| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000088 | Aggregation & Comparison | Extremum Reasoning | For item Ours in group RGB, what is the maximum value across all metrics? Exclude missing entries. Output only the number as a string. | "0.959" |
| 2 | HB_real_000096 | Aggregation & Comparison | Extremum Reasoning | For item Ours in group RGB, what is the minimum value across all metrics? Exclude missing entries. Output only the number as a string. | "0.648" |
| 3 | HB_real_000426 | Consistency & Missingness | Column Missingness | In column Mean, list the row labels whose entry is missing. If none, output []. Output only a JSON array of strings. | [] |
| 4 | HB_real_000825 | Markup Grounding | Marked Value Listing | List all bold numbers in row Ours within group RGB. Output only a JSON array of strings. | ["0.896", "0.919", "0.648"] |
| 5 | HB_real_000833 | Markup Grounding | Marked Value Listing | List all underlined numbers in row Ours within group RGB. Output only a JSON array of strings. | ["0.767", "0.850"] |
| 6 | HB_real_001305 | Aggregation & Comparison | Best Selection | Within group RGB + 3D, which row label in the leftmost row-label column is best on column Mean? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours", "value": "0.945"} |
| 7 | HB_real_001306 | Aggregation & Comparison | Rank Selection | List the top-3 items by column Mean, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. Example: [{"item":"...","value":"..."}]. Exclude missing entries. | [{"item": "Ours (RGB + 3D)", "value": "0.945"}, {"item": "AST [28] (RGB + 3D)", "value": "0.937"}, {"item": "AST [28] (RGB)", "value": "0.880"}] |
| 8 | HB_real_001307 | Aggregation & Comparison | Rank Selection | Within group RGB, who ranks 2nd in column Mean? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours", "value": "0.850"} |
| 9 | HB_real_001308 | Consistency & Missingness | Cell Missingness | Is the cell at row Ours in group RGB, column Mean missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000076 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column Mean? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "PatchCore + FPFH [17] (RGB + 3D)", "value": "0.959"} |
| 2 | HB_real_000081 | Aggregation & Comparison | Rank Selection | Who ranks 1st in column Mean? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "Ours (RGB + 3D)", "value": "0.964"} |
| 3 | HB_real_000414 | Aggregation & Comparison | Rank Selection | Who ranks 3rd in column Mean? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "Ours (RGB)", "value": "0.942"} |
| 4 | HB_real_000794 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Mean. Output only a JSON array of strings. | ["0.906", "0.930", "0.959"] |
| 5 | HB_real_000798 | Markup Grounding | Marker Counting | How many numbers are underlined in column Mean? Output only the count as a string. | "3" |
| 6 | HB_real_001319 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Mean? Exclude missing entries. Output only the number as a string. | "0.143" |
| 7 | HB_real_001320 | Markup Grounding | Marked Value Listing | List all bold numbers in column Mean. Output only a JSON array of strings. | ["0.924", "0.942", "0.964"] |
| 8 | HB_real_001321 | Aggregation & Comparison | Pairwise Comparison | For Ours in group 3D, compare Bagel and Mean. Which metric value is larger? Return JSON {"item":"<metric name>","value":"<string>"}. | {"item": "Bagel", "value": "0.943"} |
| 9 | HB_real_001322 | Aggregation & Comparison | Best Selection | Within RGB, which row label in the leftmost row-label column is best on Mean? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours", "value": "0.942"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000142 | Constrained Retrieval | Cell Retrieval | From this table, find the value at (r=4, c=2) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and left text columns. Output only the number as a string. | "0.956" |
| 2 | HB_real_000145 | Aggregation & Comparison | Extremum Reasoning | For row w/ UFF, what is the minimum value across I-AUROC, AUPRO, and P-AUROC? Output only the number as a string. | "0.898" |
| 3 | HB_real_000150 | Constrained Retrieval | Value Filtering | What is the difference in I-AUROC between row w/ UFF and row w/o UFF? Output only the number as a string rounded to three decimal places. | "0.041" |
| 4 | HB_real_000168 | Markup Grounding | Marker Counting | From this table, how many bold numbers are in the table? Output only the count as a string. | "3" |
| 5 | HB_real_000304 | Markup Grounding | Marked Value Listing | Based on the table image, list all bold numbers. Output only a JSON array of strings. | ["0.945", "0.964", "0.992"] |
| 6 | HB_real_000313 | Markup Grounding | Underline Presence | Is value 0.956 underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 7 | HB_real_000335 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 8 | HB_real_000470 | Local Relations | Neighbor Indexing | From this table image, using index (r=4,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.956", "up": "0.944", "down": "0.953", "left": "0.898", "right": "0.990"} |
| 9 | HB_real_001337 | Markup Grounding | Marked Column Listing | List all bold numbers in column P-AUROC. Output only a JSON array of strings. | ["0.992"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001300 | Aggregation & Comparison | Argmax Reasoning | In column mAP(val), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "M3Net (transformer)", "value": "69.0"} |
| 2 | HB_real_001301 | Aggregation & Comparison | Argmax Reasoning | In column NDS(val), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "M3Net (transformer)", "value": "72.4"} |
| 3 | HB_real_001302 | Aggregation & Comparison | Argmax Reasoning | In column Drivable, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "M3Net (mamba)", "value": "90.6"} |
| 4 | HB_real_000929 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 19, "col": "Drivable", "value": "90.6"} |
| 5 | HB_real_001639 | Markup Grounding | Marked Value Listing | List all bold numeric values in row M3Net (transformer). Output only a JSON array of strings. | ["69.0", "72.4", "69.6", "63.4"] |
| 6 | HB_real_000943 | Markup Grounding | Marker Counting | How many bold numeric cells are in row M3Net (mamba)? Output only the count as a string. | "5" |
| 7 | HB_real_000804 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. On the remaining cells in column Divider, what is the maximum value? Output only the number as a string. | "61.1" |
| 8 | HB_real_000300 | Markup Grounding | Bold Presence | Is the value at position (r=19,c=3) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001303 | Aggregation & Comparison | Argmax Reasoning | In column mIoU, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "M3Net (mamba)", "value": "24.1"} |
| 2 | HB_real_001304 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column drive. suf.? Output only the number as a string. | "33.2" |
| 3 | HB_real_000288 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric metric columns, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 10, "col": "pedestrian", "value": "34.6"} |
| 4 | HB_real_000124 | Markup Grounding | Bold Presence | Is the value at position (r=10,c=1) in the numeric metric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 5 | HB_real_000683 | Markup Grounding | Marked Value Listing | List all bold numeric metric values in row M3Net (mamba). Output only a JSON array of strings. | ["24.1", "28.7", "20.1", "27.3", "28.6", "34.6", "20.5", "18.0", "23.0", "21.7", "22.5", "22.0", "24.7", "27.2"] |
| 6 | HB_real_000622 | Markup Grounding | Marker Counting | How many bold numeric metric cells are in row M3Net (transformer)? Output only the count as a string. | "2" |
| 7 | HB_real_000442 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. On the remaining cells in column bus, what is the maximum value? Output only the number as a string. | "23.1" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000085 | Aggregation & Comparison | Extremum Reasoning | For item TDFusion (Ours) in block LLVIP dataset, what is the maximum value across metrics? Output only the number as a string. | "16.38" |
| 2 | HB_real_000094 | Aggregation & Comparison | Extremum Reasoning | For item TDFusion (Ours) in block LLVIP dataset, what is the minimum value across metrics? Output only the number as a string. | "0.46" |
| 3 | HB_real_000473 | Markup Grounding | Marked Value Listing | List all light-red highlighted values in row TDFusion (Ours) within block LLVIP dataset. Output only a JSON array of strings. | ["7.36", "16.38", "1.75", "0.46", "0.70", "0.67"] |
| 4 | HB_real_000480 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in row TDFusion (Ours) within block LLVIP dataset? Output only the count as a string. | "6" |
| 5 | HB_real_000829 | Aggregation & Comparison | Best Selection | Within LLVIP dataset, which row label is best on SF↑? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "TDFusion (Ours)", "value": "16.38"} |
| 6 | HB_real_001309 | Aggregation & Comparison | Best Selection | Within LLVIP dataset, which row label is best on EN↑? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "TDFusion (Ours)", "value": "7.36"} |
| 7 | HB_real_001310 | Markup Grounding | Highlight Presence | Is the number with value 15.00 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 8 | HB_real_001311 | Aggregation & Comparison | Best Selection | Within LLVIP dataset, which row label is best on SSIM↑? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "TDFusion (Ours)", "value": "0.67"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000433 | Consistency & Missingness | Column Missingness | Based on the table image, list all missing entries in column mAcc (block MSRS under Semantic Segmentation) (N/A or —). Output only a JSON array of strings. | [] |
| 2 | HB_real_001312 | Aggregation & Comparison | Argmax Reasoning | In column mIoU (block FMB under Semantic Segmentation), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "TDFusion", "value": "60.50"} |
| 3 | HB_real_001313 | Markup Grounding | Highlight Presence | Is the number with value 96.03 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001314 | Constrained Retrieval | Value Filtering | List all highlighted numeric values greater than 80 in the table. Output only a JSON array of strings, preserving reading order. | ["96.03", "85.99", "83.71", "86.04", "86.27", "95.00"] |
| 5 | HB_real_001031 | Markup Grounding | Marked Value Listing | List all values highlighted in pink. Output only a JSON array of strings. | ["96.03", "72.07", "86.04", "75.09", "67.17", "60.50", "86.27", "59.71"] |
| 6 | HB_real_000714 | Constrained Retrieval | Color-Anchor Retrieval | For the value 65.97, list all values highlighted with the same light-blue color as the anchor. Output only a JSON array of strings. | ["65.97", "58.41", "58.23", "85.99", "83.71", "74.50", "95.00", "69.18"] |
| 7 | HB_real_001003 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in the table? Output only the count as a string. | "16" |
| 8 | HB_real_000715 | Aggregation & Comparison | Grouped Aggregation | Exclude highlighted cells. Among the remaining values in column mAP50, what is the maximum value? Output only the number as a string. | "83.61" |
| 9 | HB_real_001742 | Markup Grounding | Bold Presence | At position (r=10,c=1) in the numeric part of the table, is the value boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left Methods column. | {"bold": false} |
| 10 | HB_real_001929 | Markup Grounding | Underline Presence | The value 96.03 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 11 | HB_real_001931 | Constrained Retrieval | Color-Anchor Retrieval | Within the Object Detection section, list values highlighted with the same pink color as the anchor 86.27. Output only a JSON array of strings. | ["96.03", "72.07", "86.27", "59.71"] |
| 12 | HB_real_001937 | Constrained Retrieval | Color-Anchor Retrieval | For the value 58.23, list values highlighted with the same light-blue color within the M3FD block. Output only a JSON array of strings. | ["58.23", "83.71"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000212 | Aggregation & Comparison | Extremum Reasoning | For item IV (w/o Fusion learning), what is the minimum value across all metrics? Output only the number as a string. | "0.41" |
| 2 | HB_real_000489 | Aggregation & Comparison | Extremum Reasoning | For item IV (w/o Fusion learning), what is the maximum value across all metrics? Output only the number as a string. | "14.07" |
| 3 | HB_real_000491 | Markup Grounding | Marked Value Listing | List all light-red highlighted numbers in row Ours. Output only a JSON array of strings. | ["6.86", "14.16", "1.76", "0.43", "0.68", "0.75"] |
| 4 | HB_real_000791 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the boldfaced cell with value 6.86. The requested neighbor is numeric. Output only the number as a string. | "14.16" |
| 5 | HB_real_000843 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in row Ours? Output only the count as a string. | "6" |
| 6 | HB_real_001315 | Aggregation & Comparison | Argmax Reasoning | In column EN, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours", "value": "6.86"} |
| 7 | HB_real_001316 | Aggregation & Comparison | Argmax Reasoning | In column SSIM, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours", "value": "0.75"} |
| 8 | HB_real_001317 | Aggregation & Comparison | Pairwise Comparison | In column SF, compare row IV and row Ours. Which value is larger? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours", "value": "14.16"} |
| 9 | HB_real_001318 | Local Relations | Marker-Anchor Neighbor | Find the left neighbor of the highlighted cell with value 14.16. The requested neighbor is numeric. Output only the number as a string. | "6.86" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000039 | Aggregation & Comparison | Grouped Aggregation | Read the table carefully and among the four +FarSight rows, which row has the maximum MMBench↑ delta? Return the row index as a string (top-to-bottom among +FarSight rows). | "4" |
| 2 | HB_real_000044 | Constrained Retrieval | Value Filtering | For the delta column MMBench↑ (scope: +FarSight rows only), what is the value at row 4? Output the full string. | "59.8 (+3.5)" |
| 3 | HB_real_000049 | Aggregation & Comparison | Pairwise Comparison | In column MMBench↑, compare LLaVA-1.5 + FarSight (Ours) and InstructBLIP + FarSight (Ours). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LLaVA-1.5 + FarSight (Ours)", "value": "66.0"} |
| 4 | HB_real_000372 | Constrained Retrieval | Value Filtering | Among the four +FarSight rows, list rows whose MMBench↑ delta is >= +3.0. Return JSON array of {"item":"...","value":"..."}. | [{"item": "InstructBLIP + FarSight (Ours)", "value": "+3.1"}, {"item": "Chat-UniVi + FarSight (Ours)", "value": "+3.5"}] |
| 5 | HB_real_000379 | Aggregation & Comparison | Extremum Reasoning | In column MMBench↑, what is the maximum value? Output only the number as a string. | "66.0" |
| 6 | HB_real_000627 | Aggregation & Comparison | Extremum Reasoning | For row LLaVA-1.5 + FarSight (Ours), what is the maximum main metric value? Ignore blue parenthesized delta values. Output only the number as a string. | "90.5" |
| 7 | HB_real_000739 | Aggregation & Comparison | Grouped Aggregation | Among the four +FarSight rows, which row has the minimum MMBench↑ delta? Return the row index as a string (top-to-bottom among +FarSight rows). | "1" |
| 8 | HB_real_000762 | Constrained Retrieval | Value Filtering | Read the table carefully and for the delta column MMBench↑ (scope: +FarSight rows only), what is the value at row 2? Output the full string. | "+3.1" |
| 9 | HB_real_001323 | Constrained Retrieval | Value Filtering | Considering only the four "+ FarSight (Ours)" rows in order, list all delta values in the MMBench↑ column (the blue parenthesized gain values). Output only a JSON array of strings. | ["+1.7", "+3.1", "+1.9", "+3.5"] |
| 10 | HB_real_001324 | Consistency & Missingness | Consistency Probing | In the MMBench↑ delta values for the four "+ FarSight (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {item,value}. | [{"item": "LLaVA-1.5 + FarSight (Ours)", "value": "+1.7"}, {"item": "InstructBLIP + FarSight (Ours)", "value": "+3.1"}, {"item": "Video-LLaVA + FarSight (Ours)", "value": "+1.9"}, {"item": "Chat-UniVi + FarSight (Ours)", "value": "+3.5"}] |
| 11 | HB_real_001327 | Consistency & Missingness | Cell Missingness | Is the cell (row LLaVA-1.5, column CHAIR_I↓) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000040 | Aggregation & Comparison | Grouped Aggregation | Please answer this question: among the four +FarSight rows, which row has the maximum ActivityNet-QA Accuracy↑ delta? Return the row index as a string (top-to-bottom among +FarSight rows). | "3" |
| 2 | HB_real_000045 | Aggregation & Comparison | Grouped Aggregation | Among the four +FarSight rows, which row has the maximum MSVD-QA Accuracy↑ delta? Return the row index as a string (top-to-bottom among +FarSight rows). | "4" |
| 3 | HB_real_000048 | Constrained Retrieval | Value Filtering | For the delta column MSVD-QA Accuracy↑ (scope: +FarSight rows only), what is the value at row 4? Output the full string. | "+2.9" |
| 4 | HB_real_000373 | Constrained Retrieval | Value Filtering | For the ActivityNet-QA Accuracy↑ delta column, considering +FarSight rows only, what is the value at row 3? Output the full string. | "+1.2" |
| 5 | HB_real_000377 | Aggregation & Comparison | Pairwise Comparison | In column ActivityNet-QA Accuracy↑, compare VILA and VILA + FarSight (Ours). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "VILA + FarSight (Ours)", "value": "51.4"} |
| 6 | HB_real_000628 | Aggregation & Comparison | Extremum Reasoning | In column ActivityNet-QA Accuracy↑, what is the maximum value? Output only the number as a string. | "51.4" |
| 7 | HB_real_000736 | Constrained Retrieval | Value Filtering | Among the four +FarSight rows, list rows whose ActivityNet-QA Accuracy↑ delta is >= +1.0. Return JSON array of {"item":"...","value":"..."}. | [{"item": "VILA + FarSight (Ours)", "value": "+1.2"}] |
| 8 | HB_real_001328 | Aggregation & Comparison | Best Selection | In metric Accuracy↑ (block MSVD-QA), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "VILA + FarSight (Ours)", "value": "74.5"} |
| 9 | HB_real_001329 | Constrained Retrieval | Value Filtering | Considering only the four "+ FarSight (Ours)" rows in order, list all delta values in ActivityNet-QA Accuracy↑ (blue parenthesized gains). Output only a JSON array of strings. | ["+0.6", "+0.5", "+1.2", "+0.5"] |
| 10 | HB_real_001331 | Consistency & Missingness | Column Missingness | List all missing entries in column Te.↑ (block Video-Based Text Generation) (N/A or —). Output only a JSON array of strings. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001339 | Constrained Retrieval | Cell Retrieval | What is the Stereo Pairs value for FSD (Ours)? Output only the cell text as a string. | "1000K" |
| 2 | HB_real_001340 | Aggregation & Comparison | Argmax Reasoning | In row Scenes, which dataset column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Spring [44]", "value": "47"} |
| 3 | HB_real_001341 | Constrained Retrieval | Value Filtering | In row Scenes, list all dataset columns with value >= 10. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Sintel [6]", "value": "10"}, {"item": "TartanAir [66]", "value": "18"}, {"item": "Spring [44]", "value": "47"}, {"item": "FSD (Ours)", "value": "12"}] |
| 4 | HB_real_001342 | Aggregation & Comparison | Pairwise Comparison | In row Scenes, compare TartanAir [66] and FSD (Ours). Which value is larger? Return JSON {"item":"<column label>","value":"<cell value>"}. | {"item": "TartanAir [66]", "value": "18"} |
| 5 | HB_real_001343 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000193 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=11,c=1) in the numeric part of the table, list all numbers highlighted with the same light-green color as the anchor. The numeric part excludes the header and left method-name column. Output only a JSON array of strings. | ["5.5", "1.8", "3.2", "4.9", "1.1", "0.5", "2.3", "2.8"] |
| 2 | HB_real_000198 | Markup Grounding | Highlight Presence | Is the value at position (r=11,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_green"} |
| 3 | HB_real_000789 | Local Relations | Marker-Anchor Neighbor | Find the down neighbor of the highlighted cell with value 5.5. The requested neighbor is numeric. Output only the number as a string. | "7.5" |
| 4 | HB_real_001040 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the highlighted cell with value 5.5. The requested neighbor is numeric. Output only the number as a string. | "1.8" |
| 5 | HB_real_001044 | Markup Grounding | Highlight Presence | Is the value at position (r=13,c=4) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 6 | HB_real_001047 | Aggregation & Comparison | Grouped Aggregation | Among the light-green highlighted cells, what is the minimum value? Output only the number as a string. | "0.5" |
| 7 | HB_real_001344 | Aggregation & Comparison | Best Selection | In metric KITTI-12 D1↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "Ours", "value": "2.3"} |
| 8 | HB_real_001345 | Aggregation & Comparison | Best Selection | In metric ETH3D BP-1↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "Ours", "value": "0.5"} |
| 9 | HB_real_001346 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 1, "col": 1, "value": "14.8"} |
| 10 | HB_real_001347 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light green? Output only the count as a string. | "8" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000061 | Markup Grounding | Marked Value Listing | Using the table only, list all numbers highlighted in light green. Output only a JSON array of strings. | ["1.26", "0.26", "0.09", "2.31", "1.52", "0.13"] |
| 2 | HB_real_000064 | Markup Grounding | Marked Value Listing | Return every value shown in bold. Output only a JSON array of strings. | ["1.26", "0.26", "0.09"] |
| 3 | HB_real_000068 | Markup Grounding | Marker Counting | How many bold numbers are in the table? Output only the count as a string. | "3" |
| 4 | HB_real_000562 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column EPE? Exclude N/A and —. Output only the number as a string. | "0.09" |
| 5 | HB_real_000563 | Constrained Retrieval | Cell Retrieval | What is the BP-1.0 value for row Ours? Output only the number as a string. | "1.52" |
| 6 | HB_real_000626 | Markup Grounding | Marker Counting | According to the table image, how many numeric cells are highlighted in light green in this table? Output only the count as a string. | "6" |
| 7 | HB_real_000773 | Aggregation & Comparison | Argmax Reasoning | In column BP-1.0, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "HITNet [56]", "value": "2.79"} |
| 8 | HB_real_000777 | Local Relations | Neighbor Value Tracing | The value 1.26 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": null, "right": "0.26"} |
