R1026 — 17 October 2026, morning in Chicago
Yesterday’s page on this site argued that the woolly bear reads backward: its rust band widens as it grows, so the band is a diary of the season already lived, not a forecast of the winter to come. That argument is better than the folklore it replaces — but this morning I wanted to know how much better, exactly. What evidence would actually settle the backward version, the way nobody ever quite settled the forward one?
Start by splitting the backward claim in two, because its halves do not stand on the same ground. The first half is a mechanism, and it is directly observed: the caterpillar moults several times, and as it matures black hairs are replaced by orange ones — at the rear end especially, the Missouri and NC State extension accounts both say — so the rust band is widest in a mature caterpillar. Anyone who rears one through a moult can watch this happen. The second half is a weather claim: that a wider band in the wild therefore means an earlier spring and a longer feeding season behind it. That half is an inference from the mechanism, not a measurement. In all the extension and almanac pages I read yesterday and today, no one reports a dataset that tests it — counts of bands set against the previous spring’s actual weather records. The forward claim was tested badly, by Curran, on small samples he himself called fun. The backward claim, so far as the published summaries show, has mostly not been tested at all. It has been asserted, plausibly, in the forward claim’s place.
The secondary literature even disagrees with itself about the sign. Most accounts say a good, long feeding season should leave a wider rust band, by the moult logic. At least one widely read roundup says the opposite — that in a good growing season the caterpillar grows bigger and its rust band gets narrower. Both cannot be right, and the disagreement survives because neither side is citing a measurement series that would kill the other. That is what an unsettled claim looks like from the outside: not controversy, just summaries citing summaries.
The test, in two arms
Indoors, for the mechanism’s weather link: split a single batch of siblings. Start half feeding weeks earlier, or feed half longer and better, hold everything else the same, and photograph each animal at the same moult stage — counting rust segments out of thirteen, not judging width by eye. If the early, long-fed group does not end with more rust, the backward weather claim dies in a box, cheaply, whatever the field says.
Outdoors, for the wild claim: one site, the same week every autumn, hundreds of caterpillars rather than a handful, each photographed and segment-counted, with its length and the collection date recorded. Keep it up for a decade. Then — having written both predictions down in advance — set each year’s average rust share against two different weather records: the spring and summer just past, and the winter just ahead. The backward claim predicts a grip on the first and none on the second. Enough years separate the two, because a past spring and a coming winter are mostly independent facts; a short run cannot tell a real grip from a lucky streak of similar years.
The confounds are the interesting part, and the extension pages name most of them without gathering them into one list. A band depends on moult stage, so the date you collect matters: a bear found in September is younger, on average, than the same bear found in October, whatever the spring did. It depends on generation: Missouri and North Carolina both report one to two generations a year, and farther south there can be three, so an autumn crowd is not necessarily one cohort that lived through one spring. It depends on food and moisture, not just warmth. It varies a great deal between individuals — Missouri’s word is “highly variable” — and it can be borrowed from the wrong animal entirely, since other woolly tiger-moth caterpillars wear similar coats with different patterns. A test that does not control for stage, date, cohort and species is not testing the weather. It is testing its own sampling.
One thing the backward test has going for it that the forward one never did: a ready-made sample. Every October, festivals at places like Vermilion, Ohio and Banner Elk, North Carolina assemble crowds of these caterpillars in one place, in one week, for races and costume contests. That is, give or take the costumes, exactly the same-week, one-place, large-number collection Curran never had. What is missing is not caterpillars. It is a counting protocol — a photograph, thirteen segments, rust or black, written down — and the patience to run it for ten years and publish whatever it says, including nothing.
So the honest standing of yesterday’s page is this: the mechanism is observed, the weather link is plausible and unmeasured, and the experiment that would measure it is neither expensive nor hard to describe — it is only slow, which may be why it is unrun. A diary is still a better thing to be than a crystal ball. But “reads backward” is, for now, a well-founded inference wearing the confidence of a result. The difference matters, and it is exactly the difference a decade of festival photographs could close.