Interview

/interview

Read brain/problem.md. If empty, refuse — "No problem hypothesis yet. Run /problem first."

Two modes: PRE and POST

Ask the founder which: "Are we (a) prepping for an upcoming interview, or (b) extracting signals from one you just did?"

PRE mode — generate the script

Produce an interview-discipline script tailored to the current problem hypothesis. The script focuses on past behavior, not opinions about the future.

Interview-discipline rules embedded in every script:

1. Talk about their life, not your idea
2. Ask about specifics in the past, not generics or opinions about the future
3. Talk less, listen more

Required sections of every script:

## Opening (2 min)
- Who they are, what they do
- Set expectation: "I'm researching, not selling. Just want to learn."

## Problem discovery (15 min) — past behavior only
- "Walk me through the last time [problem context happened]."
- "What did you do when that happened?"
- "How often does that come up?"
- "What's the worst part of that for you?"
- "What have you tried to fix it?"
- "How much does it cost you — in time or money?"

## Believer test (5 min) — earlyvangelist signals
- "Have you ever paid for anything to solve this?"
- "What would 'solved' look like for you?"
- "Who else has this problem?"

## Close (3 min)
- "Anything else I should be asking?"
- "Can I follow up in [N] weeks?"
- "Can you connect me with one other person who has this?"

Add 2-3 questions specific to THIS founder's hypothesis (e.g., if hypothesis is about late invoices, add: "Walk me through the last invoice you had to chase. What did that day look like?").

Anti-questions (warn the founder NOT to ask these):

• "Would you pay for a product that does X?" (opinion about future, useless)
• "Do you think this is a problem?" (compliments, useless)
• "Would you use [your solution]?" (sells, doesn't learn)

POST mode — extract signals

Founder pastes interview notes or transcript. Extract these atomic signals only:

Verbatim quotes — exact words about the pain, current alternatives, or willingness to pay.
Observed behaviors — what they DO, not what they SAY (already paid for X, tried Y three times, runs a workflow on Sundays).
Money already spent — on workarounds, adjacent tools, manual labor.
Frequency mentions — every Friday, every quarter, twice a week.

Earlyvangelist test (Blank) — score the participant

Five criteria. Yes/No each:

• Has the problem
• Knows it's a problem (named it specifically, not led)
• Has tried to solve it (workarounds, other tools, hiring)
• Has budget (already spends on adjacent tools)
• Would pay for the solution (in their own words, not led)

5/5 = BELIEVER. 3-4 = NEUTRAL. 0-2 = INFIDEL.

Write to brain/interviews/N-<short-name>.md

---
date: <today>
participant: <name, role, company, segment>
recruited_via: <how you found them>
classification: BELIEVER | NEUTRAL | INFIDEL
---

## Pain admitted (verbatim quotes only)
> "..."
> "..."

## Behavior observed
- ...
- ...

## Earlyvangelist test (Blank)
- Has the problem: yes/no
- Knows it's a problem: yes/no
- Has tried to solve it: yes/no
- Has budget: yes/no
- Would pay: yes/no
→ Classification: BELIEVER / NEUTRAL / INFIDEL

## Hypothesis impact
- Confirms: ...
- Contradicts: ...
- Action: ...

Update related files

• Append the participant to brain/believers/index.md (create if missing) with their classification and one-line summary
• If the interview contradicts brain/problem.md, suggest /problem to refine

Re-evaluate the hypothesis

After every 3 interviews, surface patterns:
"3 of 5 mentioned X — your script doesn't ask about it. Add?"
"60% of believers are in segment Y. You might have a niche, not a market. Lean in or expand?"

Stage gate check

If brain/believers/ now contains 5+ entries marked BELIEVER, suggest advancing brain/stage.md from 1 to 2. Don't auto-advance — surface the option:
"5 believers identified. Stage 1 → 2 gate met. product/ writes will become allowed (still coached against evidence). Advance?"