01

Passing units do not prove a working journey

Unit tests are valuable because they provide fast, precise feedback about small parts of a system. Contract and integration tests expand that confidence across boundaries. None of them alone proves that a real user can complete a critical journey through the deployed product.

End-to-end automation operates at that product boundary. It verifies that interfaces, services, data stores, identity and integrations cooperate to produce the accepted behaviour.

02

Begin with Story behaviour

End-to-end scenarios should not be invented after implementation from whatever the interface happens to do. They should be derived from the accepted behaviour of the Product Story.

The Story identifies the valuable journeys, their preconditions, expected outcomes and important failure paths. Automation then becomes traceable product evidence rather than a collection of brittle browser scripts.

  • Critical user and system journeys
  • Authentication and authorisation boundaries
  • Data creation and state transitions
  • External service interactions
  • Failure, recovery and retry behaviour
  • Accessible and responsive interaction
03

Use a layered validation strategy

End-to-end tests are intentionally broader and usually slower than lower-level tests. They should complement, not replace, a layered strategy.

  • Unit tests for isolated rules and transformations
  • Component tests for bounded UI or service behaviour
  • Contract tests for interfaces between independently changing systems
  • Integration tests for real adapters and persistence
  • End-to-end tests for complete product journeys
  • Health and smoke checks for deployed environments
04

Playwright as an evidence provider

Playwright can exercise a product through the browser while collecting traces, screenshots, video and structured results. These artifacts are especially valuable when a scenario fails because they shorten the path from an observed symptom to a useful finding.

The tool is not the testing method. Reliable automation still requires controlled data, explicit environment assumptions, stable selectors, bounded retries and scenarios written around product behaviour rather than implementation detail.

05

Validate the release candidate in staging

A meaningful release decision should be based on the artifact intended for production. The pipeline builds an immutable candidate, deploys that candidate to staging and runs the required integration, end-to-end and health profiles there.

If the candidate changes after validation, it is a different candidate and its evidence no longer proves the same thing. Production promotion should reuse the exact staging-approved artifact rather than rebuilding from source after approval.

06

Evidence supports—but does not make—the decision

A passing suite records a fact about a particular artifact, environment, profile and observation time. A human reviewer still evaluates coverage, known risks, unresolved findings and operational readiness.

This separation prevents automation from approving itself. The pipeline may make a review gate ready; an authorised person decides whether the evidence is sufficient for release.

07

Failure becomes structured input

When an end-to-end scenario fails, the result should preserve the scenario, environment, candidate identity, trace and observed difference. A reviewer or engineer turns that evidence into a precise finding.

After correction, the affected validation is rerun against a new candidate. Earlier evidence remains historical fact rather than being overwritten by the successful attempt.

08

Production completes the evidence chain

Promotion is not the end of validation. Production smoke tests confirm that critical paths are reachable, while health checks and observability show whether the released capability behaves within expected boundaries.

End-to-end automation is therefore part of a broader evidence chain: acceptance behaviour, local validation, CI, staging journeys, human review, production smoke tests and continued observation. That chain makes release confidence explainable and repeatable.