research-optimize

user_invocable: true name: research-optimize description: Evidence-based personal optimization framework. Use when user wants to optimize a metric (health, wealth, attractiveness, authority, productivity, etc.) using academic research. Decomposes abstract goals into researchable variables, gathers peer-reviewed evidence with effect sizes, calibrates to user's baseline, and produces ranked interventions by expected value.

Research Optimize

Evidence-based framework for personal/business optimization using academic research.

Workflow

Phase 1: Context & Decomposition

Step 1.1: Gather Context

Use AskUserQuestion to collect critical info for calibration:

Context Questions (CRITICAL - ask via options):

• What contexts matter most for this metric? (e.g., investors, team, clients, public, partners)
• What specific situations trigger the need for this metric? (e.g., pitches, negotiations, team meetings)
• Who are you typically interacting with? (age, seniority, culture)

Self-Assessment:

• What do you perceive as your current strengths related to this metric?
• What do you perceive as your weaknesses/bottlenecks?
• Any physical factors relevant? (height, voice, appearance, age relative to peers)

Sub-topics:

• Any specific sub-topics you want to explore?
• Any you want to exclude?

Also check memory/conversation for existing user context.

IMPORTANT: Check Obsidian vault for existing research:

Grep -i "[topic]" ~/Documents/MyBrain/

Previous research may already exist - merge, don't duplicate.

Step 1.2: Propose Decomposition

Break down the target metric into 3-5 researchable sub-components.

CRITICAL: Use the Skills-Signals-Perceptions Framework

When decomposing ANY human optimization metric, ensure coverage across THREE dimensions:

Why this matters: Focusing only on skills misses high-effect-size factors like:

• Network associations (r = .90)
• Authority symbols (d = 1.19)
• Speaking time (r = .82-.93)

These are SIGNALS and PERCEPTIONS, not skills - but often have HIGHER effect sizes.

For each sub-component evaluate:

• Researchability: Is there academic literature? (strong/moderate/limited)
• Research question: How would academics phrase this?
• Outcome contribution: What % of total outcome does this drive?
• Dimension: Skills / Signals / Perceptions

Step 1.3: Confirm with User

Present proposed sub-components and ask:

• "Есть ли ещё топики, которые стоит исследовать?"
• "Какие-то из этих топиков неактуальны для тебя?"
• Allow user to add/remove/modify sub-components

Output format:

## Decomposition: [METRIC]
**User context**: [key details affecting research]

| Sub-component | Researchability | Research Question | Outcome % |
|---------------|-----------------|-------------------|-----------|
| ... | strong/moderate/limited | "What factors affect X?" | ~N% |

Есть ли ещё топики для исследования? Какие-то из этих неактуальны?

After user confirms, proceed to Phase 2.

Phase 2: Evidence Gathering

FIRST: Create output folder:

mkdir -p ~/Documents/MyBrain/Research/research-optimize-{topic}-{date}/

THEN: Launch parallel Task agents (subagent_type: "general-purpose") for each sub-component with strong/moderate researchability.

Agent prompt template:

Research: [SUB-COMPONENT] for [USER CONTEXT SUMMARY]

Find peer-reviewed studies, meta-analyses, and RCTs showing:

1. CAUSAL FACTORS (ranked by effect size):
- Effect size with 95% CI
- Evidence type: RCT / natural experiment / IV / twin study / longitudinal
- Sample size and study duration
- Replication status
- Population studied (check match to user's situation)

2. BASE RATES & DISTRIBUTIONS:
- Median outcome, not mean
- Percentile distribution (25th, 50th, 75th, 90th)
- "What % of people who try X achieve Y" - not just "X can lead to Y"

3. WHAT DOESN'T WORK:
- Debunked interventions with citations
- Survivorship bias examples
- Correlational findings that don't survive controls

4. EVIDENCE GAPS:
- "No evidence" vs "evidence of no effect" distinction
- Where findings are extrapolated vs directly applicable

Output as ranked table:
| Factor | Effect Size (95% CI) | Evidence Quality | Sample Size | Citation |

WebSearch strategy for agents:

• Search: "[topic] meta-analysis", "[topic] RCT", "[topic] systematic review"
• Domains: scholar.google.com, pubmed.ncbi.nlm.nih.gov, cochranelibrary.com
• Prioritize: Cochrane reviews, JAMA, Lancet, Nature, Science, domain-specific top journals

AFTER all agents complete: MUST save combined results to evidence.md using Write tool: ~/Documents/MyBrain/Research/research-optimize-{topic}-{date}/evidence.md

Required format:

# Evidence: [METRIC]
Date: [DATE]
User context: [BRIEF SUMMARY]

## Sub-component 1: [NAME]

### Causal Factors (Ranked)
| Factor | Effect Size | CI | Evidence | Sample | Citation |

### Base Rates
[Distribution data]

### What Doesn't Work
[Debunked interventions]

### Evidence Gaps
[Limitations]

## Sub-component 2: [NAME]
...

Phase 3: Personal Calibration

CRITICAL: After finding evidence, calibrate to user's specific baseline.

