# Rai Research

> Conduct epistemologically rigorous research to inform decisions. Use before ADRs, when evaluating competing approaches, entering unfamiliar domains, or resolving parking lot items. Produces evidence catalogs with triangulated claims and actionable recommendations.

- Skill: `stefancgillberg/rai-research` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add stefancgillberg/rai-research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/stefancgillberg/rai-research/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- License: MIT
- Author: stefancgillberg (https://skillmd.com/u/stefancgillberg)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/stefancgillberg/rai-research

---


# Research

## Purpose

Conduct epistemologically rigorous research to inform decisions. Standing on the shoulders of giants, not reinventing wheels.

## Mastery Levels (ShuHaRi)

- **Shu**: Follow all steps with full evidence catalog and research prompt template
- **Ha**: Scale depth to decision importance; adapt prompt template
- **Ri**: Create domain-specific research protocols and custom prompts

## Context

**When to use:** Before ADRs, when evaluating competing approaches, entering unfamiliar domains, or resolving parking lot items.

**When to skip:** Decision is low-stakes and reversible, or prior research exists in `work/research/`.

**Inputs:** Clear research question(s), decision context, depth constraint (quick/standard/deep).

| Depth | Time | Sources | Use when |
|-------|------|---------|----------|
| Quick scan | 1-2h | 5-10 | Low-stakes, familiar domains |
| Standard | 4-8h | 15-30 | Most ADRs, technology evaluation |
| Deep dive | 2-5d | 50-100+ | Strategic decisions, unfamiliar domains |

## Steps

### Step 1: Frame the Question

Define: primary question, secondary questions, decision this informs, depth constraint.

**Epistemological principles:** Seek disconfirming evidence (falsifiability), require 3+ sources per claim (triangulation), primary > secondary > tertiary sources.

<verification>
Question is specific and falsifiable.
</verification>

<if-blocked>
Question too vague → decompose into sub-questions.
</if-blocked>

### Step 2: Select Tool & Survey

**Tool selection:**

| Tool | Best for |
|------|----------|
| `ddgr "query"` | Quick scans, no API key needed |
| `llm -m perplexity "query"` | Deep research with citations |
| WebSearch | Reliable fallback |

Gather sources: academic papers, official docs, GitHub repos (stars/activity), engineering blogs, community discussions.

<verification>
10+ sources collected (scaled to depth).
</verification>

### Step 3: Build Evidence Catalog

Per source: type (primary/secondary/tertiary), evidence level, key finding, relevance.

| Evidence level | Criteria |
|---------------|----------|
| Very High | Peer-reviewed, production-proven at scale, >10k stars |
| High | Expert practitioners at established companies, >1k stars |
| Medium | Community-validated, emerging consensus, >100 stars |
| Low | Single source, unvalidated, <100 stars |

Save to `work/research/{topic}/sources/evidence-catalog.md`.

<verification>
Evidence catalog created with levels rated.
</verification>

### Step 4: Triangulate & Synthesize

Per major claim: find 3+ independent confirmations, note consensus vs disagreement, assign confidence (HIGH/MEDIUM/LOW), acknowledge contrary evidence explicitly.

Extract patterns: convergence points, gaps, RaiSE-specific vs general findings.

<verification>
Major claims have 3+ sources. Contrary evidence documented.
</verification>

### Step 5: Recommend & Link

Produce: recommendation with confidence level, trade-offs, implementation implications, risks.

Connect to governance: create/reference ADR if architectural, update backlog if actionable, update parking lot if deferred.

<verification>
Recommendation is actionable and traces to evidence.
</verification>

## Output

| Item | Destination |
|------|-------------|
| Report | `work/research/{topic}/{topic}-report.md` |
| Evidence catalog | `work/research/{topic}/sources/evidence-catalog.md` |
| Navigation | `work/research/{topic}/README.md` |
| Next | ADR, backlog item, or parking lot update |

## Quality Checklist

- [ ] Research question is specific and falsifiable
- [ ] 10+ sources consulted (scaled to depth)
- [ ] Evidence catalog created with levels rated
- [ ] Major claims triangulated (3+ independent sources)
- [ ] Confidence level explicitly stated on recommendation
- [ ] Contrary evidence acknowledged (not hidden)
- [ ] Governance linkage established (ADR, backlog, or parking lot)
- [ ] NEVER present single-source findings as consensus

## References

- Research prompt template: `references/research-prompt-template.md`
- Existing research: `work/research/`
- Epistemology: falsifiability, triangulation, source hierarchy