A local fintech warehouse lab: replay, reconciliation, and a SQL bug hunt

[antony502189-max]

I maintain a free English fintech data engineering learning repository. I am sharing the runnable warehouse and SQL practice here for people looking for projects they can try without a cloud account.

The core example uses Python 3.12+ and SQLite with synthetic payment events. On a fresh run it ingests six events; replay inserts zero new events and identifies six duplicates. An independent reconciliation fixture contains a discrepancy of 100 USD minor units and an unexpected external payment.

git clone https://github.com/antony502189-max/fintech-data-engineering-roadmap.git
cd fintech-data-engineering-roadmap
python -m fintech_lab --out work/first-demo
python challenges/run.py init C07
python challenges/run.py grade C07

C07 starts with runnable but incorrect SQL. Edit work/challenges/C07.py to fix the report, then grade again. Its three cases check independent captures, fees, and payment attempts: equal-value events, join fan-out, currency boundaries, and captures that have no fees or attempts. SUM(DISTINCT amount) is deliberately insufficient because distinct payment identities may have equal amounts. References are available after an independent attempt.

The other scenario labs cover multi-batch rollback, late SCD2 joins, grouped payout reconciliation, schema migration, bitemporal reporting, actual SQLite backup restoration with atomic checkpoints, and features with delayed labels. The repository also includes 100 lessons, 40 foundation exercises, article and video links, and capstone specifications.

I would appreciate feedback from anyone who tries a lab: which input contract is unclear, and what realistic edge case would you add? A reproducible failure with the lab ID, command, and expected result is especially useful.

Disclosure: the first edition and this introduction were prepared with AI assistance using Codex (GPT-6). The provenance documents the intentionally rescheduled commit timestamps. These examples teach principles locally; they are not a production bank, Kafka deployment, cloud IAM implementation, or compliance certification.