{"record": {"aliases": [], "artifacts": [], "categories": [], "description": {}, "documents": [{"document_type": "registry_page", "id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "title": "SuperGLUE"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "benchmark-saturation:19", "modality": null, "name": "SuperGLUE", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"crawl_bundle": "benchmark_saturation_2026-11-04", "crawled_at": "2026-11-04T06:47:00Z", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv"}, "publisher": null, "released": "2019-05-03", "released_reference": {"basis": "benchmark_release", "source_key": "benchmark-saturation:19", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv"}, "schema_version": 1, "sizes": [], "slug": "benchmark-saturation-19", "source": "benchmark_saturation", "source_benchmark_id": "19", "source_metadata": {"benchmark_saturation": {"dataset_issues": "Other data-related issues", "saturation_index": "0.94941868", "saturation_index_interpreted": "Very high", "test_samples": "17,641"}}}, "schema_version": 1, "scores_by_source": {"benchmark_saturation": {"rows": [{"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:19", "measured_by": null, "model_id": null, "model_name": "Hairuo", "obs_id": "benchmark_saturation:19:unidentified-ae2a6800f7619b5545b9", "organization": null, "rank_in_source_response": 1, "raw_value": "91.4", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:19:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 91.4, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:19", "measured_by": null, "model_id": null, "model_name": "Vega v2", "obs_id": "benchmark_saturation:19:unidentified-9298f50972bc6682e0cc", "organization": null, "rank_in_source_response": 2, "raw_value": "91.3", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:19:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 91.3, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:19", "measured_by": null, "model_id": null, "model_name": "ST-MoE-32B", "obs_id": "benchmark_saturation:19:unidentified-76945f507efd0c7af25e", "organization": null, "rank_in_source_response": 3, "raw_value": "91.2", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:19:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 91.2, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:19", "measured_by": null, "model_id": null, "model_name": "Turing NLR v5", "obs_id": "benchmark_saturation:19:unidentified-b9d38ce416968a678cc7", "organization": null, "rank_in_source_response": 4, "raw_value": "90.9", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:19:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 90.9, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:19", "measured_by": null, "model_id": null, "model_name": "ERNIE 3.0", "obs_id": "benchmark_saturation:19:unidentified-dab743c4541a21f185a9", "organization": null, "rank_in_source_response": 5, "raw_value": "90.6", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:19:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 90.6, "value_kind": "number"}], "series": {"bounds": {"basis": "aggregator_declared", "max": null, "min": null}, "declared_max": null, "direction": "higher_is_better", "direction_basis": "source_rank_descending", "display_scale": null, "first_score_record": null, "first_score_report": null, "key": "benchmark-saturation:19", "max_score_contradicted": false, "metric": "F1, Accuracy, Correlation", "observation_count": 5, "observed_max": 91.4, "score_summary": {"display_max": 91.4, "display_multiplier": 1, "model_count": null, "model_count_basis": null, "numeric_count": 5, "raw_max": 91.4, "raw_min": 90.6, "source_reference": {"crawled_at": "2026-11-04T06:47:00Z", "obs_id": "benchmark_saturation:19:unidentified-ae2a6800f7619b5545b9", "reported_date": null, "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv"}, "unit": null}, "series_id": "benchmark_saturation:19:default"}}}, "siblings": []}