{
  "caveats": [
    "Lab workload, not a KTS-1 block and not customer acceptance. One repository and one fictional shop; Keep's own minimal agent loop, not a product coding harness. The model may have seen this open-source repository in training; the planted bugs and shop data are new. Engines do not guarantee identical output under concurrent batching. Original publisher name with variant; exact variant repository and revision retained privately.",
    "Weights file SHA-256 was not recorded for public reproduction."
  ],
  "engine": {
    "name": "NVIDIA NGC vLLM",
    "recipe": "with multi-token prediction",
    "version": "0.29.0+5013de39.dev"
  },
  "format": "olir/0.1",
  "history": {
    "corrections": [],
    "supersedes": []
  },
  "id": "1bdb1c2c257c141430090354",
  "measured_at": "2026-10-21T08:07:59.563191+00:00",
  "metrics": [
    {
      "basis": null,
      "n": null,
      "name": "tasks_passed",
      "scope": "machine_total",
      "source_field": "agent_team.tasks_passed",
      "statistic": "single",
      "unit": "count",
      "value": 38
    },
    {
      "basis": null,
      "n": null,
      "name": "tasks_total",
      "scope": "machine_total",
      "source_field": "agent_team.tasks_total",
      "statistic": "single",
      "unit": "count",
      "value": 40
    },
    {
      "basis": null,
      "n": null,
      "name": "tool_calls",
      "scope": "machine_total",
      "source_field": "agent_team.tool_calls",
      "statistic": "single",
      "unit": "count",
      "value": 702
    },
    {
      "basis": null,
      "n": null,
      "name": "protocol_errors",
      "scope": "machine_total",
      "source_field": "agent_team.protocol_errors",
      "statistic": "single",
      "unit": "count",
      "value": 0
    },
    {
      "basis": null,
      "n": null,
      "name": "engine_errors",
      "scope": "machine_total",
      "source_field": "agent_team.engine_errors",
      "statistic": "single",
      "unit": "count",
      "value": 0
    },
    {
      "basis": null,
      "n": null,
      "name": "loops",
      "scope": "machine_total",
      "source_field": "agent_team.loops",
      "statistic": "single",
      "unit": "count",
      "value": 0
    },
    {
      "basis": null,
      "n": null,
      "name": "model_requests",
      "scope": "machine_total",
      "source_field": "agent_team.model_requests",
      "statistic": "single",
      "unit": "count",
      "value": 599
    },
    {
      "basis": null,
      "n": null,
      "name": "peak_concurrent_requests",
      "scope": "machine_total",
      "source_field": "agent_team.peak_concurrent_requests",
      "statistic": "max",
      "unit": "count",
      "value": 5
    },
    {
      "basis": null,
      "n": null,
      "name": "mean_concurrent_requests",
      "scope": "machine_total",
      "source_field": "agent_team.mean_concurrent_requests",
      "statistic": "mean",
      "unit": "count",
      "value": 2.9560184109560557
    },
    {
      "basis": null,
      "n": null,
      "name": "ttft_p50_s",
      "scope": "per_person",
      "source_field": "agent_team.ttft_p50_s",
      "statistic": "median",
      "unit": "s",
      "value": 1.019965103827417
    },
    {
      "basis": null,
      "n": null,
      "name": "ttft_p95_s",
      "scope": "per_person",
      "source_field": "agent_team.ttft_p95_s",
      "statistic": "p95",
      "unit": "s",
      "value": 2.2109051721636206
    },
    {
      "basis": null,
      "n": null,
      "name": "per_request_decode_p50_tps",
      "scope": "per_person",
      "source_field": "agent_team.per_request_decode_p50_tps",
      "statistic": "median",
      "unit": "tok/s",
      "value": 42.89717956971035
    },
    {
      "basis": null,
      "n": null,
      "name": "task_p50_seconds",
      "scope": "per_person",
      "source_field": "agent_team.task_p50_seconds",
      "statistic": "median",
      "unit": "s",
