Boardroom STORM Research

Use this skill when the user mentions:

• "Run Boardroom STORM on [topic]"
• "Deep research on [topic]"
• "Multi-perspective analysis of [topic]"
• "Run STORM on [topic]"
• "Research [topic] from multiple angles"
• "Diligence report on [company/category]"
• "Map this market"
• "Research [topic] for a memo / thesis / article"
• "I need a decision-grade analysis of [topic]"
• "What are the bull and bear cases for [topic]?"
• Any research request where the user wants perspective diversity, contradiction mapping, or source-grounded synthesis rather than a single-perspective answer

Do NOT trigger when:

• The user needs a single factual answer (use web search)
• The user is validating an early-stage idea without needing multi-source depth (use idea-validation)
• The user needs a pitch deck or fundraising materials (use fundraising, but Boardroom STORM can feed into it)
• The user needs a PRD or product roadmap (use product, but Boardroom STORM can provide the research layer)
• The user wants a pre-mortem on a plan they already have (use future-failure-audit)

If the research output will feed into a downstream deliverable (pitch deck, PRD, GTM plan, failure audit), note the handoff at the end of the final output.

Apply these standards throughout every stage, not just at the end.

A successful Boardroom STORM output should:

• Surface non-obvious questions, not just summarize common knowledge.
• Show where informed perspectives disagree and why.
• Separate facts from interpretations from implications.
• Tie major claims to sources when possible.
• Produce an actionable decision memo, not a generic overview.
• Make uncertainty useful by turning it into next steps.

Writing quality rules for the draft and final output:

• No vague superlatives: avoid "transformative," "game-changing," "rapidly evolving," "significant growth," "innovative approach," "cutting-edge" unless the evidence justifies the specific word.
• No filler transitions: avoid "it's worth noting," "interestingly," "moreover," "furthermore," "in today's landscape."
• Prefer concrete over abstract: "47% YoY growth" beats "strong growth." Name the source.
• Use an investor-grade, practical tone: confident but not promotional, direct but not blunt.
• Vary sentence length. Short sentences for emphasis. Longer ones when complexity requires it.

1. Separate research from writing. Never draft before the outline is source-mapped.
2. Generate perspectives before retrieving. The questions shape the search, not the reverse.
3. Use source-grounded simulated interviews to produce better questions.
4. Build a source ledger before synthesis. STORM is retrieval plus question asking, not brainstorming with academic branding.
5. Build a source-mapped outline before drafting. The Stanford implementation explicitly separates pre-writing from writing.
6. Maintain a contradiction map and uncertainty ledger throughout.
7. Run a Co-STORM-style moderator pass before finalizing.
8. Mark unsupported claims as [Unverified] rather than presenting them as facts.
9. Optimize for decision usefulness, not encyclopedic completeness.
10. Never let one persona dominate. The value comes from discourse among perspectives, not a single authoritative voice.
11. Never confuse Claude's confidence with evidence. The user must inspect citations and decide whether evidence is adequate for their decision.

Before starting, determine the appropriate depth. If the user specifies a mode, follow it. Otherwise, default to Full Research for serious decisions and Quick Research for time-constrained or narrower questions.

For Full and Board-grade research with 7+ perspectives, batch interviews into groups of 3-4 perspectives per pass to avoid context degradation. Complete one batch, extract key claims, then proceed to the next batch with the accumulated claim set as context.

Use Claude's capabilities strategically at each stage:

Stage 0: Research charter

If the user has not provided enough context, make reasonable assumptions and proceed. Do not stall unless the task is unsafe or impossible. State assumptions clearly so the user can correct them.

Create a research charter with:

• Topic
• Decision context (what decision this research will inform)
• Audience
• Time horizon
• Geography or segment
• Source hierarchy (what sources to prioritize)
• Output format
• Research depth (Quick / Full / Board-grade)
• Key assumptions

Restate the charter in 8 bullets and confirm before proceeding.

Stage 1: Perspective discovery

Generate perspectives based on the chosen research depth (4-5 for Quick, 7-10 for Full/Board-grade).

Always consider these lenses and select the most relevant:

• Practitioner/operator who deals with this daily
• Customer/user or buyer
• Incumbent competitor
• Startup challenger
• Regulator or policy expert
• Economist/business model analyst
• Technical architect or engineer
• Skeptic/bear-case analyst
• Historian/comparable-cycles analyst
• Investor/capital allocator

For each perspective, produce:

• Persona name (specific, not generic — "Series B fintech CFO" not "finance person")
• Why this lens matters for the decision
• What this persona knows that others miss
• Five sharp questions this persona would ask
• Three trusted source types for this persona
• One blind spot this persona is likely to have

End with a ranked list of the 12 highest-leverage questions across all perspectives.

