FlowDiagram
AI & ML pipelinesMermaid

A/B model evaluation & shadow deployment template

Safely promoting a new ML model: offline evaluation, shadow traffic, canary, online A/B test with guardrail metrics and automatic rollback — for MLOps and data science teams. This data flow diagram template contains 17 labelled components, branches such as “No”, “Yes”, “Inconclusive”.

Open in the editor →
  • Free — no account, no trial, no watermark
  • Fully editable shapes and connectors, not a screenshot
  • Export PNG, SVG or PDF, or share it as a link
  • Private: the diagram never leaves your browser

Components in this data flow diagram

  • Candidate model trained
  • Offline evaluation on holdout set
  • Beats champion offline?
  • Archive in MLflow registry
  • Shadow deploy: mirror live traffic
  • Compare predictions, latency & errors
  • Latency & error budget ok?
  • Canary: 5% of users
  • Guardrail metrics healthy?
  • Automatic rollback
  • A/B test: 50 / 50 split
  • Collect conversion & revenue metrics
  • Statistically significant win?
  • Extend test duration
  • Promote to champion
  • Route 100% traffic
  • Drift & performance monitoring

How to use this data flow diagram template

  1. Click Open in the editor. The data flow diagram opens on a new page in FlowDiagram — nothing is uploaded and there is no sign-up.
  2. Rename, add or delete shapes and drag them where you want; connectors stay attached as things move.
  3. Drop in vendor logos from the library — 1,160 of them, including every AWS, Azure and Google Cloud service icon — or restyle it with the colour and theme controls.
  4. Prefer text? Edit the Mermaid source below and paste it in through ☰ → Import diagram (see the Mermaid editor); it imports as an editable diagram.
  5. Export to PNG, SVG or PDF, or press Ctrl+L for a link that carries the whole diagram.

Mermaid source

The whole template is the text below. Paste it into FlowDiagram, keep it in Git next to the code it describes, or hand it to an AI assistant to extend.

flowchart TD
  NEW([Candidate model trained]) --> OFFLINE[Offline evaluation on holdout set]
  OFFLINE --> BEAT{Beats champion offline?}
  BEAT -->|No| ARCHIVE[Archive in MLflow registry]
  BEAT -->|Yes| SHADOW[Shadow deploy: mirror live traffic]
  SHADOW --> COMPARE[Compare predictions, latency & errors]
  COMPARE --> SAFE{Latency & error budget ok?}
  SAFE -->|No| ARCHIVE
  SAFE -->|Yes| CANARY[Canary: 5% of users]
  CANARY --> GUARD{Guardrail metrics healthy?}
  GUARD -->|No| ROLLBACK[Automatic rollback]
  GUARD -->|Yes| AB[A/B test: 50 / 50 split]
  AB --> STATS[Collect conversion & revenue metrics]
  STATS --> SIG{Statistically significant win?}
  SIG -->|No| ROLLBACK
  SIG -->|Inconclusive| EXTEND[Extend test duration]
  EXTEND --> STATS
  SIG -->|Yes| PROMOTE[Promote to champion]
  PROMOTE --> FULL[Route 100% traffic]
  FULL --> MONITOR([Drift & performance monitoring])
  ROLLBACK --> ARCHIVE

Related: Data pipeline diagram tool · Mermaid editor · All ai & ml pipelines templates · All 521 templates