{
  "checked_at": "2026-10-16T12:15:53.404007+00:00",
  "source_revision": "e885130962f61fb64ac305497996a46cfc7d18f0",
  "registered_records": 715,
  "atomic_l1_templates": 33,
  "shared_compositional_workflows": 341,
  "compositional_level_variants": 682,
  "levels": {
    "L1": 33,
    "L2": 341,
    "L3": 341
  },
  "independent_registry": {
    "repository": "https://github.com/ServiceNow/BrowserGym",
    "revision": "e71419aef5759a11cb7896a26c36e5f7e92d82b2",
    "url": "https://github.com/ServiceNow/BrowserGym/blob/e71419aef5759a11cb7896a26c36e5f7e92d82b2/browsergym/experiments/src/browsergym/experiments/benchmark/metadata/workarena.csv",
    "sha256": "013b99b2712abe931c792083b10ed9e0cb097a64215650d2700d3b0541629a47",
    "canonical_records_sha256": "4b91beecab17102e2eb3a26ee76584ba9cfddb1dbf2f93f421cf82aa412d7762",
    "comparison": "Exact task IDs, levels and categories match; metadata cross-check only, not execution validation"
  },
  "curriculum": {
    "instances_per_level": 235,
    "categories": {
      "planning_and_problem_solving": {
        "workflow_classes": 66,
        "buckets": 7,
        "instances_per_level": 44
      },
      "information_retrieval": {
        "workflow_classes": 39,
        "buckets": 8,
        "instances_per_level": 56
      },
      "data_driven_decision_making_and_reasoning": {
        "workflow_classes": 156,
        "buckets": 13,
        "instances_per_level": 55
      },
      "sophisticated_memory": {
        "workflow_classes": 28,
        "buckets": 6,
        "instances_per_level": 48
      },
      "contextual_understanding_infeasible_tasks": {
        "workflow_classes": 52,
        "buckets": 8,
        "instances_per_level": 32
      }
    },
    "counted_as_extra_records": false
  },
  "prior_records_unchanged": 2996,
  "other_source_archives_unchanged": [
    "osworld",
    "windowsarena",
    "webarena",
    "visualwebarena",
    "macosworld",
    "androidworld",
    "theagentcompany",
    "devrev-enterprise",
    "iosworld"
  ],
  "new_execution_runs": 0,
  "validation_observation": {
    "scope": "Generic CompositionalTask.validate; some workflow classes override validation",
    "behavior": "Skip unvalidated subtasks; validate one current checkpoint; retain progress after a pass without rechecking earlier checkpoints",
    "interpretation": "Historical checkpoint success does not by itself prove simultaneous final-state correctness. No environment exploit or failure rate was measured.",
    "source_url": "https://github.com/ServiceNow/WorkArena/blob/e885130962f61fb64ac305497996a46cfc7d18f0/src/browsergym/workarena/tasks/compositional/base.py#L217"
  },
  "paper": "https://arxiv.org/html/2407.05291v2#S3.SS1",
  "paper_count_note": {
    "section": "Appendix C.4",
    "heading_workflows": 29,
    "described_navigation_workflows": 26,
    "described_user_management_workflows": 2,
    "pinned_registry_workflows": 28,
    "interpretation": "The heading differs from its 26 + 2 description. The import follows the pinned code and independent BrowserGym metadata, both 28; no missing workflow is invented."
  }
}
