TODO:
- Auto-similarity check by cosine-similarity between all files.
- Finer checks on most "vaguely similar" ones on certain special aspects (keywords, intricacies)
- Checkers that include things that are odd-at-large (usage of advanced material too early)
- (Opt. LLM-based similarity or per-line comparisons to avoid comments)
- Instruction/Rubric-parser + "focus-point" creator.
- auto-rubric-generator based off of instructions & use such to "grade" subsequent work if rubric not provided
- regrades based on different levels of "strictness"
- per-line-instructions on deductions to "ensure" that LLM-nonsense can be "verified"
focus first on CSS