Duplicate content is widely described as something search engines punish. What actually happens is a selection process, and the harm comes from losing control of which version is chosen.
There is no penalty, there is a choice
When a search engine finds several addresses with substantially the same content, it picks one to represent them and ranks that one.
The others are not penalised, they are set aside, and their signals are usually consolidated onto the chosen version.
Nothing is deducted from the site's standing, which is why the widespread fear of a duplicate content penalty misdescribes the mechanism.
The damage is in which version wins
The engine may select a printable version, a parameterised address, or a page on a syndication partner's site rather than the original.
When that happens the traffic goes to a page that may lack navigation, conversion elements, or in the syndication case, your site entirely.
This is the practical cost, and it is a routing problem rather than a punishment.
Most duplication is generated by the site itself
Tracking parameters, session identifiers, sort orders and pagination all create distinct addresses serving nearly identical content.
Ecommerce catalogues generate the largest volumes, particularly where products appear under several categories with different paths.
None of this is deliberate, and it is why duplication is far more often a technical artefact than an editorial decision.
It also means the volume grows silently with the catalogue, so a site that had no issue at launch can accumulate a serious one without anything being changed.
Canonical tags are a signal, not an instruction
A canonical tag states which version you consider primary, and search engines treat it as strong evidence rather than as a command.
Where the tag conflicts with other signals, such as internal links pointing consistently at a different address, the engine may disregard it.
Making the canonical choice consistent across links, sitemaps and redirects matters more than the tag alone, because the signals are evaluated together.
Syndication needs handling before publication
Republishing an article elsewhere means the two versions compete, and a larger partner site frequently wins the selection.
Agreements that specify a canonical reference back to the original, or a delay before republication, protect the position at no cost to the partner.
Where neither is possible, the realistic expectation is that the partner ranks and the original does not, which should inform whether the syndication is worth doing.