Visual regression (VRT)

Screenshot every page, compare it to an approved baseline, and get a pixel diff when a deploy shifts the layout — plus baselines, accepting changes, weighted zones, and the variance learner.

Visual regression testing (VRT) catches the failures that never throw an error: a broken header, a shifted layout, a third-party script that mangles the page. DeadWatch screenshots each page and compares it to a baseline you approved.

How VRT works

Every VRT scan captures a screenshot and diffs it against the site's baseline. The diff is scored as a percentage and bucketed:

• Unchanged (≤1%)
• Minor (1–5%)
• Major (5–10%)
• Critical (>10%)

Set a baseline

A site needs an approved baseline before it can diff anything.

1. Open a site and go to the VRT tab.
2. Click Capture Baseline Screenshot (or Run VRT Scan to capture, then Set as Baseline).
3. To do both in one step, use Scan + Set Baseline.

You'll know it worked when… the warning "No baseline set" disappears and future scans show a diff percentage.

Read a diff

1. On the VRT tab, open a result to see the Screenshot Comparison.
2. Switch comparison modes: Side-by-side, Overlay (default), Diff (highlights the changed pixels), and Zones.
3. Compare across viewports (e.g. desktop-hd) using the Compare Viewports panel, and switch Capture Modes between Hero (above-fold) and Full page.

Accept a change as the new baseline

When a diff is a real, intended change (a redesign, new content), promote it so it stops flagging.

1. On the flagged result, click Accept (or Accept Change).
2. That screenshot becomes the new baseline and older results are dismissed.

You'll know it worked when… the result shows as Accepted and its diff is treated as resolved (struck-through, effective 0%).

Dismiss false positives

When a diff is just noise — an ad, a rotating banner, a mid-animation frame — mark it so DeadWatch learns to ignore similar diffs.

1. Click FP (Mark FP) on the result.
2. Pick a False Positive Reason: Rendering Noise, Dynamic Content, Timing Issue, or Other (add a note).

You'll know it worked when… the toast reads "Marked as false positive (N similar auto-suppressed)" — the noise filter is now trained.

Weighted zones

Not every part of a page matters equally. Weight the areas you care about and mask the ones you don't.

1. On the VRT tab, click Manage Zones.
2. Use Pick mode to click directly on the baseline and define weighted zones, mask regions, and per-zone alert thresholds.

History rows can then be filtered by which weighted zones a diff touched.

Train the variance learner

Some pages always churn a little (carousels, live feeds). Instead of a fixed "low-fidelity" capture, DeadWatch learns each page's normal variance from multiple reloads.

1. Click Run Learner Pass — it captures several reloads to learn where the page naturally moves.
2. When it finishes, review the suggested zones (classified as steady-high, episodic, or shift-only) and approve them.

Approved variance is auto-suppressed, so only real regressions page you. The VRT Archive page gives you the same picture across every site at once — filter by Changed, Critical, Failed, or Unchanged.

Next: Speed, SEO & SSL → — the quiet signals that tank a site without an error page.