No model tonight. The notebook works on your repository. A reviewer reads a paper looking for the load points, the sentences the conclusion rests on, and asks four questions of each: what is the number, what is the interval, what is the control, what would have falsified it. A sentence written from the impression of a plot, with no number ever extracted from it, is how over-claims get into papers without anyone being dishonest.
You learn the five-move shape most good short papers follow, generate the methods section from the artifacts you saved in week three rather than from memory, and spend half the session reviewing another team’s draft. Reading someone else’s draft as a reviewer is the fastest way anyone has found to learn what a claim needs.
What you will do
- Mark the load-bearing sentences in your own draft and answer the four questions for each.
- Write the question in one sentence a non-specialist could repeat back to you.
- Generate the methods section from your saved eval JSON, model name, seed, and item counts.
- Review another team’s draft and return the four questions they could not answer.
What you leave with
- An introduction and methods draft.
- A written review of another team’s work, and one of yours.
Materials
Notebooks and slides are released to each cohort at the start of the session.