idea-validation

Idea validation

Map the category, then compare candidate ICPs against it, and pick the strongest wedge. Read lib/research-protocol.md and lib/playwright-auth.md first. Budget: ~15 fetches for the market map + ~6–8 fetches per candidate, ~60 fetches total.

What this skill IS and IS NOT

• IS: a two-layer analysis at idea stage. Layer 1 maps the category (who plays here, what they charge, who is funding this space). Layer 2 compares 2–5 candidate personas/segments head-to-head against that category map. Output is a wedge recommendation with explicit confidence.
• IS NOT: validation through real customers (that's dudu:customer-discovery). IS NOT a deep one-wedge rehearsal (that's dudu:market-problem). IS NOT a defensible TAM (that's dudu:market-sizing). IS NOT a deep competitor diligence on a chosen company (that's dudu:competitive-landscape) — the market layer here is category-shaping for wedge selection, not company-vs-company combat analysis.

State these distinctions in the artifact every time. The point of this skill is which wedge in which category, not is there real demand — only real interviews settle that.

When to use

• Founder has an idea but hasn't chosen a wedge.
• VC suspects the founder picked the wrong wedge and wants to test alternatives before deep-diving.
• Pre-flight before dudu:market-problem when the ICP input would otherwise be a guess.

If the founder already has paying customers or strong customer-discovery signal, skip this and run dudu:market-problem directly.

Inputs

Required (prompt if missing):

• Deal slug
• Company name (or working title for the idea)
• One-line problem statement (what hurts, for whom, when)
• Either:
  • A list of 2–5 candidate personas/segments to compare, OR
  • Permission to generate 3 candidates from the problem statement

Optional:

• Geographic scope (default: global English-speaking)
• Hard constraints (e.g., "must be SMB, not enterprise" / "must not be regulated industries")
• Pitch deck text

Pre-flight

Idempotency check. Artifact: deals/<slug>/idea-validation.md. Candidate profiles live under deals/<slug>/candidates/. The market map lives at deals/<slug>/market-map.md.

If deals/<slug>/idea-validation.md exists and --force was not passed, print "Artifact already exists at deals//idea-validation.md. Pass --force to overwrite." and stop.

Method

Step 1 — Map the category (runs once, before any candidate work)

Goal: understand the market the idea would enter, independent of which persona it targets. This is the category-shaping pass. Write deals/<slug>/market-map.md. Cap at ~15 fetches across:

1. Direct competitors — products solving the same job. Use Product Hunt category search, Crunchbase category, G2/Capterra category. For each, capture: positioning one-liner, pricing tier(s), funding stage, last public activity, primary ICP they target.
2. Indirect competitors / substitutes — what people use today instead (spreadsheets, agencies, internal tools, the status quo). Pull from Reddit/HN/forum threads where the pain is discussed — what do people say they currently do?
3. Pricing benchmarks — concrete ACV anchors across the category: lowest-tier, mid-tier, enterprise. At least 3 data points if the category has them. Note whether the category is freemium-dominant, seat-based, usage-based, or services-priced — this constrains which candidates can support a venture outcome.
4. Funding and tailwinds — Crunchbase or news search for the category over the last 18 months: who raised, what stage, what thesis. A funded, growing category is proof of demand; an empty category is a red flag or a true greenfield (call which).
5. Structural shifts — is something in the world (regulation, AI capability, platform shift, demographic) changing what's possible here? One paragraph with citations, or "no clear shift identified."

Write market-map.md with these sections:

# Market map: <Category>

**Generated:** <ISO timestamp>

## Category boundary
[1–2 sentences: what's in scope, what's out. Name the category in the founder's language and in analyst language if they differ.]

## Direct competitors

| Product | Positioning | Pricing | Stage / last raise | Primary ICP | Source |
|---------|-------------|---------|--------------------|-------------|--------|
| ... | ... | ... | ... | ... | [link] |

## Indirect competitors / status quo

| Substitute | Why people use it today | Why it's inadequate (verbatim quotes if available) | Source |
|------------|--------------------------|----------------------------------------------------|--------|
| ... | ... | ... | [link] |

## Pricing benchmarks

- Low tier: $X/seat/mo or $X/yr — [product, link]
- Mid tier: ...
- Enterprise: ...
- **Pricing model dominant in this category:** freemium / seat / usage / services
- **Implication for WTP:** [one sentence]

## Funding and tailwinds (last 18 months)

[3–6 bullet points with citations: who raised, what they pitched, what stage]

## Structural shifts

[One paragraph with citations, or "no clear shift identified."]

