Add files using upload-large-folder tool

tasks/0000_369_369503_qa_1/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- database.sqlite
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What percentage of all matches have a goal difference of zero (i.e., draws)?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Express the value as a percentage (e.g. 95.5), not a fraction.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0000_369_369503_qa_1/task.toml

@@ -0,0 +1,58 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-eval-v1/0000_369_369503_qa_1"
+description = "What percentage of all matches have a goal difference of zero (i.e., draws)?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0000/369/369503.ipynb_qa_1"
+kaggle_dataset_name = "hugomathien/soccer"
+gold_answer = "25.4%"
+reward_mode_initial = "flexible"
+package_tier = 3
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 2
+memory_mb = 4096
+storage_mb = 10240
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "hugomathien__soccer"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "hugomathien/soccer"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "25.4%"
+QUESTION = "What percentage of all matches have a goal difference of zero (i.e., draws)?"
+REWARD_MODE = "flexible"
+
+[agent]
+timeout_sec = 900.0
+
+[solution.env]

tasks/0000_465_465850_qa_5/instruction.md

@@ -0,0 +1,16 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- Iris.csv
+- database.sqlite
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which species exhibits the highest average sepal length according to the aggregated dataset statistics?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0000_465_465850_qa_5/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0000_465_465850_qa_5"
+description = "Which species exhibits the highest average sepal length according to the aggregated dataset statistics?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0000/465/465850.ipynb_qa_5"
+kaggle_dataset_name = "uciml/iris"
+gold_answer = "virginica"
+reward_mode_initial = "exact_short"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__iris"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/iris"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "virginica"
+QUESTION = "Which species exhibits the highest average sepal length according to the aggregated dataset statistics?"
+REWARD_MODE = "flexible"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0000_804_804467_qa_3/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- mushrooms.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the lowest RMSE value observed in the train_test_split results, and which models achieved it?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as a comma-separated list: <value>, <model1>, <model2>, <model3> (value first, then exact model names).
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0000_804_804467_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0000_804_804467_qa_3"
+description = "What is the lowest RMSE value observed in the train_test_split results, and which models achieved it?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0000/804/804467.ipynb_qa_3"
+kaggle_dataset_name = "uciml/mushroom-classification"
+gold_answer = "0, DecisionTree, RandomForest, SVM"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 4
+difficulty_tier = "hard"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__mushroom-classification"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/mushroom-classification"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "0, DecisionTree, RandomForest, SVM"
+QUESTION = "What is the lowest RMSE value observed in the train_test_split results, and which models achieved it?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_074_1074738_qa_1/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- database.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which U.S. state has the highest number of recorded "Murder or Manslaughter" cases, and what is the exact count of such incidents in that state?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as: <state>, <count> (comma-separated, state first, plain number).
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_074_1074738_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_074_1074738_qa_1"
+description = "Which U.S. state has the highest number of recorded \"Murder or Manslaughter\" cases, and what is the exact count of such incidents in that state?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/074/1074738.ipynb_qa_1"
+kaggle_dataset_name = "murderaccountability/homicide-reports"
+gold_answer = "California, 98994"
+reward_mode_initial = "list"
+package_tier = 3
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "murderaccountability__homicide-reports"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "murderaccountability/homicide-reports"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "California, 98994"
+QUESTION = "Which U.S. state has the highest number of recorded \"Murder or Manslaughter\" cases, and what is the exact count of such incidents in that state?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_137_1137361_qa_2/instruction.md

@@ -0,0 +1,16 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- dataset_TSMC2014_NYC.csv
+- dataset_TSMC2014_TKY.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the total number of check-ins recorded in the New York City dataset?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_137_1137361_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_137_1137361_qa_2"
+description = "What is the total number of check-ins recorded in the New York City dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/137/1137361.ipynb_qa_2"
+kaggle_dataset_name = "chetanism/foursquare-nyc-and-tokyo-checkin-dataset"
+gold_answer = "227428"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "chetanism__foursquare-nyc-and-tokyo-checkin-dataset"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "chetanism/foursquare-nyc-and-tokyo-checkin-dataset"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "227428"
+QUESTION = "What is the total number of check-ins recorded in the New York City dataset?"
+REWARD_MODE = "numeric"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_202_1202888_qa_1/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- Pokemon.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which generation has the highest probability of producing a legendary Pokémon in the dataset?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_202_1202888_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_202_1202888_qa_1"
+description = "Which generation has the highest probability of producing a legendary Pokémon in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/202/1202888.ipynb_qa_1"
+kaggle_dataset_name = "abcsds/pokemon"
+gold_answer = "Generation 3"
+reward_mode_initial = "exact_short"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "abcsds__pokemon"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "abcsds/pokemon"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Generation 3"
+QUESTION = "Which generation has the highest probability of producing a legendary Pokémon in the dataset?"
+REWARD_MODE = "exact_short"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_231_1231918_qa_1/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- menu.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What percentage of McDonald's menu items contain zero sugar based on the dataset?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_231_1231918_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_231_1231918_qa_1"
+description = "What percentage of McDonald's menu items contain zero sugar based on the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/231/1231918.ipynb_qa_1"
+kaggle_dataset_name = "mcdonalds/nutrition-facts"
+gold_answer = "9.61"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "mcdonalds__nutrition-facts"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "mcdonalds/nutrition-facts"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "9.61"
+QUESTION = "What percentage of McDonald's menu items contain zero sugar based on the dataset?"
+REWARD_MODE = "numeric"
+ATOL = "0.05"
+RTOL = "0.01"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_233_1233959_qa_2/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- mushrooms.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the most common cap shape in the dataset based on the feature frequency analysis?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_233_1233959_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_233_1233959_qa_2"
+description = "What is the most common cap shape in the dataset based on the feature frequency analysis?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/233/1233959.ipynb_qa_2"
+kaggle_dataset_name = "uciml/mushroom-classification"
+gold_answer = "convex"
+reward_mode_initial = "exact_short"
+package_tier = 2
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__mushroom-classification"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/mushroom-classification"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "convex"
+QUESTION = "What is the most common cap shape in the dataset based on the feature frequency analysis?"
+REWARD_MODE = "exact_short"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_233_1233959_qa_5/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- mushrooms.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the most common cap color in the dataset based on the feature frequency analysis?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_233_1233959_qa_5/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_233_1233959_qa_5"
+description = "What is the most common cap color in the dataset based on the feature frequency analysis?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/233/1233959.ipynb_qa_5"
+kaggle_dataset_name = "uciml/mushroom-classification"
+gold_answer = "brown"
+reward_mode_initial = "exact_short"
+package_tier = 2
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__mushroom-classification"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/mushroom-classification"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "brown"
+QUESTION = "What is the most common cap color in the dataset based on the feature frequency analysis?"
+REWARD_MODE = "exact_short"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_257_1257061_qa_1/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- kc_house_data.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the skewness of the original SalePrice distribution before any transformation?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_257_1257061_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_257_1257061_qa_1"
+description = "What is the skewness of the original SalePrice distribution before any transformation?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/257/1257061.ipynb_qa_1"
+kaggle_dataset_name = "harlfoxem/housesalesprediction"
+gold_answer = "4.024069"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "harlfoxem__housesalesprediction"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "harlfoxem/housesalesprediction"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "4.024069"
+QUESTION = "What is the skewness of the original SalePrice distribution before any transformation?"
+REWARD_MODE = "numeric"
+ATOL = "0.001"
+RTOL = "0.005"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_277_1277058_qa_2/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- us_companies.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which year had the highest number of companies founded, and how many companies were founded that year?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as a comma-separated pair: <year>, <count>, with the year first and the count as a plain number.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_277_1277058_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_277_1277058_qa_2"
+description = "Which year had the highest number of companies founded, and how many companies were founded that year?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/277/1277058.ipynb_qa_2"
+kaggle_dataset_name = "govlab/open-data-500-companies"
+gold_answer = "2011, 51"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "govlab__open-data-500-companies"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "govlab/open-data-500-companies"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "2011, 51"
+QUESTION = "Which year had the highest number of companies founded, and how many companies were founded that year?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_323_1323152_qa_1/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- IMDB-Movie-Data.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which movie generated the highest revenue in the dataset, and what was the exact revenue amount?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as: <movie title>, <revenue amount> (comma-separated, title first, keep decimals).
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_323_1323152_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_323_1323152_qa_1"
+description = "Which movie generated the highest revenue in the dataset, and what was the exact revenue amount?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/323/1323152.ipynb_qa_1"
+kaggle_dataset_name = "PromptCloudHQ/imdb-data"
+gold_answer = "Star Wars: Episode VII - The Force Awakens, 936.63"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "PromptCloudHQ__imdb-data"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "PromptCloudHQ/imdb-data"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Star Wars: Episode VII - The Force Awakens, 936.63"
+QUESTION = "Which movie generated the highest revenue in the dataset, and what was the exact revenue amount?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_354_1354131_qa_1/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- Iris.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Is the distribution of species in the Iris dataset balanced across all classes?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_354_1354131_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_354_1354131_qa_1"
+description = "Is the distribution of species in the Iris dataset balanced across all classes?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/354/1354131.ipynb_qa_1"
+kaggle_dataset_name = "uciml/iris"
+gold_answer = "yes"
+reward_mode_initial = "exact_bool"
+package_tier = 1
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__iris"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/iris"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "yes"
+QUESTION = "Is the distribution of species in the Iris dataset balanced across all classes?"
+REWARD_MODE = "exact_bool"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_364_1364936_qa_3/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- (see /home/user/input)
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which movie has the lowest total count of entries in the dataset?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_364_1364936_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-eval-v1/0001_364_1364936_qa_3"
+description = "Which movie has the lowest total count of entries in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/364/1364936.ipynb_qa_3"
+kaggle_dataset_name = "fivethirtyeight/cuss-words-and-deaths-in-quentin-tarantino-films"
+gold_answer = "Kill Bill: Vol. 2"
+reward_mode_initial = "exact_short"
+package_tier = 0
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "fivethirtyeight__cuss-words-and-deaths-in-quentin-tarantino-films"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "fivethirtyeight/cuss-words-and-deaths-in-quentin-tarantino-films"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Kill Bill: Vol. 2"
+QUESTION = "Which movie has the lowest total count of entries in the dataset?"
+REWARD_MODE = "exact_short"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_364_1364936_qa_4/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- tarantino.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many movies in the dataset were released after the year 2004?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_364_1364936_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-eval-v1/0001_364_1364936_qa_4"
+description = "How many movies in the dataset were released after the year 2004?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/364/1364936.ipynb_qa_4"
+kaggle_dataset_name = "fivethirtyeight/cuss-words-and-deaths-in-quentin-tarantino-films"
+gold_answer = "2"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "fivethirtyeight__cuss-words-and-deaths-in-quentin-tarantino-films"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "fivethirtyeight/cuss-words-and-deaths-in-quentin-tarantino-films"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "2"
+QUESTION = "How many movies in the dataset were released after the year 2004?"
+REWARD_MODE = "numeric"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_367_1367107_qa_1/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- battles.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which pair of variables in the dataset has the strongest positive correlation according to the correlation matrix?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as a comma-separated list of the exact column names.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_367_1367107_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_367_1367107_qa_1"
+description = "Which pair of variables in the dataset has the strongest positive correlation according to the correlation matrix?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/367/1367107.ipynb_qa_1"
+kaggle_dataset_name = "mylesoneill/game-of-thrones"
+gold_answer = "battle_number, year"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "mylesoneill__game-of-thrones"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "mylesoneill/game-of-thrones"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "battle_number, year"
+QUESTION = "Which pair of variables in the dataset has the strongest positive correlation according to the correlation matrix?"
+REWARD_MODE = "list_csv"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_374_1374329_qa_2/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- Suicides in India 2001-2012.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the most common cause of suicide in the dataset according to the analysis?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_374_1374329_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_374_1374329_qa_2"
+description = "What is the most common cause of suicide in the dataset according to the analysis?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/374/1374329.ipynb_qa_2"
+kaggle_dataset_name = "rajanand/suicides-in-india"
+gold_answer = "Family problems"
+reward_mode_initial = "exact_short"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "rajanand__suicides-in-india"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "rajanand/suicides-in-india"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Family problems"
+QUESTION = "What is the most common cause of suicide in the dataset according to the analysis?"
+REWARD_MODE = "exact_short"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_374_1374329_qa_3/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- Suicides in India 2001-2012.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which demographic group (based on social status) has the highest total number of suicides in the dataset?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_374_1374329_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_374_1374329_qa_3"
+description = "Which demographic group (based on social status) has the highest total number of suicides in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/374/1374329.ipynb_qa_3"
+kaggle_dataset_name = "rajanand/suicides-in-india"
+gold_answer = "Married"
+reward_mode_initial = "exact_short"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "rajanand__suicides-in-india"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "rajanand/suicides-in-india"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Married"
+QUESTION = "Which demographic group (based on social status) has the highest total number of suicides in the dataset?"
+REWARD_MODE = "exact_short"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_380_1380018_qa_1/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- rainfall in india 1901-2015.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which Indian subdivision has the highest average annual rainfall, and what is the value in millimeters?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as: <subdivision>, <value> (comma-separated, label first, keep decimals).
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_380_1380018_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_380_1380018_qa_1"
+description = "Which Indian subdivision has the highest average annual rainfall, and what is the value in millimeters?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/380/1380018.ipynb_qa_1"
+kaggle_dataset_name = "rajanand/rainfall-in-india"
+gold_answer = "Arunachal Pradesh, 3418.86"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "rajanand__rainfall-in-india"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "rajanand/rainfall-in-india"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Arunachal Pradesh, 3418.86"
+QUESTION = "Which Indian subdivision has the highest average annual rainfall, and what is the value in millimeters?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_413_1413239_qa_1/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- database.sqlite
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many teams in the dataset have missing FIFA API IDs?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_413_1413239_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_413_1413239_qa_1"
+description = "How many teams in the dataset have missing FIFA API IDs?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/413/1413239.ipynb_qa_1"
+kaggle_dataset_name = "hugomathien/soccer"
+gold_answer = "11"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "hugomathien__soccer"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "hugomathien/soccer"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "11"
+QUESTION = "How many teams in the dataset have missing FIFA API IDs?"
+REWARD_MODE = "numeric"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_425_1425114_qa_5/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- data_set_ALL_AML_train.csv
+- actual.csv
+- data_set_ALL_AML_independent.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the number of principal components used in the PCA analysis for dimensionality reduction?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_425_1425114_qa_5/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_425_1425114_qa_5"
+description = "What is the number of principal components used in the PCA analysis for dimensionality reduction?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/425/1425114.ipynb_qa_5"
+kaggle_dataset_name = "crawford/gene-expression"
+gold_answer = "2"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "crawford__gene-expression"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "crawford/gene-expression"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "2"
+QUESTION = "What is the number of principal components used in the PCA analysis for dimensionality reduction?"
+REWARD_MODE = "numeric"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_426_1426219_qa_4/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- Uniqlo(FastRetailing) 2012-2016 Training - stocks2012-2016.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the total number of trading days recorded in the year 2012?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_426_1426219_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_426_1426219_qa_4"
+description = "What is the total number of trading days recorded in the year 2012?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/426/1426219.ipynb_qa_4"
+kaggle_dataset_name = "daiearth22/uniqlo-fastretailing-stock-price-prediction"
+gold_answer = "248"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "daiearth22__uniqlo-fastretailing-stock-price-prediction"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "daiearth22/uniqlo-fastretailing-stock-price-prediction"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "248"
+QUESTION = "What is the total number of trading days recorded in the year 2012?"
+REWARD_MODE = "numeric"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_448_1448587_qa_3/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- FDI_in_India.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What were the FDI inflows for the METALLURGICAL INDUSTRIES sector in the four years 2013, 2014, 2015, and 2016?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as a comma-separated list of the four values in the order 2013, 2014, 2015, 2016, with numbers as given (e.g., 567.63).
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_448_1448587_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_448_1448587_qa_3"
+description = "What were the FDI inflows for the METALLURGICAL INDUSTRIES sector in the four years 2013, 2014, 2015, and 2016?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/448/1448587.ipynb_qa_3"
+kaggle_dataset_name = "rajanand/fdi-in-india"
+gold_answer = "567.63, 359.34, 456.31, 1440.18"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "rajanand__fdi-in-india"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "rajanand/fdi-in-india"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "567.63, 359.34, 456.31, 1440.18"
+QUESTION = "What were the FDI inflows for the METALLURGICAL INDUSTRIES sector in the four years 2013, 2014, 2015, and 2016?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_520_1520172_qa_4/instruction.md

@@ -0,0 +1,15 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- KaggleV2-May-2016.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the total number of unique appointment dates in the dataset?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_520_1520172_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-eval-v1/0001_520_1520172_qa_4"
+description = "What is the total number of unique appointment dates in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/520/1520172.ipynb_qa_4"
+kaggle_dataset_name = "joniarroba/noshowappointments"
+gold_answer = "27"
+reward_mode_initial = "numeric"
+package_tier = 1
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "joniarroba__noshowappointments"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "joniarroba/noshowappointments"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "27"
+QUESTION = "What is the total number of unique appointment dates in the dataset?"
+REWARD_MODE = "numeric"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]

tasks/0001_598_1598981_qa_3/instruction.md

@@ -0,0 +1,17 @@
+You are a data-analysis agent working in a sandbox. Use your code-execution tool to inspect the files and compute the answer.
+
+Files (in /home/user/input, no subfolders):
+- student-mat.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which student address type (urban 'U' or rural 'R') has the highest count in the dataset, and what is the specific number of students for that address type?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as: <address_type>, <count> (comma-separated, label first, plain number).
+
+Answer with a single clean value: a bare number (no commas or units, e.g. 95293), a short label, yes/no, or a comma-separated list. Keep decimal precision. If there's no applicable answer, write: Not Applicable
+
+Write only that value to /workdir/answer.txt (e.g. `echo -n "<value>" > /workdir/answer.txt`), then stop.

tasks/0001_598_1598981_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_598_1598981_qa_3"
+description = "Which student address type (urban 'U' or rural 'R') has the highest count in the dataset, and what is the specific number of students for that address type?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/598/1598981.ipynb_qa_3"
+kaggle_dataset_name = "uciml/student-alcohol-consumption"
+gold_answer = "U, 307"
+reward_mode_initial = "list"
+package_tier = 0
+difficulty_level = 1
+difficulty_tier = "easy"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__student-alcohol-consumption"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/student-alcohol-consumption"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "U, 307"
+QUESTION = "Which student address type (urban 'U' or rural 'R') has the highest count in the dataset, and what is the specific number of students for that address type?"
+REWARD_MODE = "list"
+ATOL = "0.0"
+RTOL = "0.0"
+
+[agent]
+# Capped at 600s (10 min) to kill the long-tail stuck-agent cases without
+# cutting off legitimate complex trials. Median Phase B trial is 60-120s;
+# legitimate L4/L5 tasks can hit 200-300s; anything past 600s is almost
+# certainly a stuck agent loop.
+timeout_sec = 600.0
+
+[solution.env]