Research Outline Synthesis

How detailed the outline should be

For a substantial survey:

• the refined outline should use at least 2 heading levels
• if the corpus includes 20+ full-text paper notes, the refined outline should usually use 3 heading levels
• each major section should usually contain 2-5 subsections
• each subsection should map to a visible evidence cluster, not only to a vague topic label

Do not stop at:

• 1. Introduction
• 2. Methods
• 3. Evaluation

That level is only acceptable for a rough outline, not for a refined survey outline.

Where good subsections come from

Good subsection titles should emerge from repeated reading and clustering, not arbitrary decomposition.

Use this loop:

1. read a batch of papers
2. extract recurring mechanisms, tensions, benchmarks, and failure modes
3. group similar notes into clusters
4. turn clusters into subsection candidates
5. merge or split subsection candidates until each one has a clear claim and evidence base

Typical natural sources of subsection boundaries:

• different memory functions
• different update strategies
• different retrieval strategies
• different evaluation setups
• different failure modes
• different application constraints

Rules

• Do not start with prose if structure is still weak.
• Do not finalize a formal survey outline before the full-text reading set is large enough.
• Do not keep overlapping sections with slightly different names.
• Prefer section titles that encode clear distinctions.
• If the task is a survey, make the outline reflect comparison axes, not just a chronological list.
• If a subsection cannot be supported by distinct evidence, merge it or remove it.
• Run outline lint before handing off to drafting for substantial reports.

Recommended artifacts

• rough_outlines.md or rough_outline_a.md + rough_outline_b.md
• outline_merge_notes.md
• merged_outline.md
• subsection_outline.md
• draft_outline.md
• refined_outline.md
• section_evidence_index.md
• outline_lint_report.md

Outline patterns

Read:

• references/outline_patterns.md

Use those patterns as heuristics, not as rigid templates.

Script

Use:

• scripts/lint_outline.py

Typical check:

• duplicate sibling headings
• empty headings without body or children
• insufficient subsections for top-level sections
• sibling heading overlap risk