## Category verdict

Pick one and justify in one sentence:
- **Empty** — no commercial precedent. New category bet.
- **Forming** — early funded entrants, no clear leader.
- **Contested** — multiple funded players, room for differentiation.
- **Saturated** — clear leader(s) own the category; differentiation must be sharp.

## Sources
- ...

Step 1 parallelization (Layer 2 — per source category)

The five source categories above are independent. Dispatch one worker subagent per category, all concurrently in a single turn. See lib/research-protocol.md § Parallelization for the cross-platform mapping (Claude Code: Agent with subagent_type="general-purpose"; Codex: spawn_agent with agent_type="worker" and multi_agent = true in config).

Group into 4 subagent batches (combine the two competitor categories so direct/indirect overlap is resolved in one place):

• Subagent A — Direct competitors (source 1): ~5 fetches. Skip Crunchbase if it requires Playwright; pull funding from public news/press instead, and flag any blanks for the main session to fill.
• Subagent B — Indirect / status quo (source 2): ~3 fetches.
• Subagent C — Pricing benchmarks (source 3): ~3 fetches.
• Subagent D — Funding tailwinds + structural shifts (sources 4, 5): ~3 fetches.

Each subagent prompt MUST include: the category name in both founder and analyst language, the deal's product description, the fetch cap for that batch, the citation and source-honesty rules from lib/research-protocol.md (pasted), and the required return shape (rows for the relevant market-map.md table or bullets for the relevant section, plus a Sources list).

In the main session: fill any Crunchbase gaps via Playwright (per lib/playwright-auth.md), pick the category verdict, and write market-map.md.

Step 2 — Generate or accept candidates

If candidates were supplied, write each to deals/<slug>/candidates/candidate-K.md using the profile template below. If not supplied, generate 3 candidates that span meaningfully different axes — e.g., different company size, different job-to-be-done, different industry — not three flavors of the same buyer.

Use market-map.md to inform candidate generation: candidates should target underserved or differently-served slices of the category, not the slice every direct competitor already owns.

Candidate profile template (candidates/candidate-K.md):

# Candidate <K>: <short label>

**Segment:** <industry / company size / role band>
**Buyer role:** <job title that signs the contract>
**User role:** <job title that uses the product daily — may be the same>
**Buying trigger:** <the moment that makes them shop>
**Why this candidate is plausible:** <one sentence>
**How this differs from the other candidates:** <one sentence — must be a real difference, not a synonym>
**Which competitors from the market map already serve this candidate:** <list, or "none identified">
**Which substitutes/status-quo from the market map this candidate uses today:** <list>

Hard rule: each candidate must differ from the others on at least one of {industry, company-size band, buyer role, buying trigger}. If two candidates collapse onto the same axis, merge them.

Step 3 — Research each candidate (parallel, ~6–8 fetches each)

The category-level research from Step 1 is shared across candidates, so per-candidate work focuses on what's specific to this slice. For each candidate, spend up to ~6–8 fetches across:

1. Persona-specific pain quotes — Reddit / HN / niche forums and segment-specific subreddits or trade-association forums. Quote complaints verbatim. Look for the segment's own phrasing of the pain.
2. Persona-specific WTP signal — does this segment already buy adjacent tools? At what price? Reviews where price is mentioned, segment-targeted product pricing pages, vendor case studies that quote contract sizes.
3. Reachability — can you actually find 100 of these people? Public lists, named communities, conferences, professional associations, industry directories, LinkedIn search counts via Playwright.
4. Population / segment size — BLS/Eurostat occupational counts, industry-association membership counts, LinkedIn filtered counts. One credible number is enough at this stage.
5. Competitor overlap with this candidate — of the direct competitors in the market map, which ones target this candidate vs. a different one? Read their landing pages, customer-logo walls, case studies, ICP statements. A competitor that loudly targets enterprise leaves the SMB candidate more open.

Apply lib/research-protocol.md citation rules. If a dimension can't be sourced for a candidate, write "Not found in public sources" — never invent. Do not re-fetch category-level data already in market-map.md; cite it.

Parallelization (Layer 2 — per candidate)

Candidates are independent. With 2 or more candidates, dispatch one worker subagent per candidate, all concurrently in a single turn. See lib/research-protocol.md § Parallelization for the cross-platform mapping (Claude Code: Agent with subagent_type="general-purpose"; Codex: spawn_agent with agent_type="worker" and multi_agent = true in config).

