Installs into .claude/skills of the current project.
Are you the author of Parameter Space Construction?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/yogsoth-ai-parameter-space-construction)
---
name: parameter-space-construction
description: "Orchestrates driver identification and parameter enumeration to build the complete morphological field"
version: 1.0.0
category: experiment-execution
type: tactic
used-by: scenario-planning
orchestrates:
- scenario-driver-identification
- parameter-enumeration
---
# Tactic: Parameter Space Construction
## Orchestration Pattern
1. **Spawn driver identification** → `scenario-driver-identification`
- Pass: research context, planning horizon, domain constraints
- Receive: ranked list of 5-8 uncertainty drivers with descriptions
2. **Validate driver set**
- Check: Are drivers independent? (no redundancy)
- Check: Do drivers span PESTEL categories? (no blind spots)
- Check: Are drivers at the right abstraction level? (not too broad, not too narrow)
- If validation fails: re-spawn with refined instructions
3. **Spawn parameter enumeration** → `parameter-enumeration`
- Pass: validated driver list, MECE requirement, value count guidance (2-4 per driver)
- Receive: complete Zwicky Box (driver × value matrix)
4. **Validate parameter space**
- Check: Are values mutually exclusive within each driver?
- Check: Are values collectively exhaustive within each driver?
- Check: Is the total configuration space manageable? (< 500 combinations)
- If validation fails: re-spawn enumeration with tighter constraints
5. **Compute space metrics**
- Total configurations: product of all value counts
- Expected surviving configs after CCA: ~10-20% of total
- Estimated downstream budget: configs × per-scenario cost
## Quality Checks
- [ ] All drivers are genuinely uncertain (not predetermined)
- [ ] No driver is a subset or consequence of another
- [ ] Values within each driver are truly MECE
- [ ] Total space is computationally tractable for CCA
- [ ] Parameter space covers the planning horizon adequately