startup-validator

@rules/evidence-and-scoring.md @rules/customer-discovery.md @rules/validation-experiments.md @rules/verdict-and-reporting.md @references/frameworks.md @references/flow-schema.md

Startup Validator

Reduce startup risk with evidence, not optimism. Score the idea, grade confidence separately, and end with the cheapest next learning step.

• Evaluate startup or product ideas with explicit evidence quality, uncertainty, and framework-backed scoring.
• Separate raw attractiveness from confidence so weak evidence cannot produce a high-confidence Go.
• Convert the biggest unknowns into customer-discovery questions, demand experiments, and kill criteria.
• Save a reusable multi-file validation report that can be resumed later.

<when_to_use>

Use this skill when:

• validating a new startup, product, feature wedge, or market entry idea
• deciding whether to proceed, narrow, pivot, stop, or run another validation sprint
• preparing for customer discovery, paid pilots, design partners, or fundraising conversations
• checking whether traction signals actually indicate PMF or only curiosity

Do not use this skill when:

• the main job is generating many new ideas
• the request is technical implementation planning or code work
• the user wants first-principles redesign rather than validation scoring
• the user only wants market research with no go/no-go, pivot, or validation decision

Boundary routing:

• Use genius-thinking for broad ideation without a concrete idea to evaluate.
• Use elon-musk for assumption teardown and first-principles redesign.
• Use research for source-backed market or trend research without a startup verdict.
• Use plan when the idea is already validated and the user wants implementation planning.

Positive examples:

/startup-validator B2B purchasing automation for mid-market finance teams
/startup-validator 이 아이디어가 진짜 고객 돈을 받을 수 있는지 검증해줘
/startup-validator PMF인지 아닌지 evidence 기준으로 평가해줘

Negative examples:

새 스타트업 아이디어 50개 뽑아줘
이 기능 구현 계획 짜줘

Boundary example:

제1원칙으로 사업모델을 완전히 다시 설계해줘
# Route to elon-musk unless the user asks for validation scoring or go/no-go judgment.

</when_to_use>

<input_check>

If the startup idea is missing, ask exactly one question:

Which startup or product idea should we validate?

If founder, market, customer, or traction evidence is missing, continue with explicit assumptions and low confidence instead of inventing certainty.

</input_check>

<owned_job>

For each run:

1. Frame the idea, customer, stage, current alternative, and desired decision.
2. Extract the riskiest hypotheses: problem, customer, value, distribution, monetization, defensibility.
3. Inventory evidence and tag each signal with the E0-E7 evidence ladder in rules/evidence-and-scoring.md.
4. Score the idea with the framework set in references/frameworks.md, keeping raw score and evidence confidence separate.
5. Apply customer-discovery quality gates from rules/customer-discovery.md.
6. Design the next validation sprint using rules/validation-experiments.md.
7. Produce a confidence-adjusted verdict with rules/verdict-and-reporting.md.
8. Run the validation checklist before marking flow.json complete.

</owned_job>

<document_shape>

Output Structure

.hypercore/startup-validator/[topic-slug]/
├── flow.json           # phase tracking, evidence confidence, next sprint state
├── thesis.md           # idea framing, ICP/persona, hypotheses, evidence inventory
├── thiel-scores.md     # venture-scale 7Q scoring with confidence and caveats
├── pmf-forces.md       # PMF stage, JTBD forces, VPC fit, customer pull signals
└── verdict.md          # raw score, confidence-adjusted verdict, sprint, kill criteria

• Use ASCII kebab-case for [topic-slug].
• If the folder exists, read existing files and resume from the last incomplete phase.
• Keep the four output files for compatibility; place richer sections inside the existing files instead of adding new top-level outputs.

</document_shape>

<flow_tracking>

Write flow.json at the start and update it as each phase completes. See references/flow-schema.md.

Phase	Output file	Completion signal
frame	thesis.md	target customer, current alternative, hypotheses, and evidence inventory exist
score	thiel-scores.md	7Q raw scores include confidence and score-change evidence
pmf	pmf-forces.md	JTBD/PMF forces and customer-pull signals are assessed
verdict	verdict.md	verdict, next 7-day sprint, and kill criteria are explicit

</flow_tracking>

Phase	Task	Output file
1	Frame idea, ICP/persona, stage, current alternative, and riskiest hypotheses	thesis.md
2	Score venture-scale potential and strategic risk with evidence confidence	thiel-scores.md
3	Evaluate customer pull, switching forces, VPC fit, and PMF stage	pmf-forces.md
4	Apply confidence gates, choose verdict, define sprint and kill criteria	verdict.md

Scoring rule:

• Raw score estimates attractiveness; confidence estimates evidence quality.
• E0-E2 evidence cannot produce high-confidence Go.
• PMF claims require qualified user behavior, not founder intuition or AI-generated personas.
• Missing evidence should lower confidence, not be filled with optimistic assumptions.

<output_contract>

Each output file must include:

• thesis.md: one-line thesis, target customer/ICP, buyer/user split when relevant, current alternative, value/growth/monetization hypotheses, top 5 riskiest assumptions, evidence inventory with E-levels
• thiel-scores.md: Engineering, Timing, Monopoly, People, Distribution, Durability, Secret scores; evidence confidence for each; score-change evidence; venture-scale caveats
• pmf-forces.md: JTBD story, Push/Pull/Habit/Anxiety, jobs/pains/gains fit, PMF stage, Sean Ellis/Superhuman readiness when active users exist, B2B/marketplace/deeptech caveats when relevant
• verdict.md: Go / Validate First / Narrow / Pivot / Stop, raw score, confidence-adjusted verdict, highest evidence level, critical weaknesses, next 7-day validation sprint, kill criteria, and “what would change my mind”

</output_contract>

Before finishing, verify:

• evidence quality is separated from opinion, enthusiasm, and AI-generated simulation
• raw score and confidence-adjusted verdict are both visible
• weak evidence cannot produce a high-confidence Go
• the score ties back to named frameworks and named evidence
• customer discovery questions avoid compliments, hypotheticals, and solution-first pitching
• the output includes concrete next validation actions, success metrics, and kill criteria
• all output files are saved under .hypercore/startup-validator/[topic-slug]/
• flow.json status is set to completed