Each subagent prompt MUST include:

• The full candidate profile (candidates/candidate-K.md contents) and the deal's one-line problem statement.
• The text of market-map.md so the subagent can cite category facts without re-fetching them.
• The per-candidate fetch cap of ~6 fetches (reserve ~2 per candidate for main-session Playwright work below).
• The citation and source-honesty rules from lib/research-protocol.md (paste, don't reference). If a dimension can't be sourced, the subagent must write "Not found in public sources" — never invent.
• Sources to consult: items 1, 2, and 5 from the list above. Skip items 3 and 4 for the subagent — LinkedIn search counts require Playwright with the VC's authenticated session.
• Required return shape: rubric scores for Pain severity, Pain frequency, WTP proxy, and Competitive room (each 1–5 or ? with one-line evidence-anchored rationale citing a source), the 3 strongest verbatim evidence quotes for that candidate (each with link), and a Sources list.

After all subagents return, in the main session:

1. For each candidate, do the LinkedIn / Playwright work for items 3 and 4 (Reachability + Population/segment size). Authenticated browsers cannot run in subagents.
2. Score Reachability and Segment size from the Playwright findings; copy Category proof from market-map.md (it's the same for every candidate); merge with the subagent rubric scores from the four dimensions above. Proceed to Step 4.

With a single candidate this skill does not apply — escalate to dudu:market-problem directly.

Step 4 — Score each candidate on the rubric

Score every candidate on the same seven dimensions, 1–5, with a one-line evidence-anchored rationale per score. Refuse to score without a citation; mark as ? if the evidence is missing and explain in the rationale.

Dimension	What scores high (5)	What scores low (1)
Pain severity	Multiple verbatim complaints calling it urgent/critical/blocker	Pain only inferred from product marketing
Pain frequency	Daily/weekly recurring	Annual or one-off
WTP proxy	Segment already pays >$X for adjacent tools; explicit price quotes; category pricing supports venture economics	No comparable spending pattern; free tools dominate; category pricing is structurally low
Reachability	Public lists, identifiable channels, named communities	Diffuse, no clear way to find or contact 100 of them
Category proof	Funded startups + paying users in the broader category (from market map)	No commercial precedent; category is inert or dead
Competitive room for this candidate	No incumbent loudly owns this candidate's slice; room to land a wedge	A clear leader already serves this exact candidate well
Segment size	Reachable population large enough that a wedge of it sustains a startup	Too small to support a venture outcome

Notes on the rubric:

• Category proof comes from market-map.md and is the same value for every candidate of a given idea. It's the "is anyone making money in this category at all?" check.
• Competitive room is per-candidate and answers "is this slice taken?" — a saturated category can still have an underserved candidate.
• A ? (insufficient evidence) is more honest than a guessed score. Multiple ?s on the same candidate is itself a finding: "we can't tell yet."

Step 5 — Cross-candidate comparison

Build the head-to-head matrix. Then write the comparison narrative: where do candidates agree, where do they diverge, and which dimensions carried the most weight? Explicitly call out where the market map made a candidate look better or worse than it would have looked in isolation.

Step 6 — Wedge recommendation

Pick one of:

• Lead candidate identified — name it, justify in 3 sentences referencing the rubric, list the 3 strongest evidence quotes from that candidate's research, and state confidence (Low / Medium / High).
• Tie / too close to call — name the top 2, explain what evidence would break the tie, and propose the next research step.
• No viable wedge yet — state the binding constraint (e.g., "no candidate showed WTP signal above $X") and what the founder would need to discover to make this idea pursuable.

Confidence floor: if 3+ dimensions are ? for the lead candidate, confidence cannot exceed Low. Also: if the market map's category verdict is Empty and there is no candidate with concrete WTP evidence, confidence cannot exceed Low regardless of other scores.

Artifact template

# Idea validation: <Company>

**Deal:** <slug>
**Generated:** <ISO timestamp>

> ⚠️ Idea-stage analysis from secondary sources only. This skill picks *which wedge looks strongest on paper* — it does not validate that real customers will buy. Real signal comes from `dudu:customer-discovery`. Deep competitor combat analysis comes from `dudu:competitive-landscape`.

## Problem statement

> [One sentence the VC/founder gave us]

## Market map (summary)

Full map: `market-map.md`.

