Evidence Validation.
Evidence is not valid in the abstract. It is valid—or invalid—for a particular claim, under a declared scope, at a recorded time and with a traceable path back to the observation from which it came.
Validation tests the connection between a claim and its support.
The method does not certify a source once for every possible use. It evaluates whether a specific evidence object can support a specific claim under the current scope and validity conditions.
Evidence validation is the controlled examination of a claim’s supporting observations to determine their identity, provenance, relevance, independence, integrity, consistency and temporal validity before assigning an evidence decision.
Evidence is the exact observation extracted from it and connected to a declared claim.
General reliability cannot replace direct support for the proposition being tested.
Confirmation depends on independent observation paths, not the number of pages repeating a statement.
Six records that must never be collapsed.
Evidence systems fail when claims, sources, observations and interpretations are treated as if they were the same object.
Claim
A proposition with explicit subject, predicate, scope, time and permitted strength.
Source
The document, dataset, record, interface or person from which an observation is obtained.
Observation
The exact passage, value, state, event or structure recorded from an eligible source.
Evidence object
An observation connected to a claim with provenance, collection and validity metadata.
Inference
A conclusion derived from one or more observations through an explicit reasoning rule.
Decision
The accepted, conditional, rejected, contradicted or unresolved state assigned after validation.
From proposition to decision through a traceable chain.
An evidence item can fail at several different gates. The method preserves which gate failed instead of compressing every failure into a vague low-confidence label.
Normalize claim
Freeze subject, predicate, scope, time and strength.
Identify source
Resolve the exact source object and eligible version.
Record observation
Preserve exact content, value, context and location.
Establish origin
Follow citations and copies to the earliest known path.
Test relevance
Determine whether the observation supports this claim.
Test independence
Detect shared origins, circular citation and copying.
Test integrity
Check identity, completeness and transformation history.
Inspect conflict
Collect supporting, limiting and contradictory evidence.
Test validity window
Confirm observation time and claim-time alignment.
Issue evidence state
Accept, condition, reject, contradict or leave unresolved.
Four tests. One evidence decision.
Every protocol remains visible without JavaScript. When scripting is available, the complete protocols become an interactive validation console.
Does this observation support this exact proposition?
Can the evidence be reconstructed back to its origin?
How many separate observation paths actually exist?
Preserve competing evidence before adjudicating it
Evidence must survive every transformation event.
The chain does not imply forensic possession. It records how a digital observation moved from source state to published analytical claim.
Five pages can still represent one observation.
The graph separates direct origins, true independent confirmation, derivative copies and contradictory evidence.
ORIGIN
INDEPENDENT
COPIES A
CONTRADICTS
Valid reading
C repeats A; it increases distribution, not evidentiary independence.
Its methods and underlying inputs do not descend from A.
Contradiction is investigated, not erased from the graph.
The raw count of four source pages is not the independent evidence count.
Evidence decisions preserve why the gate passed or failed.
This synthetic matrix demonstrates claim-level validation. It does not score or certify real external sources.
| ID | OBSERVATION | CLAIM FIT | PROVENANCE | INDEPENDENCE | TIME | CONFLICT | DECISION |
|---|---|---|---|---|---|---|---|
| E-001 | Direct recorded value for C-001 | Direct | Complete | Primary origin | Valid | None found | ACCEPT |
| E-002 | Summary of another source | Direct | Traceable to E-001 | Dependent | Valid | None found | CONTEXT ONLY |
| E-003 | Related statistic for different population | Contextual | Complete | Independent | Valid | Scope mismatch | LIMIT CLAIM |
| E-004 | Undated statement with no origin | Possible | Broken | Unknown | Unknown | Not testable | REJECT |
| E-005 | Earlier state of same object | Direct | Complete | Independent | Earlier window | Apparent only | TIME-SEPARATE |
| E-006 | Same-time incompatible value | Direct | Complete | Independent | Aligned | Unresolved | CONTRADICTED |
Six measures. No universal credibility score.
Each measure answers one operational question. The final decision remains claim-specific and cannot be transferred automatically to unrelated propositions.
Direct-fit rate
direct-fit evidence ÷ reviewed evidenceShows how much of the reviewed set directly addresses the normalized claim.
Provenance completion
records with complete origin paths ÷ reviewed recordsCounts evidence objects whose material chain can be reconstructed.
Origin independence
independent origin clusters ÷ supporting evidence objectsExposes repetition and syndication hidden by raw source counts.
Temporal validity
time-aligned evidence ÷ time-eligible evidenceTests whether evidence supports the claim during its declared validity window.
Open-conflict rate
unresolved material conflicts ÷ tested claimsKeeps contradictions visible instead of removing them from accepted narratives.
Claim support ratio
accepted direct origins ÷ required support thresholdCompares valid support with a threshold declared before the final decision.
Four cases where “credible source” is not enough.
All examples are synthetic and demonstrate validation logic rather than factual claims about real sources.
Eight shortcuts that produce false certainty.
These failure states remain attached to the evidence record and constrain the language of the final claim.
Source halo
Publisher reputation is treated as proof for every claim it contains.
Nearest-copy origin
The page where evidence was found is mistaken for its original source.
Citation circularity
Sources reference one another without an identifiable underlying observation.
Context stripping
A passage is detached from conditions that materially limit its meaning.
Temporal leakage
Evidence outside the claim window is used without modeling state change.
Transformation opacity
A derived value lacks its inputs, denominator or calculation history.
Conflict deletion
Contradictory evidence disappears because it complicates the desired conclusion.
Certainty inflation
Conditional or single-origin support is published as settled fact.
Every accepted claim should be reconstructable.
The manifest preserves the claim, source identities, observations, transformation chain, dependencies, conflicts and final decision as separate records.
Every method node. One controlled research route.
MTH/09 validates evidence decisions. MTH/10 preserves the complete method, inputs and transformations so another execution can reconstruct the result.