Why does the score say 100 % for two records with different names?
Because the criteria are weighted: the same VAT number, phone and e-mail outweigh a name written differently. The review screen shows each criterion with its similarity, weight and points, and the field comparison marks what is equal and what differs.
Does Dolibarr already merge third parties?
Yes, from the third party card. DataCleaner finds the pairs, explains them, previews the result and then calls the core merge. On Dolibarr 21 to 24 the native merge deletes the absorbed party’s document directory before moving it; the module copies the documents first, a setting kept on by default.
And contacts?
Dolibarr has no contact merge at all. The module relinks the known tables (events, categories, document contacts, mailings, notifications…), deduplicates conflicting keys, offers a hook for third-party modules and refuses the merge if a table it does not know references the contact.
Can I run it on a large base?
Candidate pairs are built in the database by blocking keys, in batches, rather than comparing every record with every other. The documented test run on 20,000 third parties computed fingerprints in 15 s and scanned in 47 s with 33 MB of memory.
What is stored and for how long?
Fingerprints, candidate pairs, scan runs and merge audit snapshots, in six module tables. Snapshots hold personal data, so a retention in days is configurable and the purge task removes older ones. Fingerprints and pairs disappear with their source record.
Can the module stop duplicates from being created?
When a user creates a third party or contact that looks like an existing one, a warning offers to open the existing record or to create anyway. An option blocks creation when the VAT number is already used.