The gold standard of Evidence is not a stack of citations, a convincing witness or a sophisticated instrument. It is a traceable connection between an exact claim and the observations, records or measurements that can actually support it. Evidence becomes meaningful through context, relevant methods, competing explanations and stated uncertainty.
This 23-chapter root guide connects research, science, statistics, education, forensics, law, digital records, engineering and public decisions across the eduKateSG ecosystem. It moves from the nature of a claim to the handling and interpretation of its sources, then through practical cases and a specialist reading library.
The cases are fictional educational examples. Evidence rules and professional duties vary across settings and jurisdictions; this guide is not a legal procedure or a substitute for specialised training. For the wider purpose, see Education and Inquiry.
Find the part you need
Choose a route from claim design to source provenance, explanation tests, worked examples or the specialist library.
Chapter index
Begin with the claim
Build an evidence trail
Test the explanation
Apply evidence responsibly
Work through cases
Choose the next route
CHAPTER 1 OF 23 · Begin with the claim
1. Evidence answers a question; it does not speak by itself
Evidence is information that bears on a particular claim. It might be a measurement, document, observation, testimony, experiment or record. Its relevance depends on the question and the reasoning that connects the item to a conclusion.
A photograph can establish what appears in a scene under specified conditions without automatically proving the date, cause or identity of everyone involved. A high test score may show success on particular questions without establishing durable conceptual understanding.
The gold standard of Evidence makes the claim precise, protects the source, selects suitable methods, compares alternatives and states uncertainty. It refuses to turn a plausible story or impressive instrument into certainty.
Start with Research, Inquiry and Reasoning if the question itself remains vague.
CHAPTER 2 OF 23 · Begin with the claim
2. Separate facts, interpretations and decisions
Write a sentence that could be checked: “The recorded water level was 12 centimetres higher at the later measurement.” Then distinguish an interpretation such as “rainfall caused the rise” and a decision such as “close the path.” Each is a different kind of claim.
Evidence can support a factual description without uniquely establishing an explanation. A policy decision can also involve values, costs and rights that are not settled by scientific measurement alone.
Label the steps. Observed fact, inference, competing account, applicable criterion and action should be readable separately even when they sit within one report.
The Evidence-Based Thinking, Judgement and Decision Making roots connect factual evidence with accountable choices.
CHAPTER 3 OF 23 · Begin with the claim
3. Identify the information needed before gathering more
More information is not always better evidence. Start by asking what observation would actually distinguish plausible answers. A carefully chosen comparison can resolve a question that a hundred general search results cannot.
If a plant grows poorly, possible explanations include insufficient light, water, unsuitable soil or disease. Gathering more photographs of healthy plants may be less informative than recording the actual conditions of the affected plant and a suitable comparison.
Define the scope and authorised sources. Respect privacy, consent, intellectual property and any professional limits; curiosity does not authorise access to restricted information.
Use Inquiry to make a question answerable and Investigation when a consequential event or discrepancy requires a formal evidential trail.
Provenance tells a reader where evidence originated and what happened to it. A statement attributed to “a report” is difficult to evaluate without the report’s author, date, method and relevant passage. A laboratory result needs sample identity and a method fit for its purpose.
Record source title, creator, date, context, link or identifier and important limitations. When working with original material, keep a clean record and distinguish it from copies and notes.
Different disciplines have different preservation duties. A student’s research notebook needs traceable citations; a forensic sample may require authorised chain of custody; a regulated organisation may face legal preservation duties.
The Digital Archiving Services and Digital Evidence Chain of Custody guides show why an intact record needs both reliable handling and meaningful context.
CHAPTER 5 OF 23 · Build an evidence trail
5. Check whether the source really supports the quotation
A quotation can be genuine while a summary misrepresents it. Read enough context to identify the author’s actual claim, conditions and qualifications. A figure may use a different denominator or time period from the article citing it.
Follow citations to the source when the claim matters. A report citing another report does not become independent confirmation merely by repeating the figure.
Keep observation separate from advocacy. Ask what the source actually measured or witnessed, what it inferred and which information it did not have.
The Verification and Research Services guides turn source checking into a repeatable process rather than a ritual of adding more hyperlinks.
