01 / ProcedureRequest the supporting paperwork.
A citation tells you where a writer points. It does not guarantee that the destination exists, contains the claim, supports it, or was checked by the writer. “Has references” and “has evidence for this claim” are separate approvals.
Pin down the exact claim.
Copy the sentence, not just the topic. Keep the number, population, place, period, and strength of wording: “causes,” “is associated with,” and “people reported” do different jobs. Save the page title, author, date, and link so you can retrace your route.
Open the citation. Yes, actually open it.
Find the relevant passage, table, or figure. Search within the page or PDF for the number or distinctive phrase, then read around the match. A search snippet, abstract, or AI summary can help locate material; it is not a substitute for checking the relevant source and its limitations.
Ask what this source contributed.
Did it collect data, witness an event, analyze records, or simply say “according to” someone else? If it passes the claim along, follow that next link. Note any changes in wording. A chain of attribution may be perfectly legitimate; the question is what evidence is at its end.
Check the evidence against the sentence.
Do the population, dates, units, comparison, and result match? Is a quotation being presented as a finding? Did a tentative suggestion become a certainty, or an association acquire a causal promotion? Check methods, limitations, and any visible correction, update, or retraction.
Look sideways before signing off.
Leave the page and investigate the author or organization through other sources. Look for coverage that independently checks the claim or explains the wider evidence. The source's own “About us” page is useful self-description, not an external audit.
Training simulation / Entirely fictional
02 / Worked exampleWatch a claim receive three promotions.
All publications, records, and numbers below are invented. This is a demonstration of how wording can drift, not an allegation about a real publisher. No actual productivity experiment is being reported.
You encounter: “Turning off notifications boosts productivity by 40%.” There is a citation. Excellent. The paperwork has entered the building.
- 01 / Social post
“40% more productive.”
Links to a news article. No observations or measurements of its own.
- 02 / News article
“Research shows a 40% productivity boost.”
Links to a vendor's blog. Does not describe an independent test.
- 03 / Vendor blog
“40% reported an improvement.”
Links to notes from the vendor's small internal pilot. The measured “improvement” needs inspection.
- 04 / Pilot notes
10 of 25 volunteers said they felt less distracted.
One week, no comparison group, no measurement of completed work. The notes describe a self-report—not a measured change in productivity.
Inspection result: what does the 40% actually describe?
10 ÷ 25 × 100 = 40%. That is the share of participants reporting less distraction, not the size of a productivity increase. The denominator quietly changed jobs. “Felt less distracted” also became “more productive,” while a small uncontrolled pilot became a general causal claim.
The chain ends in some evidence: the recorded responses. It just does not support the headline. Multiple links have not repaired that mismatch.
A defensible summary: “In this fictional pilot, 10 of 25 volunteers said they felt less distracted after a week with notifications off. It does not establish a 40% productivity gain or show that disabling notifications caused the reported change.”
You do not need to prove who introduced the mistake to reject the unsupported conclusion. Intent requires its own evidence. The conclusion can be wrong before anyone's motives have been investigated.
Reference department / Known failure modes
03 / Citation failure modesThe bibliography has excellent posture.
Unfortunately, posture is not a research method. Watch for these different problems; they do not all justify the same conclusion.
The relay
Article A cites B, which repeats a claim from C. Useful attribution, perhaps—but A and B have not supplied two additional tests. Follow the relay to whatever C actually provides.
The circle
A cites B; B cites C; C cites A. For the claim you followed, the references return you to the starting point without adding underlying evidence. Record the loop rather than counting each stop as confirmation.
The source-shaped decoration
A reference leads to a homepage, a broken page, or a paper you cannot locate. Search its exact title, author, date, or DOI and check the publisher's record. It may be a typo or moved page; it may be nonexistent. An access failure alone does not tell you which.
The related-but-not-supporting source
The paper exists and discusses the topic, but does not establish this claim. It might concern another population, report a different outcome, or explicitly caution against the conclusion. Relevance is not the same as support.
The borrowed assertion
A paper's introduction repeats an older claim with a footnote. That statement is not necessarily a result of the paper's own research. Follow its reference; scholarly formatting does not turn a quotation into a new experiment.
The certainty upgrade
“May help” becomes “works.” A laboratory result becomes a claim about everyday use. A press release drops the limitations and the next article cites the release. Compare the wording at each handoff, not just the number of citations.
Corroboration audit
04 / Independent evidenceFive websites. Possibly one witness.
Several outlets may republish the same wire story, quote the same spokesperson, or summarize the same study. That can make information easier to find. It does not create five independent observations.
Look for a shared byline or syndication credit, identical quotations, the same unusual phrasing, or repeated links to one press release. These are clues to common sourcing—not proof of copying or misconduct. Reputable reports can accurately share the same source.
Ask “independent of what?” A second analyst checking the same dataset can independently assess the calculation. They have not collected a second dataset. Two reporters interviewing the same witness have not found a second witness. A new study with new observations can add another kind of evidence, though its methods still need scrutiny.
Useful secondary reporting adds context: checking a document, consulting relevant independent expertise, explaining limitations, or comparing the finding with other work. Do not discard good reporting merely because it is not the original source. Do not mistake its repeated citation for replication, either.
Keep a small evidence receipt.
- Claim: the exact sentence you checked.
- Route: post → article → report, including any dead end or loop.
- Evidence: the relevant passage, page, table, recording timestamp, or data definition.
- Contribution: new observations, analysis of shared data, quotation, or repetition.
- Conclusion: supported, supported only in a narrower form, contradicted, or not established by the sources checked.
Save titles, dates, and links. Short notes are enough; this is a receipt, not a compulsory dissertation.
Investigation budget / Finite
05 / Limits and conclusionsYou may close the seventeenth tab.
Match the effort to the stakes. For a casual repost, reliable coverage that shows its work may be enough. For a consequential health, financial, or public accusation claim, seek stronger evidence and relevant expertise. A quick source trace is not a substitute for specialist advice.
If the original is missing or paywalled, try the publisher's abstract, an author or institutional repository, a library, or an archived version. An archive can establish what a page said at a particular time; it does not certify that the page was correct. Do not bypass access controls or fill in unseen methods from a headline.
Stop when you have enough evidence for the decision at hand—or can identify what remains unavailable. You may withhold sharing without resolving the entire subject. “I could not verify this claim from the accessible sources” is a useful result. It is not the same as “the claim is false.”
Final directive: Count evidence, not echoes. A citation should make a claim easier to inspect, not harder to question. Apply this procedure to Suscorp, too. Our references do not receive executive immunity.
06 / Supporting sourcesInspect this guide's sources.
This guide draws on Mike Caulfield's SIFT (The Four Moves), June 19, 2019: stop, investigate the source, find better coverage, and trace claims to their original context. His discussion also explains why tracing every claim exhaustively is not always necessary.
Our sequence, fictional notification example, failure-mode labels, and corporate commentary are Suscorp's teaching material—not quotations from Caulfield or findings from a real study. The method reference supports the tracing approach; it is not evidence for the invented productivity claim. Original SIFT material is CC BY 4.0; our wording and organization differ. No affiliation or endorsement is implied.
Published September 13, 2026. Educational public preview; independent editorial review remains outstanding.
07 / Document recordRevision is an expected outcome.
· Revision 01: introduced the standard manual format and monochrome interface. Existing lesson and fictional example retained.
Revision numbering begins with this format edition; earlier unnumbered content was published September 13, 2026. Independent editorial review remains outstanding. Department names are fictional editorial identities, not claims of staffing or review.
Return to document header ↑ · Method and institutional disclosures