For each TOP 5-10 factors by effect size, determine user's current state:

Step 3.1: Present Top Factors Show user the top factors ranked by effect size:

| Rank | Factor | Effect Size | Your Baseline Matters Because |
|------|--------|-------------|------------------------------|
| 1 | [Factor] | r = .XX | [Why baseline changes ROI] |

Step 3.2: Ask Baseline Questions (via AskUserQuestion)

For each high-effect factor, ask:

• Current level (1-10 or specific metric if measurable)
• Distance to target (how much room for improvement?)
• Feasibility (any blockers to improving this?)

Example:

To rank interventions for YOUR situation, I need your baseline on top factors:

1. Network/Prestige Associations (r = .90):
   - Current network quality: [weak/moderate/strong/excellent]
   - Access to known names for advisory: [none/some/many]

2. Voice Pitch (ρ = -.51):
   - Self-perceived pitch: [high/average/low]
   - Ever measured Hz?

Why this matters:

• If baseline is already high → factor has LOW marginal ROI for this user
• If baseline is low + effect size high → factor is TOP priority
• Same effect size can mean different ROI based on user's starting point

Step 3.3: Calculate Marginal ROI

Rank interventions by: Effect Size × Room for Improvement × Feasibility

Put highest-baseline factors in TIER 4 (Already Have), not TIER 1.

Phase 4: Intervention Ranking (ROI Tiers Format)

Analyze evidence + user profile to produce ROI-tiered actions.

Ranking criteria:

• ROI = Effect size / (Cost + Time + Effort)
• Convert effect sizes to approximate % improvement for readability
• Flag evidence quality: Causal (RCT/experimental) vs Correlational vs Survey
• Identify bottlenecks (must-do before other interventions work)

Output format:

# ROI: [METRIC]
**Профиль**: [USER BASELINE SUMMARY]

---

## TIER 1: КРИТИЧЕСКИЙ ROI (Must-do, убирает bottleneck)

| Фактор | Твой baseline | Target | Effect | Стоимость | Время | Evidence |
|--------|---------------|--------|--------|-----------|-------|----------|
| **[Factor]** | [current] | [target] | **+X-Y%** (effect size) | $X-Yk | N мес | **Causal/Correlational**: [source] |

**Почему Tier 1**: [Why this is a bottleneck that blocks other gains]

---

## TIER 2: ВЫСОКИЙ ROI (Differentiation)

| Фактор | Твой baseline | Target | Effect | Стоимость | Время | Evidence |
|--------|---------------|--------|--------|-----------|-------|----------|
| ... |

---

## TIER 3: MODERATE ROI (Optimization)

| Фактор | Твой baseline | Target | Effect | Стоимость | Время | Evidence |
|--------|---------------|--------|--------|-----------|-------|----------|
| ... |

---

## TIER 4: ALREADY HAVE (Don't over-invest)

| Фактор | Твой baseline | Action | Risk |
|--------|---------------|--------|------|
| **[Factor]** | [Already high] | [Maintain/leverage] | **Risk**: [What happens if over-signal] |

---

## STOP DOING (Negative ROI)

| Action | Expected Effect | Evidence |
|--------|-----------------|----------|
| [Bad action] | **-X-Y%** | **[Evidence type]**: [Why it backfires] |

---

## КОНТЕКСТ: [Relevant context switches]

| Если цель | Приоритеты меняются |
|-----------|---------------------|
| [Context A] | [How priorities shift] |
| [Context B] | [How priorities shift] |

---

## TIMELINE

| Период | Фокус | Ожидаемый прогресс |
|--------|-------|-------------------|
| Week 1-4 | Quick wins | [Immediate gains] |
| Month 1-6 | Primary | [Main intervention progress] |
| Month 6-12 | Secondary | [Optimization gains] |

---

## CONFIDENCE LEVELS

| High (Causal) | Medium (Correlational) | Extrapolated |
|---------------|------------------------|--------------|
| [Factor → outcome] | [Factor association] | [Your situation vs research pop] |

---

**Ключевой инсайт**: [One-sentence summary of most important finding for this user]

MUST save using Write tool to: ~/Documents/MyBrain/Research/research-optimize-{topic}-{date}/interventions.md

Key Principles

1. Causal > Correlational - Always flag evidence type
2. Effect sizes with ranges - Never vague claims
3. Base rates matter - "Works for 5% of people" is different from "can work"
4. What doesn't work - As important as what does
5. Calibrate to user - Generic advice is worthless for outliers
6. Save everything - Research is expensive, results should persist