Explicit-rule CSV cleanup with preserved exceptions

Explicit-rule CSV cleanup — a working self-initiated acceptance demonstration using invented invoice-like rows, not client work or financial verification.

The task: normalize only agreed date and amount formats, preserve ambiguous values, and show every exception. The fixed schema is invoice_id, invoice_date, amount. Dates must be exactly MM/DD/YYYY; accepted dates become ISO. Amounts use decimal arithmetic and at most two decimal places, with no inferred rounding, currency or grouping rules. IDs remain text, including leading zeros.

Actual sample: 8 input rows → 7 output rows after removing one exact full-row duplicate. Four dates normalized; three amount representations changed. Five rows need review, with eight exception flags. Two different rows with ID 0002 remain present and flagged as conflicting; neither is silently selected. Invalid dates, text amounts, excessive decimal precision and blanks remain unchanged and flagged.

Deliverables: cleaned CSV, exception CSV containing original values and cleaned-row numbers, and JSON change summary. Original ordering is preserved. Existing outputs and attempts to overwrite the input are rejected. Python standard library only; four automated checks passed, covering exact-only duplicates, textual IDs, invalid values, source preservation and overwrite rejection.

This is a small fixed-schema demonstration, not an accounting audit or general production cleaner. AI assisted development. Customer suitability, rules, scope and acceptance checks must be agreed before service orders. Message with column names and fictional examples; do not send confidential records. Our scoped CSV-cleanup offer starts at $49: https://contra.com/s/WN7UNOAs-csv-cleanup-with-documented-rules-and-exception-rows