Messy CSV Reconciliation

Build a pipeline that reconciles transaction CSVs and emits matched, unmatched, and summary outputs.

Requirements

1. 01Ingest multiple transaction CSV files while handling UTF-8 BOMs, column-name variance, and malformed rows explicitly.
2. 02Normalize identifiers, dates, and currency amounts into documented canonical fields using decimal-safe arithmetic.
3. 03Match each source record at most once and classify duplicate, matched, and unmatched records deterministically.
4. 04Export matched and unmatched rows plus a summary whose counts and totals reconcile to the inputs.
5. 05Preserve source file and row provenance for every emitted record and validation error.

Constraints

• • Submit one compact engineering spec, not implementation code.
• • The generated artifact must use the declared Python pipeline adapter.
• • Scoring rewards observable behavior and prompt efficiency, not runtime.