The QA angle you mentioned is interesting - makes me think of these tools as data sources with different schemas and freshness. Elicit gives you a clean, transformed table (the summary), but you don't see the raw data or joins. Inciteful shows you the messy, raw relationship tables (the citation edges), but you have to build the query.
For foundational papers, I've found Inciteful's graph is better *if* you treat your seed papers like test data. You wouldn't use biased sample data for a pipeline, right? So maybe don't use Elicit's top results as seeds. I try to find a couple of known survey papers first, extract their references as a .bib file, and use those as the seed set. It's slower but feels more repeatable.
Anyone tried building a small Airflow DAG to chain these steps? Like, scrape a textbook bibliography -> seed Inciteful -> export graph -> load to Neo4j for centrality queries? Might be overkill for literature review, but feels solid.
Forget reliable. Both will give you garbage in, garbage out if you're new.
>seminal papers
Inciteful's your only shot at those, but you need the right seeds. Pop the most obvious recent paper into Elicit, you'll get a summary of the current hype. Feed that hype into Inciteful, you'll get a graph of the same hype.
Go find a random, old textbook chapter. Use *those* references as your seeds. That's your systematic start. Otherwise you're just automating your own ignorance.
The textbook tip is solid. I'd add: dig into the seminal papers' own references too. Use those as new seeds in Inciteful and the graph quickly shows you the real foundations.
Also, modern textbooks sometimes update references. Find an edition from the 90s if you can.
—cp
That's a solid iterative approach, but the textbook references you're pulling from are themselves a curated sample. You're just moving the bias upstream. The author of a 90s textbook already had their own citation graph in mind, favoring the papers that supported their narrative or framework.
It's better than using a recent Elicit summary, sure. But it's still building a map from a single cartographer's notes. You might find a foundational paper, but is it foundational to the field, or just foundational to that particular textbook's chapter structure?
Data skeptic, not a data cynic.