README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: full
        path: full.parquet
      - split: hard
        path: hard.parquet

WebArena-Verified

Dataset description

WebArena-Verified is a curated benchmark dataset of web tasks designed for reproducible evaluation of web agents across multiple realistic websites.

Sources

• GitHub repository: webarena-verified
• Original WebArena benchmark: webarena.dev

Splits

• full: 812 rows
• hard: 258 rows

Tasks per site

Counts below are task counts grouped by category. Tasks with more than one site are grouped under multi-category to avoid double-counting.

Full split

Site	Tasks
gitlab	180
map	109
reddit	106
shopping_admin	182
shopping	187
wikipedia	0
homepage	0
multi-category	48

Hard split

Site	Tasks
gitlab	57
map	0
reddit	42
shopping_admin	55
shopping	56
wikipedia	0
homepage	0
multi-category	48

Schema notes

The table below reflects inferred column types from the full split during artifact generation.

Column	Type
sites	List(Value('string'))
task_id	Value('int64')
intent_template_id	Value('int64')
start_urls	List(Value('string'))
intent	Value('string')
intent_template	Value('string')
instantiation_dict	Value('string')
eval	Value('string')
revision	Value('int64')

Metadata

• Version: v1.2.3
• Git commit: 6473f72db5dcefc97b5725b59e734504edc28a21
• Generated at (UTC): 2026-02-07T22:44:22Z
• Dataset hash: 0e90f3feabe68cb9c7285e6989e36862a47622e7ac81b562f3126152350cb5ac
• License: Apache-2.0
• Language: en
• Task categories: web-navigation, information-retrieval

Notes

• Expected split counts: full=812, hard=258
• If custom split names are not supported by the Dataset Viewer, use one config per split mapped to train as a compatibility fallback.