{
  "schema_version": "oersi-local-model-bakeoff-public-v1",
  "measured_at_utc": "2026-10-29T10:15:08+00:00",
  "scope": "3 synthetic Persian tasks, one scored sample per candidate, same prompts and decode options; model-component comparison, not agent-level competition",
  "configuration": {
    "task_count": 3,
    "samples_per_candidate": 3,
    "temperature": 0.1,
    "seed": 42,
    "context_tokens": 4096,
    "task_output_budgets": {
      "arithmetic": 16,
      "formal_logic": 48,
      "evidence_boundary": 48
    },
    "shared_system_prompt": true,
    "warmup_unscored": true
  },
  "summary": {
    "status": "COMPLETED",
    "scored_calls": 6,
    "tasks_per_candidate": 3,
    "all_requests_ok": true,
    "responses_received": 6
  },
  "candidates": [
    {
      "candidate": "production_default_7b",
      "task_count": 3,
      "rubric_passed": 6,
      "rubric_total": 7,
      "rubric_ratio": 0.857,
      "median_latency_ms": 12602,
      "max_latency_ms": 19006,
      "per_task": {
        "arithmetic": {
          "rubric_passed": 1,
          "rubric_total": 1,
          "latency_ms": 2536,
          "request_status": "OK"
        },
        "formal_logic": {
          "rubric_passed": 3,
          "rubric_total": 3,
          "latency_ms": 12602,
          "request_status": "OK"
        },
        "evidence_boundary": {
          "rubric_passed": 2,
          "rubric_total": 3,
          "latency_ms": 19006,
          "request_status": "OK"
        }
      },
      "responses_received": 3
    },
    {
      "candidate": "fast_candidate_3b",
      "task_count": 3,
      "rubric_passed": 2,
      "rubric_total": 7,
      "rubric_ratio": 0.286,
      "median_latency_ms": 6153,
      "max_latency_ms": 17310,
      "per_task": {
        "arithmetic": {
          "rubric_passed": 0,
          "rubric_total": 1,
          "latency_ms": 6153,
          "request_status": "OK"
        },
        "formal_logic": {
          "rubric_passed": 1,
          "rubric_total": 3,
          "latency_ms": 5196,
          "request_status": "OK"
        },
        "evidence_boundary": {
          "rubric_passed": 1,
          "rubric_total": 3,
          "latency_ms": 17310,
          "request_status": "OK"
        }
      },
      "responses_received": 3
    }
  ],
  "limits": [
    "One sample per task; no statistical significance or broad quality claim.",
    "Regex rubrics are narrow proxies, not semantic adjudication or proof of useful real-world task completion.",
    "These are local model routes, not independent long-running agents.",
    "No external agent was compared; no superiority, general intelligence, or consciousness claim is supported.",
    "The endpoints have different runtime routing/overhead; treat latency as observed route latency, not pure model latency."
  ],
  "interpretation": "Rubric ratios count matched regex checks over seven checks and three fixed tasks; they are not verified task-success or model-accuracy rates."
}