Benchmark Baseline — Orchestration Performance

Established: 2026-03-06

Purpose

This document records the initial performance baseline for the orchestration subsystem. All future benchmark runs are compared against these thresholds to detect regressions.

Default Thresholds

Scenario	Metric	Threshold	Direction
dag_throughput	duration_ms	10000.0	lower is better
dag_throughput	memory_peak_mb	100.0	lower is better
policy_evaluation	duration_ms	5000.0	lower is better
policy_evaluation	ops_per_sec	100.0	higher is better
orchestration_e2e	duration_ms	5000.0	lower is better
report_generation	duration_ms	5000.0	lower is better
report_generation	ops_per_sec	10.0	higher is better

Scenarios

dag_throughput

Fan-out DAG with 50 parallel tasks depending on a root node, collected by a single sink node. Measures thread pool scheduling overhead and dependency resolution speed.

policy_evaluation

1000 iterations of PolicyEngine.check_command() across 10 different command strings (mix of allowed, denied, and approval-required).

orchestration_e2e

5-step sequential DAG pipeline executed 3 times. Measures end-to-end orchestration overhead.

report_generation

50 iterations of JSON + Markdown rendering for a report with 5 sections and 50 items.

How to Run

claw benchmark run              # run all and save
claw benchmark run --scenario dag_throughput   # single scenario
claw benchmark report           # show last results
claw benchmark list             # list scenarios

Initial Baseline Results

Run claw benchmark run to populate this section with actual numbers. Results are saved to ~/.claude-superpowers/benchmarks/.

Scenario	Duration (ms)	Ops/sec	Peak Memory (MB)
(run benchmarks to fill)