Skip to content

MAINTENANCE.md

Latest commit

 

History

History
46 lines (34 loc) · 2.01 KB

MAINTENANCE.md

File metadata and controls

46 lines (34 loc) · 2.01 KB
Raw

Maintenance notes

Search Restore changes a page it does not control. Keep these assumptions in the release checklist.

Current Google result structure

On August 23, 2026, a read-only browser check of https://www.google.com/search?q=example&hl=en&gl=us&pws=0 returned:

  • one #rso result container;
  • seven h3 elements with a closest external link;
  • one .MjjYud wrapper per linked result;
  • no unusual-traffic or captcha page.

The extension uses #rso and linked h3 elements as its main result signals. .MjjYud is only the preferred pagination wrapper. A generic single-heading ancestor is the fallback because generated Google class names can change.

The production core.js and content.js files were also run in a fresh logged-out tab. They added seven archive rows and one load control. One click fetched start=10, found nine new result blocks, raised the linked-title count from 7 to 16 and changed the button to Load results 21-30.

That test found a false block signal in the first detector: Google's normal page scripts contain literal /sorry/index checks. looksBlocked() now strips script and style blocks before using its HTML fallback. The response status and an actual redirect to /sorry/ remain the primary checks in content.js.

Known failure cases

The extension intentionally does nothing when /search has no #rso block. This includes some AI Mode, consent and zero-result layouts. A Google captcha or HTTP 429 stops pagination and leaves a status message. The extension does not attempt to bypass those checks.

If Google removes start= pagination, the load-more feature cannot reconstruct pages that Google no longer serves. If result titles stop using linked h3 elements, archive rows will need a new reviewed selector.

Release check

Use an unsigned, logged-out browser profile. Confirm #rso, linked h3 titles, archive-row placement and one next-page fetch. Check a normal web query and a query with an AI Overview. Record limitations instead of marking an untested layout as working.