CLI Reference

reclink includes a command-line interface for common fuzzy matching tasks. It is available as the reclink command after installation, or as python -m reclink.

reclink --help
reclink --version
python -m reclink --help

---

reclink dedupe

Deduplicate records in a CSV file. Builds a pipeline internally with preprocessing, sorted-neighborhood blocking, string comparison, and threshold classification.

reclink dedupe --input data.csv --field name --threshold 0.85 --output results.csv

Options

Flag	Short	Type	Default	Description
--input	-i	str	<em>required</em>	Input CSV file path.
--field	-f	str	<em>required</em>	Column name to match on.
--threshold	-t	float	0.85	Minimum similarity score to consider a match.
--scorer	-s	str	"jaro_winkler"	Similarity metric name.
--id-column		str	"id"	Name of the ID column. If absent, sequential IDs are auto-assigned.
--output	-o	str	<em>None</em>	Output file path. If omitted, results are printed to stdout.
--format		csv or json	csv	Output format (only used when --output is specified).

Output

When writing to a file, the output contains these columns:

Column	Description
left_id	ID of the first record in the pair.
right_id	ID of the second record in the pair.
score	Similarity score.
match_class	Classification label (if applicable).

Example

# Deduplicate with a strict threshold, output as JSON
reclink dedupe \
  --input customers.csv \
  --field full_name \
  --threshold 0.92 \
  --scorer jaro_winkler \
  --id-column customer_id \
  --output dupes.json \
  --format json

When no --output is given, results are printed to the terminal:

  cust_001 <-> cust_047  score=0.9412
  cust_003 <-> cust_089  score=0.9156

2 duplicate pair(s) found.

---

reclink link

Link records between two CSV files. Both files must share the field name specified by --field.

reclink link --left a.csv --right b.csv --field name --threshold 0.8

Options

Flag	Short	Type	Default	Description
--left	-l	str	<em>required</em>	Left CSV file path.
--right	-r	str	<em>required</em>	Right CSV file path.
--field	-f	str	<em>required</em>	Column name to match on (must exist in both files).
--threshold	-t	float	0.85	Minimum similarity score.
--scorer	-s	str	"jaro_winkler"	Similarity metric name.
--id-column		str	"id"	ID column name. If absent, IDs are auto-assigned as left_0, right_0, etc.
--output	-o	str	<em>None</em>	Output file path.
--format		csv or json	csv	Output format.

Example

reclink link \
  --left vendors_a.csv \
  --right vendors_b.csv \
  --field company_name \
  --threshold 0.75 \
  --output links.csv

Found 12 links. Results written to links.csv

---

reclink match

Find the best matches for a single query string against a file of candidates (one string per line).

reclink match --query "John Smith" --candidates-file names.txt --limit 5

Options

Flag	Short	Type	Default	Description
--query	-q	str	<em>required</em>	The query string to match.
--candidates-file	-c	str	<em>required</em>	Path to a text file with one candidate per line.
--threshold	-t	float	<em>None</em>	Minimum score filter.
--scorer	-s	str	"jaro_winkler"	Similarity metric name.
--limit	-n	int	10	Maximum number of results to return.

Output

Results are printed to stdout, sorted by score descending:

  0.9333  John Smyth  (index=42)
  0.8889  Jon Smith   (index=7)
  0.8500  John Smith  (index=0)

3 match(es) found.

Example

# Find top 3 matches above 0.8
reclink match \
  --query "Acme Corporation" \
  --candidates-file companies.txt \
  --scorer cosine \
  --threshold 0.8 \
  --limit 3

---

reclink explain

Show a per-algorithm score breakdown for a pair of strings. Useful for understanding which metric best fits your data or debugging unexpected match results.

reclink explain "John Smith" "Jon Smyth"

Arguments

Argument	Type	Description
string_a	str	First string (positional).
string_b	str	Second string (positional).

Output

A table of every available algorithm sorted by score, with a visual bar chart:

  Comparing: 'John Smith'  vs  'Jon Smyth'

  jaro_winkler                      0.8322  #########################
  token_sort_ratio                  0.8000  ########################
  jaro                              0.7972  #######################
  smith_waterman_similarity         0.7600  ######################
  partial_ratio                     0.7500  ######################
  cosine                            0.5000  ###############
  levenshtein_similarity            0.7000  #####################
  ...

Example

# Quick comparison from the command line
reclink explain "Mueller" "Muller"
reclink explain "Robert" "Bob"

---

Pipeline Behavior

The dedupe and link commands build an internal pipeline with the following steps:

1. Preprocess -- fold_case and normalize_whitespace on the target field
2. Block -- sorted-neighborhood blocking with a window of 5
3. Compare -- string comparison using the specified --scorer
4. Classify -- threshold-based classification at the specified --threshold

For more control over pipeline configuration (custom blocking strategies, multi-field comparison, Fellegi-Sunter classification, etc.), use the Python API directly. See the Pipeline reference.

---

Exit Codes

Code	Meaning
0	Success
1	Input error (file not found, missing field, empty file)
2	Argument parsing error

---

See Also

• Pipeline -- full Python pipeline API with custom blocking, comparison, and classification
• String Metrics -- all available scorer names for --scorer
• Export -- programmatic export of results to CSV/JSON
• Quick Start -- get started with the Python API