How to use Multi-Field Record Linker
- Provide two CSV tables with distinct id values and the fields to compare, using pasted text or their files.
- Declare two to eight weighted rules and a minimum similarity. Review which fields use text, exact, number or date comparison and set tolerances where required.
- Run the comparison and inspect each candidate's field scores. The combined score is rule similarity, not a probability that two people or records are identical.
- Download record-review-template.csv, enter accept or reject beside complete candidate IDs, then paste it into Reviewed decisions CSV and rerun with unchanged data and settings.
- Download the reviewed pair report and unlinked-records.json. Accepted links are one-to-one; source records are never merged or deleted.
Example: Multi-Field Record Linker
The exact name match and a one-year age difference produce a weighted similarity of 0.9333.
Supported inputs and limits
- Up to 150 rows per side, 80 CSV columns and 2 MiB per input; at most 3,000 candidate pairs.
- Text similarity uses NFKC normalization, outer trim, lowercase and character-bigram Dice scores, with 300 characters per compared text field.
- Numeric/date similarities fall linearly to zero at the positive entered tolerance.
- Blank fields score zero.
- Weights are positive.
More details
Accept only one counterpart per record.
Scores do not estimate identity probability and no source rows are merged.
JSON numeric literals are limited to 128 characters and must retain their decimal value through browser Number conversion; integers outside the safe integer range and unsupported precision are refused.
Where your input is processed
This tool does not upload your inputs. See the limits for anything saved on your device.
Questions about Multi-Field Record Linker
How does this differ from matching a single key?
Every pair is scored across at least two declared fields. Each field has its own method and weight, so a near name can be assessed alongside a date or a numeric tolerance.
What does a score of 0.9333 mean?
It is the weighted average of the rule similarities, rounded for display. It does not mean there is a 93.33 percent chance that the records identify the same person.
Can I accept two counterparts for one record?
This review uses one-to-one accepted links. An overlapping approval is refused so that an ambiguous match cannot silently resolve several source records.