not the prestige or volume of the source.
Is the expected information worth the cost and delay?
Burden includes time, money, privacy exposure, legal constraints, technical risk and opportunity cost. High burden does not forbid collection; it raises the evidence value required to justify it. Preserve the rationale when a material gap is accepted rather than resolved.
Material claims are supported within scope
Every decision-bearing proposition has a valid evidence state: supported, contradicted, uncertain or explicitly unknown. Remaining gaps concern peripheral detail or values that cannot move the declared action. The conclusion stays bounded to the observed population, time and environment.
A plausible unknown can reverse the decision
The current conclusion sits close to a threshold and the missing state could move it across. Collection should target the observation with the strongest discriminatory power, not merely the easiest additional source. Until resolved, decision readiness must remain conditional.
New sources repeat existing lineages
Additional records add mentions but not independent support, temporal coverage or methodological diversity. Search saturation is reached for that evidence family. Continue only if a new channel can expose a different failure mode or observe a missing decision interval.
No feasible observation can recover the state
The event left no durable trace, the relevant interval was never captured or access cannot lawfully be obtained. Mark the state unknowable, document attempted routes and narrow the conclusion. Repeated searching must not transform impossibility into invented confidence.
Most false certainty enters through an unnamed missing state.
These failures do not merely reduce completeness. They change what the evidence appears to prove. Each control below converts the hidden gap into a visible limitation or a specific next research action.
Replacing the unknown with the most convenient explanation
A plausible narrative is inserted where no observation exists. Plausibility can guide hypothesis design, but it is not evidence. Preserve rival explanations and seek a test that predicts different observable outcomes.
Treating repeated publication as independent confirmation
Ten pages may derive from one press release, survey or dataset. Collapse shared lineage before counting support. Distribution breadth can show circulation, while evidential weight remains tied to the originating record and its method.
Extending one snapshot across an unobserved interval
A state observed today does not automatically establish yesterday, tomorrow or continuous presence. Preserve exact capture times and mark uncovered intervals. Use matched versions when change, duration or sequence matters.
Inferring inaccessible contents from references
A citation can establish that another source referred to a record, not everything that record contained. Seek lawful access or an independent proxy. If neither exists, state the content as unverified.
Using local evidence to support a global claim
A result observed in one market, language, device or population is expanded beyond its envelope. Narrow the claim to the studied space or collect a sample designed for the wider universe.
Treating an indicator as the target state itself
A ranking, score, mention or estimated metric may correlate with the construct without measuring it directly. Declare the proxy relationship, its assumptions and the conditions under which it can fail.
Combining missing, zero and not applicable
These states have different meanings. Zero is an observed value, missing is an absent observation and not applicable means the variable does not belong to the case. Preserve separate codes so analysis cannot manufacture a false numerical result.
Collecting without a decision-linked stop rule
More documents can increase cost and apparent authority without reducing the decisive gap. Define the required claim coverage, acceptable uncertainty and threshold before collection begins, then stop when additional evidence cannot alter action.
A credible answer can include a precisely stated unknown.
Evidence-gap analysis is strongest when it changes how a claim is written, tested and used. These practical answers define the difference between honest limits and vague disclaimers.
What exactly counts as an evidence gap?
A gap exists when a material claim requires an observation or support relation that is missing, inaccessible, stale, methodologically unsuitable or unable to discriminate the relevant state. A topic not mentioned in a document is not automatically a gap; it becomes one only when that topic is required by the declared claim or decision.
Is uncertainty the same as an unknown?
No. Uncertainty describes limited confidence, precision or variability around evidence that exists. An unknown means the relevant value or state has not been resolved by the available evidence. A measured estimate can be uncertain, while an unobserved population denominator is unknown. Both should remain visible, but they require different controls.
Which gap should be investigated first?
Start with the gap that can most strongly change a material decision and can feasibly be reduced by a discriminating observation. Do not prioritize by how visually large the missing area appears. A narrow causal gap may matter more than broad descriptive incompleteness when it determines whether an intervention is justified.
