Summary

A 39-workflow study showing that natural-language enterprise workflow descriptions can be compiled into deterministic, governed execution artifacts with 100% compilation, execution, and replay determinism.

Key findings

• The system compiled 39 unique workflows from natural-language descriptions with 100% compilation and execution success across 2,040 governed runs.
• Determinism held across the downstream execution chain, verified by 10,200 independent replay executions producing byte-identical output.
• The corpus demonstrated breadth across decision, classification, routing, computation, validation, and structured generation workflows, while explicitly identifying aggregation, iteration, and grouping as the next missing primitives.

Abstract

Enterprise operations run on rules, thresholds, exceptions, and institutional knowledge that typically live in people’s heads, policy documents, and operational memory. Turning that operational intelligence into software has historically required engineers, creating a costly translation layer between what operators know and what enterprises can execute deterministically. This paper presents empirical evidence that a compiled execution architecture can bridge that gap by transforming natural-language workflow descriptions into deterministic, governed execution artifacts without per-workflow engineering implementation. Across 39 unique enterprise workflows, the system achieved 100% compilation success, 100% execution success, and 100% determinism across 2,040 governed pipeline executions spanning six workflow categories and four domains. Determinism was further verified through 10,200 independent replay executions, each producing byte-identical output from stored confirmed specifications. The results show that operational intelligence described in natural language can be reliably encoded into governed infrastructure through a human-supervised compilation pipeline, extending compiled execution from authored specifications to real intent-to-infrastructure workflows.