Skip to content

Latest commit

 

History

History
29 lines (22 loc) · 1.5 KB

File metadata and controls

29 lines (22 loc) · 1.5 KB

Benchmark Sites

This directory is the single local host for deterministic websites used by browser-agent benchmark tasks. Each simulated website owns one route inside the application, while installation, build, port, and process lifecycle stay shared.

Rules

  • Keep each site under app/<site-slug>/, served at /<site-slug>.
  • Site routes must not import from ego_bench/ or frontend/.
  • Benchmark datasets reference a site by slug ("local_site": "<slug>"), never by URL and never by site internals. ego_bench/task_resources.py joins the slug to this host's origin, so the port below is the only place it is written down.
  • Observable success and failure states must remain visible for rubric grading.
  • Do not add shared infrastructure until a second site proves the same invariant.
  • Local harness startup owns one fixed origin: http://127.0.0.1:4374. Override it for both sides with BENCHMARK_SITES_ORIGIN plus the matching port flag.

Adding a site is therefore a directory and a dataset field, with nothing to register in between. tests/test_task_resources.py fails if a dataset names a slug with no directory here, or if the port this host binds drifts from the one the harness expects.

make dev and make serve start this host with the rest of the local harness stack. make benchmark-sites-dev and make benchmark-sites-serve run it alone.

Sites

  • /ticket-rush — an English-language U.S. concert ticket sale with a navigation-relative two-minute wait and a two-second claim window.