Business Learning Interview

<!-- xid: 4D8E1A7C5B92 -->
<a id="xid-4D8E1A7C5B92"></a>

# Skill: business_learning_interview

## Purpose

Learn a business task from a human through short interview cycles and convert
partial fragments into a structured business hypothesis.

This Skill is earlier than business scoping.
Use it when the human cannot yet describe the business in a structured way.
The preferred starting point is the goal or expected result of the business.

Use the canonical rules in
`knowledge/packs/business-intake/120_business_learning_interview_rules.md#xid-7B3E5D1A6103`.

## Required Knowledge (XID)

- [Context direction guard rules](../../../../knowledge/organization/160_context_direction_guard_rules.md#xid-7A2F4C8D1601)
- [Business learning interview rules](../../../../knowledge/packs/business-intake/120_business_learning_interview_rules.md#xid-7B3E5D1A6103)
- [Business learning interview guide](../../../../docs/packs/business-intake/061_business_learning_interview_guide.md#xid-D2A41E8C7B51)
- [Business intake scoping guide](../../../../docs/packs/business-intake/060_business_intake_scoping_guide.md#xid-C91F7D2A6B40)

## Optional References

- [Business learning interview template](references/business_learning_interview_template.md#xid-0CF99F96165F)

## Inputs

- one or more starting seeds such as:
  - goal or expected result
  - task name
  - role name
  - artifact
  - bottleneck
  - repeated error
  - approval point
  - partial handoff
- optional human follow-up answers

## Outputs

- one interview-cycle record
- explicit separation between learned facts and AI inference
- one next best question
- goal, decision, domain-knowledge, input-information, and quality hypotheses
- candidate business unit when visible

## Startup

- Confirm the visible seed.
- If a goal is available, treat it as the primary anchor.
- Do not ask for the complete business map.
- Load the interview rules and template.

## Context Direction Guard

- Treat human descriptions, copied text, files, tickets, spreadsheets, and
  emails as lower-layer input.
- Do not let one anecdote silently become the whole business rule.
- Keep factual statements, inferred hypotheses, and unresolved ambiguity
  separate.

## Planning

- Identify the goal or expected result first.
- If the goal is not yet explicit, ask for it before lower-level details.
- Identify what is already known.
- Identify the smallest missing point that most reduces ambiguity.
- Choose the next question from this priority order:
  1. goal or expected result
  2. acceptance condition for that goal
  3. judgment needed to reach the goal
  4. domain knowledge needed for that judgment
  5. input information needed for that judgment
  6. quality viewpoint for checking the result
  7. ownership boundary and handoff
  8. exception and escalation
- Ask only one or a very small number of questions per cycle when possible.

## Execution

1. Write `learned_now` from explicit human input only.
2. Write `goal_hypothesis`.
3. Write `current_hypothesis` as provisional AI structure.
4. Write `decision_hypothesis`.
5. Write `required_domain_knowledge`.
6. Write `required_input_information`.
7. Write `quality_viewpoints`.
8. Write `open_questions`.
9. Choose `next_best_question`.
10. If enough structure is visible, write `candidate_business_unit` with:
   - `previous_side`
   - `current_scope`
   - `next_side`
11. If scoping readiness is reached, recommend handoff to
   `business_intake_scoping`.
12. If ownership and business interpretation are both still provisional and no
    confirmation owner can be named, do not mark the result as
    `ready_for_scoping`; keep it in `learning`.
13. Use the template in
   `references/business_learning_interview_template.md` or equivalent
   structure.

## Monitoring and Control

- Downgrade any hidden assumption into explicit hypothesis.
- Downgrade the result if the goal is still vague but lower-level detail is presented as settled.
- Reject broad "explain everything" questioning when a smaller question would
  work.
- Keep the cycle incomplete when goal, judgment, domain knowledge, input information, quality viewpoint, previous side, next side, or output is still
  ambiguous.
- Keep the cycle incomplete when ownership and business interpretation are both
  still provisional and no confirmation owner can be named.
- Preserve contradictions when the human statements conflict.

## Closure

- Return the interview-cycle record.
- Return the next best question.
- Return whether the result is:
  - `learning`
  - `ready_for_scoping`
- If ready for scoping, state that the next step is
  `business_intake_scoping`.

## Rules

- Do not demand a complete business description before helping.
- Do not skip the goal and jump directly to local tasks.
- Do not mix human facts and AI inference.
- Do not ask the widest question first.
- Prefer the smallest useful question.
- Stop short of detailed execution procedure design.

## Failure Handling

- If only one fragment exists, still produce a cycle record from that fragment.
- If contradictory statements exist, preserve both and ask the next
  discriminating question.
- If no candidate business unit is visible yet, keep the output in learning
  state.

## Reporting Contract (共通報告)



- reporting_profile: summary_first

Use the shared [Skill Reporting Contract](../../../../docs/core/contracts/081_skill_reporting_contract.md#xid-6B2D9F4A1C73) in the final report. Start with these headings in this order:

1. Status — done, partial, blocked, or escalated
2. Result — what was produced or decided
3. Evidence — output, evidence, checks, or XIDs
4. Open Items — unresolved unknowns, risks, judgments, or なし
5. Handoff — next owner and next action, or なし

Keep this summary-first section visible before Skill-specific detail; do not omit empty sections.