Claims and replication

CI smoke-tests runners. It does not publish numbers.

Published dashboard payloads are packed on a developer machine:

./scripts/run-all-benchmarks.sh --mode full --analyze
python3 dashboard/scripts/sync-data.py

Then commit dashboard/public/data/ and deploy docs.

What we claim

Rankings are within one language and one communication category (thread, async, process, shared memory, or durable), on the payload types and patterns in the run config, on one machine, one session.
Warmup index 0 is in the CSV; analysis drops it.
Cross-language times are directional only.

“The fastest queue.”
Rankings across languages.
Rankings that mix thread queues with async queues.
Broker (Redis/Kafka/ZeroMQ) performance on the same chart as an in-process queue.
Production tail latency under load.