# Margin Evidence — full text

> Margin Evidence is an independent educational publication about personal research notes, source tracking, and everyday reasoning.

# How to Evaluate an Argument Step by Step

> Identify the conclusion, list the reasons offered for it, and check both those reasons and the connection between them. Make any added assumption explicit. Record what remains unresolved instead of forcing a yes-or-no verdict. For medical, legal, financial or accusatory decisions, use the relevant professional or institutional process; this general reading exercise cannot establish those conclusions.

Source: https://s-ehrlich.com/evaluate-argument-step-by-step/  
Publisher: Margin Evidence  
Published: 2026-10-10  
Updated: 2026-10-10

## How do you evaluate an argument?

Identify the conclusion, list the reasons offered for it, and check both those reasons and the connection between them. Make any added assumption explicit. Record what remains unresolved instead of forcing a yes-or-no verdict. For medical, legal, financial or accusatory decisions, use the relevant professional or institutional process; this general reading exercise cannot establish those conclusions.

A persuasive paragraph can feel finished before its reasoning has been inspected. Slowing it down is useful: put the statements on separate lines and ask what each one contributes. The aim is a fair account of the argument, followed by a precise assessment of its support.

This guide uses an original fictional catalogue project. The people, records, rules and quantities below are invented teaching material. No research experiment or improvement in readers' reasoning is claimed.

## What exactly are you being asked to accept?

An argument offers reasons, called premises, in support of a conclusion. The conclusion is the claim the writer wants accepted. [OpenStax's explanation of arguments](https://openstax.org/books/introduction-philosophy/pages/5-3-arguments) separates checking whether premises are true from checking whether they support the conclusion. A true conclusion can still be reached through poor reasoning.

Begin with this fictional proposal:

> We should print the catalogue on Friday. All twelve item descriptions are complete, and every image has permission for this print run. We promised the group a finished catalogue this month.

Write the conclusion without enlarging it: “We should print the catalogue on Friday.” Do not replace it with “The whole project is ready” or “Friday is the best possible publication date.” Those are different claims.

Now list the offered reasons: completed descriptions, image permission and the month's commitment. Keep the original paragraph beside your reconstruction. If someone supplied it, ask whether your version captures their intended claim before criticizing it.

## Which part is a record, and which part is a decision?

Use this original project sheet. Each entry is a detail supplied by the exercise, not a verified fact about a real project.

| Entry | Supplied information | What still needs attention |
|---|---|---|
| Descriptions | Twelve are marked complete | What “complete” means in this project |
| Images | Permission covers this print run | Whether other required checks are finished |
| Commitment | The group expects the catalogue this month | Why Friday, rather than another date |
| Proposed action | Print on Friday | The rule for deciding when to print |

The sheet does not accuse anyone of an error. It identifies what the proposal actually gives you and what it leaves open. In particular, a monthly commitment does not itself choose a day within that month.

Our [observation, inference and assumption guide](/observation-vs-inference-vs-assumption/) provides a separate vocabulary for labeling notes. Here the task is narrower: determine how the supplied statements are meant to support this particular action.

## What assumption would connect the reasons to the decision?

The [HKU tutorial on hidden assumptions](https://philosophy.hku.hk/think/arg/hidden.php) explains how making an implicit premise visible allows its meaning and justification to be examined. An added premise needs assessment too; writing it down does not establish it.

For our example, propose this possible decision rule:

> If descriptions are complete and image permission is confirmed, we should print on Friday.

Label that sentence **proposed assumption**, because the original paragraph did not state it. Ask the fictional project lead whether it is the actual rule. Do not attribute your reconstruction to the speaker as though you had discovered a quotation.

Now imagine the lead supplies the project's agreed checklist: descriptions complete, permission confirmed, page proof approved, and printer availability confirmed. That changes the task. The original paragraph addresses only two of the four listed requirements. It does not tell us whether the other two are satisfied.

The useful response is a specific request: “Please confirm the page proof and printer availability.” It is not “You have failed to prepare,” which the supplied records do not establish.

## Is the support meant to guarantee the conclusion?

Some arguments aim at a conclusion that must follow if their premises hold. Others offer support that remains uncertain. [OpenStax's inference overview](https://openstax.org/books/introduction-philosophy/pages/5-4-types-of-inferences) distinguishes deductive reasoning from inductive reasoning, which does not guarantee its conclusion. Do not demand certainty from a claim that offers only a provisional expectation.

Compare two invented statements about the same project:

- **Rule application:** Our agreed rule says to approve printing when all checklist requirements are satisfied. All those requirements are satisfied. Therefore, printing should be approved under that rule.
- **Expectation:** The last few catalogue jobs finished on their planned dates. This job may also finish on its planned date.

The first makes its decision rule part of the premises. Evaluating the rule's wisdom remains a separate question. The second looks to earlier jobs; we would want to know how comparable they were before relying on that expectation.

For this exercise, keep the decision-rule version. We have supplied a finite checklist, so we can inspect its entries without turning the article into a forecasting lesson.

## Have you checked the premises separately from the reasoning?

In deductive logic, validity means the premises cannot all be true while the conclusion is false. Soundness additionally requires true premises. [HKU's soundness tutorial](https://philosophy.hku.hk/think/arg/sound.php) explains why validity alone does not establish that the premises are true.

Suppose our fictional project rule requires every item on the checklist, and the printer booking is still unconfirmed. Someone writes, “All requirements are satisfied; therefore approve printing.” The inference follows the stipulated rule, but the stated record does not establish its key premise. Mark that premise unresolved rather than treating the tidy format as approval.

For real factual claims, follow our [source-credibility guide](/how-to-evaluate-source-credibility/). A reasoning worksheet cannot substitute for examining the actual record. Equally, collecting more records will not repair a connection you have never explained.

## Does the proposed repair add any independent support?

Consider a replacement sentence: “Friday is the right date because it is the date we ought to choose.” It repeats the decision in different words and supplies no additional project information.

[HKU's good-arguments tutorial](https://philosophy.hku.hk/think/arg/goodarg.php) discusses why circular support does not give independent reasons for accepting a conclusion. A formal-looking arrangement is not enough to make an argument useful to someone who needs reasons.

For our project, a useful repair would supply the missing checklist records or change the proposed action while those checks remain open. Adding “clearly,” “obviously” or a second copy of the conclusion would leave exactly the same gap.

## Can you revise the project argument yourself?

Use these newly supplied fictional facts:

1. The checklist has four requirements: descriptions, permission, proof approval and printer availability.
2. The first three are confirmed.
3. Printer availability has not been confirmed.
4. The group's stated rule requires confirmation of all four before approving printing.

Draft a conclusion that fits these facts. Keep the project rule visible and do not invent a reason for the missing confirmation.

