Deep Survey

Deep Survey

Purpose: Precise search, full-depth reading — precise investigation of a specific sub-problem. Not broad, not exhaustive — targeted and thorough.

When to use: User knows exactly what they're looking for and needs precise, detailed understanding. E.g., "How exactly does DPO handle the reward model?" or "What are all the variants of LoRA rank adaptation?"

Budget

State Ledger

Print this table before each major iteration decision:

| SOP            | Target | Current | % Complete |
|----------------|--------|---------|------------|
| web-search     | 30     | ???     | ???%       |
| web-research   | 5      | ???     | ???%       |
| paper-overview | 40     | ???     | ???%       |
| paper-search   | 40     | ???     | ???%       |
| paper-research | 20     | ???     | ???%       |

Do not exit the strategy until all rows reach ≥90%.

Available Tactics

None mandatory — CC composes directly from SOPs.

Available SOPs

Import (strict protocol execution):

• web-search → web-browsing/skills/web-search/SKILL.md
• web-research → web-browsing/skills/web-research/SKILL.md
• paper-overview → literature-engine/skills/literature-overview/SKILL.md
• paper-search → literature-engine/skills/literature-search/SKILL.md
• paper-research → literature-engine/skills/literature-research/SKILL.md

Subagent (CC decides when to invoke):

• extract-data — structured comparison tables from deep-read papers
• gap-identification — find what the literature hasn't addressed
• survey-synthesis — produce final structured output

Execution Guidance

• Search is focused and precise — few but highly targeted queries
• High ratio of paper-research to paper-overview (20/40 = 50% deep-read rate)
• paper-research is the core operation — read raw full text, extract equations, hyperparameters, specific claims
• extract-data produces detailed comparison of approaches
• End with precise, authoritative conclusions — not "overview" but "definitive understanding"

Output Format

Detailed Technical Analysis containing:

• Specific methods compared side-by-side
• Equations and algorithms extracted verbatim
• Hyperparameters and implementation details
• Precise conclusions with evidence citations
• Remaining open questions at this level of detail