Overview
/silver:benchmark is the Silver Bullet orchestrator for performance measurement and comparison — latency profiling, throughput tests, memory analysis, and before/after optimization validation. It maps to catalog entity WF-SILVER-BENCHMARK (specialized type) in the v0.48.3 APO catalog.
Benchmarking is evidence-gathering, not feature delivery. The workflow orients on the system under test, executes controlled benchmarks, verifies result reproducibility, and documents findings for downstream planning or optimization work.
Standard composition:
For technology selection decisions based on benchmark evidence, hand off to /silver:deep-research. For implementing optimizations, route to /silver:feature.
AF-VERIFY requires consistent results across runs with documented environment, hardware, and dataset constraints — not a single lucky measurement.
When to Use
Entry trigger signals for /silver:benchmark and benchmark:
- Latency or throughput regression investigation
- Before/after comparison for an optimization proposal
- Load testing and capacity planning evidence
- Memory profiling or CPU hotspot analysis
- Competitive performance baseline for architecture decisions
Catalog Composition
| Atomic flow | Command | Outcome |
|---|---|---|
AF-ORIENT | /silver:scan, /silver:context | System topology, hot paths, existing benchmarks, environment constraints |
AF-EXECUTE | /silver:execute | Benchmark scripts, load tests, profiling runs, dataset preparation |
AF-VERIFY | /silver:verify | Reproducibility checks, statistical validity, outlier analysis |
AF-DOCUMENT | /silver:ensure-docs | Durable benchmark report with methodology, results, and recommendations |
Workflow Steps
AF-ORIENT — Establish measurement context
Maps the code paths, infrastructure, and existing performance baselines. Defines what to measure, acceptable variance, and environmental constraints (hardware, network, dataset size).
AF-EXECUTE — Run benchmarks
Implements or invokes benchmark harnesses. Controls variables: warmup iterations, sample size, isolation from concurrent load. Captures raw metrics and system telemetry.
AF-VERIFY — Validate reproducibility
Re-runs critical benchmarks, checks statistical significance, flags environmental noise. Results that cannot reproduce are marked inconclusive, not reported as fact.
AF-DOCUMENT — Persist findings
Writes a durable report under .planning/ with methodology, raw data references, charts, and actionable recommendations for optimization or architecture decisions.
Example Invocation
Silver Bullet routes to WF-SILVER-BENCHMARK. Orientation maps the search hot path and baseline metrics. Execution runs controlled load tests against both configurations. Verification confirms reproducible p99 delta. Documentation captures the comparison report for the optimization PR.