expert-interview-synthesis
Plan decision-focused expert interviews or synthesize supplied interview transcripts and notes into traceable findings, contradictory views and evidence gaps. Use for diligence, customer discovery and
SKILL.mdskills/expert-interview-synthesis/SKILL.md
GitHubExpert Interview Synthesis
Connect what was actually said to a bounded finding and a decision. Work from supplied material; preparation does not authorize interviews, recording, outreach or publication.
Inputs and Modes
- Prepare: decision question, hypotheses, participant roles, access constraints and confidentiality requirements. Produce a guide, not invented interview results.
- Synthesize: notes/transcripts, source IDs, participant roles, dates and the decision. Flag missing timestamps or incomplete notes; do not create verbatim quotes from a summary.
Read references/evidence-method.md for coding, provenance, contradictions and confidence. Read references/source-notes.md when checking the research basis or updating guidance. Copy assets/interview-guide.md and assets/interview-evidence.csv for editable outputs.
Workflow
- Establish the decision and the few questions that could change it. Recruit relevant contrasting roles if recruitment is requested; record convenience sampling, conflicts of interest and missing perspectives.
- Prepare open, neutral questions about a specific recent event. Probe sequence, consequences, numbers and alternatives without embedding the desired answer. Include a contrary-case question and confirm scope, consent and recording arrangements.
- Extract atomic evidence. For each item, retain source ID and page/timestamp/paragraph, role, period, exact quote or labelled paraphrase, and whether it is firsthand experience, estimate or hearsay.
- Code evidence into themes without collapsing raw material into conclusions. Count unique participants, not repeated comments. Deduplicate accounts of the same event or vendor claim before calling evidence independent.
- Separate observation, participant claim and analyst inference. Keep contradictory and minority views in a conflict register; investigate segment, period, wording and incentives before choosing an interpretation.
- Assess confidence from directness, independence, coverage and corroboration. Several experts repeating one source are not several independent confirmations. Interview agreement alone does not verify a financial or market number.
- Write a short synthesis: finding, evidence IDs, counterevidence, scope/confidence, decision implication and next verification. Carry unverified claims forward as unverified, even if they support the recommendation.
Output and Handoff
| Artifact | Required distinction |
|---|---|
| Interview guide | Research question versus neutral prompt and follow-up |
| Evidence ledger | Literal quote versus paraphrase; claim versus observation/inference |
| Conflict register | Disagreeing sources, plausible explanation, unresolved test |
| Findings memo | Scoped finding, supporting/contrary IDs, confidence and decision consequence |
For downstream market analysis or diligence, send anonymized source IDs, definitions, dates, participant counts and unresolved verification tasks—not a cleaned-up consensus that hides conflict.
Guardrails
- Never fabricate interviews, participants, quotes, timestamps or corroborating sources.
- Do not infer market percentages from a small purposive sample. “3 of 5 interviewees” describes that sample only.
- Keep firsthand knowledge, estimates and hearsay distinct; do not upgrade a quotation into verified fact.
- Do not treat “I don't know” as misconduct. Accept the limit, seek a narrower question or another source.
- Do not request trade secrets, confidential employer/client material or personal data irrelevant to the decision. Preserve consent and anonymization constraints; seek specialist advice for regulated or investment-sensitive information.
- Do not send follow-up messages, record sessions or share identifiable notes without explicit permission.