A suitable answer is: “Under the group's stated rule, printing cannot yet be approved because printer availability remains unconfirmed. Ask for that confirmation before applying the approval rule again.”

Notice the precision. Three confirmed requirements out of four leave one unconfirmed. That is a checklist count, not a 75% probability of success. Nor does “unconfirmed” mean the printer is unavailable. The answer identifies the unresolved requirement and the next information needed.

An unsuitable answer is: “The catalogue will be late because the printer cannot do the job.” The exercise says neither of those things. Replacing “will” with “probably will” would still introduce a forecast unsupported by these supplied facts.

## What should your final evaluation say?

Our suggested final note is short enough to reread beside the original paragraph:

| Field | Example entry |
|---|---|
| Conclusion assessed | Approve Friday printing |
| Support supplied | Checklist records and the group's approval rule |
| Remaining issue | Printer availability unconfirmed |
| Present verdict | Approval not yet established under that rule |
| Next check | Obtain the missing confirmation |

Keep a link or locator for each real supporting record in your [source log](/keep-source-log-personal-research/). If your assessment changes, retain the earlier wording and identify which new information changed it. This is an editorial recordkeeping suggestion, not a validated scoring system.

Do not assign motives, publish private material or use this worksheet to establish misconduct, diagnosis, legal liability or financial suitability. Follow the qualified process that governs a consequential decision.

A useful evaluation ends with an identifiable reason: a premise needs verification, a connection needs explanation, or the supplied support establishes only a narrower conclusion. That gives the next reader something concrete to inspect and the original writer something concrete to answer.

## Sources

- [OpenStax: Arguments](https://openstax.org/books/introduction-philosophy/pages/5-3-arguments).
- [HKU: Hidden assumptions](https://philosophy.hku.hk/think/arg/hidden.php).
- [OpenStax: Types of inferences](https://openstax.org/books/introduction-philosophy/pages/5-4-types-of-inferences).
- [HKU: Soundness](https://philosophy.hku.hk/think/arg/sound.php).
- [HKU: Good arguments](https://philosophy.hku.hk/think/arg/goodarg.php).

## Frequently asked questions

**What should I write down first when evaluating an argument?**

For the catalogue exercise, begin with the exact proposed action: printing on Friday. Keep the original paragraph beside that note. If you instead write that the entire project is ready or that Friday is the best date, you have enlarged the claim before assessing it.

**Does an unconfirmed requirement mean it has failed?**

No. In the fictional project, an unconfirmed printer booking tells us that the required confirmation is missing. It does not establish that the printer is unavailable or that the catalogue will be late. Ask for the missing record and reassess the proposal under the stated project rule.

**Can I add a missing assumption to improve the argument?**

You can propose an assumption for examination, but label it as your reconstruction and check whether the speaker accepts it. The HKU tutorial treats implicit premises as claims whose meaning and justification need examination. Adding a sentence is not evidence that its content is correct.

**Does a valid argument automatically have true premises?**

No. Validity concerns the connection between premises and conclusion; soundness also requires true premises. A neatly arranged argument may still rely on a false or unresolved claim. Check the factual premises separately, using the actual supporting records rather than treating the logical format as verification.

**Can this worksheet settle a consequential disagreement?**

It can help you identify a precise question, but it cannot establish medical, legal, financial or accusatory conclusions. Preserve the relevant records and their context, protect private information, and use the qualified professional or institutional process for that decision. Do not turn an unresolved worksheet entry into an allegation.


---

# How to Evaluate Source Credibility

> To evaluate source credibility, first state the exact claim and its stakes. Identify the creator, publisher, source type, date, purpose, and relevant access. Then leave the page to investigate the creator independently, trace the claim to its closest evidence, inspect the method and scope, and seek genuinely independent corroboration. Record a bounded verdict: usable, usable with limits, not usable, or unresolved.

Source: https://s-ehrlich.com/how-to-evaluate-source-credibility/  
Publisher: Margin Evidence  
Published: 2026-09-05  
Updated: 2026-09-07

## How do you evaluate whether a source is credible?

Evaluate a source against the job you need it to do. Identify who created it, what kind of information it contains, when and why it was produced, and what evidence supports the exact claim you want to use. Then leave the page, investigate the creator through independent sources, trace the claim upstream, and compare it with other relevant evidence. Record a bounded verdict: usable for this claim, usable with limits, or not usable.

Credibility is not a gold star permanently attached to a website. A manufacturer's manual may be the strongest source for its product dimensions and a weak source for declaring itself the market's best choice. A first-person diary may be direct evidence of what its writer reported and incomplete evidence of what everyone experienced. The question is not “Is this source good?” It is “What can this source actually support here?”

## Begin with the claim, not the page

Write the sentence you hope to support before evaluating the source. If the sentence contains three factual parts, split it into three claims.

For example:

> The program launched in 2022, serves 40 locations, and reduced waiting time by 30 percent.

That sentence needs evidence for a date, a count, and an outcome. A dated launch announcement might support the first. A current directory might support the second. The third needs a defined measurement, comparison period, and method. One impressive-looking page does not automatically support all three.

Give each claim a stake level:

- **Low:** a reading list, a hobby decision, or background context
- **Moderate:** a purchase, public post, presentation, or workplace recommendation
- **High:** health, law, safety, finance, an accusation, or a decision that can materially affect another person

Higher stakes need stronger, more current, and more directly matched evidence. This article is a general information-literacy method, not a substitute for the standards of a clinician, lawyer, regulator, institution, or professional fact-checker.

## Identify what the source actually is

Record five plain facts before judging anything:

1. **Creator:** the named person, team, organization, agency, or company
2. **Publisher or host:** who made the item public
3. **Format:** study, law, dataset, manual, press release, news report, commentary, transcript, social post, or something else
4. **Date and version:** publication, revision, data period, and superseded status where relevant
5. **Purpose and audience:** to report, instruct, sell, persuade, entertain, comply, document, or recruit

Do not infer these from typography. A formal PDF can be a sales brochure. A plain government page can be the controlling current instruction. A `.org` ending, a polished logo, or a long reference list is not a credibility verdict.

Our [primary-versus-secondary-source guide](/primary-vs-secondary-sources/) helps classify the source's relationship to the question. That classification matters, but it does not settle quality. A primary source can be incomplete, self-interested, mistaken, or useful only for a narrow fact. A careful secondary source can synthesize a field better than one isolated primary item.

## Read laterally before reading deeply

When the source or creator is unfamiliar, open new tabs and find out what independent sources say about them. This is lateral reading: investigating the page from outside the page.

A Stanford Graduate School of Education account of research on [how professional fact-checkers evaluate websites](https://ed.stanford.edu/news/stanford-scholars-observe-experts-see-how-they-evaluate-credibility-information-online) reports that fact-checkers left unfamiliar sites to seek context from other sources. In that study, they reached more warranted conclusions faster than historians and Stanford undergraduates who tended to remain on the original site.

