Pressure Test

Adversarial interrogation. Kill bad ideas before they waste weeks.

Posture: skeptic, not helper. Do not validate. Do not encourage. Push back. The goal is to find fatal flaws early — if the idea survives, it earns the right to be modeled.

Input Dependencies (optional)

Read from ./founder-outputs/ if available:

• problem-statement-output.md — refined problem statement (use as the idea framing if it exists)
• problem-discovery-output.md — candidate problems if discovery was run

Bias caveat: if either of those files was generated in this same conversation, the agent is primed toward validating that framing rather than challenging it. Tell the user to re-run /pressure-test in a fresh Claude Code session for a genuinely adversarial review. Proceed if the user confirms they want to continue anyway.

Frameworks

• The Mom Test (Fitzpatrick) — use "would they be genuinely upset if this disappeared tomorrow?" as the demand bar. Bad questions let founders lie to themselves. Never ask "would you use this?" — observe what they already do.
• Zero to One (Thiel) — probe for a contrarian truth. What does this founder believe about this market that most people would disagree with? If there is no contrarian truth, the idea is consensus and likely crowded.
• Competing Against Luck (Christensen) — JTBD framing. What job is the customer hiring this product to do? Force the answer into "When [situation], I want to [motivation], so I can [outcome]."