{
  "schema": "metatron.intelligence.entity.v1",
  "id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
  "kind": "benchmark",
  "subtype": "software_issue_resolution_benchmark_subset",
  "external_identifiers": [
    {
      "scheme": "benchmark_definition_identifier",
      "value": "SWE-bench_Verified",
      "issuer_id": null,
      "canonical_uri": "https://raw.githubusercontent.com/SWE-bench/SWE-bench/02e7a74ffd0b707aab73d203fe87bdc7c76afc8e/README.md",
      "source": {
        "source_id": "BMRK-5",
        "url": "https://raw.githubusercontent.com/SWE-bench/SWE-bench/02e7a74ffd0b707aab73d203fe87bdc7c76afc8e/README.md"
      },
      "valid_from": null,
      "valid_to": null,
      "observed_at": "2026-09-03",
      "status": "observed"
    }
  ],
  "sources": [
    {
      "source_id": "BMRK-5",
      "url": "https://raw.githubusercontent.com/SWE-bench/SWE-bench/02e7a74ffd0b707aab73d203fe87bdc7c76afc8e/README.md",
      "title": "SWE-bench README at 02e7a74f",
      "publisher": "SWE-bench",
      "source_date": null,
      "effective_date": null,
      "retrieved_at": "2026-09-03",
      "content_sha256": "06841624d7b5130dfe3cc6714642e9a8fd2a608c95c66ae0180e409f12b73aaa",
      "capture_artifact": "rs8000:/opt/metatron/shared/agent-research/metatron-world-innovation-trend-atlas/raw/research/2026-09-03-wave83-benchmark-intelligence/source-captures/BMRK-5-swe-bench-readme.md",
      "capture_method": "curl --location --fail with identified browser user agent; decoded response body preserved byte-for-byte",
      "final_url": "https://raw.githubusercontent.com/SWE-bench/SWE-bench/02e7a74ffd0b707aab73d203fe87bdc7c76afc8e/README.md",
      "http_status": 200,
      "media_type": "text/plain",
      "byte_count": 13302
    }
  ],
  "statements": [
    {
      "statement_id": "STM-394E635EA1C0189144B92452BF726234",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "name",
      "value_or_object_id": "SWE-bench Verified",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README news entry dated 2024-08-13 and pinned repository revision",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-ECEBB64C42E7463DA23DDD34C931AD38",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "benchmark_identifier",
      "value_or_object_id": "SWE-bench_Verified",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README news entry dated 2024-08-13 and pinned repository revision",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-B05FEC8E5D317C473A7109224692CDDB",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "version",
      "value_or_object_id": "Verified subset; source revision 02e7a74f",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README news entry dated 2024-08-13 and pinned repository revision",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-951A5B7E03012ADEB8A7242B32F9F6C6",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "publisher",
      "value_or_object_id": "SWE-bench",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README news entry dated 2024-08-13 and pinned repository revision",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-101A58B8965366051E6CAE392B4D4199",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "task_scope",
      "value_or_object_id": "Subset of 500 software-repository problems confirmed by software engineers as solvable.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README news entry introducing SWE-bench Verified",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-9C92DDB537B4AF9A80D82C0D8E3FD3F8",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "metric_contract",
      "value_or_object_id": "Per-instance resolved/not-resolved outcome produced by the selected SWE-bench evaluation harness; no cross-harness aggregate is represented.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README evaluation commands and generated results report",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-912A77EFCD63DEB57362FE0E27494136",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "evaluation_conditions",
      "value_or_object_id": "Result identity requires exact dataset revision, task repository state, prediction patch, container image/harness version, run ID and grading logs.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README Docker, CLI, evaluation-log and result-cache instructions",
      "valid_from": null,
      "valid_to": null,
      "confidence": "source_reported",
      "status": "observed"
    },
    {
      "statement_id": "STM-BD1FEA937C25A726CB8581957E2AA95B",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "comparability_boundary",
      "value_or_object_id": "Do not compare results across dataset, task-repository, image, harness or grading revisions without an explicit bridge; cached run IDs can reuse earlier results even when predictions change.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README Docker, CLI, evaluation-log and result-cache instructions",