CHAPTER 6 OF 23 · Build an evidence trail
6. Distinguish original data from processed results
A raw record, cleaned dataset, chart, explanation and executive summary are not interchangeable. Each transformation can introduce assumptions or errors. Good evidence practice leaves the path from source to conclusion traceable.
A graph may omit zero from its axis for a justified reason, but the choice changes the visual impression. An average may conceal individual variation. A software-generated transcript may contain mistakes even when it looks polished.
Keep the original values or document an appropriate authorised preservation method. Record corrections and the reason for them. Do not quietly replace a disputed number simply because it makes the final narrative coherent.
Study Data, Information, Statistics and Digital Archiving Services for complementary routes.
A number needs a quantity, unit, measurement method, relevant conditions and uncertainty. “The substance measured 5” says little without identifying what was measured, whether 5 is milligrams per litre, a score or a count, and how the figure arose.
Distinguish precision from accuracy. An instrument may show several decimal places without being calibrated sufficiently for the inference being made. A precisely recorded value can still describe the wrong sample.
For scientific results, consider detection limits, controls, sampling, calibration and whether the measurement represents the intended population or object.
The Measurement, Laboratory Testing Companies and Laboratory Technicians guides explain how legitimate source identity and valid methods combine.
CHAPTER 8 OF 23 · Build an evidence trail
8. Understand witness accounts as evidence with conditions
Human accounts can provide information not captured by instruments, but memory depends on attention, time, stress, questions asked and later conversations. A consistent account is not automatically complete; an inconsistent account is not automatically dishonest.
Record what the person says they directly observed separately from what they heard later or inferred. Avoid leading questions and preserve corrections where appropriate. Formal witness procedures depend on the setting and applicable law.
Evaluate statements alongside other relevant sources. Ask whether different witnesses had the same vantage point or access to the same information rather than demanding identical narratives.
See Witness Statement Preparation, Forensic Science and Investigation. Treat people fairly while testing the evidential claim.
Evidence must be interpreted through a method that fits the question. An interview may reveal a user’s experience; it cannot alone establish a population frequency. A laboratory test may detect a substance; detection alone need not prove where it came from.
State the intended inference, the data required, the method and its known limitations. A sophisticated method used outside its validated conditions can make a result look more authoritative without making it more reliable.
Where a question involves comparisons, consider a suitable control, baseline or alternative case. Where controlled testing is not possible, be explicit about which explanations remain open.
The Science, Mathematics and Research Services roots connect evidence strength to appropriate methods.
CHAPTER 10 OF 23 · Test the explanation
10. Distinguish association, causation and coincidence
Two events happening together do not prove that one caused the other. A school whose results improve after a new policy may also have different student cohorts, questions, staff and external support.
For a causal claim, consider temporal order, mechanism, alternative causes and what a meaningful comparison would show. Confounding can make unrelated factors look causally connected.
A correlation can still be useful evidence of a pattern. The error is claiming more than the pattern permits. State whether the evidence is descriptive, associative or causally persuasive.
The Evaluation and Root Cause Analysis guides show how causal questions differ from simple verification.
CHAPTER 11 OF 23 · Test the explanation
11. Test alternatives that could defeat a favourite explanation
Evidence becomes stronger when an explanation survives serious alternatives. If a system fails after a software update, the timing is a clue. Network outages, configuration errors or an unrelated hardware fault may still explain the event.
Write several plausible accounts and identify the observations each predicts. Seek evidence that distinguishes them rather than collecting only examples compatible with the preferred one.
A missing record is meaningful only if the system would reliably have created and retained it. Absence of evidence is not automatically evidence of absence.
The Critical Thinking, Investigation and Forensic Science roots develop methods for careful competing-explanation analysis.
CHAPTER 12 OF 23 · Test the explanation
12. Communicate probability and uncertainty correctly
Numbers can express uncertainty, but only if their meaning is justified. The probability of observing evidence under an assumption is not necessarily the probability that the assumption is correct after observing it.
An estimate may have sampling error, measurement limits, unknown confounders or model uncertainty. Do not turn those conditions into a spurious number such as “98 per cent certain” without a valid basis.
Explain how uncertainty affects the claim. The evidence may support a bounded range, a direction of change, a tentative association or an inability to decide.
