research-methodology

name	research-methodology
description	Structured research process for investigating topics and producing findings. Use when exploring unfamiliar domains, evaluating options, or gathering evidence to inform decisions.
metadata	{"type":"protocol","invocation":"both","practice":null}

Research Methodology

Process

1. Frame — state the question clearly. What do we need to know? What would change our decision?
2. Survey — broad search across multiple sources. Don't commit to first result.
3. Evaluate — assess source quality. Primary > secondary. Recent > old. Specific > general.
4. Synthesize — combine findings into a coherent answer with citations.
5. Gap — explicitly state what you couldn't confirm or find.

Source Evaluation

When assessing an unfamiliar source, apply these criteria to place it in the authority hierarchy (see source-authority steering):

Criterion	Question	Effect
Accountability	Named author/org with reputation at stake?	Anonymous → cap at L6
Recency	Updated within the domain's decay window?	Stale → downgrade confidence
Specificity	Addresses your exact version/context?	Generic → note in relevance
Corroboration	Do independent sources agree?	Single-source → cap at "reported"
Incentive	Informing or selling?	Vendor marketing → note bias
Traction	Stars, forks, dependents, citations?	Prioritize investigation order; evaluate fit independently

Rules

• Check at least 3 sources before concluding
• Prefer primary sources over summaries
• Use traction signals to prioritize investigation order, not as quality verdicts
• Tag all citations per source-authority steering: [L{n}:{confidence}]

Budget

• Cap at 8-10 web searches per research question
• After 3 consecutive searches returning no new information, stop and synthesize
• Prefer cloning repos (--depth 1 to .scratch/references/) over fetching individual files via HTTP
• State gaps explicitly rather than searching indefinitely

Output Format

Use the research-output skill's template — it is the canonical format for all research findings (frontmatter with topic/date/status/confidence; Summary, Findings, Sources with [L{n}:{confidence}] tags, Related Topics, Related Tools, Open Questions). Do not invent an inline structure; one format keeps findings reusable across sessions and consumable by other agents.

Anti-Patterns

• Answering from memory without checking sources
• Citing one source as definitive
• Presenting "likely" as "confirmed"
• Stopping at the first result that seems right
• Mixing findings with recommendations without separating them

References

• For parallel research dispatch pattern (subagent per topic), read references/dispatch-pattern.md