Skip to main content

Tutorials

Use tutorials when you want a guided build rather than an API catalog. The recommended first path is:

  1. Getting Started
  2. Capture Evidence
  3. Artifact Handling
  4. Compare Runs
  5. Query And Diagnostics
  6. Common Workflows
  7. Batch, Sample & Deployment

Choose A Path

GoalStart hereWhat you will learn
Get one local workspace workingGetting Startedinstall, capture evidence, inspect local output
Decide which execution evidence to captureCapture Evidencemetrics, events, and execution context
Connect produced files to a runArtifact Handlingreal outputs, hashes, and registration evidence
Select between candidate executionsCompare Runsmeasured comparison and resulting artifacts
Read and investigate stored evidenceQuery And Diagnosticsquery, comparison, and diagnostics roles
Learn the daily workflowCommon Workflowsinspect, report, diagnose, trace lineage
Track repeated workBatch, Sample & Deploymentbatches, samples, deployments, report sections
Use notebooksNotebook Surfaceinline summaries, display helpers, DataFrame conversion
Connect external toolsAdaptersStdout, OpenTelemetry, and MLflow sink boundaries
Map Contexta to real incidentsCase Studiescomplete scenario-specific programs and the evidence each one needs

Example Policy

Workflow examples should be executable by default. When a page teaches a domain workflow, examples should be presented with Docusaurus tabs for:

  • Machine Learning
  • Deep Learning
  • LLM

Short API snippets are acceptable on reference pages, but tutorial examples should create a workspace, emit evidence, and print something visible such as a run ref, report path, comparison result, or capture file.

Workflow examples should compute evidence from the work they execute, rather than insert illustrative output values.

Prerequisites

Most tutorials assume:

  • Python >=3.14
  • contexta installed in the active environment
  • a writable local directory

Optional framework packages belong to the tutorial that needs them. Contexta's core runtime stays local-first and does not require a cloud account.