@securescientist @ct_bergstrom Yes, I didn’t describe the problem well. Sorry for the confusion.

When the set of references is small, you do get useful context from where in the article the citation is referenced. But it’s in a very inconvenient form, and the signal is often pretty vague.

But with large lists, it falls apart entirely. A single sentence may have 10 references, and half may leave you scratching your head. Some may conflict with the paper’s result or with each other.

With over-frequent references sprinkled through the text, it reduces any contextual clue to either extremely fine grain, suggesting narrow value, or very little signal.

Going forward from reference to the citation works poorly. You start with context but lack title, which is usually a stronger clue about why unit was cited.

I find it more productive to work from the list of references (after reading the paper (usually)). Rather than searching for the citing context, I find it faster to just click through to read the abstract and make my own assessment..

But none of it is organized to support my task. It’s organized to support the argument, which is fine, but checking that isn’t usually my task.

Back when references weren’t hyperlinked, my strategy was different. Even with Wikipedia, I make more use of the citing context, either clicking through or searching back.

But Wikipedia, the reference titles are often quite uninformative, and the lists are usually short.

With unhyperlinked articles, the labor to track them down was the limiting step. It was worth the additional effort to narrow my search, and reference lists were usually not overwhelming.

And often I’m looking for background on a topic I don’t know well, not the specific point a reference is supporting.

I don’t quite fit the target audience, so often the contextual clues of the original point of reference, are quite useless to me. Even if I understand them perfectly, the clues don’t relate to my goal.

So a large fraction of the time, I just click on articles with promising-sounding titles, avoiding meta-analysis if possible (for reasons). It works pretty well if the references aren’t being padded.

This has probably left you no less confused before. But people approach reading papers with different strategies for different goals, and that holds true for perusing the references as well.