ICMAI 2026 Presentation for Proof-Carrying Streamed Workflow Graphs

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

An AI system can build a workflow one apparently reasonable step at a time and still arrive at a process that cannot be completed. Each task may be valid. Each connection may pass its local checks. Yet one committed addition can remove the last way to satisfy the workflow’s requirements. When a workflow is still under construction, what evidence should accompany each addition to establish that an acceptable completion remains possible? And what information must a system retain if it is to answer that question correctly as construction continues? Our motivating case is the Opus Work Knowledge Graph, or WKG. It organizes reusable tasks, context and evidence so that workflows can be assembled for particular business settings.