ds-rebuttal

Start by normalizing reviewer comments into atomic items with stable IDs, then classify each as editorial, text-only, evidence gap, experiment gap, or claim scope. Route each item to text revision, literature positioning, or supplementary experiments, and assemble the answers into a point-by-point response letter.

Classify each comment first: wording gaps, positioning issues, and evidence-repackaging needs rarely require new runs. Only comments with a genuine experiment gap should trigger supplementary experiments, and each run must be tied to a named reviewer item with a hypothesis and minimal success criterion.

Yes. When manuscript_edit_mode is set to latex_required, the workflow treats the LaTeX source tree as the preferred editing surface and produces LaTeX-ready replacement text. If LaTeX source is unavailable, it generates replacement text plus an explicit blocker note instead of pretending edits were made.

Mark the item as cannot_fully_address and state the honest limitation explicitly in the response letter. Never invent results or promise future work unless the response format allows it; downgrade the claim scope if the existing evidence no longer supports it.

Do not use it when the paper does not yet exist, when there are no concrete review materials, or when the task is fresh ideation or baseline experiments. It is designed specifically for review, revision, and rebuttal stages of an existing manuscript.