Business Idea Council

Business Idea Council

Help a founder decide what to do next with a business idea. Research what buyers do, challenge the attractive story, and recommend a commitment proportional to the evidence. A compelling pitch, busy community, or unanimous simulated council is not market validation.

1. Frame the decision

Read relevant supplied research and existing product context first: .agents/product-marketing.md, .claude/product-marketing.md, or the legacy product-marketing-context.md. Do not infer the business from an unrelated current folder or previous project.

Capture a short decision brief:

• Proposed customer, user, payer, geography and buying situation.
• Problem, current workaround, promised outcome and proposed offer/price.
• Stage and evidence already available: idea, interviews, prototype, payments, repeat use.
• Founder access, capabilities, time, budget, runway and constraints.
• Decision at stake: next experiment, pilot, build, launch, expansion or comparison.

Ask only for missing information that materially changes the research. If the idea itself is absent, ask for it. For optional gaps, state assumptions and proceed; keep distinct segments separate rather than averaging incompatible buyers together.

Default to detailed validation across all eight research lenses below. A requested quick screen may compress the report and combine roles; a requested deep dive expands only the decisive uncertainties. State the scope, research date, geography, evidence limitations and any user budget/source cap. Do not turn a bounded request into an exhaustive market study.

2. Plan the research council

Read research roles to assign distinct questions and avoid duplicated searches. These are analytical roles, not claims that real experts participated.

The chair synthesizes; it is not a ninth vote. Add a specialist only for a material uncertainty: marketplace liquidity, regulated operations, AI reliability, procurement, supply chain, local density, or retention. Explain why a seat is added or a topic is excluded.

Use subagents only when the user's request or governing instructions authorize delegation and tools support it. Give each a bounded question, assigned sources/segments, output ownership and a shared brief. Run independent research in parallel within the live capacity, then give the skeptic the merged evidence for challenge. Do not hard-code models, create app tasks, or let workers overwrite one another. When delegation is unavailable or unauthorized, perform the lenses sequentially and disclose that they were not independent reviews.

3. Gather and reconcile evidence

Read evidence and decision rules before searching. Use current web research by default; respect a user's offline or supplied-evidence-only scope. Discover actual callable tools rather than assuming access to paid databases or private social networks. Built-in search and page opening are sufficient for a first pass. Optional specialist skills such as customer-research or competitor-profiling may help a specific seat; do not automatically launch their entire workflows.

Search the buyer's problem, triggers, workarounds and buying language, not just the proposed product category. Inspect the underlying sources behind search results. Prefer official product/pricing pages for competitor claims, original buyer conversations for expressed pain, and authorized first-party records for commercial behavior. Cite current claims near the finding and connect them to evidence IDs.

Each seat returns:

1. A finding tied to a specific segment and decision.
2. Supporting and contrary evidence IDs, including dates and source limitations.
3. Observed facts, interpretations and untested assumptions kept distinct.
4. Confidence with a reason, not a source-count shortcut.
5. The next cheapest test that could change the recommendation.

Maintain an evidence ledger. Deduplicate syndicated articles, copied posts, repeated accounts and shared datasets before counting corroboration. Merge the seats' work by claim, not by concatenating eight essays. Audit every verdict-driving factual claim against its source. Missing access or unsuccessful searches mean unknown, not zero demand or no competitors.

With no browsing or insufficient evidence, complete a clearly labeled provisional review using what is available, list the missing checks and propose tests. Do not invent fresh social observations, customer quotes, revenue, search volumes, interviews or third-party tool results to fill the template.

4. Convene the marketing strategy debate

After research, load the installed marketing-council skill using its active catalog path (or a verified sibling marketing-council/SKILL.md). Load only the dossiers for 3–5 relevant advisors, including a genuine dissenter. Reuse that skill's documented frameworks and grounding rules rather than duplicating persona dossiers here.

