Update Croissant: v2 → v1.1, new RAI annotation/social-impact text

croissant.json

@@ -49,13 +49,13 @@
   },
   "@type": "Dataset",
   "conformsTo": "http://mlcommons.org/croissant/1.0",
-  "name": "core-bench-v2-ood",
-  "description": "CORE-bench v2 out-of-distribution split: 19 held-out scientific-reproducibility tasks. Drawn from disciplines and capsule shapes not represented in the mainline split, intended for measuring generalization of AI agents trained or tuned on the mainline tasks. Disjoint from the mainline split.",
-  "version": "2.0.0",
+  "name": "core-bench-v1.1-ood",
+  "description": "CORE-bench v1.1 out-of-distribution split: 19 held-out scientific-reproducibility tasks. Drawn from disciplines and capsule shapes not represented in the mainline split, intended for measuring generalization of AI agents trained or tuned on the mainline tasks. Disjoint from the mainline split.",
+  "version": "1.1.0",
   "license": "https://creativecommons.org/licenses/by/4.0/",
-  "url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood",
+  "url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood",
   "datePublished": "2026-05-03",
-  "citeAs": "@inproceedings{corebench-v2-2026, title={CORE-bench v2: A Reliability Benchmark for AI Agents on Scientific Reproducibility Tasks}, author={TODO}, booktitle={NeurIPS Datasets and Benchmarks Track}, year={2026}}",
+  "citeAs": "@inproceedings{corebench-v1-1-2026, title={CORE-bench v1.1: A Reliability Benchmark for AI Agents on Scientific Reproducibility Tasks}, author={TODO}, booktitle={NeurIPS Datasets and Benchmarks Track}, year={2026}}",
   "creator": {
     "@type": "Organization",
     "name": "TODO: lab/institution name",
@@ -76,7 +76,7 @@
       "name": "core_test.json",
       "description": "JSON array of CORE-bench tasks. One element per task.",
       "encodingFormat": "application/json",
-      "contentUrl": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/core_test.json",
+      "contentUrl": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/core_test.json",
       "sha256": "ad09c0cc164fd6d73b3891e2c42e8e04eee28aed861cb5a8e57a2a9f74f668c5"
     },
     {
@@ -217,7 +217,7 @@
           "tasks/task_prompt": "Execute 'fibrous-media-analysis.ipynb'. Save the results in latex format in ../results. For all the runs, disable the cell execution timeout and allow errors.",
           "tasks/expected_answer": "[{\"fig From the cumulative pore size distribution (PSD) comparison plots measuring pore diameter vs. cumulative frequency, report the plot label of the model with the lowest cumulative frequency at pore diameter 25 on plot (a).\": \"Lombard model (1989)\"}, {\"fig From the cumulative pore size distribution (PSD) comparison plots measuring pore diameter vs. cumulative frequency, report the plot label of the model with the lowest cumulative frequency at pore diameter 25 on plot (a).\": \"Lombard model (1989)\"}, {\"fig From the cumulative pore size distribution (PSD) comparison plots measuring pore diameter vs. cumulative frequency, report the plot label of the model with the lowest cumulative frequency at pore diameter 25 on plot (a).\": \"Lombard model (1989)\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.6709443.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-9026204.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-9026204.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-8467067",
@@ -227,7 +227,7 @@
           "tasks/task_prompt": "Run the following Python files: 'Example_1.py', 'Example_2.py', 'Example_3.py', 'Example_4.py', 'Example_5.py'.",
           "tasks/expected_answer": "[{\"fig From the example 4 plots, report from the main sequence subplot the letter of the Saccade with the lowest saccade peak velocity at a saccade amplitude of 10 degrees.\": \"E\"}, {\"fig From the example 4 plots, report from the main sequence subplot the letter of the Saccade with the lowest saccade peak velocity at a saccade amplitude of 10 degrees.\": \"E\"}, {\"fig From the example 4 plots, report from the main sequence subplot the letter of the Saccade with the lowest saccade peak velocity at a saccade amplitude of 10 degrees.\": \"E\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.b86bad0a-58ff-4648-9d9f-3bd88ce831f3",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-8467067.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-8467067.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-9419423",
@@ -237,7 +237,7 @@
           "tasks/task_prompt": "Run 'RandomForest.py'.",