Use a short lateral pass:

- Search the creator's full name and organization.
- Look for an official identity page and independent coverage.
- Check relevant professional, institutional, regulatory, or publication records.
- Search the exact title or a distinctive sentence to find earlier versions and copies.
- Identify ownership, funding, sponsorship, or a commercial relationship when it matters to the claim.
- Look for corrections, retractions, disputes, and responses from the people or institutions being described.

The point is not to collect gossip about a source. It is to establish identity, incentives, expertise, and track record using records that the source does not control.

## Match authority to the question

Authority is contextual. The Association of College and Research Libraries frames this directly as [“Authority Is Constructed and Contextual”](https://www.ala.org/acrl/standards/ilframework), one of six ideas in its information-literacy framework. Expertise relevant to one question does not transfer automatically to another.

Ask three questions:

1. **What does this creator have direct access to?** A company controls its own current terms; a regulator controls its rules; a researcher can describe the study performed.
2. **What relevant competence is visible?** Look for field-specific training, methods, institutional role, or a documented body of work—not status in an unrelated subject.
3. **What incentive shapes the presentation?** Selling, advocacy, reputation, and institutional duty do not automatically invalidate a source. They tell you what needs independent checking.

A source can be authoritative for a narrow record and still overreach in its interpretation. Preserve that boundary in the note.

## Trace the claim to the closest evidence

Secondary reports often point upstream. Follow the links, footnotes, dataset names, document numbers, quotations, and embedded screenshots until you reach the material that actually carries the claim.

At each step, compare four things:

- **Wording:** Did “may” become “will,” or an association become a cause?
- **Population:** Does a finding about one age group, country, product, or sample now describe everyone?
- **Time:** Is an old figure presented as current?
- **Magnitude:** Was a range rounded, a relative change turned into an absolute one, or a conditional number detached from its denominator?

If the upstream item is unavailable, say that. A source saying “a study found” is evidence that the source made that statement; without the study, it is not a checked account of what the study did.

For quotations, compare the exact words, speaker, date, and surrounding passage. For images, find the earliest available version and its caption or catalog record. For statistics, locate the table, field definition, unit, and period. For rules and product specifications, prefer the current official document and record its version.

## Inspect the evidence and method

Once the source is worth attention, read the part that can support the claim. The relevant checks depend on format.

| Source type | Credibility questions |
|---|---|
| Research study | What question, sample, comparison, measurement, uncertainty, and limitation are stated? |
| Dataset | Who collected it, under what definitions, over which period, with what missingness or revisions? |
| Law, regulation, or policy | Which jurisdiction, effective date, version, exceptions, and official status apply? |
| Product page or manual | Which exact model, region, revision, measurement condition, and included accessory? |
| News report | Is the claim witnessed, documented, attributed, or repeated? Are subjects given a response? |
| Historical item | Who created it, for which audience and purpose, at what time, and with what surviving context? |
| Commentary | Which parts are evidence, which are inference, and which are the author's judgment? |

The Library of Congress teaches primary-source analysis through creator, purpose, audience, point of view, context, and comparison. Its [getting-started guidance](https://www.loc.gov/programs/teachers/getting-started-with-primary-sources) explicitly notes that primary sources can be incomplete and need other resources to reveal patterns and construct knowledge.

Methods do not need to be perfect. They need to be visible enough that you can decide what the result means. If a page presents a precise outcome but gives no source, denominator, comparison, or measurement window, the precision is decorative.

## Separate bias from usefulness

“Biased” is often used as an eject button. That loses information.

Every source has a position: a purpose, perspective, selection process, and limit. The useful questions are more exact:

- What would this creator notice first?
- What might the creator omit, lack access to, or prefer not to emphasize?
- Which words are observation and which are interpretation?
- Is the perspective itself part of what the research needs to understand?
- What independent evidence could challenge or narrow the account?

A campaign leaflet is poor neutral evidence that its claims are true and strong primary evidence of what the campaign chose to say. A customer testimonial does not establish typical product performance and can document one published account. Usefulness follows the research question.

The existing [observation, inference, and assumption guide](/observation-vs-inference-vs-assumption/) provides labels for keeping these layers apart. Apply those labels to both the source and your own note.

## Corroborate without counting copies

Three pages repeating the same sentence are not necessarily three independent sources. They may all copy one press release, wire report, database, or social post.

Build a small evidence map:

| Source | Underlying origin | Independent access? | Supports which part? | Limit |
|---|---|---:|---|---|
| A | Official record | Yes | Date and filed amount | No outcome data |
| B | Interview with participant | Yes | First-person experience | One perspective |
| C | Article repeating A | No | No additional support | Adds interpretation only |
| D | Independent dataset | Yes | Trend over defined period | Different population |

Look for sources with different access routes: a direct record, a witness, a dataset, a method paper, or an independent investigation. Agreement matters more when it is genuinely independent. Disagreement is not a nuisance to hide; it tells you to inspect definitions, dates, incentives, or methods.

The Library of Congress describes historians comparing sources to find agreement, disagreement, and productive discrepancies before constructing an account. That is a better model than voting by browser-tab count.

## Check date, version, and update path

“Recent” is not automatically better. A foundational historical document does not expire because a blog was posted yesterday. But a price, law, software interface, officeholder, safety instruction, schedule, or medical guideline can become wrong while its page still looks polished.

Record:

- Publication date and last substantive revision
- Data-collection period
- Effective date for a rule or policy
- Model, edition, jurisdiction, and language version
- Whether a newer item supersedes it
- Whether the page has a correction or update policy

An undated page can still be useful for a stable concept, but it should not carry a current claim without another date signal. An archived page can document what was true at that time. Label it as historical rather than quietly upgrading it to the present.

## Use a claim-matched source card

Copy this into your notes:

| Field | Entry |
|---|---|
| Claim being checked | One sentence, split if necessary |
| Stakes | Low / moderate / high |
| Creator and publisher | Named identities |
| Source type | Study / record / manual / report / commentary / other |
| Date, version, jurisdiction | Exact when available |
| Creator's relevant access | What they can know directly |
| Purpose and audience | Report / instruct / sell / persuade / document |
| Closest evidence | Link, table, passage, dataset, or document number |
| Method or provenance | How the information was produced |
| Independent context | What lateral reading found |
| Corroboration | Independent support, disagreement, or none found |
| Limits | Population, time, scope, uncertainty, conflicts |
| Verdict | Use / use with limits / do not use / unresolved |
| Safe wording | The narrow sentence the evidence supports |

Put the source into a [personal research source log](/keep-source-log-personal-research/) once it clears. The card is the evaluation; the log is the retrieval system. Combining them into one enormous template usually produces a magnificent blank table.

