Original fictional data for Claude AI Made Practical by Serhii Komar.
No private business records or actual client data are included.

The chapter explains a manual route. Readers do not need Python.
verify_reference.py is a production-side reference check using Python's standard library only.
It reads both weeks' CSV and notes files, expected.json, and resolved_variant_jobs.csv from its own directory.
It independently groups the CSV rows and compares the result against manually calculated expected values.
It also checks that an intentionally incorrect all-status sum is rejected as completed revenue.
Execution output records this local reference calculation, not a Claude response.

Week 1: completed 4 / USD1080; scheduled 1 / USD420; cancelled 1 / USD120.
Week 1 completed client values: Cedar420, Pine360, Birch300.
Week 2: completed 4 / USD1160; scheduled 2 / USD600; cancelled 1 / USD130.
Week 2 completed client values: Cedar410, Pine410, Birch340.
Completed value difference: USD80. Percent difference, using week 1 as baseline: 7.407407...%.

Use only the files for the requested week when drafting that week's report.
The date field selects records in the requested period; the status is recorded at that week's noon cutoff. No individual completion timestamps are supplied.
For a comparison, supply both weeks and state which is the baseline.
No configuration file or external package is required.
