Problem
Several WebHarbor mirror sites expose generic local /search?q=... URLs from their own search UI even when the upstream sites use more specific URL shapes.
Examples:
- Amazon upstream uses
/s?k=<query>.
- Booking upstream uses
/searchresults.html?ss=<query>.
- Google Maps upstream uses
/maps/search/<query>.
- ESPN upstream uses
/search/_/q/<query>.
- Apple upstream uses
/search/<query>.
The current generic URLs still work for local benchmark tasks, but they reduce realism when agents inspect, copy, or navigate search URLs.
Expected
The site UI should emit realistic upstream-shaped search URLs as the canonical path, while old /search?q=... routes remain available as backward-compatible aliases for existing tasks and trajectories.
Proposed Fix
- Add canonical search routes for sites with known upstream URL forms.
- Update site-owned search forms to emit canonical URL shapes.
- Preserve existing
/search?q=... compatibility routes.
- Add documentation and a regression check for canonical vs legacy search URL policy.
Problem
Several WebHarbor mirror sites expose generic local
/search?q=...URLs from their own search UI even when the upstream sites use more specific URL shapes.Examples:
/s?k=<query>./searchresults.html?ss=<query>./maps/search/<query>./search/_/q/<query>./search/<query>.The current generic URLs still work for local benchmark tasks, but they reduce realism when agents inspect, copy, or navigate search URLs.
Expected
The site UI should emit realistic upstream-shaped search URLs as the canonical path, while old
/search?q=...routes remain available as backward-compatible aliases for existing tasks and trajectories.Proposed Fix
/search?q=...compatibility routes.