What it is and what it is notIt is a simulator. Its rules are ported from the code of RabbitMQ 4.0.5 and 4.3.6 and Kafka 4.3.1, and calibrated against those real brokers: 5,000 random routings per version, redelivery and dead-letter scenarios, 2,000 partition keys, Java consumer group assignments, what is given up in a rebalance, where a consumer resumes after a crash (with per-batch commit and with auto-commit) and how many times a share group delivers with REJECT and with RELEASE. Also RabbitMQ's spread across consumers and, in share groups, what happens when a consumer hangs (the record comes back when the lock expires) and when it dies (released right away). SQS is calibrated against ElasticMQ 1.7.1, an EMULATOR, not against AWS: redrive, visibility, FIFO ordering by group and receipt handles; SNS comes from the AWS documentation. NATS JetStream, against real NATS 2.15.0: max_deliver, nak, ack_wait, late acks, term, distribution among clients, max_ack_pending, deliver_policy, delayed nak and backoff. Redis Streams, against real Redis 8.0.2: pending entries list, XAUTOCLAIM, group-wide XACK, rereading, $ and 0, distribution and stream trimming (MAXLEN); the delivery limit and its dead-letter queue are the usual application recipe, because Redis does not provide them. MQTT, against real Mosquitto 2.0.21: what is resent and when, persistent and clean sessions, the window of 20 in flight, shared subscriptions, QoS 0, 1 and 2 and retained messages. It does not measure timing or throughput: time is logical, in ticks (1 tick = 1 s for Kafka's defaults).