{
  "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": "2610.26826",
  "title": "RefCraft: Benchmarking Visual-Reference-Conditioned Agentic Generation through Web Development",
  "summary": "RefCraft replaces idealized webpage-generation evaluations (single-screenshot reconstruction or text-only instructions) with 300 multimodal tasks combining instructions and one to five reference screenshots, scored via keypoints plus programmatic and VLM judges. It costs nothing in model deployment but requires API or GPU budget to run candidate models through render-inspect-refine loops, and the VLM-judge component adds evaluation variance and API cost. It can fail as a selection signal if the reader's production tasks differ from webpage development, if VLM judge scores are noisy or gameable, or if benchmark overfitting substitutes for real user evaluation.",
  "rationale": "For a team shipping agentic web-development features, this benchmark is directly usable via third-party APIs to compare candidate models, but its value depends on the reader's workload actually matching the webpage-development domain.",
  "technique": "Visual-reference-conditioned webpage generation benchmark (RefCraft)",
  "family": "avaliacao_de_agente",
  "recommendation": "testar",
  "published": "2026-10-19",
  "as_of": "2026-10-20",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.26826",
  "independent_implementations": 2,
  "total_implementations": 3,
  "stars": 132,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "RefCraft replaces idealized webpage-generation evaluations (single-screenshot reconstruction or text-only instructions) with 300 multimodal tasks combining instructions and one to five reference screenshots, scored via keypoints plus programmatic and VLM judges. It costs nothing in model deployment but requires API or GPU budget to run candidate models through render-inspect-refine loops, and the VLM-judge component adds evaluation variance and API cost. It can fail as a selection signal if the reader's production tasks differ from webpage development, if VLM judge scores are noisy or gameable, or if benchmark overfitting substitutes for real user evaluation.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.26826",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Best of nine frontier models scores 52.12/100 on the benchmark; no method-level improvement is claimed.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.26826",
      "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": ""
}