Use Statistics and Judgement for the relationship between calculated uncertainty and proportionate human conclusions.
CHAPTER 13 OF 23 · Test the explanation
13. Check whether repeated sources are actually independent
Three news articles can repeat the same press release. Several charts can derive from one dataset. A model and an analysis tool may share the same incorrect input. Agreement is less informative when the underlying evidence is the same.
Draw a simple source map: origin, transformations, distributors and independent checks. A genuinely new observation contributes more than another polished restatement.
If sources disagree, check whether they use different dates, definitions, populations or methods. A conflict can reflect different valid questions rather than a simple error.
The Information and Verification guides support this habit of checking the route by which a claim travelled.
CHAPTER 14 OF 23 · Apply evidence responsibly
14. Respect legal, scientific and ethical boundaries
Evidence in a classroom, published research paper, workplace inquiry and court does not operate under one universal set of rules. Scientific reliability, legal admissibility, fair procedure and ethical permission are related but distinct questions.
Avoid implying that an item becomes truthful just because it was admitted, formally stored or handled through a documented procedure. Equally, a relevant observation may require special authorised handling before it can be used in a formal case.
Privacy, consent, protection of vulnerable people and legitimate disclosure limits can restrict collection or publication even when information would be interesting.
Follow Law, Regulatory Investigation Governance, Legal Hold Management and Legal Privilege Review for specialised contexts and consult qualified procedures where required.
CHAPTER 15 OF 23 · Apply evidence responsibly
15. Teach evidence through independent checking
Students can learn to ask “How do you know?” without treating every answer as equally uncertain. Start with a claim, locate its source, identify an observation that could support it and explain what the source cannot establish.
In mathematics, an answer may be tested by substitution, a counterexample or a second method. In science, a child can compare conditions and record measurements. In English and history, a student can identify a passage, its context and a reasoned inference.
The Clementi Mathematics guide uses error diagnosis to build independent understanding. School Reading Research, School Science Inquiry and JC Knowledge and Inquiry extend the practice across ages.
Evidence literacy is not cynicism. It is the ability to build justified confidence and change one’s view when better information arrives.
CHAPTER 16 OF 23 · Work through cases
16. Worked case: a vocabulary list is mistaken for writing ability
A fictional student memorises 200 advanced words, and a tuition brochure claims this alone guarantees stronger composition marks. The vocabulary count is genuine; the claimed writing outcome remains untested.
Good evidence would examine whether the learner uses appropriate words accurately, develops ideas, organises paragraphs and adapts language to purpose. A controlled or at least carefully compared writing task can reveal more than list recall alone.
The result might show valuable vocabulary growth while exposing a gap in transferring words into coherent sentences. That is a reason to redesign practice, not dismiss the learning entirely.
The question to transfer is simple: does the chosen metric represent the skill we claim to improve? Read Evaluation and Learning.
CHAPTER 17 OF 23 · Work through cases
17. Worked case: labels break the meaning of laboratory results
A fictional school field study collects samples from two ponds. The laboratory produces reliable results, but two container identifiers were switched before arrival. The team plans to average the readings and claim one pond is more polluted.
Valid testing cannot repair an unknown sample identity. Investigate the collection and transfer records. If provenance cannot be established, report the affected site-specific inference as unsupported.
The numbers can remain in the research notebook with the explicit limitation; they should not be attached to invented certainty. Improve identifiers and receipt checks for the next investigation.
This case combines Laboratories, Coordination and Investigation.
CHAPTER 18 OF 23 · Work through cases
18. Worked case: a relative percentage disguises the denominator
A fictional public notice reports that a certain event doubled from one year to the next. Readers assume a large absolute rise. The original data show four events in the first year and eight in the second.
The relative increase is 100 per cent, but the absolute difference is four. Neither is automatically the best summary without a population denominator, time period and context.
Reconstruct the rate where data permit, disclose small-number uncertainty and explain what the comparison can support. Avoid implying that the event has become common simply because a relative figure sounds dramatic.
A stronger report distinguishes counts, rates and practical significance rather than choosing the most attention-grabbing arithmetic.
CHAPTER 19 OF 23 · Work through cases
19. Worked case: an authentic photograph carries a false caption
A fictional history project finds an old photograph and attaches a confident caption naming a date and town. The image itself is authentic, but its archive catalogue says the location remains uncertain.
