[Brainstorming] Potential LLM Enrichments

* Key Takeaways
- [x] "what are some of the authors' decisions which they make a point to justify? do these decisions represent novelty? standard practice? corroboration/disagreement with cited work?"
- "summarize the paper's main result concisely and directly. this will be an entry in a space-constrained 'crib sheet'."
- [x] "I am a graduate student who is preparing for an open book test that will be partly on the material in this paper. I am collecting content to populate a 'crib sheet' of quick facts for reference. I have a fixed, limited capacity available for this crib sheet, and many papers that I am being tested on, so I must be mindful to only fill it with the most important information. generate crib sheet entries for this paper, ordered by priority for inclusion in the crib sheet."

* citation classification
- NB: this is likely a pre-requisite for "standard practice/SOTA" methodology/recommendation mining
- checking a box (e.g. related work) / "our work cites the earlier work because `reviewer #2` demanded it"
- standard practice / "our work corroborates the earlier paper's choice to do it this way"
- improves upon / "our work demonstrates the earlier paper was doing it the wrong way"
* proposed future work extraction
- could create an "open problems" feed
* "Figure 2" identification
- i.e. identifying the most important figure in the paper which describes the core result on fundamental "how it works" diagrammatically
* Figure - to - text description
- feed image + caption to VLM for text description that could be used in place of image (e.g. for LLM consumption)

* "idea development" graph
- probably not actually useful, but hey it's an idea - https://old.reddit.com/r/MachineLearning/comments/1iw5lgj/p_see_the_idea_development_of_academic_papers/

The issue names no implementation files, tests, or entry points. Start by narrowing the listed LLM/VLM enrichments and citation-classification ideas into a defined scope with acceptance criteria. Done should be a selected, agreed enrichment with an identified implementation and validation path.

Written by the indexing model from the issue text.