analysis-retro

/analysis retro

Generate an automatic retrospective report from archived analyses.

Instructions

Step 1: Parse period parameters

Check the user's arguments:

• --last-month (default) — previous calendar month
• --last-quarter — previous 3 calendar months
• --range {from} {to} — custom range (e.g., --range 2026-01 2026-03)
• --all — all archived analyses

If no arguments provided, default to --last-month.

Tell the user: "Generating retrospective for {period}. Scanning archives..."

Step 2: Scan archived analyses

Read all analyses in analyses/archive/{YYYY-MM}/ folders within the specified period.

For each analysis, extract:

• ID and title (from folder/file name)
• Type: Investigation / Modeling / Simulation (from file header or ASK stage)
• Mode: Full / Quick
• Tags (from file header or status.md)
• Key Insight (from summary.md or EVOLVE one-sentence insight)
• Confidence levels (from VOICE: count of 🟢/🟡/🔴)
• Impact Tracking table (from EVOLVE):
  • Each recommendation's Decision (Accepted/Rejected/Modified/Pending)
  • Each recommendation's Status (Not started/In progress/Done)
  • Outcome notes
• Follow-up proposals (from EVOLVE "Follow-up Analyses" section)
• Duration (from ID date to archive date)

Also scan analyses/active/ for any follow-up analyses that reference archived ones.

Step 3: Generate retro report

Create the file at analyses/.retro/retro_{period}.md (e.g., retro_2026-02.md or retro_2026-Q1.md).

Create the .retro/ directory inside analyses/ if it doesn't exist.

Use this template:

# Analysis Retrospective — {Period}
> Generated: {today}
> Scope: {N} analyses from {period}

---

## Summary

{2-3 sentence overview of the period's analysis activity, major themes, and key outcomes.}

---

## Analysis Activity

| Metric | Count |
|--------|-------|
| Total analyses | {N} |
| Full | {n} |
| Quick | {n} |
| Investigation 🔍 | {n} |
| Modeling 📈 | {n} |
| Simulation 🔮 | {n} |

Average duration: {X} days (Full), {X} days (Quick)

---

## Impact Tracking

### Decision Outcomes
| Status | Count | % |
|--------|-------|---|
| ✅ Accepted | {n} | {%} |
| ❌ Rejected | {n} | {%} |
| 🔄 Modified | {n} | {%} |
| ⏳ Pending | {n} | {%} |

### Top Wins
{List the recommendations that were Accepted + Done with positive outcomes. Max 5.}

1. **{ID}** — {Recommendation} → {Outcome}
2. ...

### Unresolved Items
{List recommendations still Pending or Not Started.}

- {ID} Rec #{n}: "{recommendation}" — {status}
- ...

---

## Patterns

### Recurring Topics
{Topics/tags that appeared in 3+ analyses.}

| Topic | Analyses | Key Trend |
|-------|----------|-----------|
| {tag} | {count} | {one-line summary of how findings evolved} |

### Common Findings
{Findings that appear across multiple analyses — convergent evidence.}

- {Finding shared by IDs}

### Confidence Distribution
| Level | Count | % |
|-------|-------|---|
| 🟢 High | {n} | {%} |
| 🟡 Medium | {n} | {%} |
| 🔴 Low | {n} | {%} |

{If 🔴 Low > 30%: "High proportion of low-confidence findings. Consider investing in better data infrastructure or longer analysis timelines."}

---

## Unresolved Follow-ups

{EVOLVE sections propose follow-up analyses. List those that haven't been started yet.}

| Source | Proposed Follow-up | Status |
|--------|--------------------|--------|
| {ID} | {follow-up question} | 🟡 Not started / ✅ Started as {new-ID} |

{If >50% unstarted: "Most follow-up proposals haven't been acted on. Consider prioritizing the top 2-3 as next analyses."}

---

## Recommendations

Based on this retrospective:

1. **{Recommendation}**: {Explanation with supporting data}
2. **{Recommendation}**: {Explanation}
3. **{Recommendation}**: {Explanation}

{Generate 2-4 recommendations based on the patterns above. Examples:}
{- If a topic recurs: "Consolidate retention insights into a meta-analysis"}
{- If follow-ups are neglected: "Schedule dedicated follow-up time each sprint"}
{- If impact tracking is sparse: "Improve impact tracking discipline — only 30% of recommendations have outcomes recorded"}
{- If confidence is mostly 🔴: "Invest in data quality or longer timelines for higher-confidence analysis"}

---

## Appendix — Full Analysis List

| # | ID | Title | Type | Mode | Tags | Key Insight | Impact |
|---|-----|-------|------|------|------|-------------|--------|
| 1 | {ID} | {title} | {type} | {mode} | {tags} | {insight} | {Accepted/Rejected/Pending} |
| 2 | ... | ... | ... | ... | ... | ... | ... |