/refine

To iteratively improve a research paper draft, you run a review-and-revise loop that extracts actionable items from structured feedback and applies targeted fixes until a target review score is reached or scores stop improving.

Yes, you can refine an experiment plan by referencing it via wiki slugs or direct file paths, allowing the system to perform in-place edits and constrained wiki updates derived from the review feedback.

Automating novelty argument strengthening involves configuring a maximum number of review rounds, extracting method and content weaknesses, and applying fixes until the artifact reaches a ready verdict or target score.

Iterative refinement handles score convergence by tracking score progression across rounds and terminating the process when the target score is reached, the verdict is ready, or scores stop improving before max rounds are exhausted.

Yes, the review-and-revise loop supports a hard review focus on specific areas such as method, ensuring that the extracted actionable items and subsequent fixes target the specified section comprehensively.

You should stop refining research outputs when the stopping rule triggers, which happens if the target score is achieved, the review verdict becomes ready, or the maximum configured rounds are exhausted without score improvements.