      "value": 66.92939303885214
    },
    {
      "basis": null,
      "n": null,
      "name": "task_p95_seconds",
      "scope": "per_person",
      "source_field": "agent_team.task_p95_seconds",
      "statistic": "p95",
      "unit": "s",
      "value": 656.0206727720797
    },
    {
      "basis": null,
      "n": null,
      "name": "max_prompt_tokens",
      "scope": "machine_total",
      "source_field": "agent_team.max_prompt_tokens",
      "statistic": "max",
      "unit": "tokens",
      "value": 51639
    },
    {
      "basis": null,
      "n": null,
      "name": "wall_seconds",
      "scope": "machine_total",
      "source_field": "agent_team.wall_seconds",
      "statistic": "single",
      "unit": "s",
      "value": 2083.3306227349676
    }
  ],
  "model": {
    "artifact_sha256": [],
    "publisher_repo": "Qwen/Qwen3.6-35B-A3B",
    "publisher_revision": "995ad96eacd98c81ed38be0c5b274b04031597b0",
    "variant": "NVFP4 variant",
    "weights_dtype": "NVFP4",
    "weights_format": "safetensors"
  },
  "outcome": {
    "gates": [
      {
        "met": true,
        "name": "agent-pass-rate",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-category-pass-rate",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-protocol-error-rate",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-engine-errors",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-loops",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-concurrency",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-complete",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-ttft-p95",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-decode-p50",
        "observed": null
      },
      {
        "met": true,
        "name": "agent-task-seconds-p95",
        "observed": null
      }
    ],
    "reasons": [],
    "status": "pass"
  },
  "publication_group": "lab workload",
  "published": null,
  "publisher": {
    "name": "Skilak Keep"
  },
  "quality": [
    {
      "passed": 38,
      "skipped": null,
      "suite": "agent-tasks",
      "total": 40,
      "version": "agent-team-v1"
    },
    {
      "passed": 10,
      "skipped": null,
      "suite": "agent-tasks-bugfix",
      "total": 12,
      "version": "agent-team-v1"
    },
    {
      "passed": 12,
      "skipped": null,
      "suite": "agent-tasks-data",
      "total": 12,
      "version": "agent-team-v1"
    },
    {
      "passed": 6,
      "skipped": null,
      "suite": "agent-tasks-feature",
      "total": 6,
      "version": "agent-team-v1"
    },
    {
      "passed": 8,
      "skipped": null,
      "suite": "agent-tasks-investigate",
      "total": 8,
      "version": "agent-team-v1"
    },
    {
      "passed": 2,
      "skipped": null,
      "suite": "agent-tasks-multistep",
      "total": 2,
      "version": "agent-team-v1"
    }
  ],
  "record_date": "2026-10-21",
  "reproduce": {
    "driver": null,
    "fixture_sha256": "6451aabd1c9994d0f73355c962a5eaf9258381b3f2554e53db916cd9a605552a",
    "harness": {
      "name": "keep-agent-team-result-v1",
      "version": "1"
    },
    "harness_sha256": "484ed3e2b82f09ee19c87be819253811cb273628abfb3c9471a59746605e537d",
    "os": null,
    "workload_name": "agent-team-v1",
    "workload_version": "1"
  },
  "requests": [],
  "scenario": {
    "id": "agent-team",
    "version": "1"
  },
  "settings": {
    "concurrency": 5,
    "configured_context_tokens": 262144,
    "kv_dtype": "fp8",
    "max_output_tokens": 8192,
    "speculative_decoding": "with multi-token prediction"
  },
  "setup_complete": false,
  "setup_id": "1613bbe6cf3a8a03e019ade665b34366",
  "standard": "lab workload",
  "system": {
    "driver": null,
    "hardware": "NVIDIA GB10, 128 GB unified memory",
    "os": null
  },
  "workload": {
    "cache": null,
    "content_tokens": null,
    "document": null,
    "input_tokens": null,
    "output_tokens": 8192,
    "repetitions": null,
    "users": 5,
    "warmup": null
  }
}