both-sides-research

name

both-sides-research

description

AI-agnostic balanced research workflow for unbiased, both-sides argument analysis. Use when an assistant or agent needs to investigate a question, compare competing claims, prepare pro/con or debate briefs, steelman opposing views, audit source quality, red-team a draft, validate a Brainlift, or produce research with explicit counterarguments, limitations, uncertainty, and evidence-grounded conclusions.

Both-Sides Research

Use this skill to turn an open question into a balanced, evidence-grounded answer. The workflow is assistant-agnostic: apply it with any AI tool, agent, research app, browser, search engine, note system, or manual research process available in the current environment.

The goal is not artificial neutrality; it is to understand the strongest credible version of each side, test the evidence, identify cruxes, and state where the balance of evidence actually lands.

For the source workflow this skill was derived from, read references/source-workflow.md when the user asks about Brainlifts, DOK levels, SPOVs, or the original 5-phase process.

Tool-Agnostic Use

• Treat named tools in examples as replaceable. "Ask the AI", "search the web", "use a connector", or "save in notes" means use whatever assistant, browser, database, research tool, or note system is available.
• Do not depend on any single model, vendor, app, or interface.
• If live research tools are unavailable, state the limitation and work from the provided sources or the user's supplied material.
• If live research tools are available and the topic is current, high-stakes, niche, or source-dependent, verify claims against current primary or reputable secondary sources.
• Preserve the user's requested output format unless it would hide important caveats, counterarguments, or source-quality issues.

Core Workflow

1. Define the question.

   • Restate the user's research question in neutral language.
   • Identify the main competing theses, including an "against" or skeptical thesis even when the user starts from one side.
   • Set scope boundaries: timeframe, geography, population, industry, policy context, or decision criteria.
   • Note what would count as strong evidence for each side.

2. Build the source base.

   • Use primary sources when possible: peer-reviewed papers, official data, legal/regulatory text, original documents, standards, transcripts, or direct statements.
   • Include reputable secondary sources for interpretation and landscape mapping.
   • Seek at least one strong source for each major side before synthesis, unless the evidence is genuinely one-sided.
   • For current, high-stakes, niche, or source-dependent claims, verify with the relevant live research tool or source collection and cite sources.
   • Track source quality as you go: authority, recency, citation/support, funding or conflicts, and primary-vs-secondary status.

3. Synthesize both sides.

   • Steelman each side before judging it.
   • Separate claims, evidence, assumptions, mechanisms, and values.
   • Identify common ground and disputed cruxes.
   • Distinguish "this side has a point" from "this side is better supported."

4. Validate before finalizing.

   • Run the Source Quality Audit.
   • Run the Counterargument Search.
   • Run the Logic and Coherence Review.
   • Run the Bias and Missing-Evidence Check.
   • Decide whether to proceed, revise, gather more sources, or reframe the question.

5. Deliver the answer.

   • Lead with the bottom line when the user needs a decision.
   • Show the strongest case for each side.
   • Explain the cruxes and what evidence would change the conclusion.
   • State confidence and uncertainty plainly.
   • Cite or name the key sources used.

Validation Checks

Source Quality Audit

For each important source, check at least three dimensions:

• Authority: recognized expert, institution, journal, official body, or direct participant.
• Recency: publication date is appropriate for the field and question.
• Support: citations, replication, independent corroboration, or data quality.
• Funding and bias: funder, sponsor, author incentives, conflicts, or advocacy stance.
• Primary vs. secondary: whether the source is original evidence or interpretation.

Flag weak sources. Replace them when the claim depends on them.

Counterargument Search

Actively look for dissent:

• What do the strongest critics say?
• What are the acknowledged limitations?
• Have key findings failed to replicate?
• Are there alternative explanations for the same facts?
• What would an informed opponent say is being cherry-picked or ignored?

Add a counterarguments and limitations section to any long-form research output.

Logic And Coherence Review

Trace conclusions back to evidence:

• Evidence-to-claim ratio: do strong claims have strong support?
• Causality: avoid treating correlation as causation.
• Representativeness: avoid overgeneralizing from narrow samples.
• Cherry-picking: check whether contrary evidence was excluded.
• Internal consistency: check whether different claims contradict each other.
• Scope creep: avoid claims beyond what the sources support.

Bias And Missing-Evidence Check

Before finalizing, ask:

• Which side did the prompt, source set, or initial framing make easier to believe?
• Are there base rates, null results, or boring explanations missing?
• Are incentives shaping the claims?
• Is one side underrepresented because it uses different language or appears in different source communities?
• Would a domain expert consider this source set adequate?

Do not force symmetry. If one side is much better supported, say so and explain why.

Decision Checkpoint

Use this checkpoint after validation:

Result	Action
All checks passed	Finalize with confidence and caveats.
Minor issues found	Revise targeted claims, citations, or wording.
Major issues found	Gather better sources before finalizing.
Fundamental flaws	Reframe the question or split it into subquestions.

Output Formats

For quick answers, use:

Bottom line:

Strongest case for:

Strongest case against:

Key cruxes:

What would change the conclusion:

Confidence:

Sources:

For full research memos or Brainlifts, use:

# Question

# Scope

# Short Answer

# Evidence Map

# Strongest Case For

# Strongest Case Against

# Counterarguments And Limitations

# Logic And Source Quality Review

# Cruxes And Uncertainties

# Conclusion

# Sources

Brainlift Adaptation

When the user asks for a Brainlift-style output:

• Treat DOK 1 as verified facts and definitions.
• Treat DOK 2 as the source-backed knowledge tree.
• Treat DOK 3 as synthesized insights and patterns.
• Treat DOK 4 SPOVs as defensible points of view tested against counterarguments.
• For every SPOV, trace back to DOK 3 insights and DOK 2 sources.
• Add a "Counterarguments and Limitations" section before finalization.

Use the source workflow reference for the original phase names and detailed Brainlift validation language.