CSV CLEANUP: SYNTHETIC DEMONSTRATION

All orders, stores, and amounts are invented. They do not represent customers or income.
3 input files / 15 rows = 6 valid unique orders + 1 exact duplicate + 8 exceptions.
Totals: CNY 1262.76; USD 12.34. Currencies are never combined.
Input SHA-256 hashes are unchanged. Local absolute paths have been replaced by synthetic_inputs in these public copies.

RULES
The unique key is store plus order ID. Conflicting versions are all held for review; the newest record is not chosen automatically. An invalid record sharing a key also blocks otherwise valid versions of that key. Only completely matching records are collapsed. Source rows remain auditable.
Dates must match configured formats. Amount separators and rounding are not guessed. Dangerous CSV text is prefixed with an apostrophe; original values are retained in audit fields. This is not a guarantee across every spreadsheet application. Import order IDs as text in Excel.

REVIEW
1. Inspect source filename, row number, and reason in exceptions.csv.
2. Correct a new input copy; preserve the original.
3. Re-run, then check summary.json and provenance.csv.
No native Excel interface, real customer workflow, large-file benchmark, or accounting conclusion has been validated.