## Work through a neutral example

Suppose a blog says a new public program “cut delays by 30 percent.” Do not begin by scoring the blog's design.

1. Write the claim: a program caused a 30 percent reduction in a defined delay measure.
2. Identify the blog's author, publisher, date, and link.
3. Search outside the blog for the program and the quoted number.
4. Find the upstream report. Check whether 30 percent is relative or absolute, which dates were compared, what counted as a delay, and whether the report claims causation.
5. Find independent coverage or data that can test the interpretation.
6. Record any mismatch. Perhaps the report shows a 30 percent relative reduction at selected locations during a pilot, without isolating the program from other changes.
7. Rewrite the usable sentence to match: “The program's pilot report recorded a 30 percent relative reduction in its defined delay measure at selected locations during the stated comparison period.”

That sentence may still need criticism or independent evidence. It no longer claims more than the identified source did.

## Know when the verdict is unresolved

Sometimes the honest result is not “credible” or “false.” It is “not enough evidence.” Use that verdict when:

- The creator cannot be identified.
- The claim cannot be traced to the item cited.
- A key document or dataset is unavailable.
- Current and archived versions conflict without an explanation.
- Independent sources repeat one origin rather than corroborate it.
- The method cannot support the precision or causal claim presented.
- The available evidence is too weak for the stakes.

Do not turn “unverified” into “disproved.” Do not publish accusations, private data, or high-stakes conclusions because a checklist produced a red row. Escalate to the relevant qualified person or institution when the decision requires professional judgment.

## A five-minute first pass

When time is limited, do this before saving or sharing:

1. State the exact claim.
2. Identify creator, source type, date, and purpose.
3. Leave the page and investigate the creator independently.
4. Trace the claim toward its original record or evidence.
5. Check whether another independent source supports the same scoped claim.
6. Record one limit and a use / limited-use / unresolved verdict.

Five minutes will not settle every claim. It can prevent a polished page from borrowing authority it has not earned. The aim is not permanent trust. It is a sentence whose evidence, scope, and uncertainty can survive being read by someone who opens the link.

## Frequently asked questions

**What are the most important tests for source credibility?**

Match the source to one exact claim, identify who created it and why, check its date and version, trace the statement to the closest evidence, inspect the method or provenance, and investigate the creator through independent sources. Then compare genuinely independent evidence and record the limits. No single surface feature, including a domain ending, polished design, or reference list, completes that test.

**What is lateral reading?**

Lateral reading means leaving an unfamiliar page and opening other sources to investigate its creator, ownership, funding, expertise, reputation, and claims. Stanford researchers observed professional fact-checkers using this approach more effectively than readers who stayed on the original site. It is an efficient first pass, but the exact claim still needs to be traced and checked against evidence.

**Is a primary source always more credible than a secondary source?**

No. Primary describes a source's relationship to the question, not its automatic quality. A primary item may offer direct evidence while remaining incomplete, mistaken, self-interested, or narrow. A strong secondary source may synthesize many records and methods. Decide what each source can support, then evaluate creator, purpose, context, provenance, and corroboration rather than awarding trust by category.

**How can I tell whether several sources are independent?**

Trace each report upstream. Several pages may repeat one press release, wire story, dataset, or social post and therefore provide one evidentiary origin, not several. Note the underlying source beside every page. Stronger corroboration comes from genuinely different access routes, such as an official record, independent dataset, direct witness, or separate investigation that supports the same carefully scoped claim.

**What if I cannot verify a source's claim?**

Use an unresolved verdict when the creator is unclear, the cited evidence is unavailable, the statement cannot be found in the source, versions conflict, or the method cannot support the claim. Unverified does not mean disproved. Narrow or omit the statement, document what is missing, and seek qualified guidance before using weak evidence for health, legal, financial, safety, or accusatory conclusions.


---

# How to Keep a Source Log for Personal Research

> Keep a source log by creating one retrievable record per source when you first consult it. Record author or organization, title, publisher or collection, dates, stable identifier or URL, access date, exact page or section, relevance, limitations, verification status, and next check. Mark quotations, paraphrases, and your own ideas separately, then place source IDs beside working claims with a note explaining what each source supports. Preserve versions, corrections, access restrictions, and privacy limits, and test whether another authorized reader can retrieve the cited passage.

Source: https://s-ehrlich.com/keep-source-log-personal-research/  
Publisher: Margin Evidence  
Published: 2026-07-19  
Updated: 2026-08-26

## Give every source one retrievable record

A source log is a running list of what you consulted, where it can be found, what it contributed, and what limits apply. Create the record when you open the source, not three days later when twelve browser tabs have formed a small, ungovernable republic.

The [research notes section](/research-notes/) starts with retrieval because a claim you cannot trace is difficult to verify, quote, update, or correct.

## Use fields that answer future questions

For each source, record:

- a unique source ID;
- author or responsible organization;
- title and publication or document type;
- publisher, journal, site, collection, or repository;
- publication and update date when stated;
- stable URL, DOI, catalog number, or file path;
- date you accessed a changeable online source;
- page, section, timestamp, table, or figure used;
- one-sentence relevance note;
- access, copyright, privacy, or reliability limits;
- verification status and next check.

Use the citation style required by the eventual recipient, but keep richer working details even when the final style omits them. A short bibliography entry may not remember which paragraph supported which sentence.

## Separate quotation, paraphrase, and your idea

Mark verbatim language with quotation marks immediately and record its exact location. Keep excerpts short and necessary; note omissions and surrounding context. A paraphrase restates the source's meaning in your own structure and language while still requiring attribution. A personal thought or synthesis should be labeled as yours.

Do not paste source language into a general note field and plan to “sort it out later.” Later has an established interest in plausible deniability and poor punctuation.

When copyright, confidentiality, participant consent, or database terms restrict copying, record only what you are permitted to keep. Store sensitive material in an appropriate authorized system, not an ordinary personal worksheet.

## Link each claim to a source ID

In the working draft, add the source ID beside every checkable claim. One paragraph may need several IDs when it combines different evidence. Conversely, one source can support several claims without becoming permission to cite it for everything near the subject.

Write a brief support note: “S014, table 2, reports annual total for 2024,” or “S022, methods section, defines inclusion criteria.” This makes the relationship inspectable. A URL alone says where you went, not what you found there.

The [observation-inference guide](/observation-vs-inference-vs-assumption/) helps distinguish what a source reports from what you conclude; [thinking tools](/thinking-tools/) covers uncertainty labels.

## Record source quality without one magic score

Evaluate authority for the particular claim. A regulator may be primary for its rule, a manufacturer for its current manual, and a research paper for its own methods and results. None becomes universally authoritative on every adjacent question.