The photograph supports claims about visible features. It does not automatically support the invented location. Check the archive record, provenance, related materials and whether any independent source establishes the caption.
Revise the project to state what is visible and what is unresolved. Retain the picture as useful evidence within a truthful boundary instead of discarding it or treating an attractive caption as proof.
Evidence needs interpretation disciplined by source context. The Research and Inquiry roots help children and adults practise that distinction.
CHAPTER 20 OF 23 · Choose the next route
20. A diagnostic for claims that sound stronger than their sources
Ask eight questions: What exact claim is being made? What evidence would matter? Where did the source come from? What did the method measure? Is the comparison suitable? What alternatives remain? Which uncertainty matters? What conclusion is justified without stretching?
Label each response as verified, provisional, contested or unknown. An important unknown may call for further work, a narrower statement or an explicit decision not to act yet.
When the evidence has consequences for people, add questions about authority, privacy and fair treatment. When the claim concerns a causal intervention, examine whether the design can distinguish it from confounding.
The best next step is not always another data collection. It may be correcting a misleading label, improving a source record or clarifying what question was actually asked.
The linked guides below map evidence across reasoning, science, technical work, law, education and organisations. Each branch has its own methods and obligations. This root is an orientation map, not a claim that one evidence hierarchy applies to every discipline.
Begin with Research, Verification, Forensic Science, Evaluation or Education according to the claim you need to examine.
Reasoning and source quality
- Research
- Inquiry
- Information
- Data
- Verification
- Evidence-Based Thinking
- Critical Thinking
- Reasoning
- Judgement
- Decision Making
- Evaluation
Scientific evidence and measurement
- Science
- Mathematics
- Statistics
- Measurement
- Biology
- Chemistry
- Laboratories
- Laboratory Testing Companies
- Laboratory Technicians
- Research Services
Forensic, digital and legal evidence
- Forensic Science
- Forensic Scientists
- Investigation
- Digital Evidence Chain of Custody
- Digital Archiving Services
- Data Recovery Services
- Law
- Regulatory Investigation Governance
- Legal Hold Management
- Legal Privilege Review
- Witness Statement Preparation
- Cross-Border Evidence Cooperation
- Criminology
Systems, learning and improvement
- Engineering Failure Analysis
- Human Factors Engineering
- Quality
- Root Cause Analysis
- Management
- Coordination
- Integrity
- Learning
- Feedback
- Teacher Education
- Higher Education
- School Science Inquiry
- School Reading Research
- JC Knowledge and Inquiry
- Clementi Mathematics
- Education
Return to The Gold Standard master directory for wider subjects, or to Research & Inquiry Hub for related learning routes.
CHAPTER 22 OF 23 · Choose the next route
22. Frequently asked questions about evidence
Is a fact the same as evidence? A fact can be evidence for a claim, but its relevance and strength depend on the question and connection between them.
Does a citation prove a claim? No. It identifies a source; the reader must still check what the source reports and whether the reasoning is valid.
Is testimony weaker than every measurement? There is no universal ranking for all questions. Reliable direct testimony and valid measurement answer different kinds of claims under different conditions.
Does chain of custody prove that a claim is true? No. It records handling history; method validity and interpretation must also be examined.
Can AI-generated text count as independent evidence? Generated text is not independent verification of a factual claim simply because it is fluent. Trace substantive claims to appropriate original sources.
What if evidence is incomplete? State the supported boundary, what remains open and the additional observation that could make a practical difference.
CHAPTER 23 OF 23 · Choose the next route
23. Transfer the evidence habit into everyday decisions
Choose a claim from an ordinary, lawful source: a school poster, a public chart, a science exercise or a consumer comparison. Write the claim in one sentence and find the most direct available source.
Record what the source shows, what it does not show, an alternative explanation and the one check that would most improve confidence. Present your conclusion in language another reader can test.
Do not confuse caution with indecision. Evidence should support action when the case is sufficient, and protect people from unsupported certainty when it is not. The Education root connects this skill to adult professional and civic life; the Clementi Mathematics reference shows its foundation in a student’s ability to justify a mathematical step.