Each connection is closed and re-established after 10 requests, forcing repeated TCP handshakes. Same mixed GET/POST workload as baseline. Measures how efficiently a framework handles connection setup and teardown — important for real-world scenarios where clients don't maintain long-lived connections.

Each request hits GET /json. The server loads a 50-item dataset at startup, computes a derived total field per item, and returns the full result as a JSON response (~10 KB). Measures object allocation, JSON serialization performance, and framework response overhead — the typical workload of a real-world API.

Each request sends a 20 MB binary payload via POST /upload. The server reads the entire body and returns its CRC32 checksum. Measures request body ingestion throughput — how efficiently a framework handles large uploads, including buffering, memory management, and I/O handling under sustained load.

Requests GET /compression with Accept-Encoding: gzip. The server loads a 6000-item dataset (~1 MB JSON) at startup and serves it gzip-compressed on the fly. Measures the throughput cost of real-time compression — only frameworks with built-in gzip support are eligible.

Valid baseline requests interleaved with malformed noise — nonexistent paths, bad content-length, and raw binary. Only 2xx responses count. Measures resilience: how well a framework maintains throughput while rejecting garbage traffic.