Both-Sides Research

An AI-agnostic research workflow for steelmanning opposing views, auditing sources, red-teaming claims, and reaching evidence-grounded conclusions without false balance.

This repo contains a portable agent skill that can be used with any AI assistant or research workflow. It is designed for questions where "give me both sides" is not enough: the answer needs strong arguments, source quality checks, counterarguments, limitations, and a clear statement of where the evidence actually lands.

What It Does

Both-Sides Research helps an assistant or human researcher:

• Restate a question in neutral, testable language.
• Build a source base across competing positions.
• Steelman the strongest credible case for each side.
• Audit source quality for authority, recency, support, funding, and primary-vs-secondary status.
• Search for counterarguments, failed replications, limitations, and alternative explanations.
• Review reasoning for cherry-picking, scope creep, correlation-causation errors, and weak evidence-to-claim ratios.
• Produce concise pro/con answers, full research memos, or Brainlift-style outputs.

The workflow is intentionally not tied to any single AI model, app, or vendor. Use it with ChatGPT, Claude, Gemini, Perplexity, Manus, Codex, local models, browser-based research, note-taking systems, or manual research.

Why It Exists

Many "balanced" research outputs are weak in one of two ways:

• They perform fake neutrality by giving equal weight to unequal evidence.
• They argue one side well and treat the other side as an afterthought.

This skill aims for the better middle: genuinely understand each side, test the quality of the evidence, identify the real cruxes, and say plainly when one side is better supported.

Repository Contents

.
|-- SKILL.md
|-- README.md
|-- assets/
|   `-- both-sides-research-banner.png
|-- agents/
|   `-- openai.yaml
`-- references/
    `-- source-workflow.md

• SKILL.md is the core portable skill.
• agents/openai.yaml is optional UI metadata for OpenAI/Codex-style skill surfaces.
• references/source-workflow.md preserves the original 5-phase Brainlift research workflow that inspired the skill.

Quick Start

Use the skill by giving an assistant the contents of SKILL.md, then ask it to research a question.

Example:

Use the Both-Sides Research workflow to evaluate whether AI tutoring systems should replace some classroom instruction time in K-12 schools. Give me the strongest case for, strongest case against, key cruxes, and what evidence would change the conclusion.

For tools that support skill folders, install or copy this folder into the tool's skills directory and invoke:

Use $both-sides-research to investigate this question with balanced evidence, steelmanned opposing views, and source-quality checks.

Recommended Output

For quick answers:

Bottom line:

Strongest case for:

Strongest case against:

Key cruxes:

What would change the conclusion:

Confidence:

Sources:

For deeper research:

# 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

Core Principles

• Steelman before judging.
• Prefer primary sources when possible.
• Treat source quality as part of the answer, not housekeeping.
• Search for counterarguments deliberately.
• Avoid false balance when the evidence is lopsided.
• Separate evidence, assumptions, mechanisms, values, and conclusions.
• State uncertainty and what would change the conclusion.

Brainlift Support

The workflow includes optional Brainlift adaptation:

• DOK 1: verified facts and definitions.
• DOK 2: source-backed knowledge tree.
• DOK 3: synthesized insights and patterns.
• DOK 4: defensible SPOVs tested against counterarguments.

For Brainlift-specific language and the original 5-phase flow, see references/source-workflow.md.

GitHub Topics

Recommended topics for this repo:

ai, ai-agents, research, evidence-based, critical-thinking, source-evaluation, argument-mapping, steelman, counterarguments, red-team, bias-audit, prompt-engineering, knowledge-management, brainlift

License

No license has been selected yet. Add a license before encouraging broad downstream reuse or redistribution.