Product Customer Discovery

Priority: P0 (CRITICAL)

Capture rigorous, non-leading customer evidence anchored in observable behavior and informed consent, without fabricating interviews, leading subjects, or misrepresenting small samples.

1. Capability Preflight

Verify product-evidence capability before validating or committing changes to docs/product/<slug>/evidence.json:

1. Skill Resolution: Load through MCP (get_skill("product-management", "product-evidence")) or resolve installed root paths <root>/product-management/product-evidence/SKILL.md, <root>/product-evidence/SKILL.md, or flattened Kiro <root>/product-management-product-evidence/SKILL.md. A missing/excluded skill is a capability gap; never infer installation from source checkout.
2. Companion Resource: Resolve scripts/validate-product-evidence.mjs relative to the loaded package directory (ordinary <root>/product-management/product-evidence/; flattened <root>/product-management-product-evidence/). Missing companion is a separate resource gap.
3. Runtime: Check Node.js >=20 separately. Missing runtime is not a missing skill/resource.
4. Offline Mode: If any prerequisite is absent, safe research/proposal work may proceed only as unvalidated; no simulated validation, readiness claim, or silent sync/install.

2. Core Protocol

1. Discovery Scoping & JTBD Framing:
   • Frame research around customer jobs-to-be-done (JTBD) and current friction points.
   • Establish explicit cohort definitions, inclusion/exclusion criteria, and target observation window (start, end).
2. Ethical Sampling & Consent Boundaries:
   • Obtain and record explicit informed consent for all participant sessions.
   • Mandate privacy redaction: never store raw PII, customer credentials, or non-redacted participant identities.
   • No Access Rule: If real customer access is unavailable, report an explicit research gap. Output proposal-only interview guides or recruitment plans; never generate synthetic interviews or pretend AI simulation is customer evidence.
3. Non-Leading Inquiry & Observation:
   • Focus questions on past behavior, actual spending, and concrete friction ("When was the last time you...?", "Show me how you currently...").
   • Ban leading or speculative prompts ("Would you buy X if we built it?", "Do you love this concept?").
   • Record observable actions, system logs, and direct quotes separately from observer interpretations.
4. Observation vs. Inference Tagging:
   • Label direct quotes and observed actions as confirmed facts only when backed by an available, de-identified transcript SRC. If transcript access is protected/unavailable, dependent claims are inference with documented rationale.
   • Label user needs, motivations, and summaries as inference with documented rationale.
   • Separate sample frequency (e.g. "4 of 5 participants") from market percentage; small qualitative cohorts never prove population distribution.
5. Quote-to-Insight Ledger Output:
   • Emit valid SRC, CLM, and candidate OPP records conforming to docs/product/<slug>/evidence.json. All training ledgers must be explicitly marked synthetic: true.

3. Anti-Patterns & Red Flags

• No Synthetic Customers: Never invent personas, quotes, or transcripts as actual customer evidence.
• No Leading Prompts: Never pitch features during discovery interviews.
• No Small-Sample Generalization: Never convert "3 out of 4 interviewed users" into "75% of the market wants X".
• Stop if Consent Missing: If session lacked consent or recording rights, discard data from ledger intake.

4. References

• Discovery Protocol — Non-leading interview techniques and JTBD inquiry.
• Sampling and Ethics — Participant recruitment, consent, and PII protection.
• Evidence Mapping — Schema mapping to SRC, CLM, and candidate OPP records.
• Citations and Sources — Conceptual citations and independent authorship statement.
• Semantic Rubrics — Outcome-based grading rubrics and false-pass rejection criteria.
• Synthetic Example Ledger — Complete valid standalone synthetic ledger.