Benchmarks

Benchmark definitions: workload, metric, baseline and the criterion for a result to count.

Required fields

• Workload definition, precise enough to reimplement
• Metric and its units
• Baseline, named and versioned
• Acceptance criterion, fixed before any run
• Known ways to game it, stated by the author

Adding one

Take the next free BENCH- number from registry/artifacts.yaml, create the page, and add the registry entry in the same commit. An artifact that exists as a file but not as a record cannot be cited, and will be re-created by someone who could not find it.