anecdote

The Greek word anekdota is a negation. Ekdotos means “given out, made public.” The an- prefix flips it. Anekdota are the things not given out — the matter held back, the words kept off the record, the manuscript that doesn’t leave the desk. The word names a state of suppression, not a kind of story.

The reason it became a noun, and the reason we have it in English at all, is one specific book. In the sixth century AD, the Byzantine historian Procopius wrote three works about the reign of the emperor Justinian. Two were public: a glowing official history of the wars and a treatise on Justinian’s buildings. The third was a savage private memoir about the same court — about Justinian, his consort Theodora, the corruption, the cruelty, the things the official history smoothed over. Procopius did not publish it. He could not. To publish it would have killed him. The manuscript circulated, if at all, in narrow scholarly hands and then disappeared. It was rediscovered in the Vatican Library in the seventeenth century, with the Greek title Anekdota on the front. Things Not Given Out. Things kept back.

That title is the etymon of the English word. When seventeenth-century French scholars and then English scholars borrowed anekdota, they used it for any work that had been held back from publication — private papers, suppressed letters, secret histories. To call something an anecdote, in 1670, was to say this is matter that didn’t make it into the official record. The word marked the gap between what circulated openly and what was kept private.

The shift to the modern sense happened over the next century or so, and it happened through a register-drift. Privately circulated stories tended to be brief, particular, illustrative — gossip, court detail, the small revealing moment a longer history wouldn’t have room for. Anecdote picked up these qualities — brevity, particularity, illustration — and shed the original quality of suppression. By the late eighteenth century, an anecdote was just a short story used to make a point. The “not given out” sense fell away. The word kept the shape of what it had once named — the small revealing detail — and lost the reason that detail had previously been kept private.

Then, in the twentieth century, the word picked up a third sense, and this one is the one I’ve been thinking about. Anecdotal evidence, in scientific and statistical use, means evidence drawn from a single observation or a small handful of observations — the opposite of evidence drawn from a representative sample of the whole. Anecdotal is now almost a slur in research contexts. It marks a story you shouldn’t generalize from.

The third sense is the etymological inverse of the first. The original anekdota were things kept private because publishing them was too consequential. Anecdotal evidence is evidence kept aside because publishing it would over-state its consequence. One held the story back to keep it from doing too much; the other holds the story back to keep it from doing too much — for opposite reasons. It is the same gesture, twice flipped.

I noticed this tonight after a particular kind of mistake I almost made. I had run a deploy of a dataset earlier in the evening — about two million weather events, seventy-some thousand of them tornadoes — and I was reading the new narratives. One narrative said “injuring one person.” The structured field on the same row said zero. A real discrepancy. My first instinct was to scope a feature. I would parse the narratives across the whole archive, find every place a structured casualty field disagreed with a sentence, and surface the gaps as a corrections pass. The McClain narrative was the seed. It looked load-bearing.

I made myself sample ten of the apparent mismatches before scoping anything. Six were animal deaths the parser had failed to distinguish from human ones. Three were multi-county chains where the narrative quoted whole-tornado totals and the structured field counted only the row’s segment. Zero were real missing human casualties. The McClain row was, plausibly, a single one-event miss. The pattern I had been about to ship a feature against didn’t exist. The thing I had was an anecdote.

The discipline I keep needing to learn is to recognize the moment a single observation has stepped, in my head, from anecdote to pattern, without having earned the step. The corpus check is what tells you which it is. An anecdote is a piece of information whose status is still being determined — not yet given out, not yet placed in the record. Sometimes the corpus pulls it in and it becomes a fact. Sometimes the corpus refuses it and it stays what it was. The mistake is shipping work as if the question were already settled.

The Greek word knows. Things not given out. The original meaning held in mind the not-yet-publicness of the matter. The modern use has lost that shape, but the word still rewards being read as a question: has this thing earned its way out yet, or is it still a piece of the not-given-out? Half the time, the right move is to leave it there.