analysis-next

Prompt file imported from with-geun/alive-analysis (.claude/commands/analysis-next.md). Copyright stays with the author.

/analysis next

Advance the current analysis to the next ALIVE stage.

Instructions

Step 1: Identify current analysis

Read .analysis/status.md to find active analyses.

• If only 1 active Full analysis → select it automatically
• If multiple active → ask user which analysis to advance (show ID + title)
• Quick analyses don't use /analysis next (all sections are in one file)
  • If user selects a Quick, remind them to fill sections in order within the file

Step 2: Determine current stage

Read the analysis folder to see which stage files exist:

• Only 01_ask.md → current stage is ASK
• Up to 02_look.md → current stage is LOOK
• Up to 03_investigate.md → current stage is INVESTIGATE
• Up to 04_voice.md → current stage is VOICE
• All 5 files exist → current stage is EVOLVE (analysis is complete)

Step 3: Review current stage checklist

Read the current stage's checklist (embedded at the bottom of the current stage file).

Check if there are any 🔴 (stop) items:

• If 🔴 items exist → warn the user: "There are stop signals in your {stage} checklist. Review before proceeding."
• If all items are unchecked → remind: "Don't forget to review the checklist before moving on."
• If items are checked → proceed

Step 4: Generate next stage file

Based on current stage, generate the next file.

ASK → LOOK (generate 02_look.md):

Note: If this is a 📈 Modeling analysis, use the Modeling-specific templates defined in analysis-new.md Section 4B instead of the Investigation templates below. This applies to LOOK, INVESTIGATE, VOICE, and EVOLVE stages. If this is a 🔮 Simulation analysis, use the Simulation-specific templates from analysis-new.md Section 4D for all stages.

# LOOK: {title}
> ID: {ID} | Type: 🔍 Investigation | Stage: 👀 LOOK | Updated: {YYYY-MM-DD}

## Data Sources
- Primary:
- Secondary:
- Access method: (MCP / exported file / manual query / BI dashboard)

## Data Quality Review
- Row count:
- Date range:
- Missing values:
- Known issues:

## Segmentation
Break down BEFORE drawing conclusions:
- By time:
- By cohort:
- By platform:
- By segment:

Key question: "Does the pattern hold across ALL segments, or is it driven by one?"

## Confounding Variables
- What else changed at the same time?
- Population comparison: Are we comparing the same groups?
- Survivorship bias risk:
- Simpson's paradox check:

## External Factors
- Calendar: holidays, seasonality, industry events
- Competitors: recent launches, pricing changes
- Platform: iOS/Android policy, algorithm updates
- Macro: economic indicators, regulation changes

## Cross-Service Check
- Related services that may be affected:
- Shared resources (auth, data pipeline, user base):
- Recent changes in adjacent services:

## Outliers & Anomalies
-

## Sampling
- Method:
- Size:
- Rationale:

## Initial Observations
-

---
{Insert LOOK checklist from .analysis/checklists/look.md}

LOOK → INVESTIGATE (generate 03_investigate.md):

# INVESTIGATE: {title}
> ID: {ID} | Type: 🔍 Investigation | Stage: 🔍 INVESTIGATE | Updated: {YYYY-MM-DD}

## Hypothesis Scorecard
| # | Hypothesis | Evidence For | Evidence Against | Status | Contribution |
|---|-----------|-------------|-----------------|--------|-------------|
| 1 | | | | ✅/❌/⚠️ | ~%  |
| 2 | | | | ✅/❌/⚠️ | ~%  |
| 3 | | | | ✅/❌/⚠️ | ~%  |

Strategy: Test the easiest-to-disprove hypotheses first.

## Multi-Lens Analysis

### Macro (market/industry)
- Industry-wide trend?
- Competitor actions?
- Economic/regulatory factors?

### Meso (company/product)
- Cross-service impact?
- Channel mix shift?
- Product changes (releases, A/B tests)?

### Micro (user/session)
- User behavior patterns:
- Cohort-specific trends:
- Edge cases:

## Causation vs Correlation
(Fill if causal claims are being made)
- Time ordering: Did cause precede effect?
- Mechanism: Plausible pathway?
- Dose-response: More cause → more effect?
- Counterfactual: Control group / unaffected segment?
- Conclusion: Causal / Correlational / Inconclusive

## Results

### Finding 1
- Description:
- Evidence:
- Confidence: 🟢 High / 🟡 Medium / 🔴 Low
- Reasoning:

### Finding 2
- Description:
- Evidence:
- Confidence: 🟢 High / 🟡 Medium / 🔴 Low
- Reasoning:

## Sensitivity Analysis
- Date range ±1 week: Same result?
- Exclude outliers: Same pattern?
- Different metric definition: Consistent?
- Minimum actionable effect size:

## Interpretation
-

## Reproducibility
- Query/notebook location: `assets/`
- Steps to reproduce:

---
{Insert INVESTIGATE checklist from .analysis/checklists/investigate.md}

INVESTIGATE → VOICE (generate 04_voice.md):