- **Category boundary:** [one line]
- **Category verdict:** Empty / Forming / Contested / Saturated
- **Direct competitors counted:** N (top 3: ..., ..., ...)
- **Pricing range observed:** $X–$Y (model: freemium / seat / usage / services)
- **Funding signal (last 18mo):** [one line — e.g., "4 Seed/A rounds totaling $X in this category" or "no funding activity found"]
- **Structural shift:** [one line, or "none identified"]
- **Implication for wedge selection:** [one sentence — e.g., "Saturated at enterprise, open at SMB" / "Empty category — burden of proof on demand evidence"]

## Candidates

| # | Segment | Buyer role | Buying trigger | Differs from others on |
|---|---------|------------|----------------|------------------------|
| 1 | ... | ... | ... | ... |
| 2 | ... | ... | ... | ... |
| 3 | ... | ... | ... | ... |

Full profiles: `candidates/candidate-1.md`, `candidates/candidate-2.md`, `candidates/candidate-3.md`.

## Comparison matrix

| Dimension | Candidate 1 | Candidate 2 | Candidate 3 |
|-----------|-------------|-------------|-------------|
| Pain severity | <1–5 or ?> — <evidence> | ... | ... |
| Pain frequency | ... | ... | ... |
| WTP proxy | ... | ... | ... |
| Reachability | ... | ... | ... |
| Category proof | <same value for all candidates — from market map> | <same> | <same> |
| Competitive room for this candidate | ... | ... | ... |
| Segment size | ... | ... | ... |
| **Composite** | <sum / mean, with `?` count> | ... | ... |

## Competitor overlap by candidate

| Direct competitor | Targets candidate 1? | Targets candidate 2? | Targets candidate 3? | Evidence |
|-------------------|----------------------|----------------------|----------------------|----------|
| ... | yes/no/partial | ... | ... | [link] |

[Use this to make the "Competitive room" scores defensible. A candidate with no competitor overlap rows is either greenfield or invisible — the narrative should say which.]

## Where candidates diverge

[2–4 dimensions where the candidates produced meaningfully different scores. These are the decision-driving dimensions — call them out.]

## Where candidates agree

[Any dimension where all candidates scored similarly. Often this means the dimension isn't decision-relevant for *this* idea; useful to retire from further analysis.]

## Strongest evidence per candidate

### Candidate 1
- [Verbatim quote or stat] — [link]
- [Verbatim quote or stat] — [link]
- [Verbatim quote or stat] — [link]

### Candidate 2
[same shape]

### Candidate 3
[same shape]

## Wedge recommendation

- **Verdict:** Lead candidate identified / Tie / No viable wedge yet
- **Lead (if any):** Candidate <K>
- **Confidence:** Low / Medium / High
- **Why:** [3 sentences anchored on the rubric]
- **What would change my mind:** [2–3 specific things that, if discovered, would flip the recommendation]
- **Recommended next step:** [Run `dudu:market-problem` against Candidate <K> / Run `dudu:competitive-landscape` for the lead candidate's slice / Run `dudu:customer-discovery prep` for Candidate <K> / Pivot the idea — see open questions]

## Open questions

[5–10 questions the next research step needs to answer. Each rooted in a `?` cell, a divergence call-out, or a gap in the market map — not generic.]

## Sources

- [Title](url) — which layer (market map / candidate K) and which dimension(s) it informed
- ...

After writing

1. Update deals/<slug>/manifest.json skills_completed["idea-validation"] to the current ISO-8601 UTC timestamp. Add the key if it doesn't exist; this skill is optional and not part of the six required by the dudu:diligence orchestrator.
2. Print the path to the artifact and the path to market-map.md.
3. If a lead candidate was identified, print a one-line suggestion: Next: dudu:market-problem (ICP = Candidate <K>).

Handoff to other skills

Two artifacts are designed to feed downstream skills:

• market-map.md — the category landscape. Hand it to dudu:competitive-landscape as starting context so it doesn't re-discover the same competitors; that skill goes deeper on the chosen company's combat position.
• candidates/candidate-<K>.md — the lead candidate profile. Structured to be pasted directly as the ICP input for downstream skills:
  • dudu:market-problem — when prompted for ICP, supply the lead candidate; pass market-map.md as additional context to skip re-research of the category.
  • dudu:market-sizing — the lead candidate's segment + buyer role + buying trigger are exactly the wedge-ICP shape that skill expects.
  • dudu:customer-discovery prep — the lead candidate's segment characteristics seed the target list when persona profiles don't yet exist.