---
name: run-product-discovery
description: Run an evidence-led product discovery sprint that turns one live product question into a traceable go, narrow, or kill decision. Use when a user needs to validate a problem, plan or conduct customer interviews, test demand, synthesize research notes, challenge a product idea, or create a demand memo before building.
---

# Run Product Discovery

Turn one product question into a bounded decision using recent behavior, observable workarounds, and costly signals. Treat `GO` as permission to fund the next small test—not permission to build the whole product.

Read [references/templates.md](references/templates.md) before starting. Reuse its artifact structures rather than inventing new formats.

## Operating contract

- Ask one setup question at a time when inputs are missing.
- Keep the problem statement free of solution language.
- Separate supplied facts, participant evidence, and interpretation.
- Cite participant IDs or source links for every synthesis claim.
- Preserve counterevidence and segment differences.
- Never invent an interview, quote, behavior, cost, or commitment.
- Never label stated preference as validated demand.
- Stop after preparation when no real evidence has been supplied.
- Let the human own consent, interviews, materiality judgments, and the final decision.

## Required inputs

Collect:

1. Decision in the form `Should we…?`
2. Target segment and explicit exclusions
3. Triggering event and recency window
4. Riskiest assumption
5. Existing evidence and constraints
6. Decision owner and decision date
7. Access to five to seven qualified participants

If access to qualified participants is missing, produce the discovery brief, screener, and outreach message, then stop. Do not synthesize imaginary evidence.

## Run the five-day sprint

### Day 1 — Frame and recruit

Create `discovery/brief.md`, `discovery/screener.md`, and a participant plan.

- Write one decision and one problem hypothesis.
- Name the segment, trigger, current workaround, material consequence, and riskiest assumption.
- Pre-register the evidence required for go, narrow, and kill before seeing results.
- Recruit six people plus two backups from one intentionally bounded segment.
- Screen for a recent trigger and actual behavior, not interest in the proposed idea.

### Day 2 — Run the first three interviews

Create one source-linked note per participant using the interview template.

- Ask about the last real occurrence.
- Reconstruct the workflow from trigger to outcome.
- Ask to see the artifact or workaround when appropriate and consented.
- Quantify time, money, rework, trust, risk, or missed opportunity.
- Do not pitch and do not ask `Would you use or pay for this?`
- Complete notes within 15 minutes of each call.

### Day 3 — Run three more and seek disconfirmation

- Repeat the same core questions.
- Recruit or prioritize someone likely to falsify the emerging pattern.
- Keep questions stable; do not rewrite them to confirm yesterday’s story.
- Record contradictions, non-events, and segment splits explicitly.

### Day 4 — Synthesize evidence

Create `discovery/evidence-ledger.md` and `discovery/demand-memo.md`.

- Atomize findings into claims.
- Link every claim to participant IDs and raw sources.
- Label each item `observed`, `quoted`, `inferred`, or `contradicted`.
- Cluster only patterns supported by three or more independent participants.
- Split unlike segments instead of averaging them.
- Score evidence strength with the ladder in the templates reference.

### Day 5 — Decide

Create `discovery/decision.md`.

Use these small-sample heuristics as decision gates, not statistical proof:

- At least five qualified interviews
- At least four recent concrete instances
- At least three active workarounds
- At least three material consequences
- At least two strong commitment signals
- No critical trust or feasibility blocker

Choose exactly one:

- `GO`: all gates pass for one coherent segment. Fund the smallest solution or feasibility test.
- `NARROW`: gates pass only for a subgroup or one critical assumption remains. Rewrite the segment or job and name one next test.
- `KILL`: fewer than three recent instances, weak workaround or cost, or a critical assumption is falsified. Stop the bet and preserve the learning.

Include evidence IDs, counterevidence, threshold results, unknowns, next owner/date, and a reversal trigger.

## Divide human and agent work

Use the agent to:

- Draft neutral screeners, outreach, and interview guides.
- Structure supplied transcripts and notes.
- Extract evidence with participant ID, timestamp, and source.
- Maintain the evidence ledger and count patterns.
- Surface contradictions and possible segment splits.
- Draft go, narrow, and kill options.

Require the human to:

- Set the decision and thresholds.
- Recruit participants and obtain consent.
- Lead interviews and probe nuance.
- Judge materiality and context.
- Make the final go, narrow, or kill decision.

## Reject these shortcuts

- Pitching during discovery
- Asking hypothetical willingness-to-use or willingness-to-pay questions
- Recruiting friends or people outside the target segment for convenience
- Treating compliments as demand
- Moving thresholds after seeing results
- Averaging different segments
- Promoting one vivid quote into a pattern
- Accepting summaries without raw evidence
- Presenting a six-person sprint as market proof

## Finish

Return:

1. Decision and pre-registered gates
2. Artifact paths created or updated
3. Evidence collected versus missing
4. Strongest pattern and strongest counterevidence
5. Current go, narrow, or kill decision—or `NOT READY` when evidence is incomplete
6. The exact next action, owner, and date