Contribute an experiment

To contribute an example, open a pull request with the workload code, its measurements and plots. You can add a package, an operation or an improvement to an existing example. Include enough information for another maintainer to repeat the comparison.

Start from examples/kitchen-sink. Keep the lifecycle small: prepare the input, perform one meaningful operation, verify the answer independently. Existing TestItems are welcome. Add explicit tags only when an item must be restricted to correctness or performance runs.

Include these facts with the result:

Evidence	Why it matters
Package versions or immutable revisions, dependency manifests	Reconstruct the compared implementations
Workload and fixture source, seed or frozen corpus	Repeat the same work
Oracle and its outcome	Reject fast but incorrect answers
Julia version, machine, operating system, threads and CPU allocation	Understand the execution conditions
Collector, sample count, units and preparation boundary	Know what the numbers measure
Raw observations or a downloadable bundle	Replot and inspect variation
Baseline, change and repeated confirmation	Separate an improvement from noise

Generate the plots from saved measurements so documentation builds stay fast. Attach large bundles and recordings to a release or another durable download location. Include notebooks as .jl files that readers can edit and run.

List unsuccessful checks with their errors, and identify tools or platforms you could not test.

For an optimization, explain the code change and compare the repeated measurements before and after it. Include the correctness-check result and machine settings so others can try the change on their own systems.