Opinion.
One of our editors asked us to examine the corroborationAgreement among multiple sources or witnesses. The assumption that if several independent sources confirm something, it is likely true. However, corroboration is unreliable when sources share a common origin, leading to false confidence. truth gap: the assumption that if multiple sources agree, the thing they agree on must be real. The question sounds academic until you realize how much of journalism, law, intelligence analysis, and everyday reasoning depends on this assumption. And how often it fails catastrophically.
Corroboration is a heuristic, not a guarantee. The distance between corroboration and truth has cost lives, destroyed careers, and started at least one war.
Why Corroboration Feels Like Truth
The logic seems airtight. One person says something happened: maybe they are lying, confused, or mistaken. Two people say it: less likely both are wrong. Five people say it: surely they cannot all be making the same error independently.
This reasoning underpins how we evaluate evidence everywhere. Journalism requires at least two independent sources before publishing a claim. Courts treat corroborating testimony as stronger evidence. Intelligence agencies cross-reference signals intelligence with human intelligence. Scientific replication is built on the same principle: if independent labs reproduce the same result, we trust it more.
The intuition is correct on one condition. The sources must be genuinely independent of one another. When they are, corroboration and truth align. When they are not, the corroboration truth equation breaks down entirely. And sources are independent far less often than anyone wants to admit.
The Independence Problem: When Corroboration Truth Claims Collapse
For corroboration to work as advertised, each source must have arrived at its conclusion through a separate chain of evidence. If Source B got its information from Source A, then “two sources confirm” really means one source said something and another repeated it. This is not corroboration. It is an echo.
The problem is that tracing the actual chain of evidence is difficult, boring, and frequently skipped. A journalist calls three experts who all say the same thing. Corroborated? Only if those experts formed their views independently. If all three are citing the same study, or the same rumor, or the same institutional assumption, three voices carry the epistemic weight of one.
Economists Sushil Bikhchandani and David Hirshleifer formalized this problem in their work on information cascadesA situation where people rationally ignore their own private information and follow the behavior of those before them. As each person conforms, they add apparent corroboration while contributing no new evidence, allowing false beliefs to spread widely.: situations where people rationally ignore their own private information and instead follow the actions of those who came before them. Once a cascade begins, each subsequent person adds apparent corroboration while contributing zero new evidence. The cascade can grow to include millions of people, all “confirming” something that rests on the original signal of one or two early actors. As the researchers noted, this can lead entire communities to converge on the wrong answer.
Iraq: The War That Multiple Agencies Corroborated Into Existence
The canonical example of the corroboration truth failure is the intelligence assessment that preceded the 2003 invasion of Iraq. The United States Intelligence Community concluded with “high confidence” that Saddam Hussein possessed weapons of mass destruction. This assessment appeared to be well-corroborated: human intelligence sources reported WMD activity, signals intelligence captured Iraqi officials discussing weapons programs, and satellite imagery showed suspicious facilities.
Multiple collection methods. Multiple agencies. Multiple data points. All converging on the same conclusion.
All wrong.
The Senate Select Committee on Intelligence’s 2004 report found that most of the major judgments in the October 2002 National Intelligence Estimate “either overstated, or were not supported by, the underlying intelligence reporting.” The problem was not a shortage of sources. It was that the sources were not independent in the way analysts assumed.
Saddam Hussein, it turned out, had been lying to his own generals about pursuing WMD, partly to maintain internal leverage. When US signals intelligence intercepted Iraqi officials discussing weapons programs, analysts took this as corroboration of human intelligence reports. But the human and signals intelligence were feeding on the same lie. The SIGINT confirmed the HUMINT, and the HUMINT confirmed the SIGINT, in a closed loop that felt like convergence but was actually a single point of failure viewed from two angles.
At the center of it was an Iraqi defector codenamed “Curveball,” handled by Germany’s BND intelligence service. Curveball claimed Iraq had mobile biological weapons laboratories. His claims appeared in over 112 US government reports and formed the backbone of Colin Powell’s 2003 address to the United Nations. In 2011, Curveball publicly admitted he had fabricated the entire story. He wanted regime changeThe deliberate replacement of a government through military, diplomatic, or economic intervention, typically by external actors. and told the BND what they wanted to hear. The BND, desperate for a source, did not press on inconsistencies. The CIA, despite internal warnings about the source’s reliability, treated Curveball’s claims as confirmed because other intelligence streams appeared to support them.
Those other streams were not independent. They were different lenses focused on the same fabrication.