           "tasks/expected_answer": "[{\"Report the mean accuracy.\": 0.927, \"Report the mean F1 score.\": 0.923}, {\"Report the mean accuracy.\": 0.925, \"Report the mean F1 score.\": 0.918}, {\"Report the mean accuracy.\": 0.926, \"Report the mean F1 score.\": 0.92}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.0414025.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-9419423.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-9419423.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-6724161",
@@ -247,7 +247,7 @@
           "tasks/task_prompt": "Run 'Main_Cparasites.py'.",
           "tasks/expected_answer": "[{\"fig From the figure depicting calculated transmission coefficient for the 100-nF VISHAY capacitor, report the label of the line with the highest |S21| at frequency 10^0\": \"Single\"}, {\"fig From the figure depicting calculated transmission coefficient for the 100-nF VISHAY capacitor, report the label of the line with the highest |S21| at frequency 10^0\": \"Single\"}, {\"fig From the figure depicting calculated transmission coefficient for the 100-nF VISHAY capacitor, report the label of the line with the highest |S21| at frequency 10^0\": \"Single\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.76709120-1c53-4fe0-bbed-7a5f8e06cf02",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-6724161.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-6724161.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-8610546",
@@ -257,7 +257,7 @@
           "tasks/task_prompt": "Run 'experiments.py'.",
           "tasks/expected_answer": "[{\"Report the cost for the optimal method.\": 3019}, {\"Report the cost for the optimal method.\": 3019}, {\"Report the cost for the optimal method.\": 3019}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.0098640.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-8610546.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-8610546.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-0811394",
@@ -267,7 +267,7 @@
           "tasks/task_prompt": "Run 'test.py'.",
           "tasks/expected_answer": "[{\"fig From figure 2e, report the label of the line with the highest change in n(0) when G is 10^18.\": \"Auger\"}, {\"fig From figure 2e, report the label of the line with the highest change in n(0) when G is 10^18.\": \"Auger\"}, {\"fig From figure 2e, report the label of the line with the highest change in n(0) when G is 10^18.\": \"Auger\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.8344712.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-0811394.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-0811394.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-4320893",
@@ -277,7 +277,7 @@
           "tasks/task_prompt": "Create a symbolic link between /data and data. Using Rscript, run 'code/master.R' from the base directory (NOT from in code).",
           "tasks/expected_answer": "[{\"fig From the heatmap measuring mortality rate relative to American born, report the name of the country with the highest mortality rate relative to American born age 55 to 64 for F.\": \"Ireland\"}, {\"fig From the heatmap measuring mortality rate relative to American born, report the name of the country with the highest mortality rate relative to American born age 55 to 64 for F.\": \"Ireland\"}, {\"fig From the heatmap measuring mortality rate relative to American born, report the name of the country with the highest mortality rate relative to American born age 55 to 64 for F.\": \"Ireland\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.8530479.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-4320893.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-4320893.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-8185407",
@@ -287,7 +287,7 @@
           "tasks/task_prompt": "From the ../data directory, run the following python scripts: '../code/extract.py', '../code/eta_freq_sweep.py', '../code/power_sweep.py', '../code/merge_s1p.py', '../code/calc_power.py'.",
           "tasks/expected_answer": "[{\"Report the load power (Pload) for Transducer F from the measured data results. Ignore units.\": 1345.61367149}, {\"Report the load power (Pload) for Transducer F from the measured data results. Ignore units.\": 1345.61367149}, {\"Report the load power (Pload) for Transducer F from the measured data results. Ignore units.\": 1345.61367149}]",
           "tasks/capsule_doi": "https://doi.org/10.1109/iscas45731.2020.9180423",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-8185407.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-8185407.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-5360076",
@@ -297,7 +297,7 @@
           "tasks/task_prompt": "Execute 'code_ocean2_of_final_feature_engineering_+GBoost+final_figures+big_data_analytics_on_smart_grid_system_.ipynb'. Save the results in html format in ../results. Disable the cell execution timeout and allow errors.",
           "tasks/expected_answer": "[{\"fig Report the name of the machine learning model used with the greatest run time for the applied big data analytics on decentralized smart grid system.\": \"Random forest\"}, {\"fig Report the name of the machine learning model used with the greatest run time for the applied big data analytics on decentralized smart grid system.\": \"Random forest\"}, {\"fig Report the name of the machine learning model used with the greatest run time for the applied big data analytics on decentralized smart grid system.\": \"Random forest\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.9201912.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-5360076.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-5360076.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-2675546",
@@ -307,7 +307,7 @@
           "tasks/task_prompt": "Run 'main.py'.",
           "tasks/expected_answer": "[{\"fig From the UE #74 Handover Executions plot, report the Proposed decision at time 25 m/s.\": 0, \"What is the original (baseline) handover success rate for UE #74? Report the value as a percentage. Do not include the percentage sign.\": 100}, {\"fig From the UE #74 Handover Executions plot, report the Proposed decision at time 25 m/s.\": 0, \"What is the original (baseline) handover success rate for UE #74? Report the value as a percentage. Do not include the percentage sign.\": 100}, {\"fig From the UE #74 Handover Executions plot, report the Proposed decision at time 25 m/s.\": 0, \"What is the original (baseline) handover success rate for UE #74? Report the value as a percentage. Do not include the percentage sign.\": 100}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.d9fecfa5-3035-458e-bd76-4c81dc8cf0fc",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-2675546.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-2675546.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-0571975",
@@ -317,7 +317,7 @@
           "tasks/task_prompt": "Using python3, run 'satcom.py'.",
           "tasks/expected_answer": "[{\"Report the max gain % of planar IRS.\": 3.80583}, {\"Report the max gain % of planar IRS.\": 3.80583}, {\"Report the max gain % of planar IRS.\": 3.80583}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.1140031.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-0571975.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-0571975.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-5007288",
@@ -327,7 +327,7 @@
           "tasks/task_prompt": "Run 'FOM.py'.",
           "tasks/expected_answer": "[{\"fig From the plots measuring counts/s for every channel, report the label of the line with the highest Counts/s at Channel 200.\": \"Spectrum\"}, {\"fig From the plots measuring counts/s for every channel, report the label of the line with the highest Counts/s at Channel 200.\": \"Spectrum\"}, {\"fig From the plots measuring counts/s for every channel, report the label of the line with the highest Counts/s at Channel 200.\": \"Spectrum\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.7695385.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-5007288.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-5007288.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-8353473",
@@ -337,7 +337,7 @@
           "tasks/task_prompt": "Using the ln bash command and the -s flag, create a new file in the current directory with the same name as ../data. From the 'notebooks' folder, create the ../../results/figs directory. Set the environment variable MPLBACKEND to \"Agg\". Execute all the notebooks in the 'notebooks' directory in parallel. Save the results in html format in ../../results. Disable the cell execution timeout and allow errors. Pause executing the script until all of the runs are completed.",
           "tasks/expected_answer": "[{\"fig Report plot label of the index with the higher adjusted closing price in 2008.\": \"HS\"}, {\"fig Report plot label of the index with the higher adjusted closing price in 2008.\": \"HS\"}, {\"fig Report plot label of the index with the higher adjusted closing price in 2008.\": \"HS\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.7419991.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-8353473.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-8353473.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-4407237",
@@ -347,7 +347,7 @@
           "tasks/task_prompt": "Run pip install in the current directory with the -v flag. Then, run the following python files: 'examples/plot_turbo.py', 'examples/plot_RienstraCutOns.py', 'examples/plot_rectangularWaveNumbers.py', 'examples/plot_turbo.py', 'examples/plot_liner.py', and 'examples/plot_higherOrderModes.py'.",
           "tasks/expected_answer": "[{\"fig From the plot measuring damping of a wave along a rectangular duct of length 1.0 m, report the label of the line with the highest dissipation % at frequency 1000 Hz.\": \"Kirchoff\"}, {\"fig From the plot measuring damping of a wave along a rectangular duct of length 1.0 m, report the label of the line with the highest dissipation % at frequency 1000 Hz.\": \"Kirchoff\"}, {\"fig From the plot measuring damping of a wave along a rectangular duct of length 1.0 m, report the label of the line with the highest dissipation % at frequency 1000 Hz.\": \"Kirchoff\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.7639723.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-4407237.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-4407237.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-6562149",
@@ -357,7 +357,7 @@
           "tasks/task_prompt": "Execute 'DataExplorationAnalysisScript.ipynb'. Save the results in html format in ../results. Disable the cell execution timeout and allow errors.",
           "tasks/expected_answer": "[{\"fig Report the p-value of the distance (m) from the plot with the new data histograms.\": 0.061}, {\"fig Report the p-value of the distance (m) from the plot with the new data histograms.\": 0.061}, {\"fig Report the p-value of the distance (m) from the plot with the new data histograms.\": 0.061}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.8303479.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-6562149.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-6562149.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-2151475",
@@ -367,7 +367,7 @@
           "tasks/task_prompt": "Using Rscript, run 'master.R'.",
           "tasks/expected_answer": "[{\"fig Report the name of the university ranked #1 by impact factor.\": \"Chicago\", \"fig Report the name of the journal with the highest 2011 impact factor from the analysis of 30 journals.\": \"JEL\"}, {\"fig Report the name of the university ranked #1 by impact factor.\": \"Chicago\", \"fig Report the name of the journal with the highest 2011 impact factor from the analysis of 30 journals.\": \"JEL\"}, {\"fig Report the name of the university ranked #1 by impact factor.\": \"Chicago\", \"fig Report the name of the journal with the highest 2011 impact factor from the analysis of 30 journals.\": \"JEL\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.0659696.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-2151475.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-2151475.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-5172670",
@@ -377,7 +377,7 @@
           "tasks/task_prompt": "Run '5.em_R.py'.",
           "tasks/expected_answer": "[{\"Report the accuracy of the analysis of the movie.\": 0.74, \"From the distribution of the HBF in the analysis of the movie, report the % positive.\": 63}, {\"Report the accuracy of the analysis of the movie.\": 0.74, \"From the distribution of the HBF in the analysis of the movie, report the % positive.\": 63}, {\"Report the accuracy of the analysis of the movie.\": 0.74, \"From the distribution of the HBF in the analysis of the movie, report the % positive.\": 63}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.7306590.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-5172670.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-5172670.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-7350043",
@@ -387,7 +387,7 @@
           "tasks/task_prompt": "Run 'example.py' and write the output to /results/example.txt. Print commands as they run and stop if any command fails.",
           "tasks/expected_answer": "[{\"Report the final accuracy of the E2E-FS method on the mnist dataset.\": 0.9726999998092651}, {\"Report the final accuracy of the E2E-FS method on the mnist dataset.\": 0.970600009}, {\"Report the final accuracy of the E2E-FS method on the mnist dataset.\": 0.975600004196167}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.4665859.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-7350043.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-7350043.tar.gz"
         },
         {
           "tasks/capsule_id": "capsule-3990498",
@@ -397,7 +397,7 @@
           "tasks/task_prompt": "Run 'test.py'.",
           "tasks/expected_answer": "[{\"fig Report the label of the line with the lowest expectation value at Time 2.\": \"<sigma z>\"}, {\"fig Report the label of the line with the lowest expectation value at Time 2.\": \"<sigma z>\"}, {\"fig Report the label of the line with the lowest expectation value at Time 2.\": \"<sigma z>\"}]",
           "tasks/capsule_doi": "https://doi.org/10.24433/CO.0647308.v1",
-          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v2-ood/resolve/main/capsules/capsule-3990498.tar.gz"
+          "tasks/capsule_url": "https://huggingface.co/datasets/agent-evals/core-bench-v1.1-ood/resolve/main/capsules/capsule-3990498.tar.gz"
         }
       ]
     }
@@ -411,14 +411,14 @@
   "rai:dataCollectionTimeframeEnd": "2026-04-30",
   "rai:dataAnnotationProtocol": "For each capsule, annotators executed the capsule three independent times in its declared environment, recorded the resulting numeric/categorical outputs, and authored questions whose answers were stable across reruns (or, where outputs were stochastic, recorded all three observed values to support fuzzy/numeric tolerance matching during evaluation).",
   "rai:dataAnnotationPlatform": "Code Ocean execution environments + manual review",
-  "rai:dataAnnotationAnalysis": "Each ground-truth answer is the median or modal value across three independent capsule reruns. Tasks where reruns disagreed beyond expected stochastic variation were either rejected or rephrased.",
+  "rai:dataAnnotationAnalysis": "Each numeric ground truth is a 95% prediction interval over three independent capsule reruns. Each non-numeric ground truth is the deterministic answer across three independent capsule reruns. Tasks where reruns disagreed beyond expected stochastic variation were either rejected or rephrased.",
   "rai:dataAnnotationDemographics": "Annotators are graduate-level researchers with a background in computer science and computational sciences. No personal data about annotators is collected or distributed.",
   "rai:dataAnnotationPerItem": "1 annotator per task; cross-checked by a second annotator.",
   "rai:dataPreprocessingProtocol": "Capsules are distributed unmodified except that (a) the 'results/' subdirectory of each capsule is scrubbed at evaluation time to prevent data leakage to the agent, and (b) each capsule is repackaged as a single .tar.gz archive named after its Code Ocean capsule ID.",
   "rai:dataUseCases": "Intended for evaluating AI agents on end-to-end scientific reproducibility tasks: reading capsule code/data, executing it in a sandboxed environment, and answering questions whose answers depend on faithfully reproduced outputs. Suitable for outcome-consistency, calibration, and tool-use studies on agentic systems.",
   "rai:dataLimitations": "Coverage is biased toward capsules that run in CPU-only Linux environments, which under-represents GPU-heavy ML papers. The benchmark measures execution + question answering, not novel scientific reasoning. Ground-truth answers are derived from the original authors' implementations and inherit any errors therein. Some capsules involve domain-specific terminology that may disadvantage agents without scientific pretraining.",
   "rai:dataReleaseMaintenancePlan": "The dataset will be hosted on HuggingFace with versioned releases. Issues, errata, and capsule additions will be tracked in the associated GitHub repository. The maintainers commit to keeping the dataset accessible for at least three years post-publication.",
-  "rai:dataSocialImpact": "The benchmark aims to improve evaluation rigor for AI agents in scientific contexts. Risks include over-fitting agent training to the included capsules and misuse as a sole proxy for general scientific competence. We recommend reporting CORE-bench v2 results alongside other agentic and scientific benchmarks.",
+  "rai:dataSocialImpact": "The benchmark aims to improve evaluation rigor for AI agents in scientific contexts. Risks include over-fitting agent training to the included capsules and misuse as a sole proxy for general scientific competence. We recommend reporting CORE-bench v1.1 OOD results alongside other agentic and scientific benchmarks.",
   "rai:personalSensitiveInformation": "None. All source capsules are publicly licensed Code Ocean artifacts; no human-subjects data, biometric data, health/medical data, or personally identifying information about end users is included. Standard author-attribution metadata (names, institutional affiliations) is preserved in the original capsules. None of the following sensitive categories are represented in the data records: gender, socio-economic status, geography of subjects, language demographics, age, culture, experience or seniority, health or medical data, or political or religious beliefs.",
   "rai:hasSyntheticData": false,
   "prov:wasDerivedFrom": [