Submissions are evaluated using **log loss** on the binary target `on_time`. Lower is better. 
https://scikit-learn.org/stable/modules/generated/sklearn.metrics.log_loss.html

## Submission File

For each `Id` in the test set, predict the probability that the delivery is on time. Your file must be a CSV with exactly two columns named `Id` and `on_time`, one row per test `Id`, and a single header row. The order of rows does not matter, but `Id` values must match the test set exactly and appear only once. Do not include extra columns, missing values, or text labels.

Example:

```
Id,on_time
100001,0
100002,1
100003,1
```

Use the provided `sample_submission.csv` as a template.