mpa-multi-plan-aggregation

name	mpa - Multi-Plan Aggregation
description	Generate, evaluate, and manage multiple alternative plans for the same goal
output	{"format":"prose"}

Multi-Plan Aggregation

Overview

Generate, evaluate, and manage multiple alternative plans for the same goal

Steps

Step 1: Divergent generation

Generate maximum variety of approaches to the goal:

1. Brainstorm obvious approaches (3-5 conventional methods)
2. Apply strategy inversion (what if we did the opposite?)
3. Cross-domain bridging (what works in other fields?)
4. Constraint removal (what if X weren't true?)
5. Stakeholder perspective (what would expert Y do?)

Target: 8-15 distinct approaches before any filtering.

SAFETY: Only generate legal, ethical approaches. Flag any approaches that might have compliance concerns.

Step 2: Quick filter

Eliminate clearly infeasible approaches:

• Violates hard constraints?
• Requires unavailable resources?
• Mathematically/physically impossible?
• Ethically problematic?
• Already tried and failed?

Keep "unlikely but possible" - only remove "impossible".

SAFETY: Document why each eliminated plan was removed. Human can review eliminations if desired.

Step 3: Rapid development

For each surviving approach, develop a quick plan sketch:

• One-paragraph summary of the approach
• Key steps (5-10 high-level steps)
• Resource requirements estimate
• Major risks identified
• Initial probability estimate (gut feel)

Time limit: 15-30 minutes per plan to avoid over-investing before scoring.

SAFETY: Do not commit any resources yet. This is planning only.

Step 4: Systematic scoring

Score each plan on 5 dimensions:

1. Probability of Success (weight: 0.30) Scale: 0-100% Question: "If executed well, what's the chance of achieving the goal?"

2. Resource Efficiency (weight: 0.20) Scale: 1-10 Question: "How well does resource investment match potential return?"

3. Robustness (weight: 0.20) Scale: 1-10 Question: "How well does plan handle unexpected problems?"

4. Speed (weight: 0.15) Scale: 1-10 Question: "How quickly can this achieve results?"

5. Learning Value (weight: 0.15) Scale: 1-10 Question: "How much will we learn even if it fails?"

Formula: Score = (Prob * 0.3) + (Efficiency * 0.2) + (Robust * 0.2) + (Speed * 0.15) + (Learning * 0.15)

SAFETY: Document reasoning for each score. Human can adjust weights based on priorities.

Step 5: Portfolio selection

Based on portfolio_strategy, select plans for the portfolio:

CONCENTRATED (all resources on best plan):

• When: High confidence in best plan, limited resources
• Selection: Top 1 plan as primary, top 2 as documented backup

DIVERSIFIED (spread across multiple plans):

• When: High uncertainty, parallel execution possible
• Selection: Top 3 plans with different approaches, run in parallel

STAGED (primary + ready backups):

• When: Serial execution, need pivot capability
• Selection: Top plan as primary, next 2 as ready backups

SAFETY: This step REQUIRES human approval before proceeding. Present portfolio recommendation and wait for confirmation.

Step 6: Full development

Develop selected plans to appropriate detail:

PRIMARY PLAN:

• Complete STEPS.md with all phases and tasks
• Detailed resource allocation
• Decision gates with criteria
• Pivot triggers (conditions that activate backup)
• Success metrics and checkpoints

BACKUP PLANS:

• Summary-level development
• Key steps identified
• Resources reserved or identified
• Activation procedure defined

STANDBY PLANS:

• Minimal development
• Core concept documented
• Can be developed quickly if needed

SAFETY: Resource commitments require human approval.

Step 7: Initialize tracking

Set up portfolio management infrastructure:

1. Create PLAN_DATABASE.md with:

   • All plans with status (active/ready/standby/archived)
   • Ranking and scores
   • Resource allocation
   • Decision log

2. Define rebalancing triggers:

   • Active plan hits major blocker
   • New information changes probabilities
   • Resource availability changes
   • Deadline approaches with active plan behind

3. Schedule regular reviews:

   • Weekly probability updates
   • Milestone-based reassessment
   • Pre-deadline final check

SAFETY: All plan status changes logged with timestamp and reason.

When to Use

• Goal has multiple valid approaches worth exploring
• Uncertainty is high and single-plan risk is unacceptable
• Stakes are significant enough to warrant backup plans
• Time permits exploration before commitment
• Competition context where multiple entries improve odds
• Past experience shows plans often need pivoting
• Resources allow parallel development of alternatives
• Complex goal with many unknown dependencies

Verification

• Portfolio has minimum 2 plans (primary + at least one backup)
• All plans satisfy hard constraints
• Scoring is documented and defensible
• Pivot triggers are specific and measurable
• Human has approved portfolio selection
• Primary plan is fully executable
• Backup plans can be activated within defined timeframe
• Resource allocation does not exceed available resources