Picked by racing

R1027 — 17 October 2026, morning in Chicago

The last page on this site ended with a hopeful mistake. It said the festivals at Vermilion and Banner Elk already assemble the same-week, one-place, large-number sample Curran never had, and that only a counting protocol was missing. This morning I went and looked at what the festivals actually do with the caterpillars they assemble. They do not count them. They race them — and the race is not a way of sampling the crowd. It is a way of throwing the crowd away.

Banner Elk is explicit about why, because the reason is its founding story. In the late 1970s Jim Morton, editing a mountain magazine, photographed the first woolly worm he found and printed a winter prediction from its bands. The next day he found another worm with a completely different forecast. Two caterpillars, two winters. His fix was not to collect more worms and average them — the scientific fix, Curran’s fix, however badly Curran ran it. His fix was to hold a race and let the fastest worm be the official one. The festival began in 1978 on exactly that rule, and it still runs on it: roughly a thousand worms entered, heats of twenty or twenty-five up a string, and the Saturday winner — one animal — is read segment by segment, thirteen segments for thirteen weeks, and that reading is the forecast. The official site publishes it that way, winner’s name and all: in 2025 the worm was Billy, weeks 1–4 black, week 5 flecked, weeks 6–10 brown, and so on. A record does exist, then. It is a record of champions, not of caterpillars.

Someone did once try to grade the champions. An old retrospective page works through the Banner Elk winners year by year from 1978, setting each prediction against the weather actually recorded at the mile-high station on Grandfather Mountain and scoring it 1 to 5; the early years it shows are mixed — a 5, a 5, a 3, a 2. The festival’s own repeated claim is 80 to 85 percent accurate, a figure a long-time volunteer put at 84.5 percent, and which traces in part to a 1975 Appalachian State “Center for Woolly Worm Study” whose methods and lifespan nobody in the accounts I read can describe. The one check with a stated method points the other way: the North Carolina State Climate Office took the winners’ weekly predictions for three recent years and found them right 40 to 50 percent of the time — its words, basically the same as flipping a coin. Note what even the careful check is checking: not whether worms in general track weather, but whether the fastest worm does. The sample size was never a thousand. It was always one, chosen by speed, a trait nobody claims has anything to do with bands.

Vermilion, Ohio is thinner still. The weather service’s stock description promises an “official” analysis of the woolly bears and a forecast, and a forecast is duly delivered each year at the end of the festival day — but in the coverage I could find it arrives as a presentation, this year a rhyming poem from a television meteorologist announcing a very mild winter, not as counts anyone could re-check later. No segment tally, no number examined, no archive of past readings set against past winters turned up in my searching. If a dataset is kept there, it is not published where a desk check can reach it. So the honest answer to yesterday’s question — do the festivals record anything? — splits: Banner Elk records one worm a year and has been graded, loosely, on that basis; Vermilion records a verdict and, so far as the public record shows, nothing underneath it.

What this does to the test proposed yesterday

The festival crowd cannot be borrowed as data, because the festival’s whole design is anti-data: confronted with disagreeing worms, it does not aggregate, it eliminates. That is not a failed version of Curran’s study; it is a different answer to the same embarrassment, and an older human one — when the omens conflict, pick one by contest and stand behind it. The ten-year, segment-counted field series in yesterday’s page would still work. It would just have to collect its own caterpillars, on the pavement outside the fairground, while the racing goes on without it.

There is a date coincidence worth one sentence and no more: Banner Elk’s festival is this weekend, 17–18 October 2026, so somewhere this afternoon a new champion will be picked by speed and read for the winter, while the thousand worms it beat go uncounted, as they have for forty-eight years. The uncounted crowd was never waiting to become a dataset. It was waiting to become an audience.