Stage 2: Source scout

Before answering any research question, build a source plan by perspective.

When web search or research mode is available, gather sources and create a source ledger with:

• Source title
• URL or file reference
• Publisher/author
• Date
• Perspective(s) informed
• Key claims or data points
• Credibility rating: High / Medium / Low
• Freshness rating: Current / Potentially stale / Historical
• Bias or limitation notes

If web access is unavailable, ask the user to provide sources or proceed with clearly marked assumptions and no factual certainty. State explicitly: "No web access available. All claims below are based on training data and should be independently verified."

Stage 3: Simulated expert interviews

For each perspective, simulate an interview between:

• Interviewer: rigorous analyst
• Expert: the persona, grounded in the source ledger

Interview depth by research mode:

• Quick Research: 3 turns per interview
• Full Research: 4-6 turns per interview
• Board-grade: 6 turns per interview, plus follow-up interviews on major contradictions

For 7+ perspectives, batch into groups of 3-4. Complete one batch, extract claims, then run the next batch with accumulated claims as context.

Rules:

• Ask one question at a time.
• Make each follow-up build on the previous answer.
• Cite sources for factual claims when available. Say [Unverified] when not.
• Distinguish facts, interpretations, and implications.
• End each interview with: strongest claim, weakest claim, what would change the expert's mind, and the question nobody else is asking.

Stage 4: Contradiction map

Using the interview briefs and source ledger, identify:

• Direct contradictions: where two perspectives make incompatible claims
• Evidence asymmetries: where one side has stronger evidence than another
• Incentive conflicts: who benefits if a given interpretation becomes accepted
• Timing conflicts: what may be true now but false in 12-24 months
• Definition conflicts: where people use the same words to mean different things
• Missing perspectives: who was not represented and why that could matter
• Unknown unknowns: questions that emerged only because the perspectives interacted

Use this table:

End with:

• What all perspectives agree on
• What nobody adequately addressed
• Five facts most likely to change the conclusion

Stage 5: Source-mapped outline

Create a hierarchical outline before drafting.

Include:

• Working title
• One-sentence thesis
• Reader promise (what the reader will be able to decide after reading)
• Detailed outline with section descriptions
• Claims under each section
• Sources supporting each claim
• Claims excluded because evidence is weak
• Open questions to verify before drafting

Do not write the full draft until this outline is complete.

Stage 6: Draft

Draft from the approved outline only.

Requirements:

• Follow all writing quality rules from the Quality bar section above.
• Cite factual claims when sources are available.
• Name uncertainty explicitly — do not hedge vaguely.
• Include concrete implications for founders, entrepreneurs/operators, builders/product leaders, and investors when relevant.
• Use tables for comparisons, use cases, risks, and action plans.

Recommended structure:

1. Executive thesis (2-3 sentences)
2. Why now
3. What the evidence says
4. Contradictions and open questions
5. Persona-specific implications
6. Action checklist
7. Final decision memo or conclusion

Stage 7: Co-STORM moderator pass

Before finalizing, act as a skeptical moderator. The moderator's job is to find what the research missed.

Check for:

• Missing stakeholder perspective
• Overweighted source cluster (too many claims from one source)
• Unsupported causal claim
• Over-association of unrelated facts
• Stale or geography-specific evidence presented too broadly
• Incentive misread
• Technical feasibility gap
• Regulatory or compliance blind spot
• Competitive response not considered
• Investor-relevant downside case

Present findings to the user before applying fixes. Format as:

Moderator findings:

1. [Finding ranked by importance]
2. ...

Revised thesis (if needed)

Confidence scores for the most important claims (High / Medium / Low with reasoning)

Final uncertainty ledger (see template below)

Diligence questions a serious investor or board member would ask next

Ask the user: "Should I incorporate all fixes, or do you want to adjust any before I finalize?" If the user does not respond or says to proceed, incorporate all fixes.

Stage 8: Final output

Finalize only after incorporating the moderator pass.

Include:

• One-page executive memo or executive summary at the top
• Persona-specific action table (see template below)
• "What would change this conclusion" section
• "Next 7 days" action plan when applicable
• Source and claim audit table when citations are available

Before delivering, confirm:

• The draft answers the original decision context from the charter.
• No major claim lacks support or a caveat.
• Contradictions are explicit rather than buried.
• The conclusion is actionable, not just informative.

If the research output should feed into a downstream skill (fundraising deck, PRD, GTM plan, failure audit), note the handoff: "This research can serve as input for [skill]. Say 'now run [skill] using this research' to continue."

Persona-specific action table

Uncertainty ledger

Diligence questions

Group diligence questions by:

• Market
• Customer
• Product/technical
• Competitive
• Regulatory/legal
• Financial/business model
• Team/execution
• Timing

Boardroom STORM produces research that often feeds into other workflows: