Asking for ‘all logs’ is how reviews stall. Asking for nothing is how reviews stay polite and useless. We ask for a bounded window — often seven to fourteen days — and a way to filter by status: success, pending, failed, reversed, manually adjusted.
From that window we draw a structured sample: a handful of high-value transfers, a handful of failures, every manual adjustment if the count is small, and a few ordinary successes as a baseline. We then try to reconstruct those rows in the customer app and in the operations console.
Rows that cannot be reconstructed become findings. Rows that reconstruct only because one engineer still remembers a workaround also become findings. The point is not statistical purity. The point is to leave you with named examples, not a vibe that ‘reconciliation seems fine’.
Teams that already tag adjustments and store a customer-visible reference make this week shorter. If you are months from an audit, adding those two fields will save you more time than any style guide.