When 5 Witnesses Remember What Did Not Happen
The corroboration truth problem is not limited to intelligence. It operates wherever human memory is involved, which is to say everywhere.
Elizabeth Loftus, the cognitive psychologist whose research on memory has reshaped how courts evaluate testimony, demonstrated that exposure to post-event information from external sources (other witnesses, interviewers, media coverage) can contaminate a witness’s subsequent memory reports. She called this the misinformation effectThe phenomenon in which post-event information alters a person's recollection of what they originally experienced. New information integrates into memory and can overwrite or render inaccessible the original memory trace.. The critical finding: people do not merely report false information. They come to genuinely believe it. The contaminated memory feels as vivid and certain as the real one.
Now apply this to corroboration. Five witnesses to a car accident all describe the same detail. Corroborated? Possibly. Or possibly one witness mentioned the detail to the others in the parking lot afterward, and now all five sincerely remember seeing it. The testimony is unanimous and useless.
Researchers have combined Loftus’s misinformation paradigm with Solomon Asch’s classic conformity experiments to study exactly this phenomenon. The results show that social influence and post-event information exposure work together to reshape memory, not just reporting. Witnesses do not just say what others said. They remember what others said.
12,000 Cases of Corroboration, Zero Evidence
The Satanic ritual abuse panic of the 1980s and 1990s is what happens when corroboration without independence scales to an entire society. Over 12,000 cases of alleged Satanic ritual abuse were reported across North America, with additional waves in the Netherlands, Sweden, New Zealand, and Australia. Children in different cities, interviewed by different investigators, described similar horrors: animal sacrifice, cannibalism, secret tunnels beneath daycare centers.
The corroboration appeared overwhelming. These were not isolated claims. They were a pattern, emerging independently across jurisdictions.
Except they were not emerging independently. Investigators had been trained by the same seminars, using the same interview techniques, asking the same leading questions. The children were not generating these narratives spontaneously. They were being guided toward them by adults who already believed the narrative and whose questioning methods reliably produced it. When researchers later examined the cases, they found that in 95 percent of adult cases, the “memories” had been recovered during psychotherapy using techniques now widely discredited.
No material evidence of organized Satanic cult abuse was ever found. No bodies, no tunnels, no physical corroboration of any kind. But the testimonial corroboration was vast, and for nearly two decades it was enough to send people to prison.
What the Corroboration Truth Gap Actually Means
None of this means corroboration is useless. It means corroboration is a question, not an answer. When multiple sources agree, the right response is not “this must be true” but rather: why do they agree?
If the sources are genuinely independent (different methods, different data, no shared assumptions, no communication between them) then agreement is powerful evidence. Scientific replication works precisely because the protocol demands this kind of independence. A chemistry result reproduced in a lab in Tokyo and a lab in Munich, using different equipment and different researchers, carries real weight. There, corroboration and truth genuinely correlate.
But if the sources share a common upstream origin (the same informant, the same institutional assumption, the same training, the same rumor, the same media report) then their agreement tells you nothing about truth. It tells you about the reach of the original source. This is the misattributed quotes problem in a different costume: a claim repeated widely enough becomes “well-known,” and well-known starts to feel like well-established.
Cass Sunstein, in his work on conformity cascades, identified the structural version of this failure. In groups, people often suppress their private doubts and follow the apparent consensus. Each person who conforms adds to the weight of “corroboration” that the next person sees. The cascade grows more convincing as it grows larger, while the actual evidentiary base remains whatever the first one or two people believed. Sunstein’s conclusion: the most serious problem with conformity is that people fail to disclose what they actually know, depriving the group of information it needs.
This is why expert disagreement, counterintuitively, is often a healthier sign than expert consensus. Disagreement means the experts are reasoning independently. Consensus might mean they are, or it might mean an information cascade has suppressed dissent. You cannot tell from the outside without examining the underlying evidence chains.
Bridging the Corroboration Truth Divide
The uncomfortable implication is that corroboration, the primary tool we use to distinguish reliable information from unreliable information, does not do what we think it does unless we also verify the one thing we almost never check: whether the sources are actually independent.
This is not a counsel of nihilism. It is a counsel of discipline. When a claim matters (when it might start a war, convict a person, or shape public policy) the question is not “how many sources say this?” but “how many independent evidence chains lead here?” Those are different questions, and treating corroboration as truth without checking independence is how societies end up believing things that are not true while feeling entirely justified in doing so.
Corroboration is a map. A useful one. But the territory it claims to represent, truth, requires you to check whether the mapmakers were all looking at the same landscape or all copying from the same original sketch.
