Contact cleanup with a review trail.

The problem

Contact cleanup needs agreed rules and visible decisions. Similar email or phone fields can suggest a review, but cannot establish that two records represent the same person.

What I did

I defined the conservative matching and review rules, directed the AI-assisted implementation and documentation, and reviewed the recorded native Excel checks. My practical owner walkthrough is deferred.

How it works

• Extract the complete ZIP and follow START-HERE to create a new disposable workspace. The helper checks the frozen inputs, sets the local DataRoot and verifies a fresh five-table refresh.
• Remove only complete exact normalized duplicates automatically. Keep ambiguous records in review and require explicit decisions; invalid or unfinished decisions block cleaned export.
• Save and close the workbook before running the export helper. Inspect all four audit/export CSVs and reconciliation; use the documented text import and inverse codec for the single apostrophe added to each data cell.

What you can verify

• The 5,000 fictional inputs reconcile to 4,870 retained records, 100 exact duplicates, 20 explicitly approved merges and 10 rejections. All 150 review records remain auditable.
• Recorded Windows Excel 16.0 checks verify relocated setup, five result tables, four byte-equal CSV exports and refusal of changed inputs or an existing destination.
• Public v0.1.3 includes the workbook, synthetic fixtures, setup/export helpers, presentation PDFs and manifest. The full ZIP was anonymously downloaded and matched its published SHA-256.