Note whether the source is primary or secondary, current enough, transparent about methods, complete enough for the claim, and independent of a relevant conflict. Save corrections, retractions, version history, and competing evidence. Do not compress those judgments into a mysterious “8/10 reliable” unless a defined project method genuinely uses that scale.

## Keep dates and versions visible

Online pages, datasets, manuals, and policies can change. Record access dates and version identifiers. When appropriate and permitted, preserve an authorized snapshot or archived reference without bypassing access controls. For a live page, revisit before publication or an important decision.

If two versions differ, do not silently overwrite the old note. Record what changed and which version supports the current claim. A clean history is especially valuable when a source later corrects itself.

## Run a retrieval test

Once the log contains ten sources, ask someone authorized—or future-you on a fresh morning—to locate three exact supporting passages using only the records. Fix every ambiguity the test reveals. Add controlled tags only after real retrieval needs appear.

Finally, back up the log according to the project's sensitivity and retention rules. The best source system is not the prettiest database; it is the one that lets a careful reader travel from sentence to evidence without packing emergency snacks.

## Frequently asked questions

**What is the difference between a source log and bibliography?**

A bibliography lists sources in a citation format for readers. A working source log usually contains more: access dates, exact passages, relevance notes, claim links, version information, limitations, permissions, and verification tasks. The bibliography may be generated from the log, but it rarely preserves every decision needed to retrace research. Follow the required final citation style while keeping richer private working records.

**Should I copy quotations into a source log?**

Only when permitted and useful. Mark verbatim text immediately with quotation marks and an exact page, section, paragraph, or timestamp, and preserve enough context to avoid distortion. Keep excerpts no longer than needed and observe copyright, database, confidentiality, and consent limits. When copying is restricted or sensitive, store an authorized locator and summary instead in the system approved for the project.

**How do I track a webpage that might change?**

Record the page title, responsible organization, URL, publication or update date when given, access date, and the exact section used. Save a version identifier or authorized snapshot when appropriate and permitted. Recheck the live page before an important decision or publication. If content changes, preserve the earlier record and note the difference rather than silently replacing the evidence trail.

**Can one source support several claims?**

Yes, when the source actually addresses each claim. Link every working claim to the relevant source ID and location, with a short note about the support. Do not cite a source for nearby assertions it never makes. Conversely, a complex paragraph may need several sources. Claim-level links make it easier to update one fact without rebuilding the entire project from memory.

**How should I rate source reliability?**

Evaluate fitness for the specific claim rather than assigning one universal score. Consider whether the source is primary for that fact, current enough, transparent about method, complete enough, independent where needed, and subject to corrections or conflicts. Record the reasons and limitations in words. A regulator, manufacturer, witness, paper, and review can each be authoritative for different questions without becoming equally reliable for all.


---

# How to Make a Synthesis Matrix for Research

> Make a synthesis matrix by placing sources on one axis and shared questions or themes on the other. In each relevant cell, record the source's contribution, its exact location and important limits. Then compare across sources before drafting a conclusion. Keep quotations, paraphrases and your own interpretation distinct. The matrix organizes evidence; it does not verify it, establish a complete literature search, or authorize sharing confidential material or drawing high-stakes professional conclusions.

Source: https://s-ehrlich.com/make-synthesis-matrix/  
Publisher: Margin Evidence  
Published: 2026-09-08  
Updated: 2026-09-08

## How do you make a synthesis matrix for research?

Make a synthesis matrix by placing sources on one axis and shared questions or themes on the other. In each relevant cell, record the source's contribution, its exact location and important limits. Then compare across sources before drafting a conclusion. Keep quotations, paraphrases and your own interpretation distinct. The matrix organizes evidence; it does not verify it, establish a complete literature search, or authorize sharing confidential material or drawing high-stakes professional conclusions.

A source log helps you find an item again. A synthesis matrix helps you see what several items say about the same question. You can use both without making either one enormous.

## What does synthesis add to a summary?

A summary describes what a source says. Synthesis examines relationships between sources and develops an account of those relationships. [Purdue OWL's explanation](https://owl.purdue.edu/owl/research_and_citation/conducting_research/research_overview/synthesizing_sources.html) distinguishes these tasks and cautions against forcing connections or ignoring outlying perspectives.

For example, three descriptions of a catalogue do not automatically make one explanation of how that catalogue changed. You must identify which dates, versions and features can actually be compared.

The question is not “How can I fit every note into a paragraph?” It is “Which relationship can these particular notes support?” Sometimes the answer is agreement. Sometimes it is a difference in scope, an unresolved contradiction or a question that remains unanswered.

## How should you choose the columns?

Start with a focused question and themes that help answer it. The University of Illinois-hosted [Writing for Inquiry and Research chapter](https://iopn.library.illinois.edu/pressbooks/writingforinquiry/chapter/chapter-3-the-literature-review/) presents a matrix as a way to track themes and relationships, rather than every detail of every article.

Our working layout puts one source in each row. Columns contain comparison questions, followed by a limits note. Keep full bibliographic details in your [source log](/keep-source-log-personal-research/) and use its stable source IDs in the matrix.

A column headed “Interesting points” provides little help with comparison. A question such as “What does this item say about revision dates?” tells you what belongs there and what does not. Do not force unrelated material into the nearest empty cell.

You may revise a theme as your reading develops. When you do, review every existing row against the new definition. Otherwise the heading changes while older notes still answer the old question.

## What information belongs in a cell?

Use a short note you can trace back to a passage:

- the source's relevant statement, accurately paraphrased;
- its page, section, table or another usable locator;
- any scope condition necessary to understand it;
- a clearly marked uncertainty if the statement is not yet checked.

NC State's [synthesis-matrix teaching page](https://faculty.chass.ncsu.edu/slatta/hi216/learning/matrix.htm) emphasizes topic-based notes and retrievable locations. That retrieval detail matters when a compressed phrase becomes the basis of a polished sentence later.

Keep the source's statement separate from your interpretation. “This edition supplies an index” and “an index would help my project” do different work. The first describes the item; the second records your judgment.

For direct quotations, retain exact wording and attribution. A paraphrase still requires attribution, as [Purdue's quotation and paraphrase guidance](https://owl.purdue.edu/owl/research_and_citation/using_research/quoting_paraphrasing_and_summarizing/index.html) explains. Changing a few words is not a substitute for writing an accurate note in your own language. Respect copying and access restrictions throughout.

## What does a small completed example look like?

Everything in the following exercise is invented: the collection, documents, dates, IDs and locators. They are teaching material, not real research or citations.

The fictional question is: “What do these three documents establish about revision information in a small collection catalogue?”

