Test Types

Smoke is the first line of defense. It should answer “is this site broken in an obvious way?” quickly and cheaply.

Best used when:

• you want a fast health check
• you just pulled main
• you want a safe first CI or local run

Main command: npm run test:smoke

Shadow is closer to production monitoring than classic regression testing. It looks for degradation, flaky navigation, and structural issues that users would feel before a full outage is obvious.

Main command: npm run test:shadow

PDT means post-deployment tests. This suite is designed for the question “did the deploy break critical behavior?”

Main command: npm run test:pdt

E2E is where the platform starts behaving like a real reader instead of a fast monitor. These tests cover actual browsing journeys: article discovery, galleries, video, social sharing, and site-specific editorial surfaces.

Main command: npm run test:e2e

User-Flows is the premium and auth suite. It exists because anonymous browsing and logged-in premium behavior break in different ways.

Main command: npm run test:user-flows

Mobile is not one flat suite. It is intentionally tiered so teams can choose the right amount of confidence for the moment.

%%{init: {'theme':'base', 'themeVariables': {'primaryColor': '#4a90d9', 'primaryTextColor': '#fff', 'primaryBorderColor': '#2c6fad', 'lineColor': '#555', 'secondaryColor': '#e8f4e8', 'tertiaryColor': '#fff8e1', 'fontFamily': 'sans-serif'}}}%%
flowchart TB
  subgraph smoke_tier["@smoke — PR Gate (9 tests)"]
    direction LR
    S1["Viewport meta"] --- S2["No overflow"] --- S3["Breakpoints"]
    S4["Images"] --- S5["Touch targets"] --- S6["Hamburger"]
    S7["Tap navigation"] --- S8["Scroll"] --- S9["Consent"]
  end
  subgraph regression_tier["@regression — Nightly (8 tests)"]
    direction LR
    R1["Sticky header"] --- R2["Form inputs"] --- R3["Video presence"]
    R4["Media playback"] --- R5["Lazy images"] --- R6["Live surface"]
    R7["Orientation"]
  end
  subgraph deep_tier["@deep — Nightly (6 tests)"]
    direction LR
    D1["Load budget"] --- D2["LCP budget"] --- D3["CLS budget"]
    D4["DOM budget"] --- D5["Subscription"] --- D6["Full journey"]
  end
  smoke_tier --> regression_tier --> deep_tier

Important characteristics:

• Chrome and Safari variants exist
• matrix jobs can run all sites
• per-site mobile seed URLs reduce content-rotation flake
• mobile is treated as a product surface, not just a viewport resize

Main commands:

Content Tests are one of the biggest changes in the platform. They moved the system from “does the page load?” to “are the editorial modules we care about still rendering and behaving?”

%%{init: {'theme':'base', 'themeVariables': {'primaryColor': '#4a90d9', 'primaryTextColor': '#fff', 'primaryBorderColor': '#2c6fad', 'lineColor': '#555', 'fontFamily': 'sans-serif'}}}%%
flowchart LR
    REG[“Content Registry\n120 entries\n231 site slots”] --> ORCH[“Orchestrator\ngroup by context\nnavigate once”]
    ORCH --> HP[“Homepage\nContext”]
    ORCH --> ART[“Article\nContext”]
    ORCH --> SEED[“SeedURL\nContext”]
    HP --> HAND[“12 Handlers\narticle, gallery, video,\nnav, search, embed, ...”]
    ART --> HAND
    SEED --> HAND
    HAND --> RES[“Results\n+ JSONL Logs\n+ History”]

Main commands:

Visual is often misunderstood. The current suite is not a full design review suite. It is a structural screenshot suite that tries to catch layout drift without being overwhelmed by constantly changing headlines, images, ads, and live content.

Important notes:

• the suite is baseline-dependent
• Linux baselines are what matter in CI
• missing baselines are a configuration problem, not a visual regression
• the suite currently uses 18 checks: 3 structural surfaces across 6 sites

Main command: npm run test:visual

Performance tests answer a different question from functional suites: “Did the site still load well enough?”

Main commands:

Ads tests exist because revenue regressions are their own category of production issue.

Main commands:

Events tests check instrumentation rather than visible UI. They answer “are the expected analytics events still firing from real flows?”

Main commands:

If you are not sure where to start, do not jump straight to the deepest suite. Start with the cheapest suite that can still falsify your assumption. That keeps feedback fast and makes failures easier to interpret.