Idea Sprint

Interview → necessity gate → 3 candidates → score → go/no-go. Output is a decision, not a brainstorm.

Detail lives in references/. The only upstream links are the Superteam ideas dataset and colosseum-copilot; see upstream-packs.md.

Context handoff

• At start: read .claude/context/idea.md and .claude/context/build.md (format, written by build-status) if present — resume from prior state instead of re-interviewing. On a re-run, also read .claude/context/positioning.md (format, written by positioning) if present: its ICP and alternatives are inputs to the Distribution and Market pull scores, not evidence (score market pull from the customer-signal rubric), and a no-go or pivot means it needs redoing — say so.
• On completion: write/update .claude/context/idea.md with the chosen idea, scores, validation evidence, and open risks. Downstream skills (pitch-deck, hackathon) read it.

Workflow

1. Blunt interview

No flattery. Short, pointed questions, one at a time. Hard gate before any idea generation:

• Edge — what the founder knows/can do that most can't (domain, distribution, tech)
• Constraint — what they are optimising for now: time, money, team, chain commitments
• Crypto touchpoint — which part of the product genuinely benefits from being onchain

Then pin the wedge: the niche entry point, not the end-state vision. "DeFi for everyone" is not a wedge. Probes by starting point: interview-framework.md.

2. Crypto-necessity gate

Kill question: "What gets worse if I remove the blockchain?" If the answer is vague, aesthetic, or marketing-driven — redirect the idea before scoring it. Pass criteria, redirect patterns and the incumbent check: crypto-necessity-test.md.

3. Exactly 3 candidates

Generate three diverse candidates (different mechanisms/markets, not three flavors of one idea; diversity rule in scoring-rubric.md). For each:

• One-line pitch + target user
• Winner case — what's true in 18 months if it works
• Bear case — the most likely way it dies

Seed from the dataset + live landscape (below), then combine with fresh research. Datasets are inspiration, not constraints.

4. Score /15

Each candidate, 0–3 per dimension (full anchors and tie-breaks: scoring-rubric.md):

5. Validate + go/no-go

Check demand signals against customer-signal-rubric.md — manual workarounds, active forks, bounties, on-chain activity = real; likes and "cool idea" replies = noise. The same file has the user-conversation questions: ask about past behaviour, never "would you use it?".

• ≥ 8/15 → go. Write idea.md, then scaffold the project — solanabr/ai-kit ships a /scaffold command for this if it is installed.
• 6–7 → conditional: name the one dimension to de-risk first.
• < 6 → strong no-go. Every no-go gets a pivot suggestion — use pivot-or-persist.md, which also gives how to state confidence.

This score is the go/no-go rule. The one override is a hard no-go from pivot-or-persist.md (demand 0, a well-funded team already shipped it, unsolved core tech, ornamental crypto), which stops the idea at any score.

6. Write .claude/context/idea.md

Chosen idea, wedge, scores table, demand evidence, bear case, next step. Format: idea-md.md.

Idea dataset

superteam-ideas.json: 240 Solana-native ideas as inert JSON (every entry carries the same published date, 2025-10-14, so it doesn't date the idea). Inspiration only.

For other idea lists, read the publishers' own pages (a16z crypto, YC's Requests for Startups, Alliance) rather than a scraped copy, and date anything you cite.

Live hackathon landscape

colosseum-copilot — 5,400+ Colosseum submissions for crowdedness checks, winner patterns, and gap analysis. CLI-backed: sign in once per machine with npx @colosseum-org/copilot-connect login (Node 20+). Install routes: upstream-packs.md. Dated list of past Grand Champions and track winners: hackathon winners.