| 9 | HB_real_000989 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 10 | HB_real_001350 | Aggregation & Comparison | Pairwise Comparison | In column EPE, compare Ours (finetuned) and Ours. Which value is smaller? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours (finetuned)", "value": "0.09"} |
| 11 | HB_real_001351 | Constrained Retrieval | Value Filtering | In column BP-0.5, list all methods with value <= 3.20. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "MoCha-Stereo [14]", "value": "3.20"}, {"item": "Selective-IGEV [67]", "value": "3.06"}, {"item": "Ours (finetuned)", "value": "1.26"}, {"item": "Ours", "value": "2.31"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000073 | Aggregation & Comparison | Rank Selection | For column BP-2, what is the 2nd largest row? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Post-hourglass", "value": "2.20"} |
| 2 | HB_real_000083 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column BP-2? Output only the number as a string. | "2.25" |
| 3 | HB_real_000415 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column BP-2? Output only the number as a string. | "1.97" |
| 4 | HB_real_000795 | Aggregation & Comparison | Argmax Reasoning | In column BP-2, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 5, "col": 1, "value": "2.25"} |
| 5 | HB_real_000797 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=2,c=1) in the numeric part of the table, list all numbers highlighted with the same light-green color as the anchor. Output only a JSON array of strings. | ["1.97", "1.97", "1.97", "1.97", "1.97"] |
| 6 | HB_real_000950 | Local Relations | Neighbor Indexing | At position (r=8,c=1) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "2.20", "up": "2.06", "down": "1.97", "left": null, "right": null} |
| 7 | HB_real_001002 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light green? Output only the count as a string. | "5" |
| 8 | HB_real_001354 | Aggregation & Comparison | Rank Selection | For column BP-2, what is the 3rd largest row? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "RoPE", "value": "2.19"} |
| 9 | HB_real_001355 | Markup Grounding | Highlight Presence | Is the value at position (r=2,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_green"} |
| 10 | HB_real_001356 | Markup Grounding | Highlight Presence | Is value 2.01 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": false} |
| 11 | HB_real_001357 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001358 | Aggregation & Comparison | Best Selection | In metric 7B (ppl: 5.12↓) at 40%, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "MoDeGPT (ours)", "value": "8.41"} |
| 2 | HB_real_001359 | Aggregation & Comparison | Best Selection | In metric 13B (ppl: 4.57↓) at 50%, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "MoDeGPT (ours)", "value": "8.95"} |
| 3 | HB_real_001360 | Markup Grounding | Bold Presence | Is value 8.41 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 4 | HB_real_001361 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001362 | Aggregation & Comparison | Best Selection | Within group "LLAMA-2 7B, Compress=30%", which method is best on Average? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "MoDeGPT-Alpaca (ours)", "value": "62.02"} |
| 2 | HB_real_001363 | Aggregation & Comparison | Best Selection | Within group "LLAMA-2 7B, Compress=40%", which method is best on WinoG.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "MoDeGPT-Alpaca (ours)", "value": "64.40"} |
| 3 | HB_real_001364 | Aggregation & Comparison | Pairwise Comparison | For MoDeGPT-Alpaca (ours) in group LLAMA-2 7B, Compress=30%, compare ARC-e and PIQA. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "PIQA", "value": "73.34"} |
| 4 | HB_real_001365 | Constrained Retrieval | Value Filtering | In column Average, list all rows with value >= 62. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Dense (LLAMA-2 7B, 0%)", "value": "69.00"}, {"item": "MoDeGPT-Alpaca (ours) (LLAMA-2 7B, 30%)", "value": "62.02"}, {"item": "Dense (LLAMA-3 8B, 0%)", "value": "72.75"}, {"item": "MoDeGPT-Alpaca (ours) (LLAMA-3 8B, 25%)", "value": "63.96"}] |
| 5 | HB_real_001366 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000260 | Constrained Retrieval | Value Filtering | In column Average, list highlighted numbers greater than 65. Output only a JSON array of strings. | ["65.51", "66.79"] |
| 2 | HB_real_000517 | Markup Grounding | Underline Presence | From this table image, is value 4.89 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 3 | HB_real_000909 | Constrained Retrieval | Value Filtering | In column Average, list highlighted numbers greater than 60. Output only a JSON array of strings. | ["63.08", "65.51", "66.79"] |
| 4 | HB_real_001011 | Constrained Retrieval | Value Filtering | In column Average, list highlighted numbers greater than 66. Output only a JSON array of strings. | ["66.79"] |
| 5 | HB_real_001367 | Aggregation & Comparison | Best Selection | In metric Wikitext-2↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Dense LLAMA-2 70B", "value": "3.12"} |
| 6 | HB_real_001368 | Aggregation & Comparison | Best Selection | In metric Average↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Dense LLAMA-2 70B", "value": "70.02"} |
| 7 | HB_real_001369 | Markup Grounding | Bold Presence | Is value 66.79 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 8 | HB_real_001370 | Markup Grounding | Marked Column Listing | List all underlined values in column Wikitext-2↓. Output only a JSON array of strings. | ["4.89"] |
| 9 | HB_real_001371 | Aggregation & Comparison | Rank Selection | In column Average, list the top 2 methods by value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "Dense LLAMA-2 70B", "value": "70.02"}, {"item": "MoDeGPT + our sparsity + Alpaca", "value": "66.79"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001372 | Aggregation & Comparison | Best Selection | In metric Angular Error ↓ Mirror (block Laval Indoor [20]), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "StyleLight (reported by the paper)", "value": "4.30"} |
| 2 | HB_real_001373 | Aggregation & Comparison | Best Selection | Within group "Poly Haven [3]", which method is best on Normalized RMSE ↓ Diffuse? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "SDXL† + iterative + LoRA (ours)", "value": "0.20"} |
| 3 | HB_real_001374 | Consistency & Missingness | Cell Missingness | Is the cell (row StyleLight (reported by the paper), column Normalized RMSE ↓ Diffuse) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_real_001375 | Markup Grounding | Highlight Presence | Is value 5.94 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 5 | HB_real_001376 | Markup Grounding | Highlight Presence | Is the value at row StyleLight (reproduced using official code), column Normalized RMSE ↓ Diffuse within the Laval Indoor block highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true, "color": "light_yellow"} |
| 6 | HB_real_001377 | Consistency & Missingness | Consistency Probing | For row StyleLight (reported by the paper), which Normalized RMSE columns are marked missing with "-"? Output only a JSON array of column labels. | ["Normalized RMSE ↓ Diffuse", "Normalized RMSE ↓ Matte", "Normalized RMSE ↓ Mirror"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000524 | Markup Grounding | Underline Presence | According to the image table, is value 128.8 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 2 | HB_real_000602 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. If the same value appears more than once, list it only once. Output only a JSON array of strings. | ["99.9", "1.44 / 18.5", "45.6", "128.8", "24.3", "0.74"] |
| 3 | HB_real_000927 | Aggregation & Comparison | Extremum Reasoning | For row MicKey (ours), what is the minimum numeric value across the row? Output only the number as a string. | "0.74" |
| 4 | HB_real_001378 | Aggregation & Comparison | Best Selection | Within group "Dense Features", which method is best on AUC? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "RoMa [25]", "value": "0.67"} |
| 5 | HB_real_001379 | Aggregation & Comparison | Extremum Reasoning | In column Time (ms), what is the minimum value? Output only the number as a string. | "21.6" |
| 6 | HB_real_001380 | Constrained Retrieval | Value Filtering | In column AUC, list all rows with value >= 0.70. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "MicKey w/ Overlap (ours)", "value": "0.75"}, {"item": "MicKey (ours)", "value": "0.74"}] |
| 7 | HB_real_001381 | Markup Grounding | Bold Presence | Is value 0.75 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 8 | HB_real_001382 | Local Relations | Marker-Anchor Neighbor | For the bold cell with value 0.75 in row MicKey w/ Overlap (ours), return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": null, "right": "49.2"} |
| 9 | HB_real_001383 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001384 | Aggregation & Comparison | Best Selection | Within group "D+O+P Signal", which method is best on Prec. (%)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "RoMa [25]", "value": "94.4"} |
| 2 | HB_real_001385 | Aggregation & Comparison | Best Selection | Within group "Pose Signal", which method is best on AUC? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "MicKey (ours)", "value": "0.99"} |
| 3 | HB_real_001386 | Aggregation & Comparison | Extremum Reasoning | In column Trans (m), what is the minimum value? Output only the number as a string. (Column "Trans (m) / Rot (°)" uses the Trans part for this question.) | "0.11" |
| 4 | HB_real_001387 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001388 | Aggregation & Comparison | Extremum Reasoning | In column Prec. (%), what is the maximum value? Output only the number as a string. | "92.9" |
| 2 | HB_real_001389 | Markup Grounding | Bold Presence | Is value 91.7 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 3 | HB_real_001390 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 4 | HB_real_000688 | Markup Grounding | Marked Value Listing | List all bold numeric values in the Our Depth row under the SuperGlue block. Output only a JSON array of strings. | ["0.99", "91.7", "0.11", "2.06"] |
| 5 | HB_real_000722 | Markup Grounding | Marker Counting | How many bold numeric values are in the Our Depth row under the SuperGlue block? Output only the count as a string. | "4" |
| 6 | HB_real_000667 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric metric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 4, "col": "Prec. (%)", "value": "92.9"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000614 | Aggregation & Comparison | Extremum Reasoning | From this table image, in column Pose (25cm, 5°) AUC, what is the maximum value? Output only the number as a string. | "0.36" |
| 2 | HB_real_001391 | Aggregation & Comparison | Best Selection | Within group "MicKey w/ Overlap", which solver is best on VCRE (90px) AUC? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Our Solver", "value": "0.75"} |
| 3 | HB_real_001392 | Aggregation & Comparison | Best Selection | Within group "MicKey", which solver is best on VCRE (90px) Prec. (%)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Our Solver", "value": "49.2"} |
| 4 | HB_real_001393 | Local Relations | Neighbor Value Tracing | The value 0.36 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "42.1", "right": "14.6"} |
| 5 | HB_real_001394 | Aggregation & Comparison | Extremum Reasoning | In column Pose (25cm, 5°) Prec. (%), what is the minimum value? Output only the number as a string. | "6.9" |
| 6 | HB_real_001395 | Aggregation & Comparison | Pairwise Comparison | In group SuperGlue [22, 69], compare Ess. Scale and PnP on Pose (25cm, 5°) Prec. (%). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ess. Scale", "value": "16.8"} |
| 7 | HB_real_001396 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000526 | Markup Grounding | Underline Presence | Based on the table image, is value 3.7 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 2 | HB_real_001399 | Markup Grounding | Bold Presence | Is value 233.7 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 3 | HB_real_001400 | Markup Grounding | Marked Column Listing | List all underlined values in column Rot (from "Rep. / Trans / Rot"). Output only a JSON array of strings. | ["75.2"] |
| 4 | HB_real_000536 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["0.08", "7.3", "3.7", "75.2"] |
| 5 | HB_real_000942 | Markup Grounding | Marker Counting | How many bold numeric cells are in the table? Output only the count as a string. | "3" |
| 6 | HB_real_001336 | Markup Grounding | Marked Value Listing | List all bold numeric values in the table. Output only a JSON array of strings. | ["0.15 / 11.1", "3.1 / 58.3", "233.7"] |
| 7 | HB_real_001026 | Markup Grounding | Marked Column Listing | In column Trans from Median Errors, list all underlined values. Output only a JSON array of strings. | ["3.7"] |
| 8 | HB_real_001841 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left method-label column. | {"underlined": false} |
| 9 | HB_real_001027 | Markup Grounding | Marker Counting | How many underlined numeric values are in the table? Output only the count as a string. | "4" |
| 10 | HB_real_001461 | Markup Grounding | Marked Column Listing | In column Rot from Median Errors, list all bold values. Output only a JSON array of strings. | ["58.3"] |
| 11 | HB_real_002000 | Markup Grounding | Bold Presence | At position (r=1,c=2) in the numeric part of the table, is the value boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left method-label column. | {"bold": false} |
| 12 | HB_real_001028 | Aggregation & Comparison | Grouped Aggregation | Exclude underlined cells. Among the remaining values in column Rot from Median Errors, what is the minimum value? Output only the number as a string. | "58.3" |
| 13 | HB_real_001037 | Aggregation & Comparison | Argmax Reasoning | Within the Median Errors section, find the largest numeric value. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and columns Rep., Trans, Rot left-to-right. | {"row": 1, "col": 1, "value": "271.8"} |
| 14 | HB_real_001338 | Local Relations | Neighbor Indexing | Within the Median Errors section, at position (r=4,c=2), return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "3.7", "up": "3.1", "down": null, "left": "233.7", "right": "75.2"} |
| 15 | HB_real_001038 | Local Relations | Neighbor Value Tracing | The value 75.2 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "3.7", "right": null} |
| 16 | HB_real_001045 | Aggregation & Comparison | Grouped Aggregation | Across all methods, what is the mean AUC value in the VCRE section? Output the value rounded to two decimal places as a string. | "0.09" |
| 17 | HB_real_001940 | Constrained Retrieval | Cell Retrieval | Within the VCRE section, find the Prec. value at position (r=2,c=2), where c=1 is AUC and c=2 is Prec. Output only the number as a string. | "2.6" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001401 | Aggregation & Comparison | Best Selection | In metric LPIPS↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "B", "value": "0.110"} |
| 2 | HB_real_001402 | Markup Grounding | Highlight Presence | Is value 36.529 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 3 | HB_real_001403 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 4 | HB_real_001086 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 36.529, list all numeric values highlighted with the same light-pink background as the anchor. Output only a JSON array of strings. | ["0.943", "0.110", "36.529"] |
| 5 | HB_real_000772 | Markup Grounding | Marked Value Listing | List all numbers highlighted in light yellow. Output only a JSON array of strings. | ["0.113", "36.294", "0.943"] |
| 6 | HB_real_000457 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light pink. What is their maximum numeric value? Output only the number as a string. | "36.529" |
| 7 | HB_real_000664 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric metric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 5, "col": "PSNR", "value": "36.529"} |
| 8 | HB_real_000625 | Markup Grounding | Marker Counting | How many numeric cells are highlighted with a background color in the table? Output only the count as a string. | "9" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001404 | Aggregation & Comparison | Best Selection | In metric SSIM↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "H", "value": "0.877"} |
| 2 | HB_real_001405 | Constrained Retrieval | Value Filtering | In column PSNR, list all rows with value >= 33.7. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "B", "value": "34.042"}, {"item": "E", "value": "33.845"}, {"item": "H", "value": "33.778"}] |
| 3 | HB_real_001406 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 4 | HB_real_001090 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 34.042, list all numeric values highlighted with the same light-pink background as the anchor. Output only a JSON array of strings. | ["34.042", "0.148", "0.877"] |
| 5 | HB_real_000775 | Markup Grounding | Marked Value Listing | List all numbers highlighted in light orange. Output only a JSON array of strings. | ["0.870", "33.845", "0.148"] |
| 6 | HB_real_000458 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light yellow. What is their maximum numeric value? Output only the number as a string. | "33.778" |
| 7 | HB_real_000665 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric metric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 2, "col": "PSNR", "value": "34.042"} |
| 8 | HB_real_000990 | Markup Grounding | Marker Counting | How many numeric cells are highlighted with a background color in the table? Output only the count as a string. | "9" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000424 | Consistency & Missingness | Column Missingness | From this table image, list all missing entries in column DA (N/A or —). Output only a JSON array of row labels. | ["(M) Human"] |
| 2 | HB_real_000918 | Consistency & Missingness | Cell Missingness | Is the cell at row (M) Human and column DA missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 3 | HB_real_001407 | Aggregation & Comparison | Extremum Reasoning | In column Accuracy, what is the maximum value? Output only the value string as shown. | "94%" |
| 4 | HB_real_001408 | Aggregation & Comparison | Extremum Reasoning | In column Duration (s), what is the minimum value? Output only the value string as shown. | "~13.37" |
| 5 | HB_real_001409 | Constrained Retrieval | Value Filtering | In column Accuracy, list all rows with value >= 30. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "(G) Swarm(3×TOT)", "value": "30.18±4.30"}, {"item": "(L) Swarm(7×TOT)", "value": "30.56±3.25"}, {"item": "(M) Human", "value": "94%"}] |
| 6 | HB_real_001411 | Markup Grounding | Marker Counting | How many underlined numeric values are in the table? Output only the count as a string. | "0" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000087 | Aggregation & Comparison | Extremum Reasoning | For item OmniThink in group Qwen-Plus, what is the maximum value across all metrics? Exclude missing entries. Output only the number as a string. | "11.66" |
| 2 | HB_real_000095 | Aggregation & Comparison | Extremum Reasoning | For item OmniThink in group Qwen-Plus, what is the minimum value across all metrics? Exclude missing entries. Output only the number as a string. | "0.7230" |
| 3 | HB_real_000472 | Markup Grounding | Marked Value Listing | List all bold numbers in the OmniThink row within group Qwen-Plus. Output only a JSON array of strings. | ["4.00", "3.92", "4.06", "3.38", "0.7230", "11.66"] |
| 4 | HB_real_000828 | Markup Grounding | Marker Counting | How many bold numeric cells are in the OmniThink row within group Qwen-Plus? Output only the count as a string. | "6" |
| 5 | HB_real_000831 | Aggregation & Comparison | Best Selection | Within group Qwen-Plus, which method is best on Knowledge Density? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "OmniThink", "value": "11.66"} |
| 6 | HB_real_001434 | Aggregation & Comparison | Best Selection | Within group GPT-4o, which row label is best on Knowledge Density? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "OmniThink", "value": "22.31"} |
| 7 | HB_real_001435 | Aggregation & Comparison | Rank Selection | List the top-3 items by column Knowledge Density, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "GPT-4o / OmniThink", "value": "22.31"}, {"item": "GPT-4o / RAG", "value": "22.11"}, {"item": "GPT-4o / oRAG", "value": "19.70"}] |
| 8 | HB_real_001436 | Aggregation & Comparison | Rank Selection | Within group Qwen-Plus, list the top-2 methods by Knowledge Density, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "OmniThink", "value": "11.66"}, {"item": "Co-STORM*", "value": "11.52"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001412 | Aggregation & Comparison | Best Selection | Within block Flowers102 (upper half, UL=✗), which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL (Ours)", "value": "84.60"} |
| 2 | HB_real_001413 | Aggregation & Comparison | Best Selection | Within block RESISC45 (upper half, UL=✓), which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL (Ours)", "value": "85.85"} |
| 3 | HB_real_001414 | Constrained Retrieval | Value Filtering | In column EuroSAT TRZSL, list all rows with value >= 90. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "GRIP (UL=✓)", "value": "92.33"}, {"item": "CPL (Ours, UL=✓)", "value": "93.78"}] |
| 4 | HB_real_001415 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000618 | Aggregation & Comparison | Pairwise Comparison | From this table image, in column DTD UL, compare CPL (Ours, UL=✗) and CPL (Ours, UL=✓). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL (Ours, UL=✓)", "value": "53.42"} |
| 2 | HB_real_001416 | Aggregation & Comparison | Best Selection | Within block Flowers102 (upper half, UL=✗), which method is best on SSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL (Ours)", "value": "70.58"} |
| 3 | HB_real_001417 | Aggregation & Comparison | Best Selection | Within block RESISC45 (upper half, UL=✓), which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL (Ours)", "value": "86.67"} |
| 4 | HB_real_001418 | Aggregation & Comparison | Best Selection | Within block EuroSAT (lower half, UL=✓), which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL (Ours)", "value": "98.31"} |
| 5 | HB_real_001419 | Consistency & Missingness | Consistency Probing | In the DTD block, compare CPL (Ours, UL=✓) against CPL (Ours, UL=✗). List the metric columns where the UL=✓ row has a higher value. Output only a JSON array of column labels. | ["SSL", "UL", "TRZSL"] |
| 6 | HB_real_001420 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000238 | Aggregation & Comparison | Rank Selection | Using the table only, in column CALTECH-101, list the top 2 rows by value in descending order. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "LaFTer + Ours", "value": "93.4"}, {"item": "LaFTer", "value": "93.3"}] |
| 2 | HB_real_000499 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "6" |
| 3 | HB_real_000596 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["71.5", "48.0", "68.2", "74.6", "73.9", "93.3"] |
| 4 | HB_real_000869 | Aggregation & Comparison | Extremum Reasoning | In row LaFTer + Ours, what is the maximum numeric value across the table? Output only the number as a string. | "93.4" |
| 5 | HB_real_001421 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column EuroSAT? Output only the number as a string. | "82.2" |
| 6 | HB_real_001422 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 7 | HB_real_000050 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=5,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-gray row background as the anchor row. Output only a JSON array of strings. | ["76.7", "71.0", "77.3", "82.2", "56.3", "93.4"] |
| 8 | HB_real_000878 | Markup Grounding | Underline Presence | Is the value at position (r=3,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 9 | HB_real_001460 | Markup Grounding | Marked Column Listing | List all underlined values in column CALTECH-101. Output only a JSON array of strings. | ["93.3"] |
| 10 | HB_real_000806 | Aggregation & Comparison | Grouped Aggregation | Exclude underlined cells. On the remaining cells in column Flowers-102, what is the maximum value? Output only the number as a string. | "76.7" |
| 11 | HB_real_000933 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 5, "col": "CALTECH-101", "value": "93.4"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001423 | Aggregation & Comparison | Best Selection | Within block UL=✗, which method is best on SSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL CAV", "value": "55.50"} |
| 2 | HB_real_001424 | Aggregation & Comparison | Best Selection | Within block UL=✓, which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CPL LW", "value": "68.29"} |
| 3 | HB_real_001425 | Aggregation & Comparison | Extremum Reasoning | In column UL, what is the maximum value? Output only the number as a string. | "52.23" |
| 4 | HB_real_001426 | Constrained Retrieval | Value Filtering | Within block UL=✓, in column SSL, list all rows with value >= 60. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "CPL Soft CE", "value": "60.83"}, {"item": "CPL CC", "value": "61.21"}, {"item": "CPL RC", "value": "60.21"}, {"item": "CPL CAV", "value": "61.06"}, {"item": "CPL LW", "value": "60.20"}] |
| 5 | HB_real_001427 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000559 | Constrained Retrieval | Value Filtering | Based on the table image, in group DTD, column SSL, list all rows with value >= 62. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Use only the main value and ignore the small subscript values. | [{"item": "CPL (UL=✗)", "value": "62.78"}, {"item": "CPL (UL=✓)", "value": "69.82"}] |
| 2 | HB_real_000634 | Constrained Retrieval | Value Filtering | In group DTD, column TRZSL, list all rows with value >= 62. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Use only the main value and ignore the small subscript values. | [{"item": "CPL (UL=✗)", "value": "62.52"}, {"item": "GRIP (UL=✓)", "value": "64.92"}, {"item": "CPL (UL=✓)", "value": "71.97"}] |
| 3 | HB_real_001428 | Aggregation & Comparison | Best Selection | Within group DTD, which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Use only the main value and ignore the small subscript values. | {"item": "CPL (UL=✓)", "value": "71.97"} |
| 4 | HB_real_001429 | Aggregation & Comparison | Best Selection | Within group RESISC45, which method is best on SSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Use only the main value and ignore the small subscript values. | {"item": "CPL (UL=✓)", "value": "87.75"} |
| 5 | HB_real_001430 | Aggregation & Comparison | Best Selection | Within group Flowers102, which method is best on TRZSL? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Use only the main value and ignore the small subscript values. | {"item": "CPL (UL=✓)", "value": "97.34"} |
| 6 | HB_real_001431 | Aggregation & Comparison | Pairwise Comparison | In group RESISC45, column UL, compare CPL (UL=✗) and CPL (UL=✓). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Use only the main value and ignore the small subscript values. | {"item": "CPL (UL=✓)", "value": "80.88"} |
| 7 | HB_real_001432 | Aggregation & Comparison | Extremum Reasoning | For row CPL (UL=✓) in group DTD, what is the minimum value across SSL, UL, and TRZSL? Output only the number as a string. Use only the main value and ignore the small subscript values. | "57.20" |
| 8 | HB_real_001433 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000427 | Consistency & Missingness | Column Missingness | By reading the table in the image, list all missing entries in column VQAv2 (N/A or —). Output only a JSON array of row labels. | ["InstructBLIP (Dai et al., 2023)", "LLaVA (Liu et al., 2023b)", "mPLUG-Owl (Ye et al., 2023b)"] |
| 2 | HB_real_000926 | Markup Grounding | Underline Presence | From this table, is value 66.7 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 3 | HB_real_001438 | Aggregation & Comparison | Extremum Reasoning | In column MMB, what is the maximum value? Output only the number as a string. | "60.9" |
| 4 | HB_real_001439 | Consistency & Missingness | Column Missingness | Which row labels have missing entries in column OKVQA? Output only a JSON array of row-label strings. | ["LLaVA (Liu et al., 2023b)", "mPLUG-Owl (Ye et al., 2023b)", "Video-LLaVA (Lin et al., 2023)"] |
| 5 | HB_real_001440 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["123.1", "82.7", "119.3", "34.0", "48.1", "66.7", "58.0"] |
| 6 | HB_real_001796 | Markup Grounding | Marked Column Listing | In column NoCaps, list the underlined values. Output only a JSON array of strings. | ["123.1"] |
| 7 | HB_real_000997 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Model and Version columns. | {"underlined": true} |
| 8 | HB_real_001704 | Markup Grounding | Marked Value Listing | List the bold numeric values in the NExT-GPT row. Output only a JSON array of strings. | ["123.7", "84.5", "124.9", "48.4", "52.1", "57.5"] |
| 9 | HB_real_001334 | Markup Grounding | Marker Counting | How many bold numeric values are in the table? Output only the count as a string. | "8" |
| 10 | HB_real_000698 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=7,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. The numeric part excludes the left Model and Version columns. | ["123.7", "84.5", "124.9", "66.7", "48.4", "52.1", "58.0", "57.5"] |
| 11 | HB_real_001748 | Constrained Retrieval | Value Filtering | In the light-blue highlighted NExT-GPT row, list the numeric values greater than 80. Output only a JSON array of strings. | ["123.7", "84.5", "124.9"] |
| 12 | HB_real_001855 | Aggregation & Comparison | Grouped Aggregation | Exclude underlined cells. Among the remaining values in column NoCaps, what is the maximum value? Output only the number as a string. | "123.7" |
| 13 | HB_real_000981 | Markup Grounding | Bold Presence | At position (r=7,c=4) in the numeric part of the table, is the value boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left Model and Version columns. | {"bold": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000518 | Markup Grounding | Underline Presence | Using the visual table content, is value 64.5 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 2 | HB_real_001441 | Aggregation & Comparison | Extremum Reasoning | In column MSR-VTT, what is the maximum value? Output only the number as a string. | "76.2" |
| 3 | HB_real_001442 | Consistency & Missingness | Column Missingness | List all missing entries in column NExTQA (N/A or —). Output only a JSON array of row labels. | ["Codi (Tang et al., 2023)", "UIO-2XXL (Lu et al., 2023)", "Video-LLaVA (Lin et al., 2023)"] |
| 4 | HB_real_001443 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["74.4*", "78.9*", "29.6", "59.2", "64.5"] |
| 5 | HB_real_001349 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 76.2*, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["76.2*", "64.5", "61.4", "50.7", "81.3*"] |
| 6 | HB_real_000873 | Markup Grounding | Underline Presence | Is the value at position (r=6,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 7 | HB_real_000897 | Markup Grounding | Marked Column Listing | List all underlined values in column MSVD-QA. Output only a JSON array of strings. | ["64.5"] |
| 8 | HB_real_000643 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light blue in the NExT-GPT row. What is their maximum numeric value? Output only the number as a string. | "81.3" |
| 9 | HB_real_001335 | Markup Grounding | Marked Value Listing | List all bold numeric values in the NExT-GPT row. Output only a JSON array of strings. | ["76.2*", "61.4", "50.7", "81.3*"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000522 | Markup Grounding | Underline Presence | From this table image, is value 11.18 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 2 | HB_real_001444 | Aggregation & Comparison | Best Selection | In metric Image FID(↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "NExT-GPT", "value": "10.07"} |
| 3 | HB_real_001445 | Aggregation & Comparison | Best Selection | In metric Audio FAD(↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "NExT-GPT", "value": "1.68"} |
| 4 | HB_real_001446 | Aggregation & Comparison | Best Selection | In metric Video CLIPSIM(↑), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "NExT-GPT", "value": "31.97"} |
| 5 | HB_real_001447 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["11.18", "1.74", "30.96"] |
| 6 | HB_real_001464 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 10.07, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["10.07", "1.68", "31.97"] |
| 7 | HB_real_000874 | Markup Grounding | Underline Presence | Is the value at position (r=7,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 8 | HB_real_000899 | Markup Grounding | Marked Column Listing | List all underlined values in column FAD. Output only a JSON array of strings. | ["1.74"] |
| 9 | HB_real_000807 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light blue in the NExT-GPT row. What is their maximum numeric value? Output only the number as a string. | "31.97" |
| 10 | HB_real_000941 | Markup Grounding | Marker Counting | How many bold numeric cells are in the NExT-GPT row? Output only the count as a string. | "3" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001448 | Aggregation & Comparison | Best Selection | In metric OSR↑ (block val seen), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Ours", "value": "78.36"} |
| 2 | HB_real_001449 | Aggregation & Comparison | Best Selection | In metric SPL↑ (block test unseen), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Ours", "value": "38.70"} |
| 3 | HB_real_001450 | Consistency & Missingness | Cell Missingness | Is the cell (row TD-STP [71], column TL in block val seen) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_real_001451 | Consistency & Missingness | Consistency Probing | List all cells marked as missing using "-" or "N/A". Return a JSON array of strings in the form "<row label> | <block> | <column>". If none, output []. | ["TD-STP [71] | val seen | TL", "TD-STP [71] | val seen | OSR", "TD-STP [71] | val seen | SR", "TD-STP [71] | val seen | SPL", "TD-STP [71] | val seen | RGS", "TD-STP [71] | val seen | RGSPL", "TD-STP [71] | val unseen | TL", "TD-STP [71] | test unseen | TL"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000434 | Consistency & Missingness | Column Missingness | From this table image, list all missing entries in column TL (block val unseen) (N/A or —). Output only a JSON array of row labels. | ["SF [3]", "AuxRN [16]", "EGP [12]", "GBE [70]", "CCC [66]", "TD-STP [71]"] |
| 2 | HB_real_001452 | Consistency & Missingness | Column Missingness | Which row labels have missing entries in column TL (block test unseen)? Output only a JSON array of row-label strings. | ["AuxRN [16]", "EGP [12]", "GBE [70]", "CCC [66]", "TD-STP [71]"] |
| 3 | HB_real_001453 | Aggregation & Comparison | Pairwise Comparison | For Ours in block val unseen, compare SR and SPL. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "SR", "value": "74"} |
| 4 | HB_real_000995 | Markup Grounding | Marked Value Listing | List all bold numeric values in row Ours. Output only a JSON array of strings. | ["2.89", "74", "62", "3.19", "73", "62"] |
| 5 | HB_real_000723 | Markup Grounding | Marker Counting | How many bold numeric cells are in row Ours? Output only the count as a string. | "6" |
| 6 | HB_real_000668 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 18, "col": "val unseen SR", "value": "74"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001454 | Aggregation & Comparison | Argmax Reasoning | In column all, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Llama-3.1-70B / Self-CoT", "value": "76.7"} |
| 2 | HB_real_001455 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the minimum value? Output only the number as a string. | "8.1" |
| 3 | HB_real_001456 | Aggregation & Comparison | Pairwise Comparison | For Qwen2-72B / Self-CoT, compare CSV and Tree. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "CSV", "value": "95.5"} |
| 4 | HB_real_001457 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000217 | Markup Grounding | Marked Column Listing | Based on the table, in column Simple 3-Shot, return the values that are underlined. Output only a JSON array of strings. | ["49.7", "40.6"] |
| 2 | HB_real_000220 | Markup Grounding | Marked Column Listing | List all underlined numbers in column 3-Shot. Output only a JSON array of strings. | ["65.6", "58.4"] |
| 3 | HB_real_000229 | Markup Grounding | Marked Column Listing | Using the table only, list all bold numbers in column 3-Shot. Output only a JSON array of strings. | ["69.5", "61.4"] |
| 4 | HB_real_000317 | Markup Grounding | Marked Column Listing | List all bold numbers in column Base. Output only a JSON array of strings. | ["51.1", "45.8"] |
| 5 | HB_real_000321 | Markup Grounding | Marked Column Listing | List all bold numbers in column w/ Hint. Output only a JSON array of strings. | ["54.2", "56.0"] |
| 6 | HB_real_000544 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Base. Output only a JSON array of strings. | ["47.3", "42.5"] |
| 7 | HB_real_000547 | Markup Grounding | Marked Column Listing | List all underlined numbers in column w/ Hint. Output only a JSON array of strings. | ["50.9", "45.3"] |
| 8 | HB_real_000603 | Markup Grounding | Marked Column Listing | List all bold numbers in column Simple 3-Shot. Output only a JSON array of strings. | ["51.7", "50.1"] |
| 9 | HB_real_000693 | Markup Grounding | Underline Presence | Is the number with value 50.9 underlined? Answer JSON {"underlined":true/false}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 10 | HB_real_000699 | Markup Grounding | Bold Presence | Is the number with value 61.4 boldfaced? Answer JSON {"bold":true/false}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 11 | HB_real_000965 | Markup Grounding | Marker Counting | How many underlined numeric cells are in the table? Output only the count as a string. | "8" |
| 12 | HB_real_001458 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=4,c=4) in the numeric part of the table. Output only the value as a string. | "11.9" |
| 13 | HB_real_001459 | Markup Grounding | Marker Counting | How many bold numeric cells are in the table? Output only the count as a string. | "8" |
| 14 | HB_real_001462 | Markup Grounding | Highlight Presence | Is value 51.1 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000523 | Markup Grounding | Underline Presence | Is value 98.7/98.9/96.2 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 2 | HB_real_000599 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=7,c=3) in the numeric/text data part of the table. Output only the value as a string. | "98.7/98.9/96.2" |
| 3 | HB_real_000852 | Constrained Retrieval | Color-Anchor Retrieval | For the value 98.9/99.0/96.6, list all values highlighted with the same light-pink color as the anchor. Output only a JSON array of strings. | ["99.7/99.9/99.2", "98.5/71.0/69.7/94.9", "98.9/99.0/96.6", "98.9/51.2/54.7/94.4", "90.5/88.1/81.5", "99.0/47.5/50.3/95.0"] |
| 4 | HB_real_000902 | Markup Grounding | Marker Counting | Read the table carefully and how many highlighted numeric data cells are in light pink in the table? Output only the count as a string. | "6" |
| 5 | HB_real_001463 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=8,c=6) in the numeric/text data part of the table. Output only the value as a string. | "99.0/47.5/50.3/95.0" |
| 6 | HB_real_001465 | Markup Grounding | Bold Presence | Is value 99.0/47.5/50.3/95.0 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 7 | HB_real_001466 | Markup Grounding | Marked Value Listing | List all underlined data entries in the Real-IAD Image-level column. Output only a JSON array of strings. | ["89.3/86.8/80.2"] |
| 8 | HB_real_001467 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 9 | HB_real_001084 | Markup Grounding | Marked Value Listing | In the pink highlighted INP-Former row, list the two MVTec-AD cells. Output only a JSON array of strings. | ["99.7/99.9/99.2", "98.5/71.0/69.7/94.9"] |
| 10 | HB_real_001048 | Constrained Retrieval | Value Filtering | In the pink highlighted INP-Former row, list the cells whose first number is at least 99. Output only a JSON array of strings. | ["99.7/99.9/99.2", "99.0/47.5/50.3/95.0"] |
| 11 | HB_real_001488 | Markup Grounding | Marked Column Listing | In the MVTec-AD Image-level column, list all bold values. Output only a JSON array of strings. | ["99.7/99.9/99.2"] |
| 12 | HB_real_002002 | Markup Grounding | Bold Presence | At position (r=1,c=1) in the numeric part of the table, is the cell boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left Method column. | {"bold": false} |
| 13 | HB_real_001944 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the cell underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Method column. | {"underlined": false} |
| 14 | HB_real_001089 | Aggregation & Comparison | Grouped Aggregation | Exclude pink highlighted cells. In the MVTec-AD Image-level column, what is the highest remaining cell value? Output the full cell string. | "99.6/99.8/99.0" |
| 15 | HB_real_001665 | Local Relations | Neighbor Indexing | At position (r=7,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part excludes the left Method column. | {"target": "98.7/98.9/96.2", "up": "94.3/94.5/89.4", "down": "98.9/99.0/96.6", "left": "98.4/69.3/69.2/94.8", "right": "98.7/53.2/55.7/94.5"} |
| 16 | HB_real_001325 | Local Relations | Neighbor Value Tracing | The cell value 90.5/88.1/81.5 appears exactly once. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "98.9/51.2/54.7/94.4", "right": "99.0/47.5/50.3/95.0"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000263 | Markup Grounding | Marker Counting | Using the table only, how many highlighted numeric data cells are in light pink in the table? Output only the count as a string. | "6" |
| 2 | HB_real_000519 | Markup Grounding | Underline Presence | In the cell value 73.0†/61.8†/61.0†, is the number 73.0† underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 3 | HB_real_000858 | Constrained Retrieval | Color-Anchor Retrieval | Based on the table, for the number with value 97.6/98.6/97.0, list all numbers highlighted with the same light-pink color as the anchor. Output only a JSON array of strings. | ["97.6/98.6/97.0", "97.0/65.9/65.6/92.9", "96.4/96.0/93.0", "97.7/49.3/54.3/93.1", "76.7/72.3/71.7", "97.3/32.2/36.7/89.0"] |
| 4 | HB_real_001468 | Markup Grounding | Marked Column Listing | In column Real-IAD Image-level, list every boldfaced data value. Output only a JSON array of strings. | ["76.7/72.3/71.7"] |
| 5 | HB_real_001470 | Markup Grounding | Bold Presence | Is value 76.7/72.3/71.7 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 6 | HB_real_001471 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=4,c=5) in the numeric/text data part of the table. Output only the value as a string. | "73.0†/61.8†/61.0†" |
| 7 | HB_real_001472 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 8 | HB_real_001735 | Markup Grounding | Marked Value Listing | In the WinCLIP row, list all underlined numeric values. Output only a JSON array of strings. | ["97.3", "94.7", "59.5", "88.8", "47.0", "87.6", "73.0", "93.8", "13.3", "21.0", "76.4"] |
| 9 | HB_real_001348 | Markup Grounding | Marked Column Listing | In the MVTec-AD Image-level column, list all underlined values. Output only a JSON array of strings. | ["97.3", "94.7", "96.6"] |
| 10 | HB_real_001973 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the cell underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Method column. | {"underlined": false} |
| 11 | HB_real_001669 | Markup Grounding | Marked Column Listing | In the MVTec-AD Image-level column, list all bold values. Output only a JSON array of strings. | ["97.6/98.6/97.0"] |
| 12 | HB_real_001353 | Markup Grounding | Bold Presence | At position (r=1,c=1) in the numeric part of the table, is the cell boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left Method column. | {"bold": false} |
| 13 | HB_real_001829 | Markup Grounding | Marked Value Listing | In the pink highlighted INP-Former row, list the two MVTec-AD cells. Output only a JSON array of strings. | ["97.6/98.6/97.0", "97.0/65.9/65.6/92.9"] |
| 14 | HB_real_001520 | Aggregation & Comparison | Grouped Aggregation | Exclude pink highlighted cells. In the MVTec-AD Image-level column, what is the highest remaining cell value? Output the full cell string. | "96.6/-/-" |
| 15 | HB_real_001666 | Local Relations | Neighbor Indexing | At position (r=4,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part excludes the left Method column. | {"target": "87.3/88.8/84.2", "up": "85.3/87.5/84.3", "down": "89.1/-/-", "left": "96.2/-/59.5/89.0", "right": "97.2/-/47.0/87.6"} |
| 16 | HB_real_rebal_000067 | Markup Grounding | Marked Column Listing | In the Real-IAD Pixel-level column, list all underlined values. Output only a JSON array of strings. | ["93.8", "13.3", "21.0", "76.4"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000429 | Consistency & Missingness | Column Missingness | From this table image, list all missing entries in column Real-IAD P-AP (N/A or —). Output only a JSON array of row labels. | ["PatchCore [38]", "RD4AD [10]", "SimpleNet [30]"] |
| 2 | HB_real_000928 | Markup Grounding | Underline Presence | Please answer this question: is value 98.5 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 3 | HB_real_001473 | Aggregation & Comparison | Extremum Reasoning | In column MVTec-AD P-AP, what is the maximum value? Output only the number as a string. | "70.2" |
| 4 | HB_real_001474 | Aggregation & Comparison | Extremum Reasoning | In row INP-Former, what is the minimum value across all dataset metrics? Output only the number as a string. | "48.1" |
| 5 | HB_real_001475 | Consistency & Missingness | Column Missingness | Which row labels have missing entries in column Real-IAD AUPRO? Output only a JSON array of row-label strings. | [] |
| 6 | HB_real_001476 | Markup Grounding | Bold Presence | Is value 95.6 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 7 | HB_real_001477 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["99.6", "68.9", "95.0", "92.0", "45.2", "95.1", "98.5", "49.2", "93.8"] |
| 8 | HB_real_001983 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=5,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-pink row background as the anchor row. Output only a JSON array of strings. | ["99.7", "70.2", "95.4", "98.5", "49.2", "93.8", "92.1", "48.1", "95.6"] |
| 9 | HB_real_000875 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 10 | HB_real_000966 | Markup Grounding | Marked Column Listing | List all underlined values in column Real-IAD AUPRO. Output only a JSON array of strings. | ["95.1"] |
| 11 | HB_real_000805 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. On the remaining cells in column Real-IAD P-AP, what is the maximum value? Output only the number as a string. | "45.2" |
| 12 | HB_real_000931 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find all cells tied for the largest numeric value in the table. Return a JSON array of {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | [{"row": 4, "col": "MVTec-AD I-AUROC", "value": "99.7"}, {"row": 5, "col": "MVTec-AD I-AUROC", "value": "99.7"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000262 | Markup Grounding | Marker Counting | Based on the table, how many highlighted numeric data cells are in light pink in the table? Output only the count as a string. | "4" |
| 2 | HB_real_000569 | Constrained Retrieval | Color-Anchor Retrieval | Based on the table image, for the number with value 98.90/99.02/96.57, list all numbers highlighted with the same light-pink color as the anchor. Output only a JSON array of strings. | ["99.67/99.88/99.20", "98.48/71.02/69.65/94.87", "98.90/99.02/96.57", "98.90/51.22/54.74/94.36"] |
| 3 | HB_real_000861 | Markup Grounding | Highlight Presence | Is the value at position (r=4,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_pink"} |
| 4 | HB_real_000987 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light pink in the MVTec-AD Image-level column. What is their max? Exclude N/A and —. Output only the number/string as shown. | "99.67/99.88/99.20" |
| 5 | HB_real_001007 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are highlighted in light pink. On the remaining cells in column MVTec-AD Image-level, what is the max? Exclude N/A and —. Output only the number/string as shown. | "99.61/99.83/99.02" |
| 6 | HB_real_001064 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column MVTec-AD Image-level. Output JSON {"row":<int>,"col":<int>,"value":"<string>"}. The numeric part of the table excludes headers and the left setting columns. | {"row": 4, "col": 1, "value": "99.67/99.88/99.20"} |
| 7 | HB_real_001478 | Aggregation & Comparison | Argmax Reasoning | In column MVTec-AD Image-level, which setting row has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "INP=✔, Lc=✔, Lsm=✔", "value": "99.67/99.88/99.20"} |
| 8 | HB_real_001481 | Markup Grounding | Highlight Presence | Is value 98.39/70.01/69.53/95.10 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": false} |
| 9 | HB_real_001482 | Markup Grounding | Bold Presence | Is value 99.67/99.88/99.20 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 10 | HB_real_001483 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000223 | Markup Grounding | Marked Column Listing | Based on the table, in column JFR within the Overall block, return the values that are bold. Output only a JSON array of strings. | ["3.50"] |
| 2 | HB_real_000226 | Markup Grounding | Marked Column Listing | List all bold numbers in column JFR (block Fake News). Output only a JSON array of strings. | ["1.46"] |
| 3 | HB_real_000316 | Markup Grounding | Marked Column Listing | List all bold numbers in column IMR (block Social Rumor). Output only a JSON array of strings. | ["10.48"] |
| 4 | HB_real_000610 | Markup Grounding | Marked Column Listing | List all underlined numbers in column IMR (block Complex Claim). Output only a JSON array of strings. | ["22.08"] |
| 5 | HB_real_000692 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Grade (block Overall). Output only a JSON array of strings. | ["7.17"] |
| 6 | HB_real_000696 | Constrained Retrieval | Cell Retrieval | What is the Overall Grade value for row GPT-4o? Output only the number as a string. | "7.21" |
| 7 | HB_real_000983 | Aggregation & Comparison | Argmax Reasoning | In column Grade↑ (block Overall), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 13, "col": 12, "value": "7.21"} |
| 8 | HB_real_001484 | Aggregation & Comparison | Best Selection | In metric IMR↓ (block Social Rumor), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT-4o", "value": "10.48"} |
| 9 | HB_real_001485 | Aggregation & Comparison | Best Selection | In metric Grade↑ (block Overall), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT-4o", "value": "7.21"} |
| 10 | HB_real_001486 | Aggregation & Comparison | Best Selection | In metric JFR↓ (block Fake News), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Qwen2.5-72B", "value": "1.46"} |
| 11 | HB_real_001487 | Markup Grounding | Underline Presence | Is value 22.08 underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001489 | Aggregation & Comparison | Argmax Reasoning | In column Airline Level 3 Acc(t), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "o1-preview, 1-shot", "value": "0.46"} |
| 2 | HB_real_001490 | Aggregation & Comparison | Pairwise Comparison | For o1-preview (1-shot) in Airline Level 3, compare AC(t) and R(t). Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "AC(t)", "value": "0.961"} |
| 3 | HB_real_001491 | Markup Grounding | Highlight Presence | Is value 0.968 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 4 | HB_real_001492 | Markup Grounding | Bold Presence | Is value 0.972 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 5 | HB_real_001493 | Consistency & Missingness | Consistency Probing | Which metric columns have entries marked as missing using "-" in the NBA Transaction section? Output only a JSON array of column labels. | ["Level 1 AC(t)", "Level 2 AC(t)", "Level 3 AC(t)"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001494 | Aggregation & Comparison | Extremum Reasoning | In column Size, what is the minimum value? Output only the number as a string. | "546" |
| 2 | HB_real_001495 | Aggregation & Comparison | Argmax Reasoning | In column # Rel., which benchmark has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "VSR (Liu et al., 2023a)", "value": "66"} |
| 3 | HB_real_001496 | Consistency & Missingness | Consistency Probing | List all cells marked as missing using "-" or "N/A". Return a JSON array of strings in the form "<row label> | <column>". If none, output []. | ["SpatialSense (Yang et al., 2019) | # Type", "Rel3D (Goyal et al., 2020) | w/o Kn.", "SpatialSense+ (Wen et al., 2024) | # Type", "SpatialVLM (Chen et al., 2024) | # Rel.", "SpatialVLM (Chen et al., 2024) | # Type"] |
| 4 | HB_real_000705 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=1,c=1) in the numeric part of the table. Output only the cell value as a string. | "17" |
| 5 | HB_real_001410 | Consistency & Missingness | Column Missingness | Which benchmark rows have missing entries marked with "-" in column # Type? Output only a JSON array of row labels. | ["SpatialSense (Yang et al., 2019)", "Rel3D (Goyal et al., 2020)", "SpatialSense+ (Wen et al., 2024)", "SpatialVLM (Chen et al., 2024)"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000431 | Consistency & Missingness | Column Missingness | According to the image table, list all missing entries in the right half column P (N/A or —). Output only a JSON array of row labels. | ["Random Choose"] |
| 2 | HB_real_000616 | Aggregation & Comparison | Extremum Reasoning | According to the table image, in the left half column Acc, what is the maximum value? Output only the number as a string. | "98.40" |
| 3 | HB_real_000915 | Aggregation & Comparison | Grouped Aggregation | Within the closed-source MLLMs 0-shot rows on the left half, what is the mean value of column Acc? Output only the number as a string, rounded to two decimal places. | "37.80" |
| 4 | HB_real_001497 | Aggregation & Comparison | Extremum Reasoning | For row GPT-4o (0-shot), what is the minimum numeric value across P, R, F1, and Acc on the left half? Output only the number as a string. | "40.19" |
| 5 | HB_real_001498 | Aggregation & Comparison | Pairwise Comparison | In the right half column Acc, compare GPT-4o with setting 2-shot and Gemini-1.5-flash with setting 3-shot. Which value is larger? Return JSON {"item":"<model and right-half setting>","value":"<cell value>"}. | {"item": "GPT-4o, 2-shot", "value": "38.40"} |
| 6 | HB_real_001499 | Consistency & Missingness | Cell Missingness | Is the cell at row Random Choose, right-half column P missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 7 | HB_real_001500 | Markup Grounding | Bold Presence | Is value 98.40 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 8 | HB_real_001501 | Consistency & Missingness | Consistency Probing | Which rows have missing entries marked with "-" in the right-half metric columns P, R, F1, or Acc? Output only a JSON array of row labels. | ["Random Choose"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001502 | Aggregation & Comparison | Argmax Reasoning | In column Q3, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Human (-)", "value": "100.00"} |
| 2 | HB_real_001503 | Aggregation & Comparison | Extremum Reasoning | In column Q2, what is the minimum value? Output only the number as a string. | "23.17" |
| 3 | HB_real_001504 | Aggregation & Comparison | Pairwise Comparison | For GPT-4o (0-shot), compare Ay and Az. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Ay", "value": "47.50"} |
| 4 | HB_real_001505 | Consistency & Missingness | Consistency Probing | In column Settings, which row labels have entries marked as "-"? Output only a JSON array of row labels. | ["Random Choose", "Human"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001506 | Aggregation & Comparison | Argmax Reasoning | Excluding the summary row Total, in column Test all, which data row has the maximum value? Return JSON {"row":<1-based data row index>,"value":"<cell value>"}. | {"row": 8, "value": "11730"} |
| 2 | HB_real_001507 | Aggregation & Comparison | Extremum Reasoning | Excluding the summary row Total, in column Train all, what is the maximum value? Output only the number as a string. | "39081" |
| 3 | HB_real_001508 | Aggregation & Comparison | Extremum Reasoning | Excluding the summary row Total, in column Test all, what is the minimum value? Output only the number as a string. | "219" |
| 4 | HB_real_001509 | Aggregation & Comparison | Pairwise Comparison | In column Test all, compare English (en) and Portuguese (pt). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "English (en)", "value": "11730"} |
| 5 | HB_real_001510 | Constrained Retrieval | Value Filtering | Excluding the summary row Total, in column Train all, list all languages with value >= 39000. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "English (en)", "value": "39026"}, {"item": "Spanish (es)", "value": "39081"}] |
| 6 | HB_real_001511 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001512 | Aggregation & Comparison | Best Selection | In metric METEOR↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Aya-101", "value": "0.195 (±0.23)"} |
| 2 | HB_real_001513 | Aggregation & Comparison | Best Selection | In metric MAUVE↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Gemini", "value": "0.025"} |
| 3 | HB_real_001514 | Aggregation & Comparison | Best Selection | In metric LangCheck↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "v5-Eagle-7B-HF", "value": "6.01%"} |
| 4 | HB_real_001515 | Aggregation & Comparison | Pairwise Comparison | In column LD↓, compare OPT-IML-Max-30b and Aya-101. Here ↓ means smaller is better. Which is better? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "OPT-IML-Max-30b", "value": "0.998 (±0.57)"} |
| 5 | HB_real_001516 | Markup Grounding | Bold Presence | Is value 0.025 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 6 | HB_real_001517 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000179 | Aggregation & Comparison | Grouped Aggregation | Based on the table, take the cells highlighted in light blue. What is their max? Exclude N/A and —. Output only the number as a string. | "0.7553" |
| 2 | HB_real_000184 | Aggregation & Comparison | Extremum Reasoning | In column AUC ROC, what is the maximum value? Output only the number as a string. | "0.9769" |
| 3 | HB_real_000447 | Aggregation & Comparison | Argmax Reasoning | In column AUC ROC, which detector has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Llama-3-8b-MultiSocial", "value": "0.9769"} |
| 4 | HB_real_000641 | Local Relations | Neighbor Indexing | At (r=8,c=1) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and the left Rank/Detector columns. | {"target": "0.7553", "up": "0.9544", "down": "0.7464", "left": null, "right": "0.3024"} |
| 5 | HB_real_000808 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=8,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| 6 | HB_real_000810 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light green in this table? Output only the count as a string. | "14" |
| 7 | HB_real_000996 | Constrained Retrieval | Value Filtering | Among light-orange highlighted cells, list the values that are > 0.7. Output only a JSON array of strings. | ["0.7464", "0.7454", "0.7418", "0.7418", "0.7248"] |
| 8 | HB_real_001004 | Markup Grounding | Marker Counting | Based on the image table, how many highlighted numeric data cells are in the table (light green, light blue, or light orange)? Output only the count as a string. | "34" |
| 9 | HB_real_001518 | Aggregation & Comparison | Extremum Reasoning | In column MacroF1 @5%FPR, what is the minimum value? Output only the number as a string. | "0.1376" |
| 10 | HB_real_001519 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.7553, list all numbers highlighted with the same light-blue color as the anchor. Output only a JSON array of strings. | ["0.7553", "0.3024", "0.7180", "0.3416", "0.4817", "0.1711", "0.4615", "0.1516", "0.3450", "0.1376"] |
| 11 | HB_real_002248 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light blue in this table? Output only the count as a string. | "10" |
| 12 | HB_real_002249 | Constrained Retrieval | Value Filtering | Among light-blue highlighted cells, list the values that are > 0.3. Output JSON array of strings. | ["0.7553", "0.3024", "0.7180", "0.3416", "0.4817", "0.4615", "0.3450"] |
| 13 | HB_real_001480 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.7464, list all numeric values highlighted with the same light-orange background as the anchor. Output only a JSON array of strings. | ["0.7464", "0.2523", "0.7454", "0.2497", "0.7418", "0.2465", "0.7418", "0.3605", "0.7248", "0.2815"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001521 | Aggregation & Comparison | Argmax Reasoning | In column all, which detector has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "BLOOMZ-3b-mixed-Detector", "value": "0.76"} |
| 2 | HB_real_001522 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the minimum value? Output only the number as a string. | "0.35" |
| 3 | HB_real_001523 | Aggregation & Comparison | Pairwise Comparison | For RoBERTa-large-OpenAI-Detector, compare cs and el. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "el", "value": "0.74"} |
| 4 | HB_real_000779 | Markup Grounding | Marked Value Listing | List all numbers highlighted in the darkest blue shade. If the same value appears more than once, list it only once. Output only a JSON array of strings. | ["0.93", "0.94", "0.92"] |
| 5 | HB_real_000813 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in the darkest blue shade. What is their maximum numeric value? Output only the number as a string. | "0.94" |
| 6 | HB_real_000930 | Aggregation & Comparison | Argmax Reasoning | In data row 7, find the largest numeric cell. Return JSON {"col":"<column label>","value":"<cell value>"}. | {"col": "hu", "value": "0.94"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001524 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the minimum value? Output only the number as a string. | "0.66" |
| 2 | HB_real_001525 | Aggregation & Comparison | Pairwise Comparison | For P/Telegram, compare en and zh. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "en", "value": "0.86"} |
| 3 | HB_real_001526 | Constrained Retrieval | Value Filtering | In column all, list all rows with value >= 0.80. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "P / Discord", "value": "0.82"}, {"item": "S / Discord", "value": "0.88"}] |
| 4 | HB_real_002261 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in the darkest blue shade. What is their maximum numeric value? Output only the number as a string. | "0.92" |
| 5 | HB_real_000935 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 1, "col": "ca", "value": "0.92"} |
| 6 | HB_real_000817 | Local Relations | Neighbor Indexing | At position (r=1,c=3) in the numeric part of the table, return the target and its four neighbors as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. Use null if a neighbor does not exist. | {"target": "0.92", "up": null, "down": "N/A", "left": "N/A", "right": "0.81"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001527 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the maximum value? Output only the number as a string. | "0.98" |
| 2 | HB_real_001528 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the minimum value? Output only the number as a string. | "0.95" |
| 3 | HB_real_001529 | Aggregation & Comparison | Pairwise Comparison | For BLOOMZ-3b-MultiSocial, compare ga★ and gd★. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "ga★", "value": "0.90"} |
| 4 | HB_real_002263 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in the darkest blue shade. What is their maximum numeric value? Output only the number as a string. | "1.00" |
| 5 | HB_real_000936 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 5, "col": "hu", "value": "1.00"} |
| 6 | HB_real_000818 | Local Relations | Neighbor Indexing | At position (r=5,c=13) in the numeric part of the table, return the target and its four neighbors as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. Use null if a neighbor does not exist. | {"target": "1.00", "up": "0.99", "down": "0.99", "left": "0.99", "right": "0.97"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000914 | Consistency & Missingness | Column Missingness | Based on the table, list all missing entries in column ga (N/A or —). Output only a JSON array of row labels. | ["en", "es", "ru", "en-es-ru"] |
| 2 | HB_real_001530 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the minimum value? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Output only the number as a string. | "0.84" |
| 3 | HB_real_001531 | Aggregation & Comparison | Argmax Reasoning | In column en, which train-language row has the maximum value? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "en", "value": "0.96"} |
| 4 | HB_real_001532 | Aggregation & Comparison | Extremum Reasoning | For train-language row en-es-ru, what is the minimum non-missing value across all test-language columns? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Output only the number as a string. | "0.86" |
| 5 | HB_real_001533 | Aggregation & Comparison | Pairwise Comparison | For train-language row en-es-ru, compare ar and bg. Which value is larger? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Return JSON {"item":"...","value":"..."}. | {"item": "bg", "value": "0.94"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001534 | Aggregation & Comparison | Extremum Reasoning | In column all, what is the minimum value? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Output only the number as a string. | "0.88" |
| 2 | HB_real_001535 | Aggregation & Comparison | Argmax Reasoning | In column Twitter, which train-platform row has the maximum value? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Twitter", "value": "0.98"} |
| 3 | HB_real_001536 | Aggregation & Comparison | Pairwise Comparison | For row Gab, compare Twitter and WhatsApp. Which value is larger? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Return JSON {"item":"...","value":"..."}. | {"item": "Twitter", "value": "0.94"} |
| 4 | HB_real_000937 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 1, "col": "Discord", "value": "0.98"} |
| 5 | HB_real_000672 | Markup Grounding | Bold Presence | Is the value at position (r=4,c=4) in the numeric part of the table boldfaced? Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Answer JSON {"bold":true/false}. | {"bold": true} |
| 6 | HB_real_000819 | Local Relations | Neighbor Indexing | At position (r=4,c=4) in the numeric part of the table, return the target and its four neighbors as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. Ignore delta/change annotations written with +/− or ±, such as "+1.15 ± 0.63". Use null if a neighbor does not exist. | {"target": "0.98", "up": "0.95", "down": "0.92", "left": "0.92", "right": "0.92"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001553 | Aggregation & Comparison | Best Selection | In metric FM(↓) in block 14−1 with 1 Step, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "CDAD†", "value": "-0.57 / -0.04"} |
| 2 | HB_real_001554 | Aggregation & Comparison | Pairwise Comparison | For CDAD† in block 3 × 5 with 5 Steps, compare A-AUROC(↑) and FM(↓). Which metric value is larger as raw number? Return JSON {"item":<metric_name>,"value":<string>}. | {"item": "A-AUROC(↑)", "value": "89.1 / 92.2"} |
| 3 | HB_real_001745 | Markup Grounding | Marked Column Listing | In block 14−1 with 1 Step, in column A-AUROC(↑), list the underlined numbers. Output only a JSON array of strings. | ["96.0", "96.5"] |
| 4 | HB_real_000879 | Markup Grounding | Underline Presence | At position (r=6,c=3) in the numeric part of the table, is the number underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left Method column. | {"underlined": false} |
| 5 | HB_real_000360 | Markup Grounding | Underline Presence | The value 2.9 appears exactly once in the numeric part of the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 6 | HB_real_000676 | Markup Grounding | Bold Presence | At position (r=11,c=4) in the numeric part of the table, is the cell boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left Method column. | {"bold": true} |
| 7 | HB_real_000679 | Markup Grounding | Bold Presence | At position (r=2,c=2) in the numeric part of the table, is the cell boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left Method column. | {"bold": false} |
| 8 | HB_real_000463 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=11,c=1) in the numeric part of the table, list all numeric cells highlighted with the same light-gray color as the anchor. Output only a JSON array of strings. The numeric part of the table excludes the left Method column. | ["96.4 / 96.6", "-0.57 / -0.04", "94.2 / 95.3", "2.05 / 2.40", "89.1 / 92.2", "3.69 / 4.65", "94.9 / 95.7", "1.01 / 0.74"] |
| 9 | HB_real_000820 | Local Relations | Neighbor Indexing | At position (r=6,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part excludes the left Method column. | {"target": "1.8 / 0.3", "up": "29.1 / 20.6", "down": "1.0 / 0.6", "left": "93.8 / 95.7", "right": "88.7 / 93.1"} |
| 10 | HB_real_000706 | Constrained Retrieval | Cell Retrieval | In the numeric part of the table, what is the cell content at position (r=9,c=4)? The numeric part excludes the left Method column. Output only the string shown in the cell. | "2.9 / 4.0" |
| 11 | HB_real_000956 | Aggregation & Comparison | Grouped Aggregation | Exclude the light-gray highlighted row. Among the remaining cells in column FM(↓) under block 10−1 × 5 with 5 Steps, what is the minimum first number? Output only the number as a string. | "3.2" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001556 | Aggregation & Comparison | Best Selection | In metric FM(↓) in block 11−1 with 1 Step, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "CDAD†", "value": "-1.28 / -0.06"} |
| 2 | HB_real_001557 | Aggregation & Comparison | Argmax Reasoning | In column A-AUROC(↑) (block 8−4 with 1 Step), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CDAD†", "value": "85.3 / 97.1"} |
| 3 | HB_real_001778 | Markup Grounding | Marked Column Listing | In block 8−1 × 4 with 4 Steps, in column FM(↓), list the underlined numbers. Output only a JSON array of strings. | ["5.4", "9.8"] |
| 4 | HB_real_000880 | Markup Grounding | Underline Presence | At position (r=7,c=5) in the numeric part of the table, is the number underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left Method column. | {"underlined": true} |
| 5 | HB_real_000542 | Markup Grounding | Underline Presence | The value 11.4 appears exactly once in the numeric part of the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 6 | HB_real_000680 | Markup Grounding | Bold Presence | At position (r=11,c=2) in the numeric part of the table, is the cell boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left Method column. | {"bold": true} |
| 7 | HB_real_000980 | Markup Grounding | Bold Presence | At position (r=3,c=3) in the numeric part of the table, is the cell boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left Method column. | {"bold": false} |
| 8 | HB_real_000581 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=11,c=1) in the numeric part of the table, list all numeric cells highlighted with the same light-gray color as the anchor. Output only a JSON array of strings. The numeric part excludes the left Method column. | ["88.3 / 97.2", "-1.28 / -0.06", "85.3 / 97.1", "1.39 / 0.29", "83.4 / 95.8", "3.79 / 1.49"] |
| 9 | HB_real_000821 | Local Relations | Neighbor Indexing | At position (r=6,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part excludes the left Method column. | {"target": "2.7 / 0.6", "up": "17.3 / 10.9", "down": "2.4 / 1.8", "left": "85.9 / 94.5", "right": "79.9 / 94.2"} |
| 10 | HB_real_000707 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=9,c=4) in the numeric part of the table. The numeric part excludes the left Method column. Output only the string shown in the cell. | "7.7 / 1.6" |
| 11 | HB_real_001630 | Aggregation & Comparison | Grouped Aggregation | Exclude the light-gray highlighted row. Among the remaining cells in column FM(↓) under block 8−4 with 1 Step, what is the minimum first number? Output only the number as a string. | "5.8" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001599 | Aggregation & Comparison | Best Selection | In metric FM, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "CDAD", "value": "1.01 / 0.74"} |
| 2 | HB_real_001600 | Constrained Retrieval | Cell Retrieval | What is the value at column 12 for row w/o CDM? Output the full string as shown in the cell. | "93.1 / 96.4" |
| 3 | HB_real_001601 | Consistency & Missingness | Cell Missingness | Is the cell (row ControlNet, column 14) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_real_001602 | Markup Grounding | Bold Presence | Is the number with value 95.8 / 97.3 boldfaced? Answer JSON {"bold":true/false}. The value appears exactly once in the table. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000554 | Aggregation & Comparison | Pairwise Comparison | From this table image, in column Win (block PKU-SafeRLHF), compare Standard RM and Standard RM w/ KL within "InfoRM w/ IBL". Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Standard RM", "value": "67.4"} |
| 2 | HB_real_000613 | Aggregation & Comparison | Extremum Reasoning | Based on the table image, in column Win (block Anthropic-Helpful), what is the maximum value? Output only the number as a string. | "66.3" |
| 3 | HB_real_001537 | Aggregation & Comparison | Grouped Aggregation | Within the bottom dashed InfoRM rows, what is the mean value of column Win in block Anthropic-Helpful? Output only the number as a string, rounded to one decimal place. | "42.6" |
| 4 | HB_real_001538 | Aggregation & Comparison | Extremum Reasoning | In column Win (block Anthropic-Helpful), what is the minimum value? Output only the number as a string. | "28.2" |
| 5 | HB_real_001539 | Aggregation & Comparison | Extremum Reasoning | In row InfoRM w/ KL in the bottom dashed section, what is the minimum numeric value across all Win/Tie/Lose cells? Output only the number as a string. | "13.0" |
| 6 | HB_real_001540 | Aggregation & Comparison | Pairwise Comparison | For row InfoRM (bottom dashed section) in block AlpacaFarm, compare Tie and Lose. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Tie", "value": "43.8"} |
| 7 | HB_real_001541 | Local Relations | Marker-Anchor Neighbor | For the blue Win cell with value 66.3 in block Anthropic-Helpful, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "6.4", "right": "24.1"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001542 | Aggregation & Comparison | Extremum Reasoning | In column Win (block Anthropic-Helpful), what is the minimum value? Output only the number as a string. | "31.5" |
| 2 | HB_real_001543 | Aggregation & Comparison | Pairwise Comparison | For row InfoRM w/ KL (bottom dashed section) in block PKU-SafeRLHF, compare Tie and Lose. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Tie", "value": "35.1"} |
| 3 | HB_real_001544 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 4 | HB_real_001697 | Constrained Retrieval | Color-Anchor Retrieval | In the same row as the number 75.5, list all numeric values highlighted with the same light-blue background as 75.5. Output only a JSON array of strings. | ["75.5", "68.6", "64.4", "74.5"] |
| 5 | HB_real_001972 | Constrained Retrieval | Color-Anchor Retrieval | In the same row as the number 15.9, list all numeric values highlighted with the same light-yellow background as 15.9. Output only a JSON array of strings. | ["15.9", "21.9", "23.6", "17.6"] |
| 6 | HB_real_000780 | Markup Grounding | Marked Value Listing | In the InFoRM block, list all numbers highlighted in light pink in the Standard RM row. Output only a JSON array of strings. | ["8.6", "9.5", "12.0", "7.9"] |
| 7 | HB_real_000814 | Aggregation & Comparison | Grouped Aggregation | In the InFoRM block, take the cells highlighted in light yellow in the Standard RM row. What is their maximum numeric value? Output only the number as a string. | "23.6" |
| 8 | HB_real_000469 | Local Relations | Neighbor Indexing | In the InFoRM block, for the Standard RM row and Anthropic-Helpful Tie cell, return the target and its four neighboring cells as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. Use null if a neighbor does not exist. | {"target": "15.9", "up": null, "down": "20.4", "left": "75.5", "right": "8.6"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001545 | Aggregation & Comparison | Extremum Reasoning | In column Win (block Anthropic-Helpful), what is the maximum value? Output only the number as a string. | "73.9" |
| 2 | HB_real_001546 | Aggregation & Comparison | Extremum Reasoning | In column Win (block Anthropic-Harmless), what is the minimum value? Output only the number as a string. | "29.7" |
| 3 | HB_real_001547 | Aggregation & Comparison | Pairwise Comparison | For row Ensemble RM (Mean) within top InfoRM section (block AlpacaFarm), compare Tie and Lose. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Tie", "value": "42.5"} |
| 4 | HB_real_001548 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001549 | Aggregation & Comparison | Extremum Reasoning | In column Win (block Anthropic-Helpful), what is the minimum value? Output only the number as a string. | "25.8" |
| 2 | HB_real_001550 | Aggregation & Comparison | Pairwise Comparison | For row InfoRM w/ KL (bottom dashed section) in block AlpacaFarm, compare Tie and Lose. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Tie", "value": "38.6"} |
| 3 | HB_real_001551 | Constrained Retrieval | Value Filtering | In column Tie (block PKU-SafeRLHF), list all rows with value >= 45. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "InfoRM / Standard RM w/ KL", "value": "48.9"}] |
| 4 | HB_real_001552 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001564 | Aggregation & Comparison | Argmax Reasoning | In row # Sample, which benchmark has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "General-Bench", "value": "325.8K"} |
| 2 | HB_real_001565 | Aggregation & Comparison | Pairwise Comparison | For General-Bench, compare # Task and # Skill. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "# Task", "value": "702"} |
| 3 | HB_real_001566 | Aggregation & Comparison | Pairwise Comparison | In row # Tested Models, compare MMT-Bench and General-Bench. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "General-Bench", "value": "172+102"} |
| 4 | HB_real_001567 | Markup Grounding | Highlight Presence | Is value 29 highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 5 | HB_real_002269 | Markup Grounding | Marked Value Listing | In the MMIU column, list all entries highlighted with a light-blue background. Output only a JSON array of strings. | ["Txt,Img,Vid,Point-Cloud,Depth", "1", "52", "Acc.", "22"] |
| 6 | HB_real_002272 | Constrained Retrieval | Color-Anchor Retrieval | In the same row as 29, list all entries highlighted with the same light-blue background as 29. Output only a JSON array of strings. | ["1", "1", "6", "1", "1", "4", "5", "29"] |
| 7 | HB_real_002274 | Constrained Retrieval | Color-Anchor Retrieval | In the # Task row, list the numbers in light-blue highlighted cells. Output only a JSON array of strings. | ["12", "20", "30", "47", "52", "162", "505", "702"] |
| 8 | HB_real_002275 | Constrained Retrieval | Color-Anchor Retrieval | In the # Domain row, list the numbers in light-blue highlighted cells. Output only a JSON array of strings. | ["1", "1", "6", "1", "1", "4", "5", "29"] |
| 9 | HB_real_002276 | Constrained Retrieval | Color-Anchor Retrieval | In the # Tested Models row, list the numbers in light-blue highlighted cells. Output only a JSON array of strings. | ["12", "21", "24", "8", "22", "30", "22", "172+102"] |
| 10 | HB_real_002277 | Markup Grounding | Marked Value Listing | In the General-Bench column, list all bold numeric entries. Output only a JSON array of strings. | ["29", "145", "702", "325.8K", "172+102"] |
| 11 | HB_real_002278 | Markup Grounding | Marked Value Listing | In the # Tested Models row, list all bold entries. Output only a JSON array of strings. | ["172+102"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001559 | Aggregation & Comparison | Argmax Reasoning | In column #12↑, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Qwen2-VL-72B", "value": "86.90"} |
| 2 | HB_real_001560 | Aggregation & Comparison | Argmax Reasoning | In column #13↑, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "InternVL-2.5-26B", "value": "22.50"} |
| 3 | HB_real_001561 | Aggregation & Comparison | Argmax Reasoning | In column #16↑, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LLaVA-One-Vision-72B", "value": "83.70"} |
| 4 | HB_real_001562 | Aggregation & Comparison | Argmax Reasoning | In column #17↑, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "InternVL-2.5-26B", "value": "20.60"} |
| 5 | HB_real_001563 | Aggregation & Comparison | Pairwise Comparison | For Qwen2-VL-72B, compare #14↑ and #16↑. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "#16↑", "value": "70.20"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001568 | Aggregation & Comparison | Argmax Reasoning | In column #Supported Tasks, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Vitron-V1", "value": "252 (92.6%)"} |
| 2 | HB_real_001569 | Aggregation & Comparison | Extremum Reasoning | In column Level-2, what is the minimum value? Output only the number as a string. | "12.21" |
| 3 | HB_real_001570 | Aggregation & Comparison | Pairwise Comparison | For Vitron-V1, compare Level-2 and Level-3. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Level-2", "value": "30.13"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000551 | Aggregation & Comparison | Extremum Reasoning | Based on the table image, in column Level-4 (Level Score on Image), what is the maximum value? Output only the number as a string. | "4.59" |
| 2 | HB_real_001571 | Aggregation & Comparison | Argmax Reasoning | In column #Supported Task, which model is best? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Vitron-V1", "value": "42 (93.3%)"} |
| 3 | HB_real_001572 | Aggregation & Comparison | Argmax Reasoning | In column Level-4 (Level Score on Image), which numeric cell has the maximum value? Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 6, "col": "Level-4", "value": "4.59"} |
| 4 | HB_real_001573 | Aggregation & Comparison | Pairwise Comparison | For Vitron-V1, compare Level-2 and Level-3. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Level-2", "value": "30.13"} |
| 5 | HB_real_001574 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "/", "-", or "N/A"? If none, output an empty JSON array []. | ["SoTA Specialist"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001575 | Aggregation & Comparison | Best Selection | Within block Video Comprehension Skill, which model is best on #Win-over-Specialist? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Sa2VA-8B", "value": "32 (25.4%)"} |
| 2 | HB_real_001576 | Aggregation & Comparison | Best Selection | Within block Video Generation Skill, which model is best on Level-2? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "VidAgent", "value": "25.00"} |
| 3 | HB_real_001577 | Aggregation & Comparison | Argmax Reasoning | In block Video Comprehension Skill, in column #Supported Task, which model is best? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Sa2VA-8B", "value": "91 (72.2%)"} |
| 4 | HB_real_001578 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001579 | Aggregation & Comparison | Argmax Reasoning | In the lower block Audio Generation Skill, column #Win-Spclst, which model is best? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "ModaVerse", "value": "2 (10.0%)"} |
| 2 | HB_real_001580 | Aggregation & Comparison | Pairwise Comparison | In the lower block Audio Generation Skill, compare Any-GPT and Unified-io-2-XXL on Level-2. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Any-GPT", "value": "29.06"} |
| 3 | HB_real_001581 | Aggregation & Comparison | Extremum Reasoning | In the lower block Audio Generation Skill, column Level-2, what is the minimum value? Output only the number as a string. | "7.22" |
| 4 | HB_real_001582 | Aggregation & Comparison | Pairwise Comparison | For Qwen-Audio-Chat (upper block), compare #1 and #2. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "#2", "value": "68.77"} |
| 5 | HB_real_001583 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | ["SoTA Specialist"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001584 | Aggregation & Comparison | Argmax Reasoning | In the upper block 3D Comprehension Skill, in column #Task-Supprt, which model is best? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "PointLLM-13B", "value": "9 (30.0%)"} |
| 2 | HB_real_001585 | Aggregation & Comparison | Pairwise Comparison | For PointLLM-13B (upper block), compare #1 and #3. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "#3", "value": "78.14"} |
| 3 | HB_real_001586 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 4 | HB_real_000938 | Aggregation & Comparison | Argmax Reasoning | In the upper block 3D Comprehension Skill, considering only columns #1 through #13, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}, with row and column counted within that block. | {"row": 1, "col": 2, "value": "98.35"} |
| 5 | HB_real_000709 | Constrained Retrieval | Cell Retrieval | In the upper block 3D Comprehension Skill, find the value at position (r=4,c=3) in the numeric part of that block. Output only the number as a string. | "78.14" |
| 6 | HB_real_000823 | Local Relations | Neighbor Indexing | In the upper block 3D Comprehension Skill, at position (r=3,c=2) in the numeric part of that block, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "7.50", "up": "0.00", "down": "10.00", "left": "46.16", "right": "72.86"} |
| 7 | HB_real_001082 | Local Relations | Neighbor Value Tracing | The value 46.37 appears exactly once in the upper block numeric part. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "0.00", "right": "0.00"} |
| 8 | HB_real_000661 | Aggregation & Comparison | Grouped Aggregation | In the lower block 3D Generation Skill, for non-SoTA model rows, what is the mean of the Level-4 column? Output the value rounded to two decimal places as a string. | "0.00" |
| 9 | HB_real_001777 | Consistency & Missingness | Column Missingness | In the upper block, column #Task-Supprt, list the row labels whose entry is shown as /. Output only a JSON array of strings. | ["SoTA Specialist"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000553 | Aggregation & Comparison | Pairwise Comparison | Based on the table image, in column #Supported Task, compare InternVL2.5-8B and Meta-Llama-3.1-8B-Instruct. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "InternVL2.5-8B", "value": "114 (99.1%)"} |
| 2 | HB_real_001587 | Aggregation & Comparison | Extremum Reasoning | In column #Supported Task, what is the maximum value? Output only the value string as shown. | "114 (99.1%)" |
| 3 | HB_real_001588 | Aggregation & Comparison | Extremum Reasoning | In column #Supported Task, what is the minimum value? Output only the value string as shown. | "72 (62.6%)" |
| 4 | HB_real_001589 | Aggregation & Comparison | Pairwise Comparison | In column #Supported Task, compare DeepSeek-VL-7B and Qwen2-VL-7B. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DeepSeek-VL-7B", "value": "114 (99.1%)"} |
| 5 | HB_real_001590 | Aggregation & Comparison | Extremum Reasoning | For row InternVL2.5-8B, what is the minimum value across #Supported Task, #Win-over-Specialist, and Level-5 Score? Output only the number as a string. | "0.00" |
| 6 | HB_real_001591 | Aggregation & Comparison | Pairwise Comparison | For DeepSeek-VL-7B, compare #Supported Task and #Win-over-Specialist. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "#Supported Task", "value": "114 (99.1%)"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001592 | Aggregation & Comparison | Argmax Reasoning | In column of Overall, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Unified-io-2-XXL", "value": "13.70"} |
| 2 | HB_real_001594 | Aggregation & Comparison | Pairwise Comparison | For Unified-io-2-XXL, compare of Audio and of Video. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "of Audio", "value": "25.63"} |
| 3 | HB_real_000939 | Aggregation & Comparison | Argmax Reasoning | Considering the five Level 2 Score columns only, find the largest numeric cell in the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and columns left-to-right within the five score columns. | {"row": 6, "col": 1, "value": "30.13"} |
| 4 | HB_real_001332 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=8,c=5) within the five Level 2 Score columns. Output only the number as a string. | "10.80" |
| 5 | HB_real_000957 | Local Relations | Neighbor Indexing | At position (r=4,c=2) within the five Level 2 Score columns, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "12.54", "up": "8.34", "down": "7.32", "left": "19.54", "right": "17.52"} |
| 6 | HB_real_001834 | Local Relations | Neighbor Value Tracing | The value 25.63 appears exactly once in the five Level 2 Score columns. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "8.56", "right": "0.00"} |
| 7 | HB_real_001766 | Aggregation & Comparison | Grouped Aggregation | Across all rows, what is the mean of the "of 3D" column? Output the value rounded to two decimal places as a string. | "0.00" |
| 8 | HB_real_000582 | Markup Grounding | Marked Value Listing | List the entries shown in green text in the Paradigm column. Output only a JSON array of strings. | ["C+G", "C+G", "C+G", "C", "C+G", "C+G", "C", "C+G"] |
| 9 | HB_real_001945 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 10 | HB_real_001984 | Markup Grounding | Underline Presence | At position (r=1,c=1) within the five Level 2 Score columns, is the value underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 11 | HB_real_001830 | Markup Grounding | Marker Counting | How many underlined numeric values are in the table? Output only the count as a string. | "0" |
| 12 | HB_real_001836 | Aggregation & Comparison | Rank Selection | In the Ranking column, which model is ranked 3rd? Return JSON {"item":"<model>","value":"<ranking>"}. | {"item": "NExT-GPT-V1.5", "value": "3"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001595 | Aggregation & Comparison | Argmax Reasoning | In column #1↑, which model has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Long-LLaVA-9B", "value": "41.30"} |
| 2 | HB_real_001597 | Aggregation & Comparison | Pairwise Comparison | For Qwen2-VL-72B, compare #5↑ and #6↑. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "#6↑", "value": "27.20"} |
| 3 | HB_real_001598 | Markup Grounding | Bold Presence | Is value 40.00 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 4 | HB_real_rebal_000057 | Markup Grounding | Marked Value Listing | In the InternVL-2-26B row, list all bold numeric values. Output only a JSON array of strings. | ["40.00", "14.10", "31.90", "34.50", "23.50"] |
| 5 | HB_real_rebal_000058 | Markup Grounding | Marked Column Listing | In column #2↑, list all bold values. Output only a JSON array of strings. | ["14.10", "10.90", "14.50", "10.80"] |
| 6 | HB_real_rebal_000059 | Markup Grounding | Marker Counting | How many bold numeric values are in the InternVL-2-26B row? Output only the count as a string. | "5" |
| 7 | HB_real_rebal_000061 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 8 | HB_real_rebal_000062 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Model column. | {"underlined": false} |
| 9 | HB_real_rebal_000064 | Constrained Retrieval | Color-Anchor Retrieval | In the InternVL-2-26B row, list all numeric values highlighted with the same light-blue row background. Output only a JSON array of strings. | ["40.00", "14.10", "31.90", "34.50", "20.30", "16.10", "22.60", "18.60", "23.50"] |
| 10 | HB_real_rebal_000065 | Constrained Retrieval | Color-Anchor Retrieval | For the value 41.30, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["41.30", "14.50", "33.20", "39.10", "16.50", "13.60", "19.60", "18.60", "18.60"] |
| 11 | HB_real_rebal_000066 | Local Relations | Neighbor Indexing | At position (r=3,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part excludes the left Model column. | {"target": "34.50", "up": "12.40", "down": "19.50", "left": "31.90", "right": "20.30"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001603 | Aggregation & Comparison | Extremum Reasoning | In row % Process errors (correct final answers), what is the minimum value across benchmarks? Output only the percentage string. | "3.5%" |
| 2 | HB_real_001605 | Aggregation & Comparison | Pairwise Comparison | For GSM8K under correct, compare % 3/4 agreement and % 3/5 agreement. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "% 3/4 agreement", "value": "3.6%"} |
| 3 | HB_real_000940 | Aggregation & Comparison | Argmax Reasoning | In the numeric part of the table, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and numeric columns left-to-right. | {"row": 1, "col": 7, "value": "759"} |
| 4 | HB_real_001667 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=3,c=5) in the numeric part of the table. Output only the number as a string. | "8.9" |
| 5 | HB_real_000959 | Local Relations | Neighbor Indexing | At position (r=6,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "2.0%", "up": "8.9%", "down": "91.9%", "left": "3.4%", "right": "9.1%"} |
| 6 | HB_real_001930 | Local Relations | Neighbor Value Tracing | The value 95.9% appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "66.7%", "right": "59.4%"} |
| 7 | HB_real_001872 | Aggregation & Comparison | Grouped Aggregation | For the # Steps row, what is the mean of the four correct columns? Output the value rounded to one decimal place as a string. | "6.8" |
| 8 | HB_real_001990 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 9 | HB_real_001995 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 10 | HB_real_001842 | Markup Grounding | Marker Counting | How many underlined numeric values are in the table? Output only the count as a string. | "0" |
| 11 | HB_real_001870 | Aggregation & Comparison | Rank Selection | In the #Samples row, which benchmark/outcome column has the third-largest value? Return JSON {"item":"<column>","value":"<cell value>"}. | {"item": "MATH error", "value": "594"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001606 | Aggregation & Comparison | Best Selection | Within block Open-source Process Reward Models (PRMs), which model is best on Average? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Qwen2.5-Math-7B-PRM800K (our trained)", "value": "56.5"} |
| 2 | HB_real_001607 | Aggregation & Comparison | Best Selection | Within block Open-source language models, prompted as Critic Models, which model is best on Average? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "★ QwQ-32B-Preview", "value": "71.5"} |
| 3 | HB_real_001608 | Aggregation & Comparison | Best Selection | Within block Proprietary language models, prompted as Critic Models, which model is best on Average? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "o1-mini", "value": "87.9"} |
| 4 | HB_real_001609 | Aggregation & Comparison | Pairwise Comparison | For Qwen2.5-Math-7B-PRM800K (our trained), compare GSM8K and MATH. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "GSM8K", "value": "68.2"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000211 | Aggregation & Comparison | Extremum Reasoning | Read the table carefully and for item Proactive Base, what is the minimum value across all numeric metrics in this row? Output only the number as a string. | "1.51" |
| 2 | HB_real_000490 | Aggregation & Comparison | Extremum Reasoning | For item Proactive Base, what is the maximum value across all numeric metrics in this row? Output only the number as a string. | "62.0" |
| 3 | HB_real_000839 | Aggregation & Comparison | Grouped Aggregation | Within the green Base/Proactive Base group, what is the mean value of column ACC↑ (block MATH-Chat)? Output only the number as a string, rounded to two decimal places. | "11.75" |
| 4 | HB_real_001610 | Aggregation & Comparison | Best Selection | In metric ACC↑ (block MATH-Chat), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Online DPO", "value": "16.5"} |
| 5 | HB_real_001611 | Aggregation & Comparison | Best Selection | In metric #Tokens(k)↓ (block BigCodeBench-Chat), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Online DPO", "value": "1.31"} |
| 6 | HB_real_001612 | Aggregation & Comparison | Best Selection | Within group "MediumDocEdit-Chat", which row label in the leftmost row-label column is best on column ITR↑? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Online DPO", "value": "92.0"} |
| 7 | HB_real_001613 | Constrained Retrieval | Value Filtering | In column ACC↑ (block MATH-Chat), list all row labels in the primary row-label column with value >= 15. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Offline DPO", "value": "15.5"}, {"item": "Online DPO", "value": "16.5"}] |
| 8 | HB_real_001614 | Aggregation & Comparison | Grouped Aggregation | Within the orange SFT/PPO/Offline DPO/Online DPO group, what is the mean value of column ACC↑ (block MATH-Chat)? Output only the number as a string, rounded to three decimal places. | "14.625" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000160 | Markup Grounding | Marker Counting | Based on the table, how many bold numbers are in the table? Output only the count as a string. | "0" |
| 2 | HB_real_000161 | Constrained Retrieval | Cell Retrieval | What is the value at row MTPO, column Causal & Objective-Aligned Reward? Output only the cell text. | "✗" |
| 3 | HB_real_000163 | Constrained Retrieval | Cell Retrieval | What is the value at row STaR-GATE, column Causal & Objective-Aligned Reward? Output only the cell text. | "-" |
| 4 | HB_real_000347 | Consistency & Missingness | Column Missingness | Which columns contain at least one missing entry marked as "-"? Output only a JSON array of column names. | ["Causal & Objective-Aligned Reward"] |
| 5 | HB_real_000349 | Constrained Retrieval | Value Filtering | List all row labels whose entry in column User-Centric is a check mark. Output only a JSON array of strings. | ["COLLABLLM"] |
| 6 | HB_real_000350 | Markup Grounding | Underline Presence | Is the mark at row COLLABLLM, column User-Centric underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 7 | HB_real_000621 | Markup Grounding | Bold Presence | Is the mark at row COLLABLLM, column User-Centric boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 8 | HB_real_001615 | Constrained Retrieval | Cell Retrieval | What is the value at row COLLABLLM, column User-Centric? Output only the cell text. | "✔" |
| 9 | HB_real_001616 | Consistency & Missingness | Column Missingness | Which row labels have missing entries marked as "-" in column Causal & Objective-Aligned Reward? Output only a JSON array of strings. | ["ClarifyGPT (Mu et al., 2023)", "STaR-GATE (Andukuri et al., 2024)"] |
| 10 | HB_real_001618 | Markup Grounding | Marker Counting | How many underlined marks or numbers are in the table? Output only the count as a string. | "0" |
| 11 | HB_real_001619 | Consistency & Missingness | Consistency Probing | List all cells marked as missing using "-" or "N/A". Return a JSON array of strings in the form "<row label> | <column>". If none, output []. | ["ClarifyGPT (Mu et al., 2023) | Causal & Objective-Aligned Reward", "STaR-GATE (Andukuri et al., 2024) | Causal & Objective-Aligned Reward"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000071 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column # Eval. Func.? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Spider2-V (Cao et al., 2024)", "value": "151"} |
| 2 | HB_real_000078 | Aggregation & Comparison | Argmax Reasoning | In column # Eval. Func., which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "OmniBench (Ours)", "value": "255"} |
| 3 | HB_real_000417 | Aggregation & Comparison | Rank Selection | Who ranks 3rd in column # Eval. Func.? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "OSWorld (Xie et al., 2024)", "value": "134"} |
| 4 | HB_real_000420 | Aggregation & Comparison | Extremum Reasoning | What is the minimum numeric value in column # Eval. Func.? Exclude missing entries. Output only the number as a string. | "5" |
| 5 | HB_real_000793 | Aggregation & Comparison | Extremum Reasoning | For row OmniBench (Ours), what is the maximum numeric value across the visible numeric task/evaluation columns? Output only the number as a string. | "36076" |
| 6 | HB_real_000949 | Aggregation & Comparison | Pairwise Comparison | For row OmniBench (Ours), compare column # Scenario and column # Eval. Func. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "# Eval. Func.", "value": "255"} |
| 7 | HB_real_001620 | Aggregation & Comparison | Argmax Reasoning | In column # Instance, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "OmniBench (Ours)", "value": "36076"} |
| 8 | HB_real_001622 | Aggregation & Comparison | Rank Selection | Who ranks 4th in column # Eval. Func.? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "CRAB (Xu et al., 2024a)", "value": "59"} |
| 9 | HB_real_001623 | Aggregation & Comparison | Pairwise Comparison | In column # Scenario, compare OmniBench (Ours) and Spider2-V (Cao et al., 2024). Which value is larger? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "OmniBench (Ours)", "value": "20"} |
| 10 | HB_real_001624 | Consistency & Missingness | Column Missingness | In column # Compl. Dimen., list the row labels whose entry is missing. Output only a JSON array of strings. | ["AitW (Rawles et al., 2024)", "Mind2Web (Deng et al., 2024)", "MoTIF (Burns et al., 2022)", "OmniACT (Kapoor et al., 2025)", "GUI Odyssey (Lu et al., 2024)", "WebArena (Zhou et al., 2023)", "CRAB (Xu et al., 2024a)"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000012 | Aggregation & Comparison | Grouped Aggregation | Read the table carefully and exclude cells that are boldfaced. On the remaining cells in column LC, what is the max? Exclude N/A and —. Output only the number as a string. | "38.0" |
| 2 | HB_real_000015 | Markup Grounding | Marked Column Listing | In column LC, return the values that are bold. Output only a JSON array of strings. | ["49.0", "37.4"] |
| 3 | HB_real_000022 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column LC. Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and the left Method column. | {"row": 8, "col": 2, "value": "49.0"} |
| 4 | HB_real_000186 | Aggregation & Comparison | Grouped Aggregation | Using the table only, take the cells that are boldfaced. What is their max? Exclude N/A and —. Output only the number as a string. | "49.4" |
| 5 | HB_real_000435 | Markup Grounding | Marked Column Listing | In column LC, list all underlined numeric values. Output only a JSON array of strings. | ["38.0", "30.1"] |
| 6 | HB_real_000439 | Aggregation & Comparison | Pairwise Comparison | In column LC, compare Gemini-2.0-Flash and Claude-3.5-Sonnet. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Gemini-2.0-Flash", "value": "38.0"} |
| 7 | HB_real_000444 | Constrained Retrieval | Cell Retrieval | What is the LC value for row Omni-UGround-V1-7B (Ours)? Output only the number as a string. | "49.0" |
| 8 | HB_real_001625 | Markup Grounding | Underline Presence | Is the number with value 38.0 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 9 | HB_real_001626 | Markup Grounding | Bold Presence | Is the number with value 28.7 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 10 | HB_real_001627 | Aggregation & Comparison | Best Selection | Within group "Supervised Fine-Tuning Agents (Screenshot)", which row label in the leftmost row-label column is best on column LC? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Omni-UGround-V1-7B (Ours)", "value": "37.4"} |
| 11 | HB_real_001628 | Aggregation & Comparison | Rank Selection | Within block closed-source Multimodal Large Language Models (A11Y+Screenshot), list the top-2 items by column DI, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "GPT-4o (Hurst et al., 2024)", "value": "35.5"}, {"item": "Claude-3.5-Sonnet", "value": "25.0"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000214 | Aggregation & Comparison | Extremum Reasoning | Based on the table, for item ShowUI-2B* (Lin et al., 2024), what is the minimum value across all non-delta numeric metrics in this row? Output only the number as a string. | "25.4" |
| 2 | HB_real_000486 | Constrained Retrieval | Value Filtering | For row ShowUI-2B* (Lin et al., 2024), list the red downward-change annotations from Dependency Comp. Medium through Hierarchy Comp. Hard, left to right. Output only a JSON array of strings. | ["↓5.7", "↓2.7", "↓8.6", "↓4.5", "↓9.3", "↓11.2", "↓5.2", "↓5.2", "↓5.6", "↓3.9"] |
| 3 | HB_real_000840 | Aggregation & Comparison | Extremum Reasoning | For item ShowUI-2B* (Lin et al., 2024), what is the maximum value across all non-delta numeric metrics in this row? Output only the number as a string. | "45.9" |
| 4 | HB_real_001631 | Aggregation & Comparison | Argmax Reasoning | In column Knowledge Comp.↑ Easy, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "UGround-7B* (Gou et al., 2024)", "value": "42.3"} |
| 5 | HB_real_001632 | Aggregation & Comparison | Argmax Reasoning | In column Branch Comp.↑ Hard, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 5, "col": 6, "value": "32.4"} |
| 6 | HB_real_001633 | Aggregation & Comparison | Pairwise Comparison | For row UGround-7B* (Gou et al., 2024), compare column Instruction Comp.↑ Easy and column Instruction Comp.↑ Hard. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Instruction Comp.↑ Easy", "value": "53.0"} |
| 7 | HB_real_001634 | Constrained Retrieval | Value Filtering | For the ShowUI-2B* row, what is the red delta annotation under Branch Comp.↑ Hard? Output the full string. | "↓4.5" |
| 8 | HB_real_001635 | Constrained Retrieval | Value Filtering | In column Hierarchy Comp.↑ Hard, list all row labels in the primary row-label column with value >= 30. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "OS-Atlas-Base-4B* (Wu et al., 2024b)", "value": "33.2"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000153 | Markup Grounding | Marked Value Listing | From this table, list all bold numbers. Output only a JSON array of strings. | ["92.49", "87.24", "83.57", "84.89", "78.97", "70.83"] |
| 2 | HB_real_000155 | Markup Grounding | Marked Column Listing | In column AndroidControl-Low SR, list every boldfaced numeric value. Output only a JSON array of strings. | ["83.57"] |
| 3 | HB_real_000162 | Markup Grounding | Marker Counting | From this table, how many bold numbers are in the table? Output only the count as a string. | "6" |
| 4 | HB_real_000166 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=6,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 5 | HB_real_000308 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=1,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 6 | HB_real_000314 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are boldfaced. On the remaining cells in column AndroidControl-Low Type, what is the max? Exclude N/A and —. Output only the number as a string. | "93.00" |
| 7 | HB_real_000351 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are boldfaced. On the remaining cells in column AndroidControl-High SR, what is the max? Exclude N/A and —. Output only the number as a string. | "70.31" |
| 8 | HB_real_000357 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=7,c=6) in the numeric part of the table. Output only the value as a string. | "70.83" |
| 9 | HB_real_000644 | Aggregation & Comparison | Grouped Aggregation | According to the table image, take the cells that are boldfaced. What is their max? Exclude N/A and —. Output only the number as a string. | "92.49" |
| 10 | HB_real_000685 | Local Relations | Neighbor Indexing | Using the table only, at position (r=6,c=1) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "92.49", "up": "92.15", "down": "92.37", "left": null, "right": "83.51"} |
| 11 | HB_real_000690 | Markup Grounding | Marked Column Listing | In column AndroidControl-High Type, list every boldfaced numeric value. Output only a JSON array of strings. | ["84.89"] |
| 12 | HB_real_000946 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 3, "col": 1, "value": "93.00"} |
| 13 | HB_real_001637 | Markup Grounding | Marker Counting | How many underlined numeric cells are in the table? Output only the count as a string. | "0" |
| 14 | HB_real_001638 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=6,c=1) in the numeric part of the table highlighted with a background color? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": false, "color": null} |
| 15 | HB_real_001641 | Local Relations | Neighbor Value Tracing | For the unique value 70.83, return its immediate horizontal neighbors in the same row as JSON {"left":"...","right":"..."}. Use null where a neighbor is absent. | {"left": "78.97", "right": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000240 | Aggregation & Comparison | Rank Selection | Within block OmniAct-Web, list the top-2 items by column SR, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Omni-UGround-7B-V1 (Ours)", "value": "80.24"}, {"item": "UGround-7B-V1 (Gou et al., 2024)", "value": "79.85"}] |
| 2 | HB_real_000494 | Markup Grounding | Marked Value Listing | List all bold numbers in the Omni-UGround-7B-V1 (Ours) row. Output only a JSON array of strings. | ["91.24", "87.35", "80.24", "97.93", "95.21", "92.10"] |
| 3 | HB_real_000598 | Markup Grounding | Marker Counting | How many bold numeric cells are in the table? Output only the count as a string. | "6" |
| 4 | HB_real_000868 | Markup Grounding | Underline Presence | Is the value at position (r=2,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 5 | HB_real_000871 | Aggregation & Comparison | Rank Selection | Within block OmniAct-Web, who ranks 3rd by column SR? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Omni-OS-Atlas-4B (Ours)", "value": "74.62"} |
| 6 | HB_real_001642 | Aggregation & Comparison | Argmax Reasoning | In column OmniAct-Desktop SR, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Omni-UGround-7B-V1 (Ours)", "value": "92.10"} |
| 7 | HB_real_001643 | Aggregation & Comparison | Best Selection | Within group OmniAct-Web, which row label in the leftmost row-label column is best on column Grounding? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Omni-UGround-7B-V1 (Ours)", "value": "87.35"} |
| 8 | HB_real_001644 | Aggregation & Comparison | Pairwise Comparison | For row Qwen2-VL-7B (Wang et al., 2024b), compare column OmniAct-Web SR and column OmniAct-Desktop SR. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "OmniAct-Desktop SR", "value": "91.77"} |
| 9 | HB_real_001646 | Markup Grounding | Bold Presence | Is the number with value 97.93 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 10 | HB_real_001647 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000254 | Markup Grounding | Marked Value Listing | Using the table only, list values whose text is colored in red in this table. Output only a JSON array of strings. | ["↓5.58", "↓10.23"] |
| 2 | HB_real_000647 | Markup Grounding | Marked Value Listing | List values whose text is colored in blue in this table. Output only a JSON array of strings. | ["↑6.1", "↑10.32"] |
| 3 | HB_real_000922 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Avg. Sensitivity? Output only the number as a string. | "12.9" |
| 4 | HB_real_001072 | Constrained Retrieval | Value Filtering | Considering only red downward delta annotations from top to bottom, what is the value at row 1? Output the full string. | "↓5.58" |
| 5 | HB_real_001648 | Constrained Retrieval | Value Filtering | Considering only blue upward delta annotations from top to bottom, what is the value at row 2? Output the full string. | "↑10.32" |
| 6 | HB_real_001650 | Consistency & Missingness | Cell Missingness | Is the cell (row Human, column Backbone) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 7 | HB_real_000751 | Aggregation & Comparison | Grouped Aggregation | Considering only red downward delta annotations from top to bottom, which row has the largest absolute delta? Return the row index as a string. | "2" |
| 8 | HB_real_001952 | Consistency & Missingness | Consistency Probing | List the model rows that have blue upward delta annotations. Output only a JSON array of strings. | ["OS-Atlas-Pro", "Aguvis"] |
| 9 | HB_real_001758 | Constrained Retrieval | Value Filtering | List all delta annotations in the Avg. Sensitivity column from top to bottom. Output only a JSON array of strings. | ["↑6.1", "↓5.58", "↑10.32", "↓10.23"] |
| 10 | HB_real_000764 | Aggregation & Comparison | Grouped Aggregation | For rows whose model name contains "Ours", what is the mean Avg. Sensitivity value? Output the value rounded to two decimal places as a string. | "3.08" |
| 11 | HB_real_001792 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=4,c=1) in the numeric part of the table. The numeric part is the Avg. Sensitivity column. Output only the number as a string. | "3.49" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000036 | Aggregation & Comparison | Grouped Aggregation | Based on the table, among the six +intent rows, which row has the maximum Long-Range Planning↑ delta? Return the row index as a string (top-to-bottom among +intent rows). | "6" |
| 2 | HB_real_000043 | Constrained Retrieval | Value Filtering | For the Overall↑ delta column, considering +intent rows only, what is the value at row 5? Output the full string. | "+5.4" |
| 3 | HB_real_000368 | Markup Grounding | Marked Value Listing | List all blue delta values in the GPT-4o + intent prompt row. Output only a JSON array of strings. | ["+7.2", "+10.6", "+8.9"] |
| 4 | HB_real_000375 | Aggregation & Comparison | Grouped Aggregation | Among the six +intent rows, which row has the minimum Long-Range Planning↑ delta? Return the row index as a string, counting +intent rows from top to bottom. | "2" |
| 5 | HB_real_000588 | Consistency & Missingness | Consistency Probing | From this table image, in the Overall↑ delta values for the six "+ intent" rows, list all items whose delta is positive (>0). Return JSON array of {item,value}. | [{"item": "Omni-OS-Atlas-Base-4B + intent tuning", "value": "+1.7"}, {"item": "Omni-UGround-V1-7B + intent tuning", "value": "+1.2"}, {"item": "Qwen-VL-Max + intent prompt", "value": "+2.6"}, {"item": "Gemini-2.0-Flash + intent prompt", "value": "+4.9"}, {"item": "Claude-3.5-Sonnet + intent prompt", "value": "+5.4"}, {"item": "GPT-4o + intent prompt", "value": "+8.9"}] |
| 6 | HB_real_000741 | Constrained Retrieval | Value Filtering | For the Long-Range Planning↑ delta column, considering +intent rows only, what is the value at row 6? Output the full string. | "+10.6" |
| 7 | HB_real_001651 | Constrained Retrieval | Value Filtering | Considering only the six +intent rows from top to bottom, list all delta values in the Long-Range Planning↑ column. Output only a JSON array of strings. | ["+1.9", "+1.3", "+2.7", "+4.0", "+4.4", "+10.6"] |
| 8 | HB_real_001655 | Aggregation & Comparison | Grouped Aggregation | Within the six +intent rows, what is the mean of the Overall↑ main scores before the blue delta annotations? Output only the number as a string, rounded to two decimal places. | "29.88" |
| 9 | HB_real_rebal_000048 | Aggregation & Comparison | Grouped Aggregation | Among the four closed-source +intent prompt rows, which row has the largest Overall↑ delta? Return the row index as a string. | "4" |
| 10 | HB_real_rebal_000049 | Aggregation & Comparison | Grouped Aggregation | Among the two open-source +intent tuning rows, which row has the largest Overall↑ delta? Return the row index as a string. | "1" |
| 11 | HB_real_rebal_000053 | Markup Grounding | Marked Value Listing | List all blue delta annotations in the GPT-4o + intent prompt row. Output only a JSON array of strings. | ["↑7.2", "↑10.6", "↑8.9"] |
| 12 | HB_real_rebal_000054 | Consistency & Missingness | Consistency Probing | In the Overall↑ column, list the closed-source +intent prompt rows with positive blue deltas. Return a JSON array of {"item":"<row label>","value":"<delta>"}. | [{"item": "Qwen-VL-Max + intent prompt", "value": "↑2.6"}, {"item": "Gemini-2.0-Flash + intent prompt", "value": "↑4.9"}, {"item": "Claude-3.5-Sonnet + intent prompt", "value": "↑5.4"}, {"item": "GPT-4o + intent prompt", "value": "↑8.9"}] |
| 13 | HB_real_rebal_000056 | Aggregation & Comparison | Rank Selection | Among the six +intent rows, which row has the second-largest Overall↑ delta? Return JSON {"item":"<row label>","value":"<delta>"}. | {"item": "Claude-3.5-Sonnet + intent prompt", "value": "↑5.4"} |
| 14 | HB_real_rebal_000070 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000425 | Consistency & Missingness | Column Missingness | Which benchmark rows have missing entries marked with "-" in column #QA? Output only a JSON array of row-label strings. | ["DocVQA (Mathew et al., 2021)", "ChartQA (Masry et al., 2022)", "MP-DocVQA (Tito et al., 2022)", "DUDE (Landeghem et al., 2023)"] |
| 2 | HB_real_001656 | Aggregation & Comparison | Argmax Reasoning | In column #Docs, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LongDocURL (Ours)", "value": "396"} |
| 3 | HB_real_001657 | Consistency & Missingness | Column Missingness | Which benchmark rows have missing entries marked with "-" in column #Avg. Tokens? Output only a JSON array of row-label strings. | ["M-Longdoc (Chia et al., 2024)"] |
| 4 | HB_real_001658 | Constrained Retrieval | Value Filtering | In column Multi-page(%), list all row labels in the primary row-label column with value >= 30. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "MMLongBench-Doc (Ma et al., 2024)", "value": "✓(33.0%)"}, {"item": "LongDocURL (Ours)", "value": "✓(52.9%)"}] |
| 5 | HB_real_001659 | Aggregation & Comparison | Pairwise Comparison | For row LongDocURL (Ours), compare column #Avg. Pages and column #QA. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "#QA", "value": "2,325"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000108 | Aggregation & Comparison | Grouped Aggregation | Based on the table, within "OCR (PyMuPDF^4) + Large Language Models (LLMs) / Proprietary Models", what is the mean value of column Total? Exclude N/A and —. Output only the number as a string, rounded to two decimal places. | "33.44" |
| 2 | HB_real_000116 | Aggregation & Comparison | Extremum Reasoning | In row Human Experts, what is the maximum numeric value across the table? Output only the number as a string. | "87.2" |
| 3 | HB_real_000270 | Local Relations | Marker-Anchor Neighbor | For the highlighted cell with value 45.8 in row O1-preview, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "33.3", "right": "24.5"} |
| 4 | HB_real_001660 | Aggregation & Comparison | Argmax Reasoning | In column Total, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Human Experts", "value": "84.8"} |
| 5 | HB_real_001661 | Aggregation & Comparison | Best Selection | Within group "Large Vision Language Models (LVLMs) / Proprietary Models", which row label in the leftmost row-label column is best on column Total? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o", "value": "64.5"} |
| 6 | HB_real_001662 | Markup Grounding | Highlight Presence | Using the table only, is the value 45.8 in row O1-preview highlighted? Output only true or false. | "true" |
| 7 | HB_real_001663 | Consistency & Missingness | Cell Missingness | Is the cell (row Human Experts, column all 612 in Understanding single-page) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 8 | HB_real_001664 | Consistency & Missingness | Consistency Probing | Within the Understanding section, which leftmost row labels have missing entries in the single-page columns? Output only a JSON array of row labels. | ["Human Experts"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000140 | Constrained Retrieval | Cell Retrieval | Based on the table, find the value at (r=20, c=4) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and left text columns. Output only the number as a string. | "64.46" |
| 2 | HB_real_000147 | Local Relations | Neighbor Value Tracing | The value 77.28 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": null, "right": "75.91"} |
| 3 | HB_real_000185 | Aggregation & Comparison | Grouped Aggregation | Based on the table, take the cells highlighted in light purple. What is their max? Exclude N/A and —. Output only the number as a string. | "77.28" |
| 4 | HB_real_000235 | Local Relations | Neighbor Indexing | According to the table, at (r=2,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and left text columns. | {"target": "74.65", "up": "51.69", "down": "75.91", "left": "75.63", "right": "72.25"} |
| 5 | HB_real_000319 | Markup Grounding | Marked Column Listing | By reading the table in the image, list all bold numbers in column Top-100. Output only a JSON array of strings. | ["70.94", "64.94", "69.46", "64.46"] |
| 6 | HB_real_000320 | Aggregation & Comparison | Grouped Aggregation | For the Child-to-Parent CLIP-hyp row, what is the mean across Top-5, Top-10, Top-50, and Top-100? Output only the number as a string, rounded to two decimal places. | "74.25" |
| 7 | HB_real_001013 | Constrained Retrieval | Value Filtering | By reading the table image, among light-purple highlighted cells, list the values that are > 70. Output JSON array of strings. | ["77.28", "75.91", "72.85", "70.94", "74.48", "73.27", "70.78"] |
| 8 | HB_real_001668 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 3, column 1 in the numeric part of the table, list all numbers highlighted with the same light-purple color as the anchor. The numeric part of the table excludes header and left text columns. Output only a JSON array of strings. | ["77.28", "75.91", "72.85", "70.94", "68.62", "67.42", "66.21", "64.94", "74.48", "73.27", "70.78", "69.46", "69.50", "68.14", "65.67", "64.46"] |
| 9 | HB_real_001670 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 10 | HB_real_002251 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light purple in this table? Output only the count as a string. | "16" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000646 | Constrained Retrieval | Color-Anchor Retrieval | From this table image, according to the table, for the number with value 74.85, list all numbers highlighted with the same light-purple color as the anchor. Output only a JSON array of strings. | ["77.00", "85.75", "91.89", "14.96", "17.57", "20.76", "74.85", "83.79", "90.61", "16.59", "18.65", "21.29"] |
| 2 | HB_real_000857 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=3,c=3) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_purple"} |
| 3 | HB_real_001014 | Constrained Retrieval | Value Filtering | Based on the image table, among highlighted cells (any highlighted color), list the values that are > 90. Output JSON array of strings. | ["91.89", "90.61"] |
| 4 | HB_real_001066 | Aggregation & Comparison | Rank Selection | Within group "MoCo-v2 / Recall↑", list the top 2 rows by column Top-250k, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "MoCo-hyp†", "value": "90.61"}, {"item": "MoCo-euc†", "value": "90.46"}] |
| 5 | HB_real_001671 | Aggregation & Comparison | Best Selection | In metric Top-250k (block CLIP ViT / Recall↑), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "CLIP-hyp†", "value": "91.89"} |
| 6 | HB_real_001672 | Aggregation & Comparison | Best Selection | In metric Top-250k (block CLIP ViT / OT Distance↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "CLIP-hyp†", "value": "20.76"} |
| 7 | HB_real_001673 | Aggregation & Comparison | Best Selection | Within group "MoCo-v2 / Recall↑", which row label in the leftmost row-label column is best on column Top-250k? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "MoCo-hyp†", "value": "90.61"} |
| 8 | HB_real_001674 | Markup Grounding | Highlight Presence | Is the number with value 66.63 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": false} |
| 9 | HB_real_001675 | Constrained Retrieval | Value Filtering | Among highlighted cells (any highlighted color), list the values that are > 85. Output JSON array of strings. | ["85.75", "91.89", "90.61"] |
| 10 | HB_real_001676 | Aggregation & Comparison | Rank Selection | Within group "MoCo-v2 / OT Distance↓", list the top 2 rows by column Top-150k, sorted from lowest to highest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "MoCo-hyp†", "value": "16.59"}, {"item": "MoCo-euc†", "value": "16.65"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000010 | Aggregation & Comparison | Grouped Aggregation | Using the table only, exclude cells that are boldfaced. On the remaining cells in column Top-100 (upper table), what is the max? Exclude N/A and —. Output only the number as a string. | "23.36" |
| 2 | HB_real_000017 | Markup Grounding | Marked Column Listing | In column Top-100 (upper table), return the values that are bold. Output only a JSON array of strings. | ["20.53", "26.81"] |
| 3 | HB_real_000020 | Markup Grounding | Marked Column Listing | In column Top-30k (block Recall↑), return the values that are bold. Output only a JSON array of strings. | ["78.85"] |
| 4 | HB_real_000436 | Markup Grounding | Bold Presence | In the upper table numeric grid, is the value at position (r=6,c=4) boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 5 | HB_real_000441 | Local Relations | Neighbor Indexing | In the upper table numeric grid, at (r=6,c=4), return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and the Model/Dist. Func. columns. | {"target": "26.81", "up": "20.43", "down": null, "left": "27.83", "right": null} |
| 6 | HB_real_000446 | Markup Grounding | Marked Value Listing | List all numbers in the light-purple highlighted rows of the upper table. Output only a JSON array of strings. | ["28.84", "27.02", "22.87", "20.53", "30.27", "29.53", "27.83", "26.81"] |
| 7 | HB_real_000799 | Aggregation & Comparison | Extremum Reasoning | In column Top-100 (upper table), what is the maximum value? Output only the number as a string. | "26.81" |
| 8 | HB_real_000811 | Aggregation & Comparison | Grouped Aggregation | Please answer this question: take the cells that are boldfaced. What is their max? Exclude N/A and —. Output only the number as a string. | "78.85" |
| 9 | HB_real_000955 | Markup Grounding | Highlight Presence | In the upper table numeric grid, is the value at position (r=6,c=4) highlighted with a background color? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_purple"} |
| 10 | HB_real_001678 | Aggregation & Comparison | Best Selection | In metric Top-20k (block OT Distance↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "CLIP-euc†", "value": "7.04"} |
| 11 | HB_real_001679 | Markup Grounding | Marker Counting | How many bold numeric cells are in the full table image, counting both the upper and lower tables? Output only the count as a string. | "14" |
| 12 | HB_real_001680 | Aggregation & Comparison | Argmax Reasoning | In the upper table numeric grid, find the single largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 6, "col": 1, "value": "30.27"} |
| 13 | HB_real_001681 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000259 | Constrained Retrieval | Value Filtering | Using the table only, among light-yellow highlighted cells, list the values that are > 20. Output JSON array of strings. | ["72.3", "86.6"] |
| 2 | HB_real_000855 | Constrained Retrieval | Color-Anchor Retrieval | Read the table carefully and for the number with value 15.2, list all numbers highlighted with the same light-yellow color as the anchor. Output only a JSON array of strings. | ["3.00", "72.3", "1.81", "86.6", "16.9", "2.45", "1.98", "15.2"] |
| 3 | HB_real_000901 | Markup Grounding | Marker Counting | From this table, how many numeric cells are highlighted in light yellow in this table? Output only the count as a string. | "8" |
| 4 | HB_real_001012 | Constrained Retrieval | Value Filtering | According to the table image, among highlighted cells (any highlighted color), list the values that are > 50. Output JSON array of strings. | ["72.3", "86.6", "56.2"] |
| 5 | HB_real_001063 | Markup Grounding | Marked Value Listing | List all numeric values highlighted in light green. Output only a JSON array of strings. | ["2.50", "1.42", "56.2", "10.7", "1.68", "1.29", "47.4", "9.61"] |
| 6 | HB_real_001682 | Aggregation & Comparison | Best Selection | In metric T60 (%)↓ (block RAF-Furnished), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Ours (1%)", "value": "9.61"} |
| 7 | HB_real_001683 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are highlighted in light green. On the remaining cells in column C50 (dB)↓ within block RAF-Furnished, what is the min? Exclude N/A and —. Output only the number as a string. | "1.98" |
| 8 | HB_real_001684 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light green. What is their max? Exclude N/A and —. Output only the number as a string. | "56.2" |
| 9 | HB_real_001685 | Markup Grounding | Marker Counting | How many highlighted numeric data cells are in the table (light yellow or light green)? Output only the count as a string. | "16" |
| 10 | HB_real_001686 | Aggregation & Comparison | Best Selection | In metric C50 (dB)↓ (block RAF-Furnished), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Ours (1%)", "value": "1.29"} |
| 11 | HB_real_002254 | Markup Grounding | Highlight Presence | Is the value at position (r=6,c=8) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_green"} |
| 12 | HB_real_002255 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=5,c=4) in the numeric part of the table. Output only the value as a string. | "16.9" |
| 13 | HB_real_002256 | Local Relations | Neighbor Indexing | Using the table only, at position (r=5,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "16.9", "up": "24.1", "down": "10.7", "left": "86.6", "right": "2.45"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000261 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are greater than 30. Output JSON array of strings. | ["39.7", "44.9"] |
| 2 | HB_real_000557 | Markup Grounding | Marker Counting | How many light-yellow highlighted numeric cells are in the table? Output only the count as a string. | "12" |
| 3 | HB_real_000856 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 39.7, list all numbers highlighted with the same light-yellow color as the anchor. Output only a JSON array of strings. | ["1.31", "1.65", "1.43", "1.88", "1.26", "39.7", "18.5", "1.87", "1.56", "44.9", "6.8", "1.15"] |
| 4 | HB_real_001062 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=6,c=1) in the numeric part of the table, list all values highlighted with the same light-green color as the anchor. Output only a JSON array of strings. | ["0.99", "1.02", "24.3", "0.98", "1.44", "10.8", "1.11", "1.45", "31.9", "0.85", "6.3"] |
| 5 | HB_real_001068 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the highlighted cell with value 1.26. The requested neighbor is numeric. Output only the number as a string. | "1.03" |
| 6 | HB_real_001687 | Aggregation & Comparison | Best Selection | In metric C50 (dB)↓ within block Hallway, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "AV-NeRF [39]", "value": "1.03"} |
| 7 | HB_real_001688 | Aggregation & Comparison | Grouped Aggregation | Among the light-yellow highlighted cells, what is the maximum value? Output only the number as a string. | "44.9" |
| 8 | HB_real_001689 | Markup Grounding | Highlight Presence | Is the number with value 1.03 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 9 | HB_real_001690 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 1, "col": 9, "value": "306.9"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000074 | Aggregation & Comparison | Rank Selection | Within block Czech to English with 16 query samples, who ranks 2nd in column BERTScore? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "+LoRD", "value": "0.906 ± 0.00"} |
| 2 | HB_real_000079 | Aggregation & Comparison | Rank Selection | Within block Czech to English with 16 query samples, who ranks 1st in column BERTScore? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Victim Model", "value": "0.957"} |
| 3 | HB_real_000416 | Aggregation & Comparison | Extremum Reasoning | For row +LoRD in block Finnish to English with 16 query samples, what is the maximum value across all metrics? Output only the number as a string. | "0.905 ± 0.00" |
| 4 | HB_real_000796 | Aggregation & Comparison | Extremum Reasoning | For row +MLE in block Czech to English with 16 query samples, what is the minimum value across all metrics? Output only the number as a string. | "0.245 ± 0.01" |
| 5 | HB_real_000948 | Aggregation & Comparison | Rank Selection | Within block German to English with 16 query sample, who ranks 2nd in column BERTScore? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "+LoRD", "value": "0.917 ± 0.00"} |
| 6 | HB_real_001691 | Aggregation & Comparison | Best Selection | Within group German to English with 16 query sample, which row label is best on column BLEU-1? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Victim Model", "value": "0.661"} |
| 7 | HB_real_001692 | Aggregation & Comparison | Pairwise Comparison | For row +LoRD in block Czech to English with 16 query samples, compare column BLEU-1 and column Rouge-L. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "BLEU-1", "value": "0.545 ± 0.01"} |
| 8 | HB_real_001693 | Aggregation & Comparison | Extremum Reasoning | For item Local Model in block German to English with 16 query sample, what is the minimum value across all metrics? Output only the number as a string. | "0.130" |
| 9 | HB_real_001694 | Constrained Retrieval | Value Filtering | In column BLEU-4 within block German to English with 16 query sample, list all row labels with value >= 0.30. Return a JSON array of {"item":"<row label>","value":"<cell value>"} objects. | [{"item": "Victim Model", "value": "0.377"}, {"item": "+MLE", "value": "0.302 ± 0.01"}, {"item": "+LoRD", "value": "0.308 ± 0.00"}] |
| 10 | HB_real_001695 | Aggregation & Comparison | Rank Selection | Within block Finnish to English with 16 query samples, who ranks 2nd in column BERTScore? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "+MLE", "value": "0.905 ± 0.00"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000054 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": false, "color": null} |
| 2 | HB_real_000188 | Aggregation & Comparison | Grouped Aggregation | Read the table carefully and take the cells highlighted in light pink. What is their max? Exclude N/A and —. Output only the number as a string. | "0.785 ± 0.01" |
| 3 | HB_real_000450 | Aggregation & Comparison | Pairwise Comparison | In the PIQA block, column Precision, compare +MLE (baseline) and +LoRD (ours). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "+LoRD (ours)", "value": "0.795 ± 0.01"} |
| 4 | HB_real_000451 | Aggregation & Comparison | Extremum Reasoning | In the PIQA block, what is the maximum value in column Accuracy? Output only the number/string as shown. | "0.828" |
| 5 | HB_real_000592 | Aggregation & Comparison | Extremum Reasoning | In the TruthfulQA block, what is the minimum value in column Recall? Output only the number/string as shown. | "0.190 ± 0.09" |
| 6 | HB_real_000763 | Local Relations | Neighbor Indexing | In the TruthfulQA numeric grid, at (r=3,c=3), return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and the left method column. | {"target": "0.190 ± 0.09", "up": "0.195", "down": "0.232 ± 0.01", "left": "0.500 ± 0.00", "right": "0.266 ± 0.09"} |
| 7 | HB_real_000765 | Local Relations | Neighbor Value Tracing | The value 0.381 ± 0.17 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": null, "right": "0.500 ± 0.00"} |
| 8 | HB_real_000769 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 9 | HB_real_001008 | Markup Grounding | Marker Counting | By reading the table image, how many numeric cells are highlighted in light pink in this table? Output only the count as a string. | "6" |
| 10 | HB_real_001696 | Markup Grounding | Marked Value Listing | List numbers highlighted in light pink. Output only a JSON array of strings. | ["0.785 ± 0.01", "0.785 ± 0.01", "0.783 ± 0.02", "0.381 ± 0.17", "0.190 ± 0.09", "0.266 ± 0.09"] |
| 11 | HB_real_001700 | Markup Grounding | Highlight Presence | Is the number with value 0.190 ± 0.09 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 12 | HB_real_002257 | Constrained Retrieval | Value Filtering | Among light-pink highlighted cells, list the values that are > 0.7. Output JSON array of strings. | ["0.785 ± 0.01", "0.785 ± 0.01", "0.783 ± 0.02"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000151 | Markup Grounding | Marked Value Listing | Based on the table, return every value shown in bold. Output only a JSON array of strings. | ["1.66", "0.16", "6.45", "2.81", "3.56", "0.71", "1.34", "3.55", "1.15", "0.79"] |
| 2 | HB_real_000158 | Markup Grounding | Marked Column Listing | In column Toxicity (DiaSafety), return the values that are bold. Output only a JSON array of strings. | ["6.45"] |
| 3 | HB_real_000164 | Markup Grounding | Marker Counting | Please answer this question: how many bold numbers are in the table? Output only the count as a string. | "10" |
| 4 | HB_real_000165 | Markup Grounding | Marked Column Listing | In column Profanity (SafeRLHF), return the values that are bold. Output only a JSON array of strings. | ["1.66"] |
| 5 | HB_real_000311 | Aggregation & Comparison | Extremum Reasoning | In column Severe Toxicity (SafeRLHF), what is the minimum value? Output only the number as a string. | "0.16" |
| 6 | HB_real_000353 | Aggregation & Comparison | Pairwise Comparison | In column Profanity (SafeRLHF), compare +MLE and +LoRD. Which value is smaller? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "+MLE", "value": "1.66"} |
| 7 | HB_real_000355 | Aggregation & Comparison | Extremum Reasoning | For row +LoRD, what is the maximum value across all numeric metrics? Output only the number as a string. | "6.45" |
| 8 | HB_real_000577 | Markup Grounding | Bold Presence | Is the value at row +LoRD, column Toxicity (DiaSafety) boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 9 | HB_real_000725 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 10 | HB_real_001705 | Aggregation & Comparison | Pairwise Comparison | For row +LoRD, compare column Toxicity (DiaSafety) and column Toxicity (SafeRLHF). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Toxicity (DiaSafety)", "value": "6.45"} |
| 11 | HB_real_001707 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000917 | Consistency & Missingness | Cell Missingness | Is the cell (row RESC [14], column DictAS (Ours)) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 2 | HB_real_001071 | Markup Grounding | Marked Value Listing | Using visual markers in the table, list numbers whose text is colored in red in this table. Output only a JSON array of strings. | ["(98.6, 95.1, 66.8)", "(98.8, 91.9, 41.8)", "(98.4, 94.9, 44.2)", "(98.4, 95.8, 42.9)", "(98.0, 83.3, 66.8)", "(98.4, 92.2, 52.5)", "(97.5, 89.7, 74.9)", "(97.3, 77.2, 59.3)", "(97.4, 83.4, 67.1)"] |
| 3 | HB_real_001729 | Markup Grounding | Marked Value Listing | List red tuple values in the Medical Datasets section only. Output only a JSON array of strings. | ["(97.5, 89.7, 74.9)", "(97.3, 77.2, 59.3)", "(97.4, 83.4, 67.1)"] |
| 4 | HB_real_001730 | Constrained Retrieval | Cell Retrieval | What is the DictAS (Ours) tuple for row MPDD [17] in block Industrial Datasets (AUROC, PRO, AP)? Output the full tuple string. | "(98.4, 95.8, 42.9)" |
| 5 | HB_real_001731 | Aggregation & Comparison | Grouped Aggregation | Within the Medical Datasets section, what is the mean AP value inside the DictAS (Ours) tuples? Output only the number as a string, rounded to one decimal place. | "67.1" |
| 6 | HB_real_001732 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000011 | Aggregation & Comparison | Grouped Aggregation | From this table, exclude cells that are boldfaced. On the remaining cells in column AP (%), what is the max? Exclude N/A and —. Output only the number as a string. | "57.5±0.6" |
| 2 | HB_real_000018 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in the DictAS row? Output only the count as a string. | "3" |
| 3 | HB_real_000128 | Markup Grounding | Bold Presence | Please answer this question: is the number at (r=8,c=2) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric columns Method/Backbone/Resolution. | {"bold": true} |
| 4 | HB_real_000138 | Markup Grounding | Underline Presence | Please answer this question: is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric columns Method/Backbone/Resolution. | {"underlined": false} |
| 5 | HB_real_000297 | Local Relations | Neighbor Value Tracing | The value 73.5 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "3.4", "right": "66.8±0.4"} |
| 6 | HB_real_000301 | Constrained Retrieval | Value Filtering | How much higher is DictAS AP (%) than PromptAD AP (%)? Output only the number as a string rounded to one decimal place. | "9.3" |
| 7 | HB_real_001017 | Constrained Retrieval | Value Filtering | According to the table image, among highlighted cells (any highlighted color), list the values that are > 0. Output JSON array of strings. | ["3.4", "73.5", "66.8±0.4"] |
| 8 | HB_real_001746 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000052 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": false, "color": null} |
| 2 | HB_real_000059 | Markup Grounding | Highlight Presence | Is the value at position (r=7,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| 3 | HB_real_000277 | Aggregation & Comparison | Grouped Aggregation | Using the visual table content, within "Module Ablation", what is the mean value of column AP? Exclude N/A and —. Output only the number as a string, rounded to two decimal places. | "63.83" |
| 4 | HB_real_000561 | Markup Grounding | Marked Column Listing | In column AP, return the values that are bold. Output only a JSON array of strings. | ["66.8"] |
| 5 | HB_real_000594 | Constrained Retrieval | Cell Retrieval | What is the PRO value for row DictAS? Output only the number as a string. | "95.1" |
| 6 | HB_real_000660 | Aggregation & Comparison | Pairwise Comparison | In column AP, compare DictAS and w/o L_TAC. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DictAS", "value": "66.8"} |
| 7 | HB_real_000766 | Aggregation & Comparison | Extremum Reasoning | For row DictAS, what is the maximum value across all numeric metrics? Output only the number as a string. | "98.6" |
| 8 | HB_real_000969 | Markup Grounding | Bold Presence | Is the number at (r=7,c=3) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left ablation item column. | {"bold": true} |
| 9 | HB_real_001767 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column AP? Output only the number as a string. | "66.8" |
| 10 | HB_real_001769 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000258 | Constrained Retrieval | Value Filtering | Among all highlighted cells in the table, list the values that are less than 0.1. Output JSON array of strings. | ["0.096", "0.089", "0.084"] |
| 2 | HB_real_000430 | Consistency & Missingness | Column Missingness | In column PSNR↑ 6-view, list the row labels whose entry is missing. Output only a JSON array of strings. | ["SparseNeRF"] |
| 3 | HB_real_000632 | Markup Grounding | Marker Counting | How many highlighted numeric cells are in the table across all highlight colors? Output only the count as a string. | "28" |
| 4 | HB_real_000859 | Constrained Retrieval | Color-Anchor Retrieval | Using the table only, for the number with value 20.79, list all numbers highlighted with the same light-pink color as the anchor. Output only a JSON array of strings. | ["20.79", "24.78", "26.36", "0.724", "0.839", "0.878", "0.183", "0.110", "0.084"] |
| 5 | HB_real_000912 | Aggregation & Comparison | Grouped Aggregation | Excluding the light-pink highlighted cells, what is the maximum value in PSNR↑ 9-view among the remaining rows? Output only the number as a string. | "26.06" |
| 6 | HB_real_001060 | Markup Grounding | Highlight Presence | Is the value at position (r=10,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 7 | HB_real_001708 | Local Relations | Neighbor Value Tracing | For the cell with value 24.49, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "24.49", "up": "24.09", "down": "24.78", "left": "20.45", "right": "26.06"} |
| 8 | HB_real_001709 | Aggregation & Comparison | Best Selection | In metric PSNR↑ (9-view), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "SE-GS", "value": "26.36"} |
| 9 | HB_real_001710 | Consistency & Missingness | Column Missingness | In column 6-view under PSNR↑, which method rows have missing entries marked as "-"? Output only a JSON array of strings. | ["SparseNeRF [39]"] |
| 10 | HB_real_001711 | Aggregation & Comparison | Argmax Reasoning | In column PSNR↑ (9-view), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 10, "col": 3, "value": "26.36"} |
| 11 | HB_real_001712 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=9,c=2) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_orange"} |
| 12 | HB_real_001713 | Aggregation & Comparison | Grouped Aggregation | Among the light-orange highlighted cells in row CoR-GS, what is the minimum LPIPS↓ value? Output only the number as a string. | "0.089" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000203 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 2 | HB_real_000205 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=9) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 3 | HB_real_000359 | Markup Grounding | Marked Column Listing | List all underlined numbers in column LPIPS↓ 9-view. Output only a JSON array of strings. | [] |
| 4 | HB_real_000367 | Markup Grounding | Highlight Presence | Is the value at position (r=5,c=9) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 5 | HB_real_000730 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=5,c=9) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_pink"} |
| 6 | HB_real_001018 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are less than 0.08. Output JSON array of strings. | ["0.050", "0.074", "0.045", "0.073", "0.043"] |
| 7 | HB_real_001065 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 18.57, list all numbers highlighted with the same light-yellow color as the anchor. Output only a JSON array of strings. | ["23.69", "26.80", "0.941", "0.158", "0.086", "0.050", "18.57", "0.822", "0.905", "0.941"] |
| 8 | HB_real_001714 | Aggregation & Comparison | Best Selection | In metric LPIPS↓ (9-view), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "SE-GS", "value": "0.043"} |
| 9 | HB_real_001716 | Aggregation & Comparison | Grouped Aggregation | Among the light-pink highlighted cells in row SE-GS, what is the minimum LPIPS↓ value? Output only the number as a string. | "0.043" |
| 10 | HB_real_001717 | Local Relations | Neighbor Indexing | At position (r=5,c=9) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.045", "up": "0.064", "down": "0.043", "left": "0.074", "right": null} |
| 11 | HB_real_001718 | Aggregation & Comparison | Grouped Aggregation | Excluding the light-pink highlighted cells, what is the minimum highlighted value in column LPIPS↓ 9-view? Output only the number as a string. | "0.045" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000252 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 19.91, list all numbers highlighted with the same light-pink color as the anchor. Output only a JSON array of strings. | ["19.91", "0.596", "0.400", "23.74", "0.745", "0.265"] |
| 2 | HB_real_000570 | Markup Grounding | Marker Counting | How many numeric cells are highlighted across all highlight colors? Output only the count as a string. | "19" |
| 3 | HB_real_000645 | Markup Grounding | Marked Value Listing | List all values highlighted with the light-yellow background. Output only a JSON array of strings. | ["18.80", "0.531", "0.418", "23.28", "0.715", "0.274"] |
| 4 | HB_real_000853 | Markup Grounding | Highlight Presence | Is the value at position (r=4,c=6) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 5 | HB_real_001059 | Aggregation & Comparison | Grouped Aggregation | Among the light-pink highlighted cells, what is the minimum LPIPS↓ value? Output only the number as a string. | "0.265" |
| 6 | HB_real_001719 | Aggregation & Comparison | Best Selection | In metric LPIPS↓ (12-view), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "SE-GS", "value": "0.400"} |
| 7 | HB_real_001720 | Aggregation & Comparison | Pairwise Comparison | For row FSGS [54], compare column PSNR↑ (12-view) and column PSNR↑ (24-view). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "PSNR↑ (24-view)", "value": "23.28"} |
| 8 | HB_real_001721 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are less than 0.30. Output JSON array of strings. | ["0.274", "0.271", "0.265"] |
| 9 | HB_real_001722 | Aggregation & Comparison | Argmax Reasoning | In column PSNR↑ (24-view), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 4, "col": 4, "value": "23.74"} |
| 10 | HB_real_001723 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000630 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are greater than 32. Output JSON array of strings. | ["32.09", "32.68"] |
| 2 | HB_real_000860 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 27.88, list all numbers highlighted with the same light-pink color as the anchor. Output only a JSON array of strings. | ["27.88", "30.19", "32.68", "0.857", "0.902", "0.937", "0.277", "0.214", "0.177"] |
| 3 | HB_real_000913 | Markup Grounding | Marker Counting | How many numeric cells are highlighted across all highlight colors? Output only the count as a string. | "27" |
| 4 | HB_real_001061 | Aggregation & Comparison | Argmax Reasoning | In column PSNR↑ (10-view), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 4, "col": 3, "value": "32.68"} |
| 5 | HB_real_001724 | Local Relations | Neighbor Value Tracing | For the cell with value 29.77, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "29.77", "up": "29.12", "down": "30.19", "left": "27.62", "right": "32.09"} |
| 6 | HB_real_001725 | Aggregation & Comparison | Grouped Aggregation | Among the light-pink highlighted cells, what is the minimum LPIPS↓ value? Output only the number as a string. | "0.177" |
| 7 | HB_real_001726 | Markup Grounding | Highlight Presence | Is the number with value 31.82 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 8 | HB_real_001727 | Constrained Retrieval | Value Filtering | Among light-pink highlighted cells, list the values that are greater than 32. Output JSON array of strings. | ["32.68"] |
| 9 | HB_real_001728 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=3,c=2) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_orange"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000098 | Markup Grounding | Marked Value Listing | Using the table only, list all underlined numbers. Output only a JSON array of strings. | ["0.68"] |
| 2 | HB_real_000106 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Avg H↓? Exclude N/A and —. Output only the number as a string. | "0.29" |
| 3 | HB_real_000527 | Aggregation & Comparison | Extremum Reasoning | In column Avg U↑, what is the maximum value? Output only the number as a string. | "0.70" |
| 4 | HB_real_000528 | Aggregation & Comparison | Pairwise Comparison | In column Avg H↓, compare GPT-3.5 and GPT-4. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-3.5", "value": "0.3"} |
| 5 | HB_real_000530 | Aggregation & Comparison | Pairwise Comparison | For row Mixtral 8x7b instruct, compare column R-BIN Utility and column R-NUM Utility. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "R-NUM Utility", "value": "0.34"} |
| 6 | HB_real_000533 | Markup Grounding | Marker Counting | How many underlined numeric cells are in the table? Output only the count as a string. | "1" |
| 7 | HB_real_000537 | Markup Grounding | Underline Presence | Is value 0.70 underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 8 | HB_real_001733 | Aggregation & Comparison | Best Selection | In metric Avg H↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT-4", "value": "0.29"} |
| 9 | HB_real_001734 | Markup Grounding | Underline Presence | Is the number with value 0.68 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 10 | HB_real_001737 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000199 | Markup Grounding | Marker Counting | Based on the table, how many underlined numbers are in the table? Output only the count as a string. | "1" |
| 2 | HB_real_000358 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["0.64"] |
| 3 | HB_real_000731 | Aggregation & Comparison | Extremum Reasoning | In row GPT-4, what is the maximum numeric value across the table? Output only the number as a string. | "1.0" |
| 4 | HB_real_001738 | Aggregation & Comparison | Best Selection | In metric Avg Utility↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT-4", "value": "0.71"} |
| 5 | HB_real_001739 | Aggregation & Comparison | Best Selection | In metric Avg H↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT-4", "value": "0.66"} |
| 6 | HB_real_001740 | Markup Grounding | Underline Presence | Is the number with value 0.64 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 7 | HB_real_001741 | Local Relations | Marker-Anchor Neighbor | For the highlighted green cell with value 0.71, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": null, "right": "0.66"} |
| 8 | HB_real_001783 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.71, list all numeric values highlighted with the same light-green row background as the anchor row. Output only a JSON array of strings. | ["0.71", "0.66", "0.29", "0.52", "0.33/0.48", "1.0", "1.0/0.0", "0.61", "0.65/0.39"] |
| 9 | HB_real_000876 | Markup Grounding | Underline Presence | Is the value at position (r=4,c=2) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 10 | HB_real_000815 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light peach in the GPT-3.5 row. What is their maximum numeric value? Output only the number as a string. | "0.96" |
| 11 | HB_real_000932 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 4, "col": "Historical Events Utility", "value": "1.0"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001749 | Aggregation & Comparison | Argmax Reasoning | In column Avg., which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "34.1"} |
| 2 | HB_real_001750 | Aggregation & Comparison | Best Selection | Within group "Closed-source LVLMs", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o [31]", "value": "26.4"} |
| 3 | HB_real_001751 | Aggregation & Comparison | Pairwise Comparison | For row Ours, compare column Ingre. and column Causal. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Ingre.", "value": "33.7"} |
| 4 | HB_real_001752 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column TFLOPs? Exclude N/A and —. Output only the number as a string. | "9.26" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000152 | Markup Grounding | Marked Value Listing | Using the table only, return every value shown in bold. Output only a JSON array of strings. | ["72.25", "61.28", "59.68", "75.73", "65.84"] |
| 2 | HB_real_000156 | Markup Grounding | Marked Column Listing | In column Overall, return the values that are bold. Output only a JSON array of strings. | ["65.84"] |
| 3 | HB_real_000159 | Markup Grounding | Marked Column Listing | In column Con., return the values that are bold. Output only a JSON array of strings. | ["75.73"] |
| 4 | HB_real_000167 | Markup Grounding | Marker Counting | Using the table only, how many bold numbers are in the table? Output only the count as a string. | "5" |
| 5 | HB_real_000305 | Aggregation & Comparison | Extremum Reasoning | In column Overall, what is the maximum value? Output only the number as a string. | "66.13" |
| 6 | HB_real_000310 | Aggregation & Comparison | Argmax Reasoning | In column Overall, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o", "value": "66.13"} |
| 7 | HB_real_000354 | Constrained Retrieval | Cell Retrieval | What is the Fac. value for row Ours? Output only the number as a string. | "61.28" |
| 8 | HB_real_000356 | Aggregation & Comparison | Pairwise Comparison | In column Con., compare Ours and GPT-4o. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "75.73"} |
| 9 | HB_real_000576 | Aggregation & Comparison | Extremum Reasoning | For row Ours, what is the maximum value across all numeric metrics? Output only the number as a string. | "75.73" |
| 10 | HB_real_000681 | Markup Grounding | Bold Presence | Is the value 65.84 in column Overall boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 11 | HB_real_000724 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 12 | HB_real_001756 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000041 | Aggregation & Comparison | Grouped Aggregation | According to the table, among rows with explicit Acc. delta gains, which row has the maximum Acc. delta? Return the row index as a string (top-to-bottom among those rows). | "1" |
| 2 | HB_real_000042 | Constrained Retrieval | Value Filtering | For the Acc. delta values, considering only rows with explicit green delta gains, what is the value at row 1? Output the full string. | "+3.80" |
| 3 | HB_real_000371 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=3,c=5) in the numeric/text data part of the table. Output only the value as a string. | "5G +14G +14G" |
| 4 | HB_real_000374 | Local Relations | Neighbor Indexing | Using the table only, at position (r=3,c=5) in the numeric/text data part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "5G +14G +14G", "up": "0G +14G +15G", "down": "3G +14G +12G", "left": "1.33", "right": "34.76+1.02"} |
| 5 | HB_real_000410 | Constrained Retrieval | Value Filtering | By reading the table in the image, for the delta column Acc. delta (scope: rows with explicit Acc. delta gains), what is the value at row 2? Output the full string. | "+1.02" |
| 6 | HB_real_000586 | Constrained Retrieval | Value Filtering | According to the table image, considering only rows with explicit Acc. delta gains (green +x values) in order, list all delta values in the Acc. column. Output only a JSON array of strings. | ["+3.80", "+1.02", "+0.29"] |
| 7 | HB_real_000738 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric Acc. value in the table, ignoring the green delta annotations. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 2, "col": 6, "value": "37.54"} |
| 8 | HB_real_000760 | Constrained Retrieval | Value Filtering | For the Acc. delta values, considering only rows with explicit green delta gains, what is the value at row 3? Output the full string. | "+0.29" |
| 9 | HB_real_001077 | Markup Grounding | Marked Value Listing | According to the table image, list text spans whose text is colored in red in this table. Output only a JSON array of strings. | ["5G", "3G"] |
| 10 | HB_real_001757 | Markup Grounding | Marked Value Listing | List all green numeric text spans in this table. Output only a JSON array of strings. | ["14G", "12G", "+3.80", "+1.02", "+0.29"] |
| 11 | HB_real_001759 | Aggregation & Comparison | Grouped Aggregation | Among rows with explicit Acc. delta gains, which row has the minimum Acc. delta? Return the row index as a string, counting those rows from top to bottom. | "3" |
| 12 | HB_real_001760 | Constrained Retrieval | Cell Retrieval | Using the table only, find the Acc. cell value at position (r=4,c=5) in the numeric/text data part of the table, including the visible green delta annotation. Output only the value as a string. | "34.13+0.29" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000255 | Markup Grounding | Marked Value Listing | From this table, list numbers whose text is colored in green