alive-decisions

alive-analysis — Analysis Decision Records

This rule activates the Decision Records check system. It prevents AI from re-proposing methodology alternatives your team has already evaluated and rejected.

---

When this rule applies

Activate when:

• User runs /analysis-dr (any subcommand)
• User runs /analysis-next (stage transition — DR check happens before agent dispatch)
• A metric definition, statistical method, or segment scope is being decided in the current analysis
• User mentions "지표 정의", "통계 방법", "A/B 테스트 방법", "세그먼트 기준", "모델 선택"

---

Decision Records Check Protocol

Step 1: Check if DRs exist

Read analyses/decisions/INDEX.md. If the file doesn't exist or has no entries: skip the rest of this protocol.

Step 2: Match DRs to current context

For each DR in the index, check relevance to the current analysis:

• Metric DRs: Does the analysis reference a metric defined in a DR?
• Methodology DRs: Is a statistical method being used or suggested?
• Scope DRs: Is a segment or time window being defined?
• Model DRs: Is an algorithm or model being selected?
• Principle DRs: Does the analysis involve causal claims or metric proposals?

Step 3: Apply matching DRs

For each matching DR:

1. Read the full DR file (analyses/decisions/DR-{NNN}.md)
2. Apply its 결정 (decision) as a constraint
3. Do NOT suggest alternatives listed as rejected in the 근거 section
4. If you believe the DR should be reconsidered, state this explicitly — do not silently override

Acknowledge applied DRs briefly:

DR-{NNN} 적용: {one-line summary}

Step 4: Suggest new DRs when appropriate

If a significant methodology decision is made during analysis that isn't yet recorded:

• After the analysis stage is complete, suggest: "이 결정을 DR로 기록하면 어떨까요? /analysis-dr new"
• Especially at EVOLVE stage: any metric proposed or method standardized is a DR candidate

---

DR Types Reference

Type	Covers	Example
Metric	Metric definitions	DAU = 핵심 기능 1회 이상 사용
Methodology	Statistical/analysis methods	A/B 테스트는 베이지안, 신뢰도 95%
Scope	Segment or time window	신규유저 = 가입 후 7일 이내
Model	Algorithm selection	이탈 예측 = gradient boosting
Principle	Analytical standards	인과관계 주장은 DiD/RD/IV 필수