Write the schema before changing a cell.
Define required columns, types, allowed blanks and normalization rules first. A date that cannot be parsed is an exception; it should not quietly become today, zero or an empty value.
Spreadsheet cleanup controls
A clean spreadsheet needs an exception ledger, not just deduplication.
A file can look tidy and still be impossible to audit. The useful result is not merely fewer rows or uniform formatting; it is a clean table plus evidence for every rejected, merged and corrected record.
Control equation: input rows = accepted rows + rejected rows + duplicate dispositions. Numeric totals need the same before-and-after reconciliation wherever the fields support it.
Give rejected rows stable reason codes.
Keep invalid records in a separate queue with source-row identity and codes such as invalid_date, missing_key or nonnumeric_amount. This makes correction possible without rerunning guesswork.
Make duplicate decisions reviewable.
Choose an explicit business key. Record whether duplicates were exact, conflicting, merged or retained for review, and identify which source row produced the accepted record.
Reconcile counts and totals.
Compare input, accepted, rejected and duplicate counts. For quantities or amounts, document which totals should remain invariant and explain any intentional difference.
Freeze the correction ledger.
Record rule versions, affected-row counts and output hashes. A second run on the same input and rules should reproduce the same result, while a correction round should leave a visible delta.
Privacy boundary
This control is designed for authorized, sanitized sales, product, inventory or similar nonregulated business records. It does not require names, emails, phones, addresses, payment data, account access, live CRM access or web/contact verification.