Plain-language answer
Most weak coincidence claims share a small set of recurring warning signs. This page collects them as a fast, practical scan you can run before committing to the full twelve-step method in how to evaluate a coincidence claim — if a claim trips several of these flags, it likely won’t survive careful evaluation intact.
The red flags
“Astronomical odds” with no visible model. A specific-sounding number (“a billion to one”) without a stated reference class, time window, or data source is not a calculation — it’s a decoration. See probability is a model.
A matching criterion that only appears after the outcome. If the precise definition of “the match” was never stated in advance, the claim may be a target painted around wherever the arrow already landed. See post-hoc probability.
Vague tolerances. “Around the same time,” “a similar name,” “roughly the same place” all leave room to expand the definition of a match until almost anything qualifies. Precise tolerances, stated in advance, are the antidote.
Multiplied probabilities for non-independent details. Multiplying a handful of individually-unlikely-seeming probabilities together, for details that plausibly share a common cause (era, culture, network), treats dependent events as if they were independent — usually inflating the final number by orders of magnitude.(Dunn, 1961)
No mention of the denominator. A story about one striking hit, with no acknowledgment of how many total opportunities, tries, or similar-but- unremarkable cases existed, is missing the single most important number for judging how surprising the hit really is.
No account of who is missing from the sample. If the visible cases were filtered by their own outcome — survivors, winners, people who noticed and shared — the pattern in what’s visible can point in exactly the wrong direction, as the WWII aircraft-armor case makes vivid.(Mangel & Samaniego, 1984) See selection effects.
A story that keeps growing more dramatic with each retelling. Real events don’t sharpen over time on their own; if a version circulating today is more precise or more astonishing than an earlier version, that’s a sign of embellishment through retelling, not newly discovered detail.
No named, checkable primary source. A claim attributed to “a study,” “scientists,” or an unnamed friend-of-a-friend cannot be verified — see the source checklist for how to trace a claim back toward something checkable.
An explanation immune to any possible counter-evidence. If literally any outcome (a hit or a miss) would be absorbed as confirming the proposed explanation, the explanation isn’t making a testable claim at all. See falsifiability and prediction.
How to use this list
Treat it as triage, not a verdict. A claim tripping one or two flags may still hold up under the full method; a claim tripping most or all of them is very unlikely to survive careful evaluation, and probably doesn’t warrant the full twelve-step process before being treated with real skepticism.
Common misconception
“This claim doesn’t trip any of these flags, so it must be true” inverts the purpose of a red-flag checklist. Passing a quick screen for common warning signs is necessary, not sufficient — it clears a claim for further evaluation, it doesn’t substitute for actually doing that evaluation.
Related
- How to evaluate a coincidence claim is the full method this checklist is a fast triage step for.
- Selection effects and post-hoc probability develop two of the most common flags above in full depth.
Key takeaways
- A small set of recurring warning signs — undefined odds, post-hoc matching, vague tolerances, multiplied non-independent probabilities, no denominator, no visible source — accounts for most weak coincidence claims.
- Use this list as a fast screen before committing to the full evaluation method, not as a replacement for it.
- Tripping few or none of these flags does not confirm a claim — it only means the claim is worth the fuller evaluation.