**Short answer: the Iris data is clean. The only real issues are one duplicated row and a handful of statistical outliers, and none of the values are impossible.** **Checks that passed** (step-0 and step-3 profiling queries on `raw.iris`) - **Completeness:** all 150 rows have non-null values in all four measurements and in `species`. - **Plausible ranges:** all four measurements are positive and in a believable range for iris flowers. | Measurement | Min | Max | |---|---|---| | Sepal length | 4.3 | 7.9 | | Sepal width | 2.0 | 4.4 | | Petal length | 1.0 | 6.9 | | Petal width | 0.1 | 2.5 | There are no zeros, negatives or sentinel values such as -1 or 999. - **Consistent formats:** `species` has exactly 3 distinct values, and the format-issue check found 0 problems, so there are no case, spelling or whitespace variants. All four measurement columns are numeric (DOUBLE), so nothing is stored as text. - **Mixed units:** I found no sign of them. The ranges are in one consistent scale, which fits centimetres. The dataset has no unit column, so the unit itself is an assumption. - **Lineage:** there is a single `_batch_id`, `_source_file` and `_source_system`, so there is no mixing of sources or loads. **Issues found** 1. **One duplicate row.** The duplicate check (step-1 and step-3) found 1 group of identical records, which means 1 extra row. It is small, but it slightly over-weights that specimen in any model or statistic. Whether to drop it depends on whether identical measurements could be genuinely separate flowers, which is plausible for measurements recorded to one decimal place. 2. **Statistical outliers, all within plausible ranges.** My outlier query (step-1) flagged 1 sepal-length value, 5 sepal-width values, 5 petal-length values and 2 petal-width values. The evidence does not show the exact cutoff used. These are most likely legitimate natural variation rather than errors: the extremes (for example sepal width of 2.0 or 4.4, and petal length of 1.0 or 6.9) all fall inside believable botanical ranges. The species differ strongly in size, so a value that looks extreme across the whole dataset may be normal within its own species. I would keep these rows and not treat them as errors. **Bottom line:** no format or unit cleaning is needed. Consider deduplicating the one repeated row, and review the flagged outliers within each species rather than removing them automatically.