Shadow mode
Shadow mode runs a candidate system alongside an active system, producing outputs without using them to affect users or decisions.
Financial AI in Practice
Shadow mode is a deployment approach in which a model processes live inputs without controlling decisions, allowing its outputs to be evaluated safely.
- How it is used
- Validates live fraud-model outputs and tunes alert thresholds before scores affect customer decisions.
- How deep it goes
- Details threshold calibration, monitoring, and a two-week validation check, without specifying shadow-mode implementation.
- Examples
- Uses a 100-alert daily investigation capacity, a 0.1% legitimate-transaction friction target, and two weeks of live validation.
- What you can do after
- Evaluate and tune fraud thresholds on live traffic without changing decisions, then monitor alerts and recalibrate as conditions shift.
Machine Learning System Design
An operating mode in which a candidate system runs alongside the active system and produces outputs without affecting users or decisions.
- How it is used
- Uses parallel production execution to validate model releases before exposing users or decisions to candidate predictions.
- How deep it goes
- Explains the release sequence and monitoring concerns but stops short of implementation details for traffic duplication or comparison logic.
- Examples
- Compares a newly validated model with the current model under real-world conditions while preserving current production behavior.
- What you can do after
- Evaluate candidate-model performance in production without allowing candidate predictions to affect users or decisions.
Read how Machine Learning System Design explains shadow mode →
Related concepts
- deploymentused by · Deployment is the process of making software, models, or other computational systems available to operate in a target environment.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.