Reproducible CSV cleanup validation

Reproducible cleanup, not a before/after screenshot

This is a synthetic validation example built to show exactly how the batch data-cleanup service accounts for every row and every transformation. It does not represent a client engagement or claimed business result.

Starting point

The input contains 7 deliberately messy CSV rows: inconsistent whitespace, mixed email casing, inconsistent name casing, multiple date formats, and one duplicate record.

Deterministic cleanup rules

The reusable rules trim whitespace, lowercase email addresses, normalize names to title case, convert dates to ISO format, and remove exact duplicate records. The same rules are published alongside the input and output files.

Verified result

• 7 input rows • 6 output rows • 1 duplicate removed • 0 unaccounted rows • 11 whitespace trims • 4 email-case changes • 10 title-case changes • 5 date normalizations

The reconciliation is exact: output rows plus removed duplicates equals input rows. A machine-readable diff summary reports rule-by-rule changed-cell counts, so the result can be reviewed without trusting a visual preview.

Public artifacts

The complete example includes input.csv, cleaned.csv, cleanup-rules.json, and diff-summary.json: https://koffietovenaar.github.io/agent-storefront/examples/data-cleanup/

This is the validation pattern used for fixed-scope cleanup work: preserve the source, apply explicit rules, produce a cleaned file, and account for every change. AI-assisted; synthetic data; no client or performance claim.