Assume these are the complete relevant details supplied:

| Fictional source | Revision date | Description of changes | Item-level detail | Limit |
|---|---|---|---|---|
| Example A: catalogue introduction, section 1 | States March 2024 revision | Says entries were revised | No individual changes listed | Describes that edition only |
| Example B: change notice, paragraph 2 | States March 2024 revision | Lists wording changes | Identifies entries 4 and 9 | Gives no reasons for the changes |
| Example C: later reader note, paragraph 1 | Does not state a revision date | Mentions seeing different wording | Does not identify entries | Observation cannot be matched to an edition |

The matrix contains three sources and three comparison questions: **3 × 3 = 9 comparison cells**. The separate limit column adds context, not a fourth research question. Counting cells checks the layout; it does not quantify evidence quality.

Notice what the exercise does not supply: an independent history of the collection, a test of reader comprehension, or proof that the revision corrected errors. You cannot obtain those conclusions by combining the available cells more confidently.

## How do you write from the example without overclaiming?

Begin with one column and compare the relevant cells. Then draft a relationship sentence, keeping the fictional source IDs attached.

A supported synthesis would be:

> In this fictional set, A and B identify a March 2024 revision, while B alone identifies the affected entries as 4 and 9. C reports different wording but supplies neither the edition nor the entries, so its observation cannot be matched to that revision from the details given.

This is our original example paragraph. It combines the records while preserving the difference between a dated document and an unmatched observation.

An unsupported version would be: “All three sources prove that the March update improved the catalogue.” The exercise supplies neither agreement from all three nor evidence of improvement. The problem is not merely missing cautious language; the proposed conclusion asks the documents to establish something they do not contain.

Do not silently repair that sentence by changing “prove” to “suggest.” First remove the unsupported relationship and outcome claim. Then state the narrower relationship the records actually support.

## Does an empty cell mean the source disagrees?

No. Use explicit working labels so an unfinished note does not become a negative finding:

| Label | What you mean |
|---|---|
| Not yet checked | You have not completed the relevant reading |
| Not discussed in the reviewed material | You checked the stated scope and found no discussion there |
| Not applicable | The question does not fit this source's role |
| Unclear | The wording or evidence needs another check |

These are our suggested labels, not a standardized coding system. Define any additional labels before using them.

“Not discussed in chapter two” is narrower than “the author never discusses this.” Record the scope of your reading. When a source is inaccessible, mark access incomplete rather than filling its row from another writer's description.

A disagreement requires statements that can actually conflict. Different questions, definitions, populations or editions may explain apparent differences; they should remain visible rather than being compressed into yes and no.

## Should a theme appear in every source?

No. The Illinois chapter's example makes room for uneven coverage. Purdue also advises against ignoring outliers or forcing relationships.

Our editorial recommendation is to assess a theme's relevance and evidentiary support, not discard it solely because only one source addresses it. One item may supply an important qualification. Equally, a detail appearing everywhere may be peripheral to your actual question.

Keep a note of why a theme stays, changes or leaves. “Outside this project's question” is a more useful reason than “only one filled cell.” You are making a reasoned selection, not a popularity count.

## How do you avoid turning the matrix into a second bibliography?

Give each tool a defined job:

- **Source log:** identity, retrieval, version and access details.
- **Matrix:** relationships across the selected comparison questions.
- **Draft:** a readable explanation with appropriate citations.

Do not paste the full log into every row. Keep the ID stable so the three tools remain connected. When you split a source into separate relevant sections, label those sections clearly without implying they are independent documents.

Before drafting, follow one matrix note back to its precise source location. Before sharing, follow one draft claim back through the matrix to that location. Fix any broken connection. This is our suggested traceability check, not a validated guarantee of accuracy.

The [source-credibility guide](/how-to-evaluate-source-credibility/) handles the separate question of whether the material is suitable evidence in the first place.

## When is the matrix ready to support a draft?

It is useful when you can explain the main relationships, locate their supporting passages and name the remaining limits. A completely filled grid is neither necessary nor sufficient.

Keep the project question and version date beside it. Preserve meaningful changes to your interpretation instead of silently overwriting the reasoning that led to an earlier draft. Store only information you are authorized to retain, in an appropriate system.

For academic work, follow the instructor's or institution's method and citation requirements. Medical, legal, financial, safety and accusatory conclusions require their own professional standards. An ordinary research worksheet does not establish those conclusions or justify publishing personal information.

Continue through [Research Notes](/research-notes/) when you need help keeping evidence attached to the sentences it supports.

## Sources

- [Purdue OWL: Synthesizing sources](https://owl.purdue.edu/owl/research_and_citation/conducting_research/research_overview/synthesizing_sources.html).
- [Charitianne Williams: Literature Review, Writing for Inquiry and Research](https://iopn.library.illinois.edu/pressbooks/writingforinquiry/chapter/chapter-3-the-literature-review/).
- [NC State: Synthesis matrix teaching page](https://faculty.chass.ncsu.edu/slatta/hi216/learning/matrix.htm).
- [Purdue OWL: Quoting, paraphrasing and summarizing](https://owl.purdue.edu/owl/research_and_citation/using_research/quoting_paraphrasing_and_summarizing/index.html).

## Frequently asked questions

**How is a synthesis matrix different from a source log?**

A source log preserves an item's identity, location, version and access details. A synthesis matrix compares what several sources contribute to shared questions or themes. Keep a stable source ID connecting the two. Neither tool replaces evaluating the evidence or checking that your eventual claim matches the cited passage.

**Do sources belong in rows or columns?**

Either orientation can work if it stays clear and consistent. This guide uses sources in rows and comparison questions in columns. Choose the layout that makes your particular notes readable, label both axes and keep locators attached. The orientation does not change what the sources can actually establish.

**Must every synthesis-matrix cell be filled?**

No. Distinguish material not yet checked from a question not discussed in the reviewed scope, an inapplicable comparison and genuinely unclear evidence. A blank cell alone explains none of those possibilities. Record the appropriate status rather than inventing content, treating silence as disagreement or claiming to have searched more than you did.

**Can I remove a theme because only one source mentions it?**

Not solely for that reason. Assess its relevance to your question and what the source actually supports. A lone qualification can matter, while a widely repeated detail may be peripheral. Explain your inclusion or exclusion decision rather than using filled-cell counts as a measure of importance or evidence quality.

**Does a completed matrix count as a systematic review?**

No. An organized grid does not by itself establish a defined search, complete coverage, selection criteria or an appropriate review method. Follow the requirements of your field, assignment and institution. Keep your scope and limitations explicit, and do not use the worksheet alone for high-stakes professional conclusions or sharing confidential information.


---

# Observation vs Inference vs Assumption

> An observation records what was directly noticed, measured, or documented under stated conditions. An assumption is a premise accepted without being established in the current task. Inference connects evidence to a conclusion through reasoning and often relies on assumptions. For example, “the clock displayed 9:07” is an observation; “the clock was accurate” is an assumption; “the meeting began seven minutes late” is an inference if 9:00 was the agreed start.

Source: https://s-ehrlich.com/observation-vs-inference-vs-assumption/  
Publisher: Margin Evidence  
Published: 2026-07-19  
Updated: 2026-09-07

## Observation records; inference interprets; assumption fills

For evidence notes, an observation records what perception or an instrument noticed under stated conditions. An inference is a conclusion drawn from observations or other evidence plus reasoning. An assumption is a premise accepted for the current task without being established there. These categories shift with context: one person’s observation may reach another only as a reported claim.

The [thinking tools section](/thinking-tools/) uses these working labels to make reasoning visible. A [NASA evidence lesson](https://imagine.gsfc.nasa.gov/educators/programs/cosmictimes/downloads/lessons/1955/Evidence_is_Clear.pdf) similarly separates observations from interpretations made through inference.

## What is the difference between an observation and an assumption?

| Test | Observation | Assumption |
|---|---|---|
| Source | Perceived, measured, or recorded under stated conditions | Accepted without being established in this task |
| Role | Supplies evidence | Lets reasoning proceed |
| Meeting example | “The displayed room clock read 9:07 when the chair began speaking” | The clock was accurate, 9:00 remained the agreed start, and speaking defined “start” |
| Next check | Inspect or repeat the record and its conditions | Verify the premise or state its uncertainty |

An observation can depend on assumptions about an instrument, definition, or record. Labeling both does not make the observation useless; it exposes what the inference needs.

## Write observations with conditions attached

“The chair was careless” is not an observation; it interprets motive or character. “The meeting started seven minutes late” is a reasonable inference only if the scheduled time and clock are accepted and “start” has a shared definition.

Record instrument, location, time, unit, and method when relevant. Photographs, sensors, documents, and transcripts are records created through systems with limits. For formal measurement, [NIST explains](https://www.nist.gov/metrology/metrological-traceability) that a result includes an estimate and evaluation of uncertainty; casual notes need proportionate context rather than pretending every record is limit-free.

## Show the bridge to an inference

> Observation: During a 40-minute watch, three scheduled trip listings did not appear on the stop display.
>
> Inference: Service may have been disrupted because several consecutive listings disappeared.

The display could have failed, the schedule could have changed, or the observer could have watched the wrong stop. The inference therefore remains provisional. Naming alternatives identifies what another check must distinguish; it does not make the inference useless. Use confidence language proportionate to evidence: suggests, is consistent with, probably, remains uncertain, or contradicts. Reserve “proves” for evidence that actually supports that strength of claim.

## Find and test the hidden assumption

The bus inference assumes the display normally reflects service and the listed trips were expected. State those assumptions, then check an official alert, timetable, direct observation, or independent record. Assumptions are not automatically errors; trouble begins when a contestable premise is disguised as fact or carries more weight than disclosed.

The [source-log guide](/keep-source-log-personal-research/) gives each check a retrievable record, while [research notes](/research-notes/) keeps quotations and paraphrases separate from interpretation.

List at least two plausible explanations and ask what evidence each predicts. Seek evidence that could weaken your view, not only details that decorate it. For the meeting, alternatives include a prior emergency, room-clock error, accessibility delay, or schedule change unseen by the observer. Those possibilities do not prove any one explanation. They prevent a timestamp from becoming an unsupported judgment about character.

## Repetition is not the same as independence

Five articles may repeat one original report. Ten social posts may quote the same screenshot. Trace claims upstream: did sources observe separately, analyze the same dataset, copy a release, or cite one another? Dependent agreement can show how information spread, but not the same confirmation as independent evidence. The [Cochrane Handbook](https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-05) applies the same principle narrowly to multiple reports of one study; our article-and-post examples are an editorial analogy.

## Use a three-column note

Create columns labeled Observation, Inference, and Assumption or Question. Put each sentence in one column; split a mixed sentence. Add a source ID and confidence note where appropriate.

Before sharing a conclusion, ask:

- Can another person locate the underlying record?
- Does the inference state its reasoning?
- Which assumption matters most?
- What alternative explanation remains plausible?
- What new evidence would change the conclusion?

This worksheet cannot establish legal fault, medical diagnosis, fraud, misconduct, intent, identity, discrimination, or another high-stakes conclusion. Do not publish accusations, private information, or identifiable material from it. Preserve the record and context, then use the qualified professional or institutional process that governs the decision.

The method does not guarantee truth. It makes assumptions visible and revision possible.

## Sources

- [NASA: Evidence Is Clear](https://imagine.gsfc.nasa.gov/educators/programs/cosmictimes/downloads/lessons/1955/Evidence_is_Clear.pdf)
- [NIST: Metrological Traceability](https://www.nist.gov/metrology/metrological-traceability)
- [Cochrane Handbook, Chapter 5](https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-05)

## Frequently asked questions

**What is the difference between an observation and an assumption?**

An observation records what perception, an instrument, or a document showed under stated conditions. An assumption is a premise accepted for the current task without being established there. Observations can rely on assumptions about instruments or definitions, so label both: inspect the observation’s record and conditions, then verify or disclose the assumption.

**Is an observation always a fact?**

An observation records what someone or a system noticed under particular conditions; it is not a guarantee of complete or error-free reality. Perception, instruments, definitions, timing, selection, and recording methods introduce limits. State those conditions and verify consequential observations independently. Another reader may receive the observation only through a report, adding source and transmission questions.

**What is an example of observation versus inference?**

“The room clock read 9:07 when the chair began speaking” is an observation with a stated indicator. “The chair delayed the meeting through carelessness” infers cause and motive. A schedule change, emergency, clock error, or accessibility need could explain the timing. The timestamp supports inquiry but cannot alone establish character or intention.

**Are assumptions always bad reasoning?**

No. Reasoning often needs background assumptions, such as accepting a calibrated instrument or stated schedule. Ask whether each assumption is contestable, disclosed, checkable, and strong enough for the conclusion. A planning premise differs from an unstated premise carrying a high-stakes accusation. Label important assumptions and verify those on which the conclusion depends most.

**Why do repeated reports not equal independent evidence?**

Several articles or posts may repeat one press release, witness, dataset, or screenshot. Their agreement shows repetition, not necessarily independent confirmation. Trace each claim upstream and identify whether sources observed separately or copied one origin. Dependent sources can reveal distribution and interpretation, but counting them as separate confirmations exaggerates the underlying evidence.

**How can I test an inference?**

List at least two plausible alternatives, identify what each predicts, and seek evidence that distinguishes them. Check the most consequential assumption, use independent sources when possible, and look for information that could weaken the preferred conclusion. State current confidence and unresolved questions. Compare explanations capable of fitting the known observations rather than inventing absurd alternatives.


---

# Primary vs Secondary Sources: A Practical Test

> A primary source provides original evidence or a firsthand record relevant to your research question. A secondary source analyzes, interprets, summarizes, or discusses primary material. The same item can change roles when the question changes: a newspaper editorial may be primary evidence of contemporary opinion but secondary commentary on the event. Primary does not automatically mean accurate, and secondary does not mean inferior; useful research often needs both.

Source: https://s-ehrlich.com/primary-vs-secondary-sources/  
Publisher: Margin Evidence  
Published: 2026-09-04  
Updated: 2026-09-07

## Primary vs Secondary Sources: The Short Answer

A primary source provides original evidence or a firsthand record relevant to your research question. A secondary source analyzes, interprets, summarizes, or discusses primary material. The important phrase is **relevant to your research question**: the same newspaper article, photograph, or book can be primary for one project and secondary for another. Classify the source by the job it does in your investigation, not by its file extension or the seriousness of its cover.

Primary does not automatically mean true or unbiased, and secondary does not mean second-rate. Good research often needs original evidence plus informed interpretation.

## What Counts as a Primary Source?

The University of Minnesota Crookston Library describes primary sources as original material where events or evidence are first recorded, including original thinking, new discoveries, and fresh information. Depending on the field and question, examples can include:

- letters, diaries, speeches, interviews, and correspondence;
- photographs, artworks, maps, recordings, and artifacts;
- laws, regulations, court records, and official documents;
- datasets, survey results, lab reports, and articles reporting original research;
- contemporary advertisements, websites, or social posts studied as objects themselves.

The category describes proximity to evidence, not reliability. A diary directly shows what its writer recorded; it does not prove every event happened as described. Ask who created it, when, for whom, for what purpose, and what it cannot show.

Put those observations in a [source log](/keep-source-log-personal-research/) beside the URL, creator, date, quotation, and limits note. A primary source without context remains primary, but becomes harder to interpret responsibly.

## What Counts as a Secondary Source?

A secondary source explains or evaluates other material. The University of New South Wales Library lists analysis, synthesis, interpretation, and evaluation as characteristic work. Common examples include:

- a history book interpreting letters and government records;
- a review article comparing several original studies;
- a biography built from interviews, correspondence, and archives;
- literary criticism discussing a novel;
- a documentary that assembles footage and testimony into an argument.

Secondary sources help readers enter an unfamiliar subject, discover vocabulary, locate primary evidence through citations, and see where informed interpretations disagree. They still require evaluation. Check the author’s expertise, evidence, publication context, date, corrections, and whether the cited material supports the conclusion.

“It has footnotes” is encouraging, not conclusive. Footnotes can lead somewhere excellent, somewhere irrelevant, or occasionally into a decorative shrubbery.

## Why Can the Same Source Be Both?

Classification depends on the question. The UNSW Library uses newspaper opinion pieces to show this context problem: an editorial may be primary evidence for studying public reaction at the time, but secondary when it interprets the event itself.

Try this table:

| Source | Research question | Role |
|---|---|---|
| 1998 newspaper review | How was a film received in 1998? | Primary evidence of contemporary reception |
| 1998 newspaper review | What caused the historical event shown in the film? | Secondary interpretation, if it discusses that event |

Do not ask, “What kind of object is this?” first. Ask, “What claim am I using this object to support?” The answer tells you its role.

## Use This Four-Question Classification Test

When a source refuses to sit politely in one box, write four answers:

1. **What exactly am I researching?** Name the event, claim, work, behavior, or period.
2. **Who created this source, and when?** Record the creator’s relationship to what you study.
3. **Does it present original evidence for my question?** Identify the specific data, testimony, object, or contemporary record.
4. **Does it mainly interpret other evidence for my question?** Find the analysis, synthesis, or restatement.

If both roles appear, label the particular passage you use. A research article may report a new experiment in its results section and summarize earlier work in its introduction. “Primary article” can be useful shorthand, but the paragraph still has a job.

This is also where [observation, inference, and assumption](/observation-vs-inference-vs-assumption/) must remain separate. “The letter is dated May 4” is an observation. “The writer delayed replying because she was angry” is an inference unless other evidence establishes the cause.

## Is a Primary Source Always the Best Source?

No. Match the source to the claim. Use the original regulation for what the regulation says, a dataset for reported values, and the study report for its methods and results. Use strong secondary scholarship to understand context, evaluate competing readings, and find related evidence.

For important legal, medical, financial, academic, or public accusations, this classification test is not professional verification. Follow the institution’s standards and consult a qualified professional where the decision requires one. Preserve private and confidential information, respect copyright and access terms, and never publish a person’s data merely because it appears in a source.

The useful research stack is rarely “primary good, secondary bad.” It is **question, evidence, interpretation, cross-check**. Once each source has a role, your notes behave like an argument.

## Sources

- [University of Minnesota Crookston Library — Primary, Secondary, and Tertiary Sources](https://crk.umn.edu/library/primary-secondary-and-tertiary-sources) (accessed September 4, 2026). Supports the definitions of original primary material, interpretive secondary material, and the cross-disciplinary example types.
- [University of New South Wales Library — Primary and Secondary Sources](https://www.library.unsw.edu.au/using-the-library/information-resources/primary-and-secondary-sources) (accessed September 4, 2026). Supports the analysis, synthesis, interpretation, and evaluation distinction and the principle that classification can change with the research context.

## Frequently asked questions

**What is the main difference between primary and secondary sources?**

A primary source supplies original evidence or a firsthand record for the question you are studying. A secondary source analyzes, interprets, summarizes, or discusses other material. The distinction describes the source's relationship to your research question, not its format, quality, or truthfulness. Start by naming the claim you want the source to support.

**Can one source be both primary and secondary?**

Yes. A newspaper editorial can be primary evidence when you study public opinion at the time, but secondary commentary when you study the event it interprets. A research paper can be primary for its new results and secondary when it reviews earlier studies. Label the role of the specific passage, data, or object you actually use.

**Is a primary source more reliable than a secondary source?**

Not automatically. A firsthand account can be mistaken, selective, or created for a particular audience. A careful secondary source can compare several records and supply valuable context. Evaluate creator, date, purpose, evidence, publication setting, corrections, and limits. Use original material for original claims and strong secondary work for interpretation and cross-checking.

**Is a research article a primary or secondary source?**

An article reporting a new experiment, dataset, survey, or other original study is usually primary for those methods and results. A review article that synthesizes earlier studies is secondary. One paper can contain both functions: its results may be primary evidence, while its literature review summarizes other work. Your field and research question determine the final classification.


---