Tactic: Cross-Consistency Filtering

Tactic: Cross-Consistency Filtering

Orchestration Pattern

1. Spawn consistency evaluation → consistency-pair-evaluation

   • Pass: Zwicky Box (full parameter space), evaluation criteria
   • Receive: CCA matrix with pairwise consistency scores

2. Apply filtering rules

   • Mark configurations with ANY inconsistent pair as eliminated
   • Mark configurations with conditionally consistent pairs as flagged
   • Retain configurations with all-consistent pairs as primary scenarios
   • If too few survive (< 3): relax to include conditionally consistent configs
   • If too many survive (> 10): tighten criteria or select representative subset

3. Rank surviving configurations

   • Score by: total consistency score (sum of pairwise ratings)
   • Score by: diversity (maximize coverage of parameter space)
   • Score by: relevance (proximity to current trajectory)
   • Select top N (typically 4-8) for narrative construction

4. Spawn narrative construction → scenario-narrative-construction (per selected config)

   • Pass: parameter configuration, consistency context, research approach
   • Receive: rich scenario narrative

5. Validate narratives

   • Check: Does narrative honor all parameter values in the configuration?
   • Check: Is the causal logic internally consistent?
   • Check: Is the scenario distinguishable from others?
   • If validation fails: re-spawn with specific correction guidance

Quality Checks

• CCA matrix is complete (all pairs evaluated)
• Filtering produces 3-8 surviving configurations
• Surviving configs span the parameter space (not clustered)
• Each narrative is internally consistent with its configuration
• Narratives are qualitatively distinct from each other