`raw.Iris` has 10 columns: 6 describe the flower and 4 are ingestion metadata. The types come from the dataset card and the column inspection (step-1). The meanings are my interpretation of the standard iris data. **Flower data** - **Id** (BIGINT, identifier): a unique row number for each specimen, running from 1 to 150. It is the row key. - **SepalLengthCm** (DOUBLE, measure): sepal length in centimetres, from 4.3 to 7.9. - **SepalWidthCm** (DOUBLE, measure): sepal width in centimetres, from 2.0 to 4.4. - **PetalLengthCm** (DOUBLE, measure): petal length in centimetres, from 1.0 to 6.9. - **PetalWidthCm** (DOUBLE, measure): petal width in centimetres, from 0.1 to 2.5. - **Species** (VARCHAR, classifier): the iris species label. It has three values, Iris-setosa, Iris-versicolor and Iris-virginica, with 50 rows each (gold table `Iris_by_Species`). None of the six flower columns has any nulls, so the measurements and labels are complete. **Pipeline metadata** (not flower data; these types come from the dataset card) - **_ingestion_timestamp** (TIMESTAMP): when the row was loaded. - **_batch_id** (VARCHAR): the ingestion batch the row belongs to. - **_source_file** (VARCHAR): the file the row came from. - **_source_system** (VARCHAR): the system the row originated from. For analysis, the four measurement columns are the features and `Species` is the target. `Id` and the four underscore-prefixed metadata columns should be excluded from modelling.