Research
The Research department answers specific questions that block product, design, or business decisions. Research runs in parallel with all other departments and can be triggered any time an open question needs evidence.
When to involve Research
Research is triggered by open questions in other depts' documents. Look for "Open Questions" sections in the PRD, SRS, or design briefs. If a question affects a key decision and no one has data, Research produces a finding doc.
What Research produces
| Output | Location | Purpose |
|---|---|---|
<topic>_finding.md |
research/mvp/ |
Evidence + recommendation for one research question |
One file per question. Each file follows templates/research_finding.md.
How to operate
Setup (one time)
cd ~/code/nepwalk
git clone <nepwalk-docs-url> nepwalk-docs
git clone <nepwalk-research-url> nepwalk-research
cd nepwalk-research
./scripts/sync-from-docs.sh
```text
> **Windows users:** Run these commands in **Git Bash** or **WSL** — not PowerShell or CMD.
Open `nepwalk-research/` in your AI tool.
### Session loop
**Check open questions across all depts:**
research_status:
The AI scans `nepwalk-docs` for "Open Questions" sections in management, design, and marketing docs and lists unanswered ones.
**Start a research question:**
research: start "do Nepal tour operators have reliable Google accounts" research: start "what is the acceptable loading time for operators on 4G"
The AI creates a finding doc from the template, filling in the research question, method suggestion, and evidence structure.
**Fill in evidence and findings:**
Run interviews, pull analytics, or do desk research. Fill in the Evidence Summary and Findings sections directly in `nepwalk-docs/docs/research/mvp/<topic>_finding.md`.
**Publish:**
research: publish research/mvp/google_accounts_finding.md
The relevant dept is notified via the PR. They update their open question with the finding reference.
---
## Research question format
Good research question: *"Do Nepal-based tour operators have reliable access to a Google account?"*
This is answerable, specific, and tied to a decision (whether to support Google OAuth only or add email/password login).
Bad research question: *"What do users want?"*
Too broad — breaks into 10 different questions, produces no actionable finding.
---
## Specialist agents
| Agent | When to use |
|---|---|
| `product-trend-researcher` | Market sizing, competitor analysis, industry data |
| `product-feedback-synthesizer` | Interview guides, usability findings, behavioural patterns |
research: agent product-feedback-synthesizer "draft interview guide for Nepal tour operator discovery"
```text
Current work
Browse mvp/ for research finding docs.