      "valid_from": null,
      "valid_to": null,
      "confidence": "editorial_synthesis",
      "status": "observed"
    },
    {
      "statement_id": "STM-A902113A250652CD1EFAD031E2FA6AED",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "contamination_boundary",
      "value_or_object_id": "The selected source does not establish absence of issue, patch or repository exposure in model training or tool retrieval.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README Docker, CLI, evaluation-log and result-cache instructions",
      "valid_from": null,
      "valid_to": null,
      "confidence": "editorial_synthesis",
      "status": "observed"
    },
    {
      "statement_id": "STM-C19F75DE244485272E4BC996BE1B52D8",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "scope",
      "value_or_object_id": "Reviewed benchmark or evaluation-protocol identity; model results and leaderboard ranks are separate, unrepresented observations.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README news entry dated 2024-08-13 and pinned repository revision",
      "valid_from": null,
      "valid_to": null,
      "confidence": "editorial_synthesis",
      "status": "observed"
    },
    {
      "statement_id": "STM-1EBE797A9006F699A689AE1A4676C904",
      "subject_id": "ENT-b2bceffe-2ce0-475c-b161-529c13553b46",
      "predicate": "boundary",
      "value_or_object_id": "Verified means the subset was reviewed for solvability; it is not verification of a submitted model result, software quality or production engineering ability.",
      "observed_at": "2026-09-03",
      "source_id": "BMRK-5",
      "source_locator": "README Docker, CLI, evaluation-log and result-cache instructions",
      "valid_from": null,
      "valid_to": null,
      "confidence": "editorial_synthesis",
      "status": "observed"
    }
  ],
  "current_projection": {
    "name": {
      "value": "SWE-bench Verified",
      "statement_ids": [
        "STM-394E635EA1C0189144B92452BF726234"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "benchmark_identifier": {
      "value": "SWE-bench_Verified",
      "statement_ids": [
        "STM-ECEBB64C42E7463DA23DDD34C931AD38"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "version": {
      "value": "Verified subset; source revision 02e7a74f",
      "statement_ids": [
        "STM-B05FEC8E5D317C473A7109224692CDDB"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "publisher": {
      "value": "SWE-bench",
      "statement_ids": [
        "STM-951A5B7E03012ADEB8A7242B32F9F6C6"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "task_scope": {
      "value": "Subset of 500 software-repository problems confirmed by software engineers as solvable.",
      "statement_ids": [
        "STM-101A58B8965366051E6CAE392B4D4199"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "metric_contract": {
      "value": "Per-instance resolved/not-resolved outcome produced by the selected SWE-bench evaluation harness; no cross-harness aggregate is represented.",
      "statement_ids": [
        "STM-9C92DDB537B4AF9A80D82C0D8E3FD3F8"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "evaluation_conditions": {
      "value": "Result identity requires exact dataset revision, task repository state, prediction patch, container image/harness version, run ID and grading logs.",
      "statement_ids": [
        "STM-912A77EFCD63DEB57362FE0E27494136"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "comparability_boundary": {
      "value": "Do not compare results across dataset, task-repository, image, harness or grading revisions without an explicit bridge; cached run IDs can reuse earlier results even when predictions change.",
      "statement_ids": [
        "STM-BD1FEA937C25A726CB8581957E2AA95B"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "contamination_boundary": {
      "value": "The selected source does not establish absence of issue, patch or repository exposure in model training or tool retrieval.",
      "statement_ids": [
        "STM-A902113A250652CD1EFAD031E2FA6AED"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "scope": {
      "value": "Reviewed benchmark or evaluation-protocol identity; model results and leaderboard ranks are separate, unrepresented observations.",
      "statement_ids": [
        "STM-C19F75DE244485272E4BC996BE1B52D8"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "boundary": {
      "value": "Verified means the subset was reviewed for solvability; it is not verification of a submitted model result, software quality or production engineering ability.",
      "statement_ids": [
        "STM-1EBE797A9006F699A689AE1A4676C904"
      ],
      "selection_policy": "single_reviewed_statement"
    },
    "current_status": "unknown"
  },
  "claims": {
    "match_rule": "exact_source_url",
    "count": 0,
    "claim_ids": []
  },
  "relations": []
}