{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "An agent looks good in demos, but you cannot tell which product tasks it fails.",
    "change": "Use the study's evaluation idea to inspect a small set of tasks with human-checked answers.",
    "baseline": "Your current evaluator and a manually reviewed reference set.",
    "steps": [
      "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.",
      "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.",
      "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.",
      "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test."
    ],
    "measure": "Agreement with human review, missed failures, false alarms, and evaluation cost per task.",
    "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation."
  },
  "builder_review": null,
  "arxiv_id": "2610.31274",
  "title": "PointerBench: A Diagnostic Benchmark for Fine-Grained GUI Grounding",
  "summary": "PointerBench replaces ad-hoc or end-to-end-only evaluation of GUI grounding models with a diagnostic suite of 1,500 pixel-exact screenshot-instruction pairs covering spreadsheets, long-text documents, and 100 professional applications. Adoption costs only evaluation compute: running candidate models against the released benchmark and code, which fits a single 24 GB GPU for local specialist models or API calls for frontier ones. It can fail if the reader's production UI distribution differs from the benchmark's applications, since aggregate scores may not predict grounding accuracy on the reader's specific interfaces.",
  "rationale": "The reader deploying computer-use or GUI agents needs a cheap, reproducible way to select and regression-test grounding models, and this benchmark directly supports that decision but must be validated against the reader's own interface workload before driving adoption choices.",
  "technique": "Diagnostic benchmark for fine-grained GUI grounding",
  "family": "avaliacao_de_agente",
  "recommendation": "testar",
  "published": "2026-10-22",
  "as_of": "2026-10-23",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.31274",
  "independent_implementations": 0,
  "total_implementations": 1,
  "stars": 0,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "PointerBench replaces ad-hoc or end-to-end-only evaluation of GUI grounding models with a diagnostic suite of 1,500 pixel-exact screenshot-instruction pairs covering spreadsheets, long-text documents, and 100 professional applications. Adoption costs only evaluation compute: running candidate models against the released benchmark and code, which fits a single 24 GB GPU for local specialist models or API calls for frontier ones. It can fail if the reader's production UI distribution differs from the benchmark's applications, since aggregate scores may not predict grounding accuracy on the reader's specific interfaces.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.31274",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "exposure_map": [
    {
      "dimension": "quality",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compute",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "latency",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "operations",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compatibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "security",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "data_and_training",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "reproducibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "risks": [],
  "minimum_test": [],
  "open_questions": [],
  "independent_tests": [],
  "lab_reproductions": [],
  "report_available": false,
  "validation_tier": "unknown",
  "evidence_tier": "unknown",
  "infrastructure_basis": "unknown",
  "equations": [],
  "equations_status": "not_fetched",
  "core_kind": "",
  "equations_fetched_at": "",
  "equations_source_url": "",
  "equations_source_sha256": "",
  "exposures_reviewed_at": "",
  "exposures_source_sha256": ""
}