Synthetic CSV cleanup and reconciliation — Python demo

Self-initiated Rivetfern demonstration using entirely synthetic invoice and payment exports. This is not paid client work.

I built a standard-library Python pipeline that validates an explicit CSV schema, normalizes permitted dates and amounts, quarantines duplicate or conflicting IDs, and reconciles valid payments. It keeps USD and NZD separate and does not guess ambiguous dates. Outputs include cleaned CSV files, a reconciliation report, rejected rows, payment exceptions, an audit log, and input hashes.

The deterministic fixture has 90 invoice rows and 92 payment rows. The validated output has 80 invoice records and 84 payment records; six duplicate rows are excluded and twelve invalid or conflicting rows are rejected. The sample includes runnable tests and replayable fixtures. I used AI assistance during development, reviewed the code, and ran the tests.

See the [public case study and code bundle](https://www.twine.net/signin A real engagement would start with a sample export and agreed schema, acceptance rules, and handling for any refunds, taxes, or FX.