When does “not found” support absence?
Only when the search universe is appropriately complete, the procedure would probably detect the target if present, the target is expected to leave an observable trace and the null result is documented. Otherwise the defensible statement is limited to non-detection within the declared corpus, interval and method.
Can evidence gaps be reduced to one score?
A score can summarize dimensions for triage, but it should not erase their structure. The same total can represent strong access with poor temporal coverage or the reverse. Decisions should inspect the component profile, dominant gap and claim sensitivity rather than treating a composite number as a universal measure of truth.
How should an unresolved gap appear in a conclusion?
Name the exact missing state, explain why it matters, state what the current evidence does support and describe whether further evidence is feasible. Avoid empty phrases such as “more research is needed.” A useful limitation tells the reader what is absent and how that absence bounds the conclusion.
How do evidence gaps apply to content and topical coverage?
Map the questions, entities, attributes and relationships required to resolve a user task. A coverage gap exists where no page answers a material question or where several pages mention it without providing a complete claim-level answer. The remedy is not automatic volume; it is the missing content object with the clearest role in the topic structure.
Can a decision be ready while unknowns remain?
Yes. Decision readiness does not require total knowledge. It requires that remaining unknowns are visible, bounded and unlikely to reverse the selected action—or that the cost of delay exceeds the expected value of resolving them. The decision record should preserve those conditions so readiness can be reconsidered when new evidence arrives.
Do not hide the limit in a footnote. Build it into the finding.
A decision-ready gap statement lets a reader distinguish the observed result, the unobserved state and the consequence of that absence. It should be specific enough for another researcher to test and short enough to remain attached whenever the conclusion is reused.
State the proposition
Name the exact atomic claim under review. Avoid combining occurrence, cause, scale and consequence in one sentence because each may have a different evidence state.
Record what was observed
Identify the evidence object, environment and time that directly support the bounded part of the claim. Do not begin with interpretation.
Name the missing state
Specify the unobserved value, interval, population, mechanism or source content. “Insufficient data” is too vague to guide review.
Explain why it is missing
Classify the gap as access, temporal, coverage, method, resolution or observation failure. Different causes imply different remedies.
Bound the conclusion
Show which inference is blocked and whether the unknown could reverse the decision. Preserve supported subclaims rather than discarding the entire finding.
Define the next test
Name the evidence capable of reducing the gap, or state why further recovery is infeasible. “Search more” is not a testable action.
Declare sufficiency
Set the condition under which collection ends: threshold crossed, rival explanation rejected, interval recovered or expected decision value exhausted.
Gaps bound the record. Synthesis decides whether it is enough.
EVD/11 exposes what remains missing and whether further collection has value. EVD/12 integrates the complete evidence state into a bounded decision-readiness judgment.
Evidence objects, claims, context and support relations.
EVD / 02CLASSIFYEvidence Types & ClassesObserved, documentary, computational and derived evidence.
EVD / 03LINEAGESource ProvenanceOrigin, custody, transformation and derivation paths.
EVD / 04QUALITYSource QualityAuthority, directness, independence and claim fit.
EVD / 05CAPTUREEvidence Collection & PreservationAcquisition context, integrity and reproducible records.
EVD / 06CORROBORATECorroboration & TriangulationIndependent convergence across sources and methods.
EVD / 07CONFLICTConflicting Evidence ResolutionScope, time, meaning and method divergence.
EVD / 08TIMETemporal Validity & Evidence DecayFreshness windows, volatility and supersession.
EVD / 09MAPPINGClaim–Evidence MappingExplicit support paths from records to propositions.
EVD / 10CONFIDENCEConfidence CalibrationBounded confidence aligned with evidence strength.
EVD / 11GAPSEvidence Gaps & UnknownsMissing observations, blind spots and unresolved states.
EVD / 12SYNTHESISEvidence Synthesis & Decision ReadinessIntegrated evidence states for bounded decisions.