Summary

A 120-workflow, 12-domain study showing that compiled execution preserves determinism, complete policy-change visibility, and bit-identical replay at enterprise scale.

Key findings

• Compiled execution maintained determinism at 1.0 across 120 workflows and 12 domains.
• The structural diff surface surfaced 219 of 219 known policy changes, giving complete visibility into governance-relevant modifications.
• Every historical decision was reproducible with bit-identical replay and a cryptographic proof chain across 2,294 test inputs.

Abstract

As enterprises deploy AI-driven decision systems into regulated operational settings, the binding constraint is not accuracy but governance. When a policy changes, three questions must be answerable: what changed, what decisions are affected, and can the result be proven? Runtime inference systems do not answer these questions structurally because their execution artifacts are generated token sequences rather than formal representations amenable to complete diff, bit-identical replay, or content-addressed verification. This paper presents an empirical evaluation of compiled execution across 120 enterprise workflows spanning 12 domains, with 2,294 test inputs and 240 compiled artifacts. The structural diff surface identifies 219 of 219 known policy changes, every historical decision is reproducible with bit-identical output and a cryptographic proof chain, and determinism holds across all 120 workflows. The findings show that the execution guarantees established in earlier experiments are structural rather than statistical, and that compiled execution operates as governance-native infrastructure at enterprise scale.