Orchestrator performance benchmarks

We evaluate Orchestrator on software engineering tasks that measure whether AI coding agents can understand real codebases, make accurate changes, and complete complex work reliably. These results show the lift Orchestrator adds on top of frontier models and coding harnesses.