# VOICE: {title}
> ID: {ID} | Type: 🔍 Investigation | Stage: 📢 VOICE | Updated: {YYYY-MM-DD}

## Executive Summary
(1-3 sentences for leadership — lead with business impact)

## So What → Now What

### Finding 1
- **Finding**:
- **So What?** (business impact):
- **Now What?** (options):
  - Option A: {action} — benefit: {X}, risk: {Y}
  - Option B: {action} — benefit: {X}, risk: {Y}
- **Confidence**: 🟢 High / 🟡 Medium / 🔴 Low
- **Reasoning**:

### Finding 2
- **Finding**:
- **So What?**:
- **Now What?**:
- **Confidence**: 🟢 High / 🟡 Medium / 🔴 Low
- **Reasoning**:

## Recommendations
1. (with trade-off analysis)
2. (with trade-off analysis)

## Guardrail Impact
- Does this recommendation affect any guardrail metrics? (reference config.md)
- Trade-off: improving {metric A} may risk {metric B}

## Audience-specific Messages

### For {stakeholder 1}
-

### For {stakeholder 2}
-

## Limitations & Caveats
(These are first-class content, not a footnote)
- What we couldn't verify:
- Where sample was small:
- Assumptions we made:
- What would change our conclusion:

## Data Sources
- Sources used:
- Query locations: `assets/`

---
{Insert VOICE checklist from .analysis/checklists/voice.md}

VOICE → EVOLVE (generate 05_evolve.md):

# EVOLVE: {title}
> ID: {ID} | Type: 🔍 Investigation | Stage: 🌱 EVOLVE | Updated: {YYYY-MM-DD}

## Conclusion Robustness Check
- What new data could **disprove** our conclusion?
- What **assumptions** did we make but not verify?
- If a colleague challenged this, what would they attack?
- In 3 months, what would make us say "we were wrong"?

## Reflection
- What went well in this analysis?
- What could be improved?
- What surprised us?

## Monitoring Setup
- Metric to track going forward:
- Threshold for re-investigation:
- Dashboard / alert: (reference data stack from config.md)
- Owner:

## Unanswered Questions
-

## Follow-up Analysis Proposals
- [ ] {description} → (not yet created)
- [ ] {description} → (not yet created)

## Impact Tracking
> Track whether this analysis led to real decisions and outcomes. Revisit this section after 2-4 weeks.

| # | Recommendation | Decision | Owner | Status | Outcome |
|---|---------------|----------|-------|--------|---------|
| 1 | {from VOICE} | Accepted / Rejected / Modified | {who} | Not started / In progress / Done | {what happened} |
| 2 | | | | | |

- **Analysis influenced a decision?** Yes / No / Pending
- **Decision date**: {when the decision was made}
- **Outcome check date**: {2-4 weeks after decision — set a reminder}
- **Retrospective**: Was the recommendation correct? What would we do differently?

## Knowledge Capture
- Reusable patterns (SQL templates, segmentation approaches):
- Data gotchas for future analyses:
- Saved queries/notebooks: `assets/`
- Data quality issues to flag:

## North Star Connection
- How does this connect to {North Star metric from config.md}?
- Does this change our understanding of what drives it?
- Should our metric framework be updated? → If yes, use the section below.

## Proposed New Metrics
> Fill this section through conversation with the AI. Say "I think we need a metric for {X}" and the AI will guide you through defining it together.

### Metric: {name}

**Background**
- What gap did this analysis reveal in the current metric framework?
- Trigger: (the finding, blind spot, or missing visibility that led to this proposal)
- Replaces or complements: (existing metric, or "new")

**Purpose**
- Decision this metric informs:
- Primary audience:
- Proposed tier: 🌟 North Star / 📊 Leading / 🛡️ Guardrail / 🔬 Diagnostic

**Definition & Logic**
- Formula / calculation:
- Data source:
- Granularity: (daily / weekly / monthly / per-cohort)
- Refresh cadence:
- Edge cases handled: (zero denominator, new users, seasonality)

**Interpretation Guide**
- Healthy range:
- Alert threshold:
- Counter-metric: (what to watch so this metric isn't gamed)
- Plain-language: "When this goes up, it means... When it drops, it means..."

**STEDII Validation**
- [ ] **Sensitive** — Can it detect real changes?
- [ ] **Trustworthy** — Is the data accurate and the definition unambiguous?
- [ ] **Efficient** — Can it be computed in a practical timeframe?
- [ ] **Debuggable** — When it moves, can you decompose WHY?
- [ ] **Interpretable** — Does the team understand it without a 5-minute explanation?
- [ ] **Inclusive** — Does it fairly represent all user segments?

**Action**
- [ ] Add to `.analysis/config.md` → {tier}
- [ ] Set up dashboard / alert
- [ ] Communicate definition to stakeholders

(Copy this block for additional metrics)

## Automation Opportunities
- Can any part of this analysis be automated/scheduled?
- Recommended cadence:

## One-Sentence Insight
> (Capture the single most important takeaway)

---
{Insert EVOLVE checklist from .analysis/checklists/evolve.md}

EVOLVE reached → Tell user: "This analysis is complete. Run /analysis archive to archive it."

Step 5: Update status.md

Update the analysis row in .analysis/status.md:

• Change the stage column to the new stage icon
• Update "Last updated" timestamp

Step 6: Confirmation

Tell the user:

• Advanced from {old stage} to {new stage}
• Show the new file path
• Remind them to fill in the content and review the checklist
• If VOICE: "After completing VOICE and EVOLVE, run /analysis archive"