{
  "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": "2610.19863",
  "title": "When Extremists Enter LLM Societies: Investigating Extreme Behavior in Multi-Agent Systems",
  "summary": "EXAM replaces average-persona or fixed-scalar treatments of tail actors with LLM-native extremist agents whose proportion, spatial distribution, interaction protocol, and measurement interface are explicitly configured. It costs additional agents, prompt-history state, API tokens or GPU inference time, scenario-specific design, and repeated multi-round simulations rather than a single-model call. It can fail when results are backbone-dependent, prompt-brittle, protocol-sensitive, or when measured collective shifts are simulation artifacts rather than transferable production behavior.",
  "rationale": "For a constrained production team this is a simulation methodology for stress-testing LLM multi-agent dynamics, not a deployable efficiency or quality improvement with a validated production gain.",
  "technique": "Configurable extremist agents for LLM societies",
  "family": "orquestracao_multiagente",
  "recommendation": "observar",
  "published": "2026-10-14",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.19863",
  "independent_implementations": 1,
  "total_implementations": 2,
  "stars": 132,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "EXAM replaces average-persona or fixed-scalar treatments of tail actors with LLM-native extremist agents whose proportion, spatial distribution, interaction protocol, and measurement interface are explicitly configured. It costs additional agents, prompt-history state, API tokens or GPU inference time, scenario-specific design, and repeated multi-round simulations rather than a single-model call. It can fail when results are backbone-dependent, prompt-brittle, protocol-sensitive, or when measured collective shifts are simulation artifacts rather than transferable production behavior.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.19863",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Reports measurable, backbone- and scenario-dependent collective shifts from small persistent extremist groups across four simulated scenarios; the abstract provides no numeric effect size.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.19863",
      "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": ""
}