ModelRefs / Shadow Mode — AI Glossary
Shadow Mode — AI Glossary
Running a new model or prompt in parallel with the production model, logging its outputs for offline evaluation without affecting users.
Overview
Shadow mode ('shadow deployment', 'dark launch') routes every request to both the current production model and the candidate model. Only production outputs are returned to users; shadow outputs are logged for offline quality analysis. Enables safe evaluation of a new model on real production traffic before committing to a rollout.
Reference details
| Topic | ecosystem |
|---|---|
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Shadow Mode — AI Glossary.
Frequently asked questions
What is Shadow Mode?
Running a new model or prompt in parallel with the production model, logging its outputs for offline evaluation without affecting users.
What concepts are related to Shadow Mode?
Closely related concepts include canary deployment, a b testing llm, evaluation harness.