{
  "agentic_oracle7": {
    "attempted": 18,
    "campaign_required_passed": 18,
    "captured_pretty_spec_file_sha256": "17e142c5c907380469c9f4f5cea7c8ab01e830f96a3095e51c5eb8c92a6b69d0",
    "identity_source": "glm-max2-agentic-artifact.json results.evidence.identities and results.evidence.run",
    "native_canonical_spec_sha256": "049d8a604d013d6d670fb1ff2a95529bf4614db796b7f53e47f0f472e3dbfa68",
    "native_observed": {
      "attempted": 18,
      "pass_rate": 1.0,
      "passed": 18
    },
    "native_required_pass_rate": 0.75,
    "request_controls": {
      "reasoning_effort": "max",
      "temperature": 1.0,
      "thinking_mode": "default",
      "top_p": 0.95
    }
  },
  "c4_long_functional_contract": {
    "actual_prompt_tokens": [
      165903,
      165905,
      165907,
      165908
    ],
    "artifact": "glm-max-c4-long-summary.json",
    "claim_boundary": "Offered C4 client overlap only, not engine occupancy. No finish_reason is retained. Ignore emitted performance_eligible and do not use this artifact for quality, speed, or throughput claims.",
    "complete_capture": true,
    "completed": 4,
    "concurrency": 4,
    "failed": 0,
    "foreign_markers": 0,
    "identity_and_health_binding": {
      "configured_context_tokens": 327680,
      "health_claim": "Managed status was running with no OOM kill; private receipt retained outside this repository.",
      "repository": "local-inference-lab/GLM-5.3-Flash-NVFP4-MXFP8-CSF-QAD",
      "revision": "4f90b74165cb1542cca35730b7dfa4d550bcf0ce",
      "runtime_image_digest": "sha256:6008f020c23065bef71390194f968af9f5ebbedf57807b1659f876632e4a1067",
      "served_model": "candidate-glm-max-tier-r1",
      "source": "private after-contract managed status plus retained startup/config evidence"
    },
    "nominal_context_tokens": 201000,
    "own_canary_first": true,
    "plan": "glm-max-c4-functional-plan.json",
    "status": "passed-functional-only"
  },
  "campaign_id": "2026-10-05-xeon-model-qualification",
  "campaign_plan_capture": {
    "current_file_sha256": "bd93a46fde9c472e664022410e11ccb641cf7c4a3b179f907eda26d2876b0973",
    "meaning": "Hash of the campaign-plan file captured during this active campaign; live-progress text is mutable and this is not a pre-first-request attestation."
  },
  "comparison_limit": {
    "allowed_interpretation": "Supports the absolute >=4/5 replacement floor only. It does not support a matched causal quantization, hardware, or coding-improvement claim versus r11.",
    "shared_controls": "The r11 and candidate scouts use the same five-task selection, scout profile, package set, mini-SWE agent, and grader pins.",
    "thinking_controls": "r11 retained control: thinking enabled with no explicit reasoning effort; Qwen: default/medium; GLM: default/max."
  },
  "context256_64_contract": {
    "actual_prompt_tokens": [
      247617,
      247617,
      247638,
      247638
    ],
    "artifact": "glm-max-context256-64-artifact.json",
    "claim_boundary": "Proves four successful requests at the retained actual prompts with 65536 accepted output headroom. It does not prove a full 327680-token request or a generated 65536-token completion.",
    "completed_count": 4,
    "configured_context_tokens": 327680,
    "max_tokens_accepted": 65536,
    "native_artifact": "glm-max-context256-64-native.json",
    "nominal_requested_tokens": 250000,
    "passed_count": 4,
    "status": "passed-with-bounded-claim"
  },
  "context319_8_contract": {
    "actual_prompt_tokens": [
      316766,
      316766,
      316774,
      316774
    ],
    "artifact": "glm-max-context319-8-artifact.json",
    "claim_boundary": "Proves four successful requests at the retained actual prompts with 8192 accepted output headroom. It does not prove a full 327680-token request or a generated 8192-token completion.",
    "completed_count": 4,
    "configured_context_tokens": 327680,
    "max_tokens_accepted": 8192,
    "native_artifact": "glm-max-context319-8-native.json",
    "nominal_requested_tokens": 319000,
    "passed_count": 4,
    "status": "passed-with-bounded-claim"
  },
  "identity": {
    "configured_context_tokens": 327680,
    "repository": "local-inference-lab/GLM-5.3-Flash-NVFP4-MXFP8-CSF-QAD",
