rule-engine-bench.

Problem solved

When a rule evaluation service stops keeping up, the usual answer is to rewrite it in a faster language. This measures whether that is where the time actually goes.

What I built

Five implementations of the same prop firm rule evaluation, run over the same synthetic data, checked against each other for identical verdicts, and measured. On this workload the language is worth about 2x and the data format about 80x: the data format beat the language 40 to 1. I built the production pipeline this comes from, but that code, its data and its numbers belong to my employer, so this is the same problem rebuilt from nothing in September 2026: my own rule set, my own generated data.