Automated sifting pattern discovery.

Provides a pluggable generate-evaluate framework for discovering narratively interesting patterns from simulation trace data.

Architecture

The system works as an iterative loop:

1. A [CandidateGenerator] proposes candidate patterns from a [TraceCorpus]
2. [PatternEvaluator]s score each candidate
3. Scored results feed back to the generator for the next round
4. A [PatternFilter] decides which patterns to keep

The [DiscoverySession] orchestrates this loop with configurable budgets.