Evidence is not one category of thing. It is a relationship between information and a claim.
A photograph, experiment, eyewitness account, sensor reading, official record, statistical pattern and expert interpretation can all function as evidence. But their value depends on what claim they are being used to support, how directly they connect to it, how reliable the source is, how independent the observations are and how much uncertainty remains.
Categorising evidence well prevents a common failure: treating all pieces of information as though they carry equal weight simply because they are “evidence”.
Quick answer: how should evidence be categorised?
- Source: who or what produced it?
- Proximity: how close is it to the event or phenomenon?
- Directness: does it bear directly on the claim or only indirectly?
- Reliability: how trustworthy is the source or method?
- Relevance: how well does it address the specific claim?
- Independence: is it genuinely separate from other evidence?
- Strength: how much does it change confidence?
- Provenance: where did it come from and how was it handled?
- Uncertainty: what remains unknown, disputed or approximate?
This is a domain application of How to Categorise Anything. The archival specialisation is covered separately in How Archival Evidence Works.
1. Start with the claim
Evidence cannot be evaluated in isolation. Ask what proposition it is supposed to support or challenge.
2. Information is not automatically evidence
A fact becomes evidence when it is relevant to a claim under a reasoning process.
3. Direct evidence connects closely to the claim
A contemporaneous measurement or recording may bear directly on what occurred, depending on authenticity and context.
4. Indirect evidence supports through inference
Indirect evidence can be powerful, but the reasoning chain should be visible.
5. Primary and secondary sources answer different questions
A primary source may be closer to an event; a secondary source may offer better synthesis, context or critical comparison.
6. Primary does not mean automatically reliable
An eyewitness can be mistaken. A sensor can be miscalibrated. An original document can be forged or incomplete.
7. Secondary does not mean weak
A high-quality synthesis may compare many primary sources and correct local errors.
8. Source identity matters
Record whether evidence comes from a person, institution, instrument, database, experiment or derived analysis.
9. Method matters as much as source name
A reputable institution using a weak method can still produce weak evidence for a specific claim.
10. Reliability concerns repeatability and trustworthiness
Ask whether similar procedures or observations would produce consistent results under comparable conditions.
11. Validity concerns whether the evidence measures the intended thing
A perfectly reliable instrument can consistently measure the wrong variable.
12. Relevance is claim-specific
Evidence can be highly reliable yet irrelevant to the exact proposition under review.
13. Proximity can be temporal
Evidence recorded immediately may preserve details later memory loses, though immediate accounts can also lack context.
14. Proximity can be physical
A nearby sensor or witness may have better access to a local event than a distant observer.
15. Proximity can be causal
Evidence closer in the causal chain may support a mechanism more strongly than a distant correlation.
16. Independence prevents double counting
Ten articles copying one original report are not ten independent sources.
17. Common-source dependence should be traced
Map citations, data lineage or witness dependence before treating apparent corroboration as independent confirmation.
18. Corroboration strengthens evidence when sources are genuinely independent
Multiple independent methods converging on the same conclusion can be stronger than repeated use of one method.
19. Contradictory evidence deserves its own state
Do not delete inconvenient observations. Record disagreement and investigate why sources diverge.
20. Negative evidence needs careful interpretation
Failure to observe something is informative only if the observation method should reasonably have detected it.
21. Absence of evidence is not always evidence of absence
Incomplete records, weak detection or poor sampling can explain non-observation.
22. Quantitative evidence is not automatically stronger
Numbers can be precise descriptions of biased samples or invalid measures.
23. Qualitative evidence can be rigorous
Structured observation, interviews, documents and case analysis can provide strong evidence when methods and limitations are explicit.
24. Experimental and observational evidence differ
Controlled interventions can help test causality, while observational evidence may better represent real-world conditions.
25. Statistical association is not mechanism
A correlation can be evidence of relationship without proving why the relationship exists.
26. Mechanistic evidence and population evidence complement each other
One explains how; the other can show whether the effect appears across cases.
27. Evidence strength is not universal
The same evidence can strongly support one narrow claim and weakly support a broader one.
28. Evidence should update confidence, not merely decorate conclusions
Ask how much a piece of evidence should change belief compared with plausible alternatives.
29. Provenance is part of evidence quality
Record origin, handling, transformations, custody and version where those affect trust.
30. Derived evidence should preserve lineage
A chart, summary or AI extraction should link back to the underlying observations or documents.
31. Evidence can be classified by status
Proposed, collected, verified, challenged, superseded and withdrawn describe evidence lifecycle rather than evidence type.
32. Uncertainty should attach locally
The source can be authentic while interpretation remains uncertain; the measurement can be precise while sampling remains weak.
33. Evidence categories can be faceted
Source type, method, proximity, relevance and confidence are independent dimensions and should not be forced into one hierarchy.
34. AI can help organise evidence
Models can extract source, claim, method and citation relationships, but automated summaries should not erase provenance.
35. AI-generated content is not self-validating evidence
Its value depends on the underlying sources, method, reproducibility and fit to the claim.
36. Evidence classification needs domain standards
Medicine, law, history, science and engineering use different evidence hierarchies because their questions and error costs differ.
37. Cross-domain evidence labels need caution
“High quality” in one domain may not mean the same thing in another. Name the framework and authority.
38. A practical evidence record
- evidence ID;
- claim supported or challenged;
- source;
- method;
- directness;
- proximity;
- reliability;
- relevance;
- independence;
- provenance;
- uncertainty;
- status;
- reviewer;
- version.
39. Good evidence classification prevents false equivalence
It stops one anecdote, one dataset, one instrument reading and one synthesis from being treated as interchangeable objects.
40. The deeper idea
Evidence has meaning because it changes what a reasonable observer should think about a claim.
Do not ask only, “What kind of evidence is this?” Ask, “Evidence for what, produced how, and strong enough to change what conclusion?”
Final answer
Categorise evidence by source, method, proximity, directness, reliability, relevance, independence, provenance, strength and uncertainty. Evaluate it relative to a specific claim, distinguish corroboration from copied sources, preserve contradictory evidence, and keep method and lineage visible.