{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "A workflow needs specialist review, but extra agents may add more cost than useful work.",
    "change": "Test the proposed division of work against a single-agent implementation.",
    "baseline": "One agent with the same tools and an equal total token and time budget.",
    "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": "Accepted outputs, coordination failures, wall-clock time, and total cost across all agents.",
    "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": "2610.29292",
  "title": "Credit Where Credit Is Due: Contrastive Attribution Topology Search for Multi-Agent Systems",
  "summary": "CATS replaces score-only heuristic or random search over multi-agent topologies with LLM-attributed per-agent credit from contrasted execution traces, pruning weak edges and agents to yield a compact deployment team. The cost is an additional search phase: matched candidate evaluations across the supergraph plus LLM attributer calls, which adds offline token spend and implementation complexity before any deployment savings are realized. Credit assignment can fail when attributer judgments are noisy or confounded, traces do not isolate individual agent effects, or the pruned topology overfits the search benchmarks and underperforms on shifted production distributions.",
  "rationale": "The method runs with small open-weight models and needs no training or cluster, so it is feasible on constrained infrastructure, but the reported gains are benchmark-specific and the attribution mechanism requires workload-specific validation before adoption.",
  "technique": "LLM-attribution topology search for multi-agent teams",
  "family": "orquestracao_multiagente",
  "recommendation": "testar",
  "published": "2026-10-21",
  "as_of": "2026-10-22",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.29292",
  "independent_implementations": 2,
  "total_implementations": 3,
  "stars": 132,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "CATS replaces score-only heuristic or random search over multi-agent topologies with LLM-attributed per-agent credit from contrasted execution traces, pruning weak edges and agents to yield a compact deployment team. The cost is an additional search phase: matched candidate evaluations across the supergraph plus LLM attributer calls, which adds offline token spend and implementation complexity before any deployment savings are realized. Credit assignment can fail when attributer judgments are noisy or confounded, traces do not isolate individual agent effects, or the pruned topology overfits the search benchmarks and underperforms on shifted production distributions.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.29292",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "On BigCodeBench, CATS improves mean accuracy by 5.3 percentage points over score-only heuristic search while using 54% fewer deployment tokens; on MBPP it matches the strongest baselines with fewer tokens, and in a HotpotQA prompt-poisoning test it excludes all adversarial agents in all three runs.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.29292",
      "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": ""
}