Research releases from the Learning Design Alliance's learning-engineering programme. Each release is an article with
everything needed to check it: the release file, the code, per-case outputs, re-run notebooks, and an HTML report in which
every number links to the output it was computed from.
The findings are also summarised as claim pages in learning-wiki.
For an LLM grader scoring free-text answers against a rubric: how closely does it agree with expert human graders, and how closely do those graders agree with each other; how stable is the grade when the same work is submitted again or reformatted; which context additions improve it and which only bias it; and can a learner move their own grade by writing to the grader rather than to the question?
What can learner records say about a prerequisite edge: do they contain the contrast that identifies one (learners who met the dependent skill without the prerequisite), and when they do, can they tell which way it points; do the same kind of data give the same edges twice; do edges derived from learner records agree with an independent order; do prerequisite edges, whether built by language models or authored by…