which pages travel with your name?
Listing the domains an answer cited is cheap and it is where source analysis usually ends. The list on its own does not answer the question that follows it, which is which one of these is worth getting into. This takes one step past the list and then stops on purpose, because the step after it is a different kind of claim and needs different data.
is it association or effect?
Association, not effect. For a given source, it compares how often the brand appears in the runs that cited it against the runs that did not. That contrast is real and it is measurable from data already collected. It is not the counterfactual: a source can travel with the brand because it causes the mention, or because both are symptoms of the model having decided to answer the question a particular way that draw.
The honest name for the number is a difference in appearance rate conditional on citation, and every printed line says so. The causal version needs the model to answer the same question with a source removed, which is an ablation on a controlled replica, and it is deliberately not faked here.
is a citation the same as influence?
No. A model reads pages it does not cite, and cites pages it barely used. What is measured is what the provider reported, which is citation. Calling it "the sources the model used" would be a claim about the inside of a system nobody outside it can see.
what is resampled?
The prompt, as everywhere else in the package. The k repeats of one question are not independent, so the pooled interval bootstraps over prompts. Pooling over runs would make every source look decisive.
No command prints this yet. It is called from the library: from lulumelon.mirror.sources import source_graph.
what does it export?
Every public name in lulumelon.mirror.sources that its source says something about, with the first paragraph of what it says.