SoftClient4ES benchmarks

Reproducible benchmarks comparing SoftClient4ES against comparable stacks. Each subdirectory is one self-contained benchmark: a Docker Compose stack, a deterministic data generator, stock clients, and documented results.

Benchmark	Compares	Status
extraction-vs-trino	Extracting large result sets out of Elasticsearch into a Python client, vs Trino's Elasticsearch connector	Complete — results · PDF report

Principles

A benchmark published by the vendor of one of the systems is read critically, and should be. Every benchmark here follows the same rules:

1. Every number comes from a run of the harness in this repository — nothing is estimated or carried over from elsewhere.
2. Correctness before timing. Runners assert the exact expected row count and cross-check that both systems returned the same data; a run that returns the wrong result is discarded, not timed.
3. Both systems use stock clients and documented defaults, pinned explicitly. Where a default is changed, the reason is recorded in the benchmark's METHODOLOGY.md and applied to both sides.
4. Biases are disclosed in the direction they run, including the ones that count against SoftClient4ES.
5. Every benchmark states where the other system is stronger, with the same prominence as the rest of the results.
6. Licence tiers are disclosed wherever a quota affects what can be measured.

Reproducing

Each benchmark's README lists its host requirements and run commands. All expect Docker and a native CPython 3.12, and none depend on anything outside their own directory.