Suitable starting lenses include Dunford on competitive alternatives, Halbert on demonstrated buyer appetite, Hormozi on the offer, Hopkins on response tests, and Godin or Sharp on audience/reach trade-offs. Choose for the case, not a fixed celebrity lineup. A thin-evidence case still needs a useful challenge, not twelve simulated endorsements.

Label this section: Simulated strategic council: applications of published frameworks, not actual reviews or endorsements. Connect each recommendation to the research ledger and identify where the framework's scope is an analogy. Paraphrase; use quotes only when verified. Do not impersonate a living advisor commenting on a named competitor or controversy. Refer to abstract alternatives and strategic choices instead.

The advisors cannot create evidence, verify demand by agreement, or override the research with authority. Surface 2–4 substantive disagreements when present, and the observation or experiment that would settle each. Do not manufacture conflict if the real uncertainty is simply missing evidence. If marketing-council or a dossier is unavailable, report the missing dependency and use clearly labeled generic strategy lenses; do not invent an advisor's position.

5. Decide and design the next test

Apply the gates in evidence and decision rules. Assess pain, willingness to pay, alternatives/wedge, distribution, economics, retention, feasibility and founder fit as supported / mixed / contradicted / unknown, each with confidence and evidence IDs. Do not average away a fatal constraint or present an arbitrary score as a probability of success.

Choose one primary recommendation, scoped to a segment and next commitment:

• Proceed to a constrained pilot: sufficient behavioral evidence and a credible delivery path justify a bounded next step; identify what remains unproven before scaling.
• Validate a critical assumption: a decisive unknown remains; run the cheapest discriminating test before substantial building or spend.
• Pivot: evidence supports a different segment, job, offer, channel or business model; explain what changes and what evidence transfers.
• Park or stop: credible evidence contradicts a critical premise or a constraint makes the plan untenable; give conditions for reopening it.

Distinguish evidence against the idea from insufficient evidence. Be willing to recommend a pilot when evidence merits it; skepticism is not automatic rejection. Never describe desk research alone as product-market fit, repeatable demand or permission to scale.

Design 2–4 prioritized experiments around the decisive uncertainties. Each specifies hypothesis, target buyer/recruitment, behavior measured, denominator, test method, cost/time cap, owner, success and failure thresholds, an inconclusive result, and the decision after each outcome. Set thresholds before execution; label proposed thresholds and input estimates as assumptions, not universal benchmarks. Fit the tests to the user's constraints.

An interview script explores a recent real event, workaround, spending and decision process rather than pitching for compliments. A stronger demand test asks for a meaningful commitment at explicit terms, followed by delivery and repeat-use checks. Do not fabricate traction, scarcity or availability in test materials. Draft outreach and experiment assets when useful; sending messages, spending money, collecting payments, publishing or handling customer data requires the relevant authorization.

6. Deliver and verify

Use the report template, adapting depth to the scope. Default deliverables are report.md and evidence.csv in a new dated idea-validation/<idea-slug>/<date>/ folder (or the user's requested location). Preserve prior reports; do not overwrite them silently. A brief chat-only request may keep the ledger in the answer instead of creating files.

Before delivery, check:

• The requested four research specialties and the material commercial/operational lenses are covered or explicitly marked unverified.
• Every decisive external claim has an inspected, relevant citation; assumptions, source dates, sample limitations and contradictions are visible.
• Social attention, stated interest, competitor supply and actual purchases are not interchangeable.
• Formulas, units, currencies, recurring periods and scenario arithmetic are consistent; unknowns remain unknown.
• The skeptic had access to the evidence, disagreements are resolved by tests, and advisor simulation is labeled.
• The recommendation follows the weakest critical gate, names its limits and provides falsifiable next actions.
• Report links and evidence IDs resolve, and no proposed experiment is claimed to have run.

Lead the user-facing response with the verdict and its strongest reason, link the detailed report, and name the next decision. Hand off execution only as appropriate: product-marketing for positioning, offers/pricing for commercial design, customer-research for interviews, ads/social for approved acquisition tests, or ab-testing for experiment analysis. This council decides whether and how to invest further; it does not automatically build or publish a business.