Research Review And Revise

Ocupación

Categorías

Académico

Then produce a revised draft or a targeted revision plan.

For substantial projects, keep critique items structured:

• feedback
• severity (high, medium, low)
• followup_query (if evidence gap exists)
• must_retrieve (true/false)
• skip_retrieval_reason (required when must_retrieve=false)

Revision rules

• Fix evidence and coverage problems before polishing prose.
• Do not silently remove nuance just to make the text smoother.
• Preserve explicit uncertainty when the sources are mixed.
• If citations are malformed, fix the citation layer after content changes stabilize.
• Keep both draft_report_raw.md and draft_report_refined.md when large revisions are applied.
• Leave revision_delta.md for claim-level and citation-level change trace.
• For each analytical subsection, enforce one explicit claim-first sentence in the revised draft.
• Rewrite caveat-heavy endings into conditional statements that qualify a clear conclusion, instead of ending on an undecided stance.
• In final publication-facing files, remove draft helper labels and workflow/meta wording from abstract, body, and conclusion.

Quality gate

Before publication export, score and gate:

• coverage
• structure
• relevance
• citation precision (and recall if available)
• unique body-cited source count (target depends on corpus size policy)

Use:

• scripts/quality_gate_scorecard.py

Typical thresholds:

• coverage >= 3
• structure >= 3
• relevance >= 3
• citation precision >= 0.60
• if total collected records are <15: prompt supplementation to at least 15, and cite all included sources
• if total collected records are 15-29: cite all included sources
• if total collected records are >=30 and screening is used: unique body-cited sources >= 30

If the corpus is below 15 collected records, record a supplementation prompt to reach at least 15 before finalizing.

Checklist

Read:

• references/review_checklist.md