Write a Python script that cleans a CSV export you get every week

Write a Python script that cleans the CSV export described below the same way every time, for me to run {{run_frequency}} on my own computer, so I stop fixing it by hand and get the same result each run.

The first 20 to 30 rows of the export, including the header and some messy rows:
"""
{{sample_rows}}
"""

The cleaning steps I do by hand now, in order:
"""
{{manual_steps}}
"""

What the cleaned file must look like (columns, order, formats, file name): {{output_spec}}
My setup (Python version, whether I can install packages such as pandas, operating system): {{setup}}
Where the input file is and where the output should go: {{file_locations}}

Requirements:
1. One script file, with a short comment above each step that matches one of my manual steps.
2. The input and output paths set at the top of the file, so I can change them without reading the code.
3. It never overwrites the original export.
4. It prints a short summary at the end: rows read, rows written, rows removed and why, and values changed per column.
5. It stops with a clear message, not a stack trace, if an expected column is missing or renamed.
6. After the script, a short section on how to run it and how to compare the first output with a file I cleaned by hand.

Keep it under {{max_lines}} lines, using only the standard library plus packages my setup allows. Prefer plain, readable code over clever one-liners, because the next person to fix it may not be a programmer.

For example, the end summary might print: "Rows read: 1,204. Removed: 12 blank rows, 3 duplicates by Order ID. Dates changed to YYYY-MM-DD in 2 columns." This example shows the form only.

Do not add cleaning steps I did not list, such as dropping columns or fixing spellings, unless the output spec needs one, and then say so in a comment. If a step in my list is ambiguous, for example "remove duplicates" without saying which columns define a duplicate, list your questions first, write the script for the steps that are clear, and leave the unclear step as a commented placeholder.