**The most counterintuitive finding: Iris-setosa is the smallest flower on almost every measure, yet it has the widest sepals.** This comes from the per-species aggregates I pulled from `Iris_by_Species` (50 specimens per species): | Measure (average) | Setosa | Versicolor | Virginica | |---|---|---|---| | Sepal length (cm) | 5.01 | 5.94 | 6.59 | | Petal length (cm) | 1.46 | 4.26 | 5.55 | | Petal width (cm) | 0.24 | 1.33 | 2.03 | | Sepal width (cm) | **3.42** | 2.77 | 2.97 | - **Size ordering:** On sepal length, petal length and petal width, the species line up as setosa < versicolor < virginica. Virginica's average petal length is about 3.8 times setosa's. - **The reversal:** Sepal width runs the other way. Setosa averages 3.42 cm, about 23% wider than versicolor's 2.77 cm. Setosa's widest sepal (4.4 cm) is also beyond the maximum of both other species (3.4 and 3.8). - **Practical consequence:** Sepal width is not a simple "bigger flower" proxy. It can even point the wrong way, so assuming every measurement scales together would be misleading. A related finding is that the separability is very uneven: - **Setosa is cleanly isolated on petals.** Its largest petal length is 1.9 cm, while the smallest for versicolor is 3.0 cm. Its largest petal width is 0.6 cm, against 1.0 cm for versicolor. A single petal threshold therefore identifies setosa perfectly. - **Versicolor and virginica overlap on every measure.** Petal length ranges are 3.0–5.1 and 4.5–6.9. Petal width ranges are 1.0–1.8 and 1.4–2.5. Sepal length ranges (4.9–7.0 and 4.9–7.9) overlap almost completely. A one-feature rule will therefore misclassify some of those two species, and a multivariate model is needed to separate them. I did not measure how often a threshold would misclassify the overlapping species. The overlap is judged from the min/max ranges only.