The Founding Sample
Every field carries a policy nobody remembers arguing for. Ask why the quota is set where it is, why the safety interval is four hours and not three, why the standard sample size is what it is, and you'll get an answer delivered with total confidence and zero memory of its origin. Push further and eventually you find a paper. Usually a small one.
Call it the seven-leopard sample: a founding study modest enough that you could list every subject in the margin, sitting under a policy broad enough to govern people who will never open it, may not know it exists, and have no ordinary occasion to go looking.
The naive model of bad evidence is that it's fragile. Something built on a small, dated, or narrow study should be easy to knock over, someone runs a bigger study, finds a gap, and the policy adjusts. That happens, sometimes. But it's not the default path, and the reason has nothing to do with the quality of the original work. It's about what citation does to a number as it moves.
A citation exists to transfer a conclusion, not a methodology. When a later paper cites the founding one, it says "this is established" and hands the reader the finding stripped of the conditions that produced it: seven leopards, one season, one valley. The next paper cites that one for the same conclusion, sometimes alongside two others that also traced back to the original, and now the finding has three citations behind it instead of one, which reads as convergent evidence even when it's the same seven leopards being counted three times under three names. By the time a policy document cites the field's "established consensus," the sample size is four hops deep in someone else's bibliography, and nobody drafting the policy has any professional reason to go get it.
This isn't a story about bad actors. Nobody chooses to launder a caveat out of a citation chain. It happens because carrying the caveat forward costs something, a sentence, a hedge, an admission that the number is smaller than the confidence around it, and dropping it costs nothing, because the next reader has no way to know it's missing. Every individual act of citing is locally reasonable. The finding was published; it wasn't retracted; other people cite it too. Stack enough locally reasonable acts and you get a structure nobody actually built on purpose, resting entirely on a number nobody currently holding the policy has seen.
That's the part that breaks the naive fragility model. A policy this thin should get weaker as it moves further from its source. It gets stronger. Every additional citation reads as additional evidence, when what it actually is is additional distance. The original study's honesty about its own limits, the sentence where the authors admit small sample, single site, further study needed, gets cut for length a citation or two in and never comes back, because nobody downstream is incentivized to reconstruct a limitation that isn't sitting in the abstract they're reading. The caveat had exactly one place to survive, and it didn't make the cut.
What actually protects a policy like this isn't correctness, it's altitude. The founding study is invisible not because someone hid it but because nothing in the normal operation of citing, drafting, or complying asks anyone to go find it. You'd have to want to doubt the policy specifically enough to go spelunking through a citation chain for a document that, by the time you found it, would look almost quaint next to the institutional weight now sitting on top of it. Auditing a policy's founding evidence is optional homework nobody assigns and no workflow rewards, so it doesn't happen, and the policy calcifies exactly where it's most vulnerable.
The fix isn't "cite better," though that would help at the margin. It's noticing that the number of citations behind a claim tells you almost nothing about its independence, and that the confidence a policy radiates is often just the compounding of its own distance from the place where you could still check it. If a rule feels unquestionable, that's not evidence it was ever tested at scale. It might just mean the paper trail got long enough that nobody's walked it end to end.
Somewhere under most of the rules you don't argue with, there's a small room with seven leopards in it, and the door hasn't been opened since the first citation walked out of it.