Hi Vercel team,

I have a reproducible issue starting a Python Workflow from a Vercel Preview deployment.

Project: raddad1/raddad Workflow entrypoint: services.workflow_retry_dispatch:app

Vercel successfully discovers and deploys the workflow function as: _py_workflows/services-workflow__retry__dispatch__app

However, calling Python Workflow start() fails before the run is created:

vercel._internal.workflow.world.WorkflowWorldError: POST /v3/runs/null/events -> HTTP 400: Bad Request

I reproduced the identical failure with both:

vercel==0.8.1 and vercel==0.10.0

I confirmed:

• VERCEL_DEPLOYMENT_ID is present and resolved by the SDK
• VERCEL_OIDC_TOKEN is present and resolved
• Workflow registration/build succeeds
• FastAPI starts normally
• /v3/runs/null/events is the SDK’s intended initial run creation request
• the response contains no useful JSON error body

Clean reproduction:

UTC: 2026-08-17T20:14:58.919896Z

Deployment ID: dpl_2vJ4zknJspUJkM3Gv6KUHFTWT9zJ

Trace:

services/workflow_retry_dispatch.py → start() → events_create() → POST /v3/runs/null/events → HTTP 400

The same error persists on vercel==0.10.0, including its updated Workflow transport/auth implementation.

Could someone from the Vercel team inspect the backend/edge logs for this deployment and timestamp and identify why the initial run_created event is rejected?

The Python SDK does not expose a detailed server-side error message, so there is no further validation detail available from the client side.