Fuzzy-match deduplication and master mapping for Excel and CSV data. No Python required to run.
- Dedupe Mode - finds near-duplicate records within a single spreadsheet
- Master Mode - maps your data to a trusted master list (many-to-many column mappings)
- Handles typos, abbreviations, and word-order differences
- Two engines - RapidFuzz (best under ~10,000 rows) or TF-IDF (best above that)
- Human-in-the-loop review - keyboard-driven (
←/→retain,Spaceskip,Fflag,Ctrl+Zundo) - Canonical registry - reuse earlier decisions and retro-update superseded ones
- Structured-code matching - tag code columns so part/SKU numbers must match exactly
- Resumable - progress is saved automatically; close and pick up where you left off
- Exports: a decisions workbook (Decisions + Flagged + Clusters sheets), an optional
write-back to your source file, and optional SQL
UPDATEstatements - all with a full audit trail
Grab the latest build from the Releases page -
unzip and double-click Conflate.exe. No install needed (Windows).
- Select Primary Data - your Excel/CSV file to clean. (Optionally add a Master List to map against.)
- Check the columns to match (e.g.
Supplier Name); set Match Strictness (85% is a good default). - Start Scan, review the score histogram, then Proceed to Review.
- Decide each pair with the keyboard -
←retain left,→retain right,Spaceskip,Fflag,Ctrl+Zundo. - On finish, pick your outputs: a Decisions workbook, a write-back to your source file, and/or SQL UPDATE statements.
A pizza shop's order system has the same customers entered several ways over time, with small differences in name and address:
| Cust_ID | Name | Address |
|---|---|---|
| C-1001 | Jonathan Meyer | 1428 Elm Street, Apt 3B |
| C-1002 | Jon Meyer | 1428 Elm St #3B |
| C-1003 | Maria Gonzalez | 76 Lakeview Dr |
| C-1004 | Maria Gonzales | 76 Lake View Drive |
| C-1005 | Antonio Russo | 9 Park Pl |
Load the file, check both Name and Address as search columns, set Cust_ID as the unique ID, and set Match Strictness to 85%:
After the scan, Conflate shows how the match scores are distributed so you can gauge data quality before reviewing:
It surfaces the likely same-person duplicates as candidate pairs:
| Score | Item A | Item B |
|---|---|---|
| 95% | Jonathan Meyer | 1428 Elm Street, Apt 3B | Jon Meyer | 1428 Elm St #3B |
| 92% | Maria Gonzalez | 76 Lakeview Dr | Maria Gonzales | 76 Lake View Drive |
Step through each pair in the review screen and keep the cleaner record as canonical with ←:
The exported Decisions sheet is a join-ready audit trail, use Primary_ID / Duplicate_ID to VLOOKUP the merges back into your order data:
| Primary_ID | Duplicate_ID | Final Selection | Action | Score | Note |
|---|---|---|---|---|---|
| C-1002 | C-1001 | Jonathan Meyer | 1428 Elm Street, Apt 3B | Retained Left | 95 | same household |
| C-1004 | C-1003 | Maria Gonzalez | 76 Lakeview Dr | Retained Left | 92 |
📖 Full User Guide - setup, every option, review workflow, outputs, troubleshooting, and glossary.
Requirements: Python 3.10+, Windows. Run from a plain Command Prompt (not Anaconda Prompt).
git clone https://github.com/iacsha/conflate
cd conflate
build.bat
The script creates a clean virtual environment, installs pinned dependencies
(scikit-learn 1.5.2), and packages a standalone Conflate.exe via PyInstaller.
Python · customtkinter · pandas · RapidFuzz · scikit-learn · PyInstaller
MIT - see LICENSE.