# Benchmarks

Current paired and single benchmark examples with scripted conversations, dependency-graph snapshots, rendered graph visualizations, and the text-state ledgers used to reconstruct conversation, theory, and memory files.

- Total benchmarks: 22
- Paired benchmarks: 3
- Single benchmarks: 19
- Paired benchmarks group linked conversations; single benchmarks show one isolated scripted run.
- Scripted conversations are reconstructed from the benchmark fixtures used to generate these runs.
- Included per benchmark: scripted conversation JSON, dependency graph JSON, rendered graph images, and any conversation/theory/memory text-state ledgers present.
- File guide: scripted_conversations.json reconstructs the scripted exchange; dependency_graph.json is the machine-readable provenance graph; dep_viz PNGs are quick visual renders; pt_conv / pt_kf / pt_mem files are the piece-table ledgers for conversation, theory, and memory text state.
- Excluded: runtime logs, event logs, knowledge files, observation snapshots, performance outputs, and graph-layout source files.