Get startedWorkflows · Design and orchestration
Build a production workflow
Compose the complete vision lifecycle instead of treating model training as the final destination.
Source reviewed 2026-09-16. Availability depends on your installation, permissions, and compatible runtimes. A supported path is not a guarantee of model quality or production readiness.
Outcome
Create a reusable workflow that can start from a pretrained model, train or fine-tune when needed, evaluate, deploy, and improve.
Start here
What you need to know first
Check the inputs below before starting. If you are new, begin with the first-project guide. A dataset holds media and labels; a model produces results; a deployment makes a selected model version callable. Creating one does not create the others.
Bring these inputs
- — Typed datasets, model versions, prompts, or upstream outputs
- — Runtime, policy, and acceptance constraints
Expected output and limits
- — An immutable workflow revision
- — Typed stage outputs and launchable workload configuration
Core workflow
Open the workflow from its project
Every project shows its lifecycle from data through monitoring. Choose Open visual workflow to open the matching graph. If the project has no graph yet, Score Studio creates one from its saved task, dataset version, and runnable model version; returning to the project preserves unsaved-change protection.
Repair setup issues
When repairs can be applied, choose Auto-fix workflow to repair and save the setup immediately, or Review issues to inspect it first. If no automatic repair can be applied, Review issues opens the steps that need your input, with a button to open the next step. Auto-fix repairs connections and fills missing settings from the project model, selected dataset, or a single compatible healthy resource. It preserves existing choices and never starts a run. The result lists applied changes and unresolved steps. Undo is available until you change the repaired graph.
Choose a system objective
Choose a task and inspect each block’s runtime availability. A template or task name does not prove an executable adapter is installed.
Bind versioned inputs
Attach datasets, pretrained registry models, prompts, policies, or upstream workflow outputs.
Add lifecycle stages
Configure optional training or fine-tuning, evaluation gates, deployment, monitoring, and feedback.
Launch a workload
Run the reusable definition against real inputs with progress, spend, logs, and outputs attached.
What this surface supports
Task-specific blocks with installed runtimes
Pretrained-to-production paths
Fine-tuning and feedback loops
Preview and release stages
Expert section
Contracts, signals, and failure modes
Use this section when you are defining acceptance criteria, automating the surface, or reviewing whether its output is safe to promote downstream.
Quality and operating signals
- — Schema compatibility
- — Evaluation-gate results
- — End-to-end latency and cost
Common failure modes
- — Incompatible block contracts
- — Unversioned dependencies
- — A training-only graph with no production outcome
Expert release checklist
□ Inputs and dependencies are pinned to immutable versions.
□ Acceptance metrics include critical classes and operating slices.
□ Failure, retry, cost, and rollback behavior are understood.
□ The resulting artifact has an owner and a downstream review path.
Engineering safeguards
- — Unknown blocks fail visibly
- — Release blocks require evidence; unavailable saved-run adapters block execution
- — Workflow revisions preserve provenance