Research Repo Setup

Research Repo Setup Skill

Teams pay for the same research repeatedly: the churn interviews from last year answer this quarter's question, but they live in a deck nobody remembers, so the question gets re-studied at full price. The repo fixes the unit — not "studies" (monolithic decks that answer only their original question) but atomic insights: one finding + its evidence + source + date + tags, individually findable by future questions the original study never anticipated. Around the unit: an intake funnel (every study deposits its insights while fresh) and the reuse norm (research requests start with a repo check), because a repo consulted by nobody is a very organized graveyard.

What This Skill Produces

• The repo structure — where it lives (searchable > beautiful), the insight-entry format, the study-index layer
• The atomic format — finding (one sentence) · evidence (the support, with its n) · source study + date · tags · confidence
• The intake funnel — the end-of-study deposit ritual (interview-synthesis themes flow in directly)
• The reuse norms — repo-check-first on new research requests, and the citation habit that keeps entries alive

Required Inputs

Ask for these if not provided:

• The research backlog — what studies exist (decks, docs, transcripts) and which still get asked about; the back-fill starts with the asked-about, per the greatest-hits rule
• The question patterns — what the team repeatedly wants to know (segments? churn drivers? feature reactions?); tags are designed from future questions, not past study titles
• The platform — wiki, database tool, docs; filterable-by-tag and full-text-searchable are the two requirements, everything else is taste
• The research producers and consumers — who deposits, who should check first; the norms name both sides

Framework: The Repo Rules

1. Atomize or archive: the unit is the insight, one per entry — "Enterprise buyers involve security review before contract (8/12 interviews, unprompted; Q3-2025 churn study)" — findable by someone asking about security, buying process, or enterprise, none of which the study title mentioned. Whole decks get indexed (the study layer) but the insights are the retrieval surface.
2. Every entry carries its epistemics: the n, the sample shape, the date, and the confidence grade (source-triangulation levels work) — because a 2023 insight from six interviews should be trusted differently than last month's survey of four hundred, and the entry must let a stranger tell.
3. Tag for future questions: the tag set comes from the team's recurring question patterns (segment, lifecycle stage, topic, product area) — small, controlled, gardened (knowledge-gardening owns the vocabulary drift). Free-form tagging balkanizes into synonyms within a quarter.
4. The funnel is the study's last step: every research effort ends with the deposit — themes in atomic format, 20 minutes while fresh — wired into the synthesis ritual the way decision-log-setup wires into meetings. Retroactive depositing never happens; the funnel or nothing.
5. Reuse is a norm with a moment: every new research request starts with the repo check ("what do we already know?") — findings that answer it get cited with their dates ("per Q3-25, revalidate if stale"), and partial answers reshape the new study to fill gaps instead of re-covering ground. The repo's KPI is re-research avoided, visible in study briefs that open with "what the repo already says."