    "revision": "4f90b74165cb1542cca35730b7dfa4d550bcf0ce",
    "runtime_engine_revision": "00a33e23142097f99fef95b1ecab6da4ea271e2e",
    "runtime_image_digest": "sha256:6008f020c23065bef71390194f968af9f5ebbedf57807b1659f876632e4a1067"
  },
  "matched_swe_plan": {
    "claim_boundary": "Original scout time is excluded. Every replica requires Oracle 18/18 and SWE >=4/5; pair and aggregate speed gates each require at least 10% candidate advantage.",
    "current_period": {
      "agent_duration_s": 3321.827074,
      "claim_boundary": "A1 fails the 4/5 quality gate. B1, B2, and A2 were not run; no comparative speed claim.",
      "long_tool": {
        "actual_prompt_tokens": 110760,
        "passed": true,
        "requested_context_tokens": 131072
      },
      "official_grader_duration_s": 95.808019,
      "oracle": {
        "attempted": 18,
        "pass_rate": 1.0,
        "passed": 18,
        "required_pass_rate": 0.75
      },
      "period": "candidate-A1",
      "preflight": "25/25 passed",
      "swe_campaign_full_attempt_metric": "2/5",
      "swe_native_summary": {
        "attempted": 5,
        "graded": 4,
        "resolve_rate": 0.5,
        "resolved": 2
      },
      "swe_officially_graded_metric": "2/4 (resolve_rate 0.5)",
      "ungraded_instance_ids": [
        "sympy__sympy-12096"
      ],
      "unresolved_instance_ids": [
        "django__django-13512",
        "pylint-dev__pylint-8898"
      ]
    },
    "early_stop_artifact": "matched-swe-early-stop.json",
    "order": [
      "candidate-A1",
      "incumbent-B1",
      "incumbent-B2",
      "candidate-A2"
    ],
    "plan": "matched-swe-plan.json",
    "primary_measure": "officially resolved SWE tasks per hour of SWE agent-stage duration_s",
    "quality_gate": {
      "oracle_attempted": 18,
      "oracle_passed": 18,
      "swe_attempted": 5,
      "swe_min_resolved": 4
    },
    "source_status_artifact": "matched-a1-source-status.json",
    "speed_requirement": "candidate/incumbent >=1.10 aggregate AND A1/B1 >=1.10 AND A2/B2 >=1.10",
    "status": "partial; aborted after candidate A1 quality gate",
    "warmup_addendum": "matched-swe-warmup-addendum.json"
  },
  "profile": {
    "name": "scout",
    "profile_sha256": "fce0e6e1e7497585bdfc6f76507d3baaff99a6b4f174d841d268de20c68de7f6"
  },
  "schema": "anvil-serving.workload-manifest/v1",
  "status": "closed; matched-SWE partial and aborted after A1 quality gate; exact r11 restoration complete",
  "swe_verified_scout5": {
    "captured_pretty_spec_file_sha256": "7ba66d9ee28ab7499838bda0a5a75995c91fe8582d10403c2c1817220c88451c",
    "dataset": "princeton-nlp/SWE-bench_Verified",
    "grader": {
      "name": "swe-bench",
      "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
    },
    "identity_source": "glm-max2-swe-scout5-spec.json parameters.identities",
    "instance_ids": [
      "sphinx-doc__sphinx-9658",
      "sympy__sympy-12096",
      "django__django-13512",
      "pydata__xarray-4356",
      "pylint-dev__pylint-8898"
    ],
    "native_canonical_spec_sha256": "f9265e8dfd098c10bdee7a953996f3fc44b65fb93862151e727d12ab8074991a",
    "native_plan_sha256": "9640c1a859c948334c7cbcc7499cc182207d0a8dfda3f988d4b5039626c4395b",
    "native_selection_sha256": "622934396fb97d950f803dfbb0791f6a398a5a6fe0fd94cc27b329f1a953cac4",
    "native_summary": {
      "attempted": 5,
      "graded": 5,
      "resolve_rate": 0.8,
      "resolved": 4
    },
    "official_grader_complete": true,
    "required_resolved": 4,
    "resolved_instance_ids": [
      "sphinx-doc__sphinx-9658",
      "sympy__sympy-12096",
      "pydata__xarray-4356",
      "pylint-dev__pylint-8898"
    ],
    "split": "test",
    "stage_duration_s": {
      "agent": 2121.469953,
      "official_grader": 109.970112
    },
    "status": "completed; official grader complete",
    "timing_interpretation": "Agent and grader stages overlapped offline CPU tests; durations are not comparative performance evidence.",
    "unresolved_instance_ids": [
      "django__django-13512"
    ]
  },
  "vision8_contract": {
    "artifact": "glm-max-vision8.json",
    "corpus": "exact eight-image ordered/reversed corpus",
    "max_images_per_request": 8,
    "passed_attempts": 4,
    "requests": 4,
    "status": "passed"
  }
}