# 🧵 Tracing Claims Back to Primary Sources

By Better Wise

You already have two moves: lateral reading, where you open new tabs to see what others say about a source, and the credibility checklist, where you judge who is behind a claim and what evidence they show. This lesson adds the move that ends the hunt: following a recycled claim upstream until you reach the document, dataset, or actual sentence where it started. Most claims you meet are retellings, so **the real question is rarely "is this true?" but "what did the original actually say?"**

By the end of this lesson you can:

- Tell primary, secondary, and tertiary sources apart in under a minute.
- Trace a claim through news reports, press releases, and reposts to the original document.
- Run a match test on the original: same number, same population, same wording, same date.
- Decide what to do when the trail ends and no original appears.

## The Chain of Retelling

Every claim you meet has a history. A researcher measures something, a press office sharpens it, a reporter compresses it, a post strips it to a headline, and your feed hands you the last link. Each hop is a small act of translation, and translations lose things: hedges, sample sizes, caveats like "in mice" or "may be linked to."

```mermaid
flowchart TD
    P["🔬<br/>Primary source<br/>study, dataset, filing"] --> PR["📄<br/>Press release"]
    PR --> N["📰<br/>News article"]
    N --> S["📱<br/>Social post"]
    S --> Y["👀<br/>Your feed"]
```

The _____ is the original: the study, the raw dataset, the court filing, the full transcript, the unedited video. A _____ reports on it, and a _____ compiles the secondary ones, like an encyclopedia entry or an aggregator thread. None of these layers is automatically dishonest. They are just lossy, and the losses compound.

**Practice 1.** A post reads: "Scientists found coffee doubles focus." Before searching anything, write down three questions the post does not answer.

**Click to reveal guidance**

Your questions should aim at the missing origin. Strong answers include: Which study, by whom, published where and when? What does "focus" mean here, and how was it measured? Doubled compared to what: a placebo, a lower dose, no coffee at all? Who was studied, and how many? Any question that points upstream is a good one.

## Reading the Trail Upstream

The trail is usually visible if you look for the most specific detail. Reporters and reposts keep the punchy number and drop the citation, but the study name, author, journal, or year often survives somewhere in the chain.

1. Copy the most specific detail: the number, the study name, the named author, the quoted phrase.
2. Search it in quotes, adding a neutral word like "study," "report," or the year.
3. Find the earliest, most specific version of the claim, not the most popular one.
4. Open that original document and run the match test: number, population, wording, date.
5. If no original turns up, label the claim untraced and say so plainly.

Hedges do not survive the journey. A paper's "was associated with a modest improvement in a small sample" becomes "doubles focus" by the third hop, so when you reach the original, expect it to be more modest than the version you were handed. That gap is the finding you are actually looking for.

**Practice 2.** Two citations sit under the same article. (a) "According to researchers at a top university..." (b) "Kühn et al., 2019, *Nature Neuroscience*, doi:10.1038/s41593-019-0000-0." Which one can you follow, and what do you type first?

**Click to reveal guidance**

Only (b) is followable: it names authors, a year, a journal, and a DOI. Paste the DOI or the title into a search engine or the publisher's site, and if the full text is paywalled, check for a preprint or an author copy. Citation (a) is an appeal to authority with no address; treat it as a signal to keep searching, not as evidence.

## Field Guide: What Counts as Primary in Each Domain

"Primary" is not a property of a website, it is a property of the relationship to the claim. The same newspaper page is primary for "the mayor said X" and secondary for "a study found Y."

| Claim about | Primary source looks like | Where to start |
| --- | --- | --- |
| Health or science | Journal article, preprint, trial registry | DOI, PubMed, Google Scholar, preprint servers |
| Official statistics | Agency dataset or statistical release | The publishing agency's own site |
| A legal matter | Court docket, filing, ruling | The court's public records |
| A public statement | Full transcript, unedited video | Official transcript, timestamped recording |
| A company's numbers | Regulatory filing, annual report | The regulator's filing database |

**Practice 3.** A viral post says "a court filing proves the company knew." What is the primary source, and what would make you doubt the post?

**Click to reveal guidance**

The primary source is the actual filing in the actual case, which you can check through the court's records using the case name or number. You should doubt the post if it shows no case number, no court, no page reference, or a screenshot with the caption cropped. A real filing gives you a document you can open and quote; a screenshot of one gives you a claim about a document.

## When the Trail Ends in Nothing

Sometimes you follow the chain and it loops: post cites article, article cites blog, blog cites "a 2014 study" that nobody can produce. That is a circular citation, and the claim is untraced, not proven false. Say "I could not find the original" and hold your judgment; the burden of proof sits with whoever made the claim.

Two more dead ends are worth checking by name. First, **retractions**: search the paper's title plus "retracted," or check the journal page, because a retracted study is no longer evidence even while news coverage of it stays online. Second, **press release inflation**: compare the university or company release with the abstract, and the abstract with the limitations section. Each step usually softens the claim.

