The Workflow Upgrade Risk (WUR) model is a data-driven approach to safely upgrading durable workflows without defaulting to pinning every in-flight run. Current engines like Temporal, Azure Durable Functions, and Vercel's Workflow SDK all recommend pinning runs to the deployment that started them, which can prevent critical bug fixes from reaching long-running workflows. WUR uses a static diff and existing engine telemetry to score upgrade risk for each in-flight run individually. Risk is decomposed into backward risk (can the recorded history rehydrate under V2, computed exactly) and forward risk (will the remainder traverse changed code, estimated probabilistically via Bayesian Markov chains). The model requires no code execution, dry-runs, or sandboxes. In evaluation across 64,920 synthetic and 18,160 real-corpus verdicts, it achieved zero false negatives, 0.933 precision on real workflows, and can classify 50,000 in-flight runs in 0.72 seconds on a single core. The result is a three-way policy: migrate, review, or pin — with pinning now a measured cost rather than the default.
