DeDupe Pro v7

A single-file, offline web app (index.html) that finds and merges duplicate contacts in vCard (.vcf) exports without losing anything — especially the multi-line Notes that an earlier tool stripped.

Live: https://dedupe-pro.vercel.app · Works equally well by opening index.html locally. No data leaves the browser.

What it does

1. Load an Outlook contacts export (.csv) or a .vcf (Synology / iPhone / Mac). Optionally load a second file of either kind to compare the two lists.
2. Find Duplicates — contacts are grouped by shared email, shared phone, and name similarity (exact, nickname, "Last, First" vs "First Last", phonetic). Every group shows a confidence score, the reasons it matched, and warnings.
3. Review — sure things (95%+ with no warnings) are pre-approved. Everything else waits for a Merge, Edit, or Skip. In an expanded group you can pick which name spelling to keep, take a card out of the group ("Not a dupe"), and see the exact merged result before deciding.
4. Export — a clean .vcf, an audit log, an Outlook-compatible CSV, a printable duplicate list, and (with two files) "only in file 2" / "only in file 1" .vcf files.
5. Browse Contacts — a read-only viewer for any loaded file (search, A–Z index, full card with notes). This is the safe way to look up a number from an export on a Mac without importing it into Contacts.

The guarantees

• Notes are never lost. Every unique note body from every duplicate is kept, joined with ---. The .vcf download is blocked if any approved merge has fewer note characters than its sources (only possible if a reviewer edits notes by hand).
• Fields are never lost. A merged card is a property-level superset of its inputs: every email, phone (with its label), address, URL, photo, category, and custom X- property survives. Where a single-valued field disagrees (two different companies), the winner is written to the field and the other is recorded in the notes under [Merged duplicates], along with alternate name spellings that carry extra words (e.g. Handyman, Al Holmes).
• Untouched contacts are written back byte-for-byte (line endings normalised to CRLF). Nothing is re-encoded unless it was merged.
• Nothing is written anywhere. The tool only produces downloads. Importing them into Synology/Outlook/iPhone is a deliberate, separate step. Keep the original export as the backup.

Outlook CSV support

• Reads classic Outlook's export (File → Open & Export → Import/Export → Export to a file → Comma Separated Values → Contacts). Columns are matched by name, so other Outlook versions' column sets work too. Comma, semicolon and tab delimiters; UTF-8 or Windows-1252 (classic Outlook's "ANSI") are detected automatically.
• Outlook's filler values (0/0/00 dates, Normal, False, Unspecified, auto-generated Initials) are treated as empty. The E-mail Display Name columns ("Name (address)") are ignored. Exchange /o=… addresses are resolved to the SMTP address in the display name.
• Clean contacts for Outlook (.csv) uses the loaded file's exact header, delimiter and encoding. Untouched rows are written back value-for-value. Merged rows put each value back in its original column, then a free column of the same kind; anything left over goes into Notes under [Extra details — no matching Outlook column]. The file is Windows-1252 when every character fits, otherwise UTF-8 with a BOM.
• If five or more notes are exactly 255 characters long, the app warns that the export itself may have been truncated.
• Import safely: back up the .pst, import the CSV into a new, non-synced contacts folder, check it, then swap it in. Importing over the live, synced folder duplicates or misses merges on every device. Don't open the CSV in Excel first.

Matching rules (scores add up; groups form at or above the threshold, default 60%)

Evidence	Points
Same email address	50
Same phone number (last 7 digits)	40
Same toll-free number (800/888/877/…)	30, or 10 with a warning if the names are unrelated
Exact name (word order ignored)	50
Nickname match (Bob/Robert, Cathy/Catherine…)	45
Same first and last name, extra words differ	42
Names share two words	35
Same last name + first initial, or last name only on one side	25
Names share one word / last name only	8 / 5 (deliberately below the floor — households share phones)
Same company	+10

Group-level warnings: Chained (AB and BC but A and C never matched directly), Large group, Different organizations, Different birthdays, Name-only match, Names look unrelated, Shared toll-free number. Any warning prevents pre-approval.

Verification

node tests/test_engine.mjs                                   # unit tests on the real engine code
node tests/test_engine.mjs ~/Documents/Contacts/ALLSynologyContactsSep292025.vcf   # + real-file smoke test
python3 -m unittest tests/test_note_roundtrip.py             # note escape/merge mirror tests

Against the September 2025 Synology export (10,208 contacts, 3,448 with notes, 546,095 note characters): every merge passes the note and field integrity checks, and a second pass over the exported file finds no 95%+ groups.

Importing the result

• Synology Contacts / iPhone / Mac Contacts import a multi-contact .vcf directly.
• Classic Outlook for Windows imports only the first contact from a .vcf. Use the Outlook CSV download (see above).
• Because contacts sync through CardDAV to every device, import into one place and let it propagate. Never edit inside the Synology DSM Contacts UI.

Files

• index.html — the whole app (engine between // ENGINE-START and // ENGINE-END, then the UI).
• tests/test_engine.mjs — Node tests that load the engine straight out of index.html (including an Outlook CSV round trip of the real export).
• tests/fixtures/outlook_export_cp1252.csv — hand-built classic-Outlook export in Windows-1252 with multi-line notes, quotes, filler values and an Exchange address.
• tests/test_note_roundtrip.py, tests/verify_synology_notes.py — Python mirrors of the note handling.
• docs/ — design spec and plan from the July 2026 rebuild; Phyllis_Contact_Dedup_Rebuild_Handoff.md — background on the environment and the original notes-loss incident.