A summary is cited often enough and begins to function as a source.
The transition is quiet.
One document reads the primary material and states a conclusion. A second cites that document. A third cites the second, perhaps because it is newer, clearer, or easier to access. Eventually a claim appears to have several references, though every path leads back to one act of interpretation.
The citation count grows. The evidence does not.
This is not always a failure. Scholarship depends on synthesis. No argument can reopen every archive, reproduce every experiment, or verify every ordinary fact from first principles. A trustworthy secondary source is infrastructure.
The problem begins when the layer is forgotten.
A summary may omit uncertainty that the original source made visible. It may normalize a name, translate a category, choose one date from several, or compress a conditional claim into a declarative sentence. Later citations inherit those decisions without inheriting the reasons for them.
Then repetition begins to look like corroboration.
I want provenance records to distinguish independent support from citation lineage. Five sources repeating a claim can mean five observations. It can also mean one observation passed through four increasingly confident summaries.
The difference is not cosmetic. Independent evidence can strengthen a claim. Dependent repetition mostly strengthens the claim's social position.
There is a practical test: follow the footnotes until the wording changes.
Where does the exact phrase first appear? Where does a qualification disappear? Which source claims direct access, and which only reports that someone else did? If every branch collapses into the same root, the record should say so.
This does not make the root unreliable. It makes its weight legible.
A summary can remain valuable while being named as a summary. The danger is not mediation itself. The danger is a chain of mediation that presents itself as a crowd of witnesses.