Add files using upload-large-folder tool

tasks/0000_526_526258_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):
+- AguaH.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many entries are categorized into the non-NA group, edge-NA group, and interrupted group based on missing value patterns?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as a comma-separated list of the three counts in the order: non-NA, edge-NA, interrupted.
+
+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_526_526258_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0000_526_526258_qa_2"
+description = "How many entries are categorized into the non-NA group, edge-NA group, and interrupted group based on missing value patterns?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0000/526/526258.ipynb_qa_2"
+kaggle_dataset_name = "marcomolina/water-consumption-in-a-median-size-city"
+gold_answer = "141205, 32568, 4824"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "marcomolina__water-consumption-in-a-median-size-city"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "marcomolina/water-consumption-in-a-median-size-city"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "141205, 32568, 4824"
+QUESTION = "How many entries are categorized into the non-NA group, edge-NA group, and interrupted group based on missing value patterns?"
+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/0000_780_780974_qa_4/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):
+- arrests.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What was the maximum number of arrests recorded at the Southwest border and in which year?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as: <value>, <year> (comma-separated, numeric value first, year as a four-digit 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/0000_780_780974_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0000_780_780974_qa_4"
+description = "What was the maximum number of arrests recorded at the Southwest border and in which year?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0000/780/780974.ipynb_qa_4"
+kaggle_dataset_name = "cbp/illegal-immigrants"
+gold_answer = "1643679 in 2000"
+reward_mode_initial = "list"
+package_tier = 0
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "cbp__illegal-immigrants"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "cbp/illegal-immigrants"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "1643679 in 2000"
+QUESTION = "What was the maximum number of arrests recorded at the Southwest border and in which 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_085_1085629_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):
+- (see /home/user/input)
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What are the top three features according to the feature importance ranking provided by the Extra Trees Classifier?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as a comma-separated list of the exact feature 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_085_1085629_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-eval-v1/0001_085_1085629_qa_2"
+description = "What are the top three features according to the feature importance ranking provided by the Extra Trees Classifier?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/085/1085629.ipynb_qa_2"
+kaggle_dataset_name = "uciml/adult-census-income"
+gold_answer = "fnlwgt, age, hours.per.week"
+reward_mode_initial = "list"
+package_tier = 0
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__adult-census-income"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/adult-census-income"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "fnlwgt, age, hours.per.week"
+QUESTION = "What are the top three features according to the feature importance ranking provided by the Extra Trees Classifier?"
+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_133_1133625_qa_4/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):
+- up_res.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+In which phase did BJP+ achieve the highest percentage of total votes, and what was that percentage?
+
+Work it out step by step — inspect the data first (head, shape, dtypes), then compute.
+
+Answer as: <phase>, <percentage> (comma-separated, phase first, percentage with two 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_133_1133625_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_133_1133625_qa_4"
+description = "In which phase did BJP+ achieve the highest percentage of total votes, and what was that percentage?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/133/1133625.ipynb_qa_4"
+kaggle_dataset_name = "ankit2106/uttar-pradesh-assembly-elections-2017"
+gold_answer = "Phase 1, 45.48"
+reward_mode_initial = "list"
+package_tier = 1
+difficulty_level = 3
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "ankit2106__uttar-pradesh-assembly-elections-2017"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "ankit2106/uttar-pradesh-assembly-elections-2017"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Phase 1, 45.48"
+QUESTION = "In which phase did BJP+ achieve the highest percentage of total votes, and what was that percentage?"
+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_160_1160639_qa_1/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):
+- banknifty.csv
+- nifty50.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the average opening price of Nifty 50 across all recorded 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_160_1160639_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_160_1160639_qa_1"
+description = "What is the average opening price of Nifty 50 across all recorded dates in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/160/1160639.ipynb_qa_1"
+kaggle_dataset_name = "ramamet4/nse-stocks-database"
+gold_answer = "7374.52"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "ramamet4__nse-stocks-database"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "ramamet4/nse-stocks-database"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "7374.52"
+QUESTION = "What is the average opening price of Nifty 50 across all recorded dates in 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_188_1188925_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):
+- UCI_Credit_Card.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What percentage of the dataset consists of credit card defaults?
+
+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_188_1188925_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_188_1188925_qa_3"
+description = "What percentage of the dataset consists of credit card defaults?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/188/1188925.ipynb_qa_3"
+kaggle_dataset_name = "uciml/default-of-credit-card-clients-dataset"
+gold_answer = "22"
+reward_mode_initial = "numeric"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__default-of-credit-card-clients-dataset"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/default-of-credit-card-clients-dataset"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "22"
+QUESTION = "What percentage of the dataset consists of credit card defaults?"
+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_189_1189227_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):
+- UCI_Credit_Card.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the mean age of credit card holders who defaulted?
+
+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_189_1189227_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_189_1189227_qa_2"
+description = "What is the mean age of credit card holders who defaulted?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/189/1189227.ipynb_qa_2"
+kaggle_dataset_name = "uciml/default-of-credit-card-clients-dataset"
+gold_answer = "35.73"
+reward_mode_initial = "numeric"
+package_tier = 2
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__default-of-credit-card-clients-dataset"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/default-of-credit-card-clients-dataset"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "35.73"
+QUESTION = "What is the mean age of credit card holders who defaulted?"
+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_189_1189227_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):
+- UCI_Credit_Card.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many samples were allocated to the training set?
+
+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_189_1189227_qa_5/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_189_1189227_qa_5"
+description = "How many samples were allocated to the training set?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/189/1189227.ipynb_qa_5"
+kaggle_dataset_name = "uciml/default-of-credit-card-clients-dataset"
+gold_answer = "24000"
+reward_mode_initial = "numeric"
+package_tier = 2
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__default-of-credit-card-clients-dataset"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/default-of-credit-card-clients-dataset"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "24000"
+QUESTION = "How many samples were allocated to the training set?"
+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_191_1191057_qa_4/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):
+- WA_Fn-UseC_-HR-Employee-Attrition.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which original features were removed from the dataset because they contained only a single unique value across all observations?
+
+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_191_1191057_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_191_1191057_qa_4"
+description = "Which original features were removed from the dataset because they contained only a single unique value across all observations?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/191/1191057.ipynb_qa_4"
+kaggle_dataset_name = "pavansubhasht/ibm-hr-analytics-attrition-dataset"
+gold_answer = "EmployeeCount, Over18, StandardHours"
+reward_mode_initial = "list"
+package_tier = 2
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "pavansubhasht__ibm-hr-analytics-attrition-dataset"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "pavansubhasht/ibm-hr-analytics-attrition-dataset"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "EmployeeCount, Over18, StandardHours"
+QUESTION = "Which original features were removed from the dataset because they contained only a single unique value across all observations?"
+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_196_1196803_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):
+- directory.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the most common ownership type among all Starbucks stores 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_196_1196803_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_196_1196803_qa_3"
+description = "What is the most common ownership type among all Starbucks stores in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/196/1196803.ipynb_qa_3"
+kaggle_dataset_name = "starbucks/store-locations"
+gold_answer = "Company Owned"
+reward_mode_initial = "exact_short"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "starbucks__store-locations"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "starbucks/store-locations"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Company Owned"
+QUESTION = "What is the most common ownership type among all Starbucks stores 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_238_1238370_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):
+- Netflix Shows.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many unique Netflix shows are present in the dataset, considering duplicate titles?
+
+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_238_1238370_qa_1/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_238_1238370_qa_1"
+description = "How many unique Netflix shows are present in the dataset, considering duplicate titles?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/238/1238370.ipynb_qa_1"
+kaggle_dataset_name = "chasewillden/netflix-shows"
+gold_answer = "496"
+reward_mode_initial = "numeric"
+package_tier = 0
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "chasewillden__netflix-shows"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "chasewillden/netflix-shows"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "496"
+QUESTION = "How many unique Netflix shows are present in the dataset, considering duplicate titles?"
+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_239_1239559_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):
+- chopstick-effectiveness.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How much does the average food pinching efficiency decrease from the optimal length (240mm) to the next length tested (270mm)?
+
+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_239_1239559_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_239_1239559_qa_4"
+description = "How much does the average food pinching efficiency decrease from the optimal length (240mm) to the next length tested (270mm)?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/239/1239559.ipynb_qa_4"
+kaggle_dataset_name = "priya2908/chopsticks-1992"
+gold_answer = "1.999"
+reward_mode_initial = "numeric"
+package_tier = 0
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "priya2908__chopsticks-1992"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "priya2908/chopsticks-1992"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "1.999"
+QUESTION = "How much does the average food pinching efficiency decrease from the optimal length (240mm) to the next length tested (270mm)?"
+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_243_1243037_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):
+- diabetes.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which feature exhibits the strongest positive correlation with the Outcome variable (diabetes status)?
+
+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_243_1243037_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_243_1243037_qa_2"
+description = "Which feature exhibits the strongest positive correlation with the Outcome variable (diabetes status)?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/243/1243037.ipynb_qa_2"
+kaggle_dataset_name = "uciml/pima-indians-diabetes-database"
+gold_answer = "Glucose"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__pima-indians-diabetes-database"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/pima-indians-diabetes-database"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Glucose"
+QUESTION = "Which feature exhibits the strongest positive correlation with the Outcome variable (diabetes status)?"
+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_293_1293142_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):
+- 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 correlation coefficient between the sqft_living feature and the log-transformed price variable 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_293_1293142_qa_5/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_293_1293142_qa_5"
+description = "What is the correlation coefficient between the sqft_living feature and the log-transformed price variable in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/293/1293142.ipynb_qa_5"
+kaggle_dataset_name = "harlfoxem/housesalesprediction"
+gold_answer = "0.70"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "harlfoxem__housesalesprediction"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "harlfoxem/housesalesprediction"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "0.70"
+QUESTION = "What is the correlation coefficient between the sqft_living feature and the log-transformed price variable in 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_349_1349978_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):
+- mushrooms.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many categorical features were originally present in the mushroom dataset before numerical encoding?
+
+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_349_1349978_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_349_1349978_qa_4"
+description = "How many categorical features were originally present in the mushroom dataset before numerical encoding?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/349/1349978.ipynb_qa_4"
+kaggle_dataset_name = "uciml/mushroom-classification"
+gold_answer = "23"
+reward_mode_initial = "numeric"
+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 = "23"
+QUESTION = "How many categorical features were originally present in the mushroom dataset before numerical encoding?"
+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_353_1353632_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):
+- dataset.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the difference between the number of "run" samples collected on the left wrist versus "walk" samples on the same wrist?
+
+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_353_1353632_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_353_1353632_qa_3"
+description = "What is the difference between the number of \"run\" samples collected on the left wrist versus \"walk\" samples on the same wrist?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/353/1353632.ipynb_qa_3"
+kaggle_dataset_name = "vmalyi/run-or-walk"
+gold_answer = "5086"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "vmalyi__run-or-walk"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "vmalyi/run-or-walk"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "5086"
+QUESTION = "What is the difference between the number of \"run\" samples collected on the left wrist versus \"walk\" samples on the same wrist?"
+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_1367483_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):
+- battles.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Did any attacker king achieve a 100% win rate in all battles fought according to 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_367_1367483_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_367_1367483_qa_4"
+description = "Did any attacker king achieve a 100% win rate in all battles fought according to the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/367/1367483.ipynb_qa_4"
+kaggle_dataset_name = "mylesoneill/game-of-thrones"
+gold_answer = "Balon/Euron Greyjoy"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "mylesoneill__game-of-thrones"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "mylesoneill/game-of-thrones"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Balon/Euron Greyjoy"
+QUESTION = "Did any attacker king achieve a 100% win rate in all battles fought according to 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_487_1487950_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):
+- database.sqlite
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the total net investment required for betting $10 on every match 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_487_1487950_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_487_1487950_qa_3"
+description = "What is the total net investment required for betting $10 on every match in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/487/1487950.ipynb_qa_3"
+kaggle_dataset_name = "hugomathien/soccer"
+gold_answer = "259790"
+reward_mode_initial = "numeric"
+package_tier = 3
+difficulty_level = 2
+difficulty_tier = "medium"
+
+[environment]
+build_timeout_sec = 600.0
+os = "linux"
+cpus = 1
+memory_mb = 1024
+storage_mb = 5120
+gpus = 0
+allow_internet = true
+mcp_servers = []
+
+# Pre-agent hook: Harbor runs the command AFTER container start and BEFORE the
+# agent setup begins. We use it to pull this task's bucket prefix into
+# /home/user/input/. See environment/pull_bucket.py.
+[environment.healthcheck]
+command = "python3 /opt/pull_bucket.py && [ -n \"$(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 = "hugomathien__soccer"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "hugomathien/soccer"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "259790"
+QUESTION = "What is the total net investment required for betting $10 on every match 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_521_1521206_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):
+- titanic_data.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+How many distinct Pclass values are present 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_521_1521206_qa_2/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_521_1521206_qa_2"
+description = "How many distinct Pclass values are present in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/521/1521206.ipynb_qa_2"
+kaggle_dataset_name = "prkukunoor/TitanicDataset"
+gold_answer = "3"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "prkukunoor__TitanicDataset"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "prkukunoor/TitanicDataset"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "3"
+QUESTION = "How many distinct Pclass values are present 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_527_1527039_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):
+- tarantino.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+Which specific expletive appears most frequently in Tarantino's films according to 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_527_1527039_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_527_1527039_qa_3"
+description = "Which specific expletive appears most frequently in Tarantino's films according to the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/527/1527039.ipynb_qa_3"
+kaggle_dataset_name = "fivethirtyeight/cuss-words-and-deaths-in-quentin-tarantino-films"
+gold_answer = "fucking"
+reward_mode_initial = "exact_short"
+package_tier = 0
+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 = "fucking"
+QUESTION = "Which specific expletive appears most frequently in Tarantino's films according to 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_532_1532154_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):
+- indian_liver_patient.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the average age of all patients 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_532_1532154_qa_5/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_532_1532154_qa_5"
+description = "What is the average age of all patients in the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/532/1532154.ipynb_qa_5"
+kaggle_dataset_name = "uciml/indian-liver-patient-records"
+gold_answer = "44.746141"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__indian-liver-patient-records"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/indian-liver-patient-records"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "44.746141"
+QUESTION = "What is the average age of all patients in the dataset?"
+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_538_1538781_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):
+- 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 professional category has the highest number of female suicides according to 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_538_1538781_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "data-agent-train-v1/0001_538_1538781_qa_4"
+description = "Which professional category has the highest number of female suicides according to the dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/538/1538781.ipynb_qa_4"
+kaggle_dataset_name = "rajanand/suicides-in-india"
+gold_answer = "House Wife"
+reward_mode_initial = "exact_short"
+package_tier = 0
+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 = "House Wife"
+QUESTION = "Which professional category has the highest number of female suicides according to 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_541_1541002_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):
+- vgsales.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+In the stacked bar plot comparing platform sales by genre, which platform has the highest sales in the Sports category?
+
+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_541_1541002_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_541_1541002_qa_4"
+description = "In the stacked bar plot comparing platform sales by genre, which platform has the highest sales in the Sports category?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/541/1541002.ipynb_qa_4"
+kaggle_dataset_name = "gregorut/videogamesales"
+gold_answer = "Wii"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "gregorut__videogamesales"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "gregorut/videogamesales"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "Wii"
+QUESTION = "In the stacked bar plot comparing platform sales by genre, which platform has the highest sales in the Sports category?"
+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_580_1580621_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):
+- Iris.csv
+
+Installed: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn, statsmodels, tabulate, sqlite3, plotly (pip install more if needed).
+
+Question:
+What is the median petal width in the entire 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_580_1580621_qa_4/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify2/0001_580_1580621_qa_4"
+description = "What is the median petal width in the entire dataset?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/580/1580621.ipynb_qa_4"
+kaggle_dataset_name = "uciml/iris"
+gold_answer = "1.3"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "uciml__iris"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "uciml/iris"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "1.3"
+QUESTION = "What is the median petal width in the entire 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_583_1583897_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):
+- xAPI-Edu-Data.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 students in the medium performance category (M) according to the Class distribution derived from the crosstab?
+
+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_583_1583897_qa_3/task.toml

@@ -0,0 +1,64 @@
+schema_version = "1.2"
+artifacts = []
+
+[task]
+name = "train-verify/0001_583_1583897_qa_3"
+description = "What is the total number of students in the medium performance category (M) according to the Class distribution derived from the crosstab?"
+authors = []
+keywords = ["data-agent", "data-analysis", "kaggle"]
+
+[metadata]
+source_dataset = "jupyter-agent/jupyter-agent-dataset"
+source_row_id = "0001/583/1583897.ipynb_qa_3"
+kaggle_dataset_name = "aljarah/xAPI-Edu-Data"
+gold_answer = "211"
+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 \"$(ls /home/user/input)\" ]"
+interval_sec = 2.0
+timeout_sec = 180.0
+start_period_sec = 5.0
+start_interval_sec = 2.0
+retries = 30
+
+[environment.env]
+HF_BUCKET = "AdithyaSK/jupyter-agent-kaggle-all"
+BUCKET_PREFIX = "aljarah__xAPI-Edu-Data"
+HF_TOKEN = "${HF_TOKEN}"
+KAGGLE_DATASET_NAME = "aljarah/xAPI-Edu-Data"
+
+[verifier]
+timeout_sec = 120.0
+
+[verifier.env]
+EXPECTED_ANSWER = "211"
+QUESTION = "What is the total number of students in the medium performance category (M) according to the Class distribution derived from the crosstab?"
+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]