Test Levels
The CD Model uses a taxonomy of test levels (L0-L4) based on execution environment and scope.
Taxonomy Overview
The following diagram provides a visual overview of the test level taxonomy, showing the L0-L4 levels with their execution environments, scope, and the recommended distribution ratio (70% L0-L1, 20% L2, 10% L3-L4).
| Level | Name | Environment | Scope | Dependencies | Determinism |
|---|---|---|---|---|---|
| L0-L1 | Unit Tests | DevBox/Agent | Source/Binary | All test doubles | Highest |
| L2 | Emulated System | DevBox/Agent | Deployable artifacts | All test doubles | High |
| L3 | In-Situ Vertical | PLTE | Deployed system | All test doubles | Moderate |
| L4 | Production Testing | Production | Deployed system | Real services | High |
Anti-Pattern (Out-of-Category):
| Level | Name | Environment | Dependencies | Determinism |
|---|---|---|---|---|
| HE2E | Horizontal E2E | Shared test env | Other teams' deployments | Lowest |
L0-L1: Unit Tests
Purpose: Validate individual functions, methods, and component interactions in isolation.
| Characteristic | Description |
|---|---|
| Speed | Milliseconds per test |
| Isolation | No external dependencies, all mocked |
| Scope | Single function/method (L0) or component interactions (L1) |
| Execution | In-process, no network calls |
CD Stages: 2 (Pre-commit), 3 (MR), 4 (Commit)
L2: Emulated System Tests
Purpose: Validate deployable artifacts in an emulated environment with test doubles.
| Characteristic | Description |
|---|---|
| Speed | Seconds per test |
| Isolation | Cross-process, but all external deps are test doubles |
| Scope | Deployable artifacts, component-to-component communication |
| Execution | Multiple processes, network calls to test doubles only |
Test doubles: In-memory databases, mock message queues, stubbed external APIs.
CD Stages: 3 (MR), 4 (Commit)
L3: In-Situ Vertical Tests
Purpose: Validate a deployed module in PLTE with vertical testing boundaries.
| Characteristic | Description |
|---|---|
| Speed | Minutes per test |
| Isolation | Single deployable module only (vertical) |
| Scope | Deployed system in production-like infrastructure |
| Execution | Cloud environment, test doubles for ALL external services |
Validates:
- Deployment procedures work
- Infrastructure configuration correct (networking, DNS, load balancing)
- Service responds correctly in cloud environment
Does NOT validate: Cross-service interactions (that's L4).
CD Stages: 5 (Acceptance), 6 (Extended)
L4: Production Testing
Purpose: Validate real cross-service interactions in production.
| Characteristic | Description |
|---|---|
| Speed | Seconds to minutes |
| Isolation | None - real production environment |
| Scope | Cross-service workflows (horizontal) |
| Execution | Production, may use live test doubles for specific cases |
Validates:
- Real cross-service interactions
- End-to-end user journeys
- Production infrastructure under real load
Includes: Synthetic monitoring, canary validation, exploratory testing.
CD Stages: 11 (Live), 12 (Release Toggling)
Horizontal E2E Anti-Pattern
Shared pre-production environments where multiple teams deploy pre-prod services linked together.
| Problem | Impact |
|---|---|
| Any team's broken deployment breaks everyone | Blocking |
| Version mismatches common | Non-deterministic results |
| Hard to isolate failures | Difficult debugging |
| Can't test until all deps deployed | Slow feedback |
Solution: Shift LEFT (L0-L3 with test doubles) and RIGHT (L4 in production).
Shift-Left and Shift-Right Strategy
The shift-left/shift-right strategy moves testing emphasis away from the fragile middle (shared pre-production environments) toward fast, deterministic tests on the left and real production validation on the right.
The following diagram shows how this strategy maps across test levels and CD stages:
Why this works:
| Strategy | Benefit |
|---|---|
| Shift LEFT | Fast feedback, high determinism, no cross-team dependencies |
| Shift RIGHT | Real production validation without pre-prod fragility |
| Avoid middle | Skip the worst of both worlds (slow, fragile, non-deterministic) |
Test Distribution
| Level | Volume | Purpose |
|---|---|---|
| L0-L2 | 95% (hundreds-thousands) | Fast, deterministic validation |
| L3 | 5% (5-20 scenarios) | Infrastructure validation |
| L4 | Continuous | Production monitoring |
External Resources
Process Isolation by Level
Each test level operates within different process boundaries. Understanding these boundaries helps design appropriate test strategies.
L0-L1: Single Process
Tests run in-process with the code artifact. No cross-process communication occurs.
L2: Multi-Process/Container
Tests involve multiple processes or containers, including the production container and L2 test containers, with test orchestration coordinating them.
L3: Cloud Environment
Tests run against distributed deployments across cloud infrastructure, with test orchestration reaching into PLTE environments.
Comprehensive Taxonomy Reference
The following diagram provides a complete view of the test taxonomy matrix, showing test levels across all CD stages with shift-left and shift-right indicators.
Next Steps
- Verification Types - IV/OV/PV for acceptance testing
- Testing Taxonomy - Tag contracts and test suites
- Environments - PLTE for L3 tests
Tutorials | How-to Guides | Explanation | Reference
You are here: Explanation — understanding-oriented discussion that clarifies concepts.