| name | boardroom-storm-research |
| description | Run decision-grade, multi-perspective research using Boardroom STORM (an adaptation of Stanford's STORM method). Use when the user wants a rigorous research report, market map, investment/diligence memo, competitive analysis, board memo, product strategy, or long-form thesis on an ambiguous topic where smart people would disagree — anything where a shallow consensus answer is costly. Triggers: "run Boardroom STORM", "STORM research", "research X thoroughly", "build a diligence/thesis/board memo", "map this market". Do NOT use for single factual lookups answerable with one or two web searches. |
Boardroom STORM
A six-stage research workflow that front-loads organization before writing: discover perspectives → scout sources → simulate expert interviews → map contradictions → outline → draft → moderator pass. Adapted from Stanford OVAL Lab's STORM (Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking) and Co-STORM. The point is not prettier prose — it is finding the questions and contradictions a generic report misses.
When to use
High when the cost of a shallow answer is high: entering a market, writing a public thesis, diligence on a startup, evaluating a protocol, mapping competitors, a board memo, or deciding if a trend is signal or noise. The sweet spot is an ambiguous topic where five smart people would disagree before converging. Skip it for a simple factual lookup — plain web search is better there.
How to run it
- Gather the charter inputs first. Ask the user (or infer and state) the bracketed fields in Prompt 0: topic, decision context, audience, time horizon, geography/segment, source hierarchy, output target. Do not skip this — it shapes every later stage.
- Enable retrieval. Use web search / research mode whenever current facts matter. Every factual claim needs a citation, or it is marked
Unverified.
- Run the nine prompts in order, from
references/prompts.md. Each prompt is one stage; substitute the bracketed values. Show the user the output of each stage and let them steer before continuing (this is the Co-STORM "human-in-the-loop" discipline). For an autonomous run, proceed stage to stage but still surface the source ledger, contradiction map, and outline as checkpoints.
- Hold the outline-before-draft line. Do not write the final piece until the source ledger, interview briefs, contradiction map, and outline are all complete.
The six stages (nine prompts)
| Stage | Prompt(s) | Produces |
|---|
| 0. Charter | Prompt 0 | Restated charter + assumptions |
| 1. Perspectives | Prompt 1 | 8–10 decision lenses + ranked top-12 questions |
| 2. Source scout | Prompt 2 | Source ledger (credibility + freshness rated) |
| 3. Interviews | Prompt 3 | 6-turn simulated expert interview per perspective |
| 4. Contradictions | Prompt 4 | Contradiction map + unknown-unknowns + 5 conclusion-movers |
| 5. Outline | Prompt 5 | Source-mapped outline (no prose yet) |
| 6. Draft → moderate → finalize | Prompts 6, 7, 8 | Grounded draft → skeptical moderator pass → final memo |
The full copy-paste text of all nine prompts is in references/prompts.md. Read it before starting and follow it stage by stage.
Non-negotiable quality rules
- Separate research from writing. No drafting before the outline is source-mapped.
- Never skip the source ledger. STORM is retrieval + question-asking, not brainstorming with academic branding.
- Never let one persona dominate. Value comes from discourse among lenses, not a single authoritative voice.
- Never hide uncertainty. Maintain an uncertainty ledger (weak claims, stale sources, contradictions, missing data). Mark unverifiable claims
Unverified rather than smoothing over them.
- Never confuse confidence with evidence. Inspect citations; the user decides if evidence is adequate for the decision.
- Watch STORM's known failure modes (flagged in the paper): source-bias transfer, and over-associating facts that merely co-occur. The moderator pass (Prompt 7) exists to catch these.
What this is not
Not a replacement for Stanford's official STORM codebase, and not a substitute for expert diligence, legal/financial advice, customer calls, or technical validation. It is a better way to build the first serious map of a problem space and decide where human work should go next.
Attribution: Method adapted from Stanford OVAL Lab's STORM and Co-STORM research. Workflow framing ("Boardroom STORM") by Linas Beliūnas (Linas's Newsletter).