**Practice 4.** A post cites "a Harvard study from 2014" with no title or author. Searching the phrase returns only other posts repeating it. What do you conclude, and what do you write?

**Click to reveal guidance**

Conclude that the claim is unsupported as presented: no original document has been produced. Write something like "I could not trace this to a study; if you have the title or link, I will read it." That keeps you accurate and non-cynical: you are not calling it false, you are reporting that the evidence is missing.

## Field Notes From the Trail

- Do not stop at the abstract. Abstracts compress findings and quietly omit limitations, sample size, and population.
- Match before you judge. Same number? Same group? Same comparison? Same year? A mismatch is your story.
- "Studies show" with no citation is a mood, not evidence. Ask for the study and keep reading.
- Search the earliest version, not the loudest. The most-shared post is rarely the closest to the source.
- Paywalls are not dead ends. Try the DOI, a preprint, an author's page, or a polite request to the author.
- Retracted, corrected, or superseded work stays in circulation. Check the status of the paper itself.
- Absence of a source is not proof of falsity. It means untraced. Say that, and stay open to the document arriving.

## What the Trail Gives You

You can now take a claim that has traveled through five hands and walk it back to one, checking each hop for the detail that got lost: _____ vanish first, numbers drift, and citations sometimes _____ back to nothing. When the original exists, you compare it; when it does not, you say so without pretending certainty in either direction. That habit is what separates a skeptic from a cynic.

Next, the checkpoint for this module puts all three moves together: SIFT, credibility judgment, and source tracing, on claims you have not seen before.

## Review Deck

Front: Primary source
Back: The original document where a claim began: the study, dataset, filing, transcript, or statement.

Front: Secondary source
Back: A report that summarizes or interprets a primary source, such as a news article.

Front: Tertiary source
Back: A compilation of secondary sources, such as an encyclopedia entry or aggregator thread.

Front: First move when a claim cites 'a study'
Back: Search the study's title, authors, journal, or DOI, not the wording of the claim.

Front: What usually disappears as a claim travels
Back: Hedges, caveats, sample details, and limits such as 'in mice' or 'small sample'.

Front: Why an abstract is a weak stopping point
Back: Abstracts compress findings and often omit limitations, population, and sample size.

Front: Press release inflation
Back: University and company releases often sharpen findings before journalists ever see the paper.

Front: Where to find a health or science primary source
Back: The journal site via DOI, PubMed, Google Scholar, or a preprint server.

Front: How to check a viral quote from a public figure
Back: Find the full transcript or unedited video and locate the exact timestamp and surrounding context.

Front: Circular citation
Back: A claim cites a source that cites another, and the chain ends with no original document.

Front: Retraction check
Back: Search the paper title plus 'retracted' or check the journal page; a retracted study is no longer evidence.

Front: What to conclude when no original exists
Back: Call the claim untraced or unsupported, not proven false, and ask for the source.

Front: The match test
Back: Compare claim and original on number, population, comparison group, wording, and date.

---

## Contents

Part of [🔍 Critical Thinking: Ask Better Questions and Verify Information in the AI Era](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era.md)

- **🧠 Claims, Evidence, and Fallacies**
  - [🧩 Facts, Opinions, and Testable Claims](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/facts-opinions-and-testable-claims.md)
  - [🪤 Logical Fallacies in Everyday Arguments](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/logical-fallacies-in-everyday-arguments.md) (sign-in required)
  - [🪞 Cognitive Biases and Motivated Reasoning](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/cognitive-biases-and-motivated-reasoning.md)
  - [✅ Checkpoint: Claims, Fallacies, and Bias](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/checkpoint-claims-fallacies-and-bias.md)
- **🧭 The Verification Toolkit**
  - [🔎 SIFT and Lateral Reading in Practice](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/sift-and-lateral-reading-in-practice.md)
  - [🏛️ Assessing Source Credibility](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/assessing-source-credibility.md)
  - [🧵 Tracing Claims Back to Primary Sources](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/tracing-claims-back-to-primary-sources.md) — this lesson
  - [✅ Checkpoint: Verification Toolkit](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/checkpoint-verification-toolkit.md)
- **🤖 Cross-Checking, AI Answers, and Calibrated Confidence**
  - [⚖️ Cross-Checking Evidence with Claim, Evidence, Reasoning](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/cross-checking-evidence-with-claim-evidence-reasoning.md)
  - [🛰️ Verifying AI-Generated Answers](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/verifying-ai-generated-answers.md)
  - [📶 Calibrated Confidence and Unsupported Conclusions](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/calibrated-confidence-and-unsupported-conclusions.md) (premium content)
  - [✅ Checkpoint: Evidence, AI, and Confidence](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/checkpoint-evidence-ai-and-confidence.md) (premium content)
- [Course cheatsheet](https://everwise.study/courses/critical-thinking-ask-better-questions-and-verify-information-in-the-ai-era/cheatsheet.md) (premium content)