Values clarification interview (agent script)

Use this checklist in order. Paraphrase; do not skip red lines. Map answers into a values profile (schema/values_profile.schema.json; see data/examples/values_interview_transcript.example.json). See also docs/discernment-flow.md.

Before you start

• State that recommendations use published evaluator evidence and that impact figures are modeled, not guaranteed.
• Do not invent cost-effectiveness numbers.
• Do not contact organizations on the donor’s behalf.

Steps

1. Outcome family — Lives saved, wellbeing (WELLBYs / mental health), income, education, animals, climate, policy, other? Ask for a ranked preference if they care about more than one.
2. Risk / evidence appetite — Top-charity certainty vs promising / higher uncertainty.
3. Geography — Unrestricted vs preferred regions / exclusions.
4. Tax & vehicles — US 501(c)(3), DAF, employer match, crypto, direct international.
5. Non-cash stewardship — Time, skills, networks, in-kind (map to non_cash_giving / gv_time_skills).
6. Red lines (required) — Causes, methods, or orgs they refuse. Confirm before ranking.
7. Budget band (optional) — Helps scale modeled estimates; never required for browsing.
8. Confirm profile — Read back the structured profile; offer edits.
9. Rank & explain — Re-rank org_* / funds with citations; attach modeled impact only from verified claims.
10. Human loop — Escalate high-dollar gifts or contested claims.

After

Persist the transcript + values_profile for the session. Prefer evaluator funds when tax/deductibility or allocation confidence is the constraint.