**The most counterintuitive finding: Iris-setosa is the smallest flower on almost every measure, yet it has the widest sepals.** The data comes from the gold table `Iris_by_Species` (50 specimens per species). - **Sepal width runs opposite to size.** Setosa's average sepal width is about 3.42 cm, against 2.77 for versicolor and 2.97 for virginica. Setosa's average sepal length is the smallest (5.01 cm, versus 5.94 and 6.59), and its average petal length is about 1.46 cm, roughly a quarter to a third of the others (4.26 and 5.55). So "bigger flower means bigger in every dimension" fails for sepal width. - **Setosa is perfectly separable on petals.** Its petal length tops out at 1.9 cm, while versicolor's minimum is 3.0 cm. Its maximum petal width is 0.6 cm, while versicolor's minimum is 1.0 cm. A single petal threshold therefore isolates setosa with no overlap in this data. - **Sepals separate the species poorly.** Versicolor and virginica both start at a sepal length of 4.9 cm, and their sepal widths overlap heavily (2.0–3.4 versus 2.2–3.8). Setosa's sepal length (4.3–5.8) also overlaps versicolor's (4.9–7.0). Sepal measurements alone would mislead a classifier. - **Versicolor and virginica overlap even on petals.** Versicolor's petal length reaches 5.1 cm and virginica's starts at 4.5 cm. Petal width overlaps too (versicolor up to 1.8 cm, virginica from 1.4 cm). A simple petal rule cleanly finds setosa but can only approximately split the other two, so that pair needs a multivariate model. Sepal width is the feature you would expect to help least as a proxy for overall size, yet it is the one where setosa stands out most in the opposite direction. The overlap ranges here are species-level min/max from the gold table. I did not compute row-level misclassification rates. {"action":"final","content":"**The most counterintuitive finding: Iris-setosa is the smallest flower on almost every measure, yet it has the widest sepals.**\n\nThe data comes from the gold table `Iris_by_Species` (50 specimens per species).\n\n- **Sepal width runs opposite to size.** Setosa's average sepal width is about 3.42 cm, against 2.77 for versicolor and 2.97 for virginica. Setosa's average sepal length is the smallest (5.01 cm, versus 5.94 and 6.59), and its average petal length is about 1.46 cm, roughly a quarter to a third of the others (4.26 and 5.55). So \"bigger flower means bigger in every dimension\" fails for sepal width.\n- **Setosa is perfectly separable on petals.** Its petal length tops out at 1.9 cm, while versicolor's minimum is 3.0 cm. Its maximum petal width is 0.6 cm, while versicolor's minimum is 1.0 cm. A single petal threshold therefore isolates setosa with no overlap in this data.\n- **Sepals separate the species poorly.** Versicolor and virginica both start at a sepal length of 4.9 cm, and their sepal widths overlap heavily (2.0–3.4 versus 2.2–3.8). Setosa's sepal length (4.3–5.8) also overlaps versicolor's (4.9–7.0). Sepal measurements alone would mislead a classifier.\n- **Versicolor and virginica overlap even on petals.** Versicolor's petal length reaches 5.1 cm and virginica's starts at 4.5 cm. Petal width overlaps too (versicolor up to 1.8 cm, virginica from 1.4 cm). A simple petal rule cleanly finds setosa but can only approximately split the other two, so that pair needs a multivariate model.\n\nSepal width is the feature you would expect to help least as a proxy for overall size, yet it is the one where setosa stands out most in the opposite direction. The overlap ranges here are species-level min/max from the gold table. I did not compute row-level misclassification rates.","charts":[{"kind":"bar","x":"Species","series":["SepalWidthCm__avg","PetalLengthCm__avg"],"title":"Average sepal width vs. petal length by species"}]}