Deep Research Brief

Deep Research Brief

Purpose

Turn a research question into an auditable brief. Separate discovery, source appraisal, extraction, synthesis, uncertainty, and citation verification.

Scale The Workflow

• Low-risk scan: 10-15 good sources, compact source map, concise synthesis.
• Serious brief: 30-40 substantive sources, source matrix, disagreements, uncertainty, and final memo.
• High-stakes or user-requested thoroughness: use explicit research gates, second-pass searches, citation verification, and a Textbook Synthesis Gate before final synthesis.

Workflow

1. Define scope.

   • Restate the research question.
   • Identify decision context, time horizon, geography/jurisdiction, freshness needs, exclusions, and success criteria.
   • If scope is unclear and affects source choice, ask or state assumptions.

2. Build a source strategy.

   • Prioritize the following sources: official docs, standards, papers (particularly well-cited ones), popular blog posts from AI Alignment Forum or LessWrong or OpenAI Alignment Blog or Anthropic Alignment Research, Wikipedia articles, textbooks (when available and relevant), filings, source repos, regulators, direct data, and original announcements.
   • Add high-quality secondary sources (here, "secondary" means: "from a partially diverging branch of the tree of human knowledge" (the point is, I want you to foster an apporach of synthesizing in-distributino perspecitves with out-of-distribution perspectives (possibly from distinct fields, possibly from divergent thinkers))) for synthesis, critique, or practitioner context.
   • Search across distinct branches so the source set is not one-dimensional.

3. Create a source matrix before final synthesis for serious briefs.

   • Columns: source, author/org, date, URL/path, key claim, evidence type, limitations, relevance.
   • Track inaccessible, weak, abandoned, or excluded sources when they affect confidence.
   • Do not treat source count as quality. Prefer fewer strong primary sources over many weak summaries.

4. Extract evidence.

   • Capture claims, data, methods, assumptions, limitations, and source incentives.
   • Separate what the source says from your inference.
   • Note recency risk for facts likely to change.

5. Map disagreements and gaps.

   • Identify whether disagreement comes from methods, definitions, dates, incentives, populations, jurisdiction, measurement, or missing data.
   • Name what evidence would change the conclusion.

6. Synthesize for the user's decision.

   • Start with the answer or decision-relevant takeaway.
   • Then present findings, evidence, caveats, and implications.
   • Avoid source-by-source summaries unless the user asked for an annotated bibliography.

7. Apply the Textbook Synthesis Gate for learning, expert-track, or course-like requests.

   • Do not point to sources as a substitute for explaining.
   • Source count is not depth; depth means the user can understand mechanisms, apply concepts, and generate better questions after reading.
   • Include mechanisms, worked examples, counterexamples, and exercises when the user wants to learn or gain fluency.
   • For field-mastery work, include taxonomies, prototype experiments, failure modes, anti-theater rules, and open research questions.
   • Cite sources as the evidence behind the teaching, not as the teaching itself.

8. Verify before finalizing.

   • Re-check load-bearing claims against the cited sources.
   • Ensure citations support the sentence or paragraph they are attached to.
   • Look for citation drift after synthesis.

Brief Structure

Lighter Brief Structure

For lighter briefs, use this structure unless the user requests another format:

# [Question] Research Brief

## Executive Answer
## Minimal Background
## Core Mental Model
## Key Findings
## Evidence And Confidence
## Disagreements / Missing Evidence
## Implications For The User
## What Would Change My Mind
## Retrieval Check
## Sources

Deeper Brief Structure

For deeper briefs, use this structure unless the user requests another format:

# [Question] Research Brief

## Executive Answer

A direct answer in 3–7 sentences. Include the bottom line, the main reason, the most important caveat, and the practical implication.

## Why This Matters

Explain why the question matters for the user’s decision, research direction, learning trajectory, or operational context.

## Scope And Method

State:
- what was included
- what was excluded
- search/source strategy
- recency requirements
- source-quality standard
- limits of the review

## Minimal Background

Define the few concepts needed to understand the rest of the brief.

Keep this short. This is pre-training, not a textbook.

## Core Mental Model

Compress the topic into a reusable model.

Use one of these forms:
- causal model
- taxonomy
- mechanism
- decision rule
- process diagram
- tradeoff map
- historical arc
- failure-mode map

## Key Findings

Present the main findings in descending order of importance.

For each finding:
1. state the finding
2. explain the mechanism
3. give the strongest evidence
4. note the practical implication

## Evidence Map

Summarize the evidential basis.

Recommended columns:
| Claim | Best Evidence | Confidence |
|---|---|---|

Keep table cells short. Put long prose below the table.

## Disagreements And Missing Evidence

Separate:
- live disagreements
- weak evidence
- absent evidence
- methodological limitations
- places where sources talk past each other

## Uncertainty And Confidence

Give confidence levels for the main claims.

Use calibrated language:
- High confidence: multiple strong sources converge
- Medium confidence: plausible but incomplete evidence
- Low confidence: speculative, early, indirect, or source-limited

## Implications For The User

Translate the research into user-specific action.

Separate:
- what to believe
- what to do
- what to watch
- what not to over-update on

## What Would Change This Conclusion

List the observations, studies, events, or arguments that would materially alter the answer.

This keeps the brief epistemically live rather than frozen.

## Retrieval Check

Include 3–5 questions that test whether the reader absorbed the model.

Prefer:
- “Explain why…”
- “Predict what would happen if…”
- “Distinguish X from Y…”
- “Apply this to your situation…”

## Sources

List all cited sources.

For very large work, create a separate research memo first, then draft the final brief from the memo.

For textbook-style work, create the research memo and source matrix first, then transform them into chapter-like teaching prose. The chapter should teach the user's target skill or mental model directly; it should not merely say which readings cover which topics.

Quality Bar

• Important claims are cited.
• Current or unstable facts are verified with current sources.
• Source limitations are explicit.
• Disagreements are explained, not smoothed over.
• Recommendations are tied to evidence and user constraints.
• The final answer is useful for a decision, not just comprehensive.
• Expert-track work passes the Textbook Synthesis Gate: it teaches mechanisms, examples, counterexamples, exercises, and research taste directly rather than outsourcing the substance to source links.

Final Response

Report:

• Source count and source types inspected.
• Where the source matrix or memo was saved, if files were created.
• Key conclusion and confidence.
• Whether the Textbook Synthesis Gate was used and what teaching substance was produced.
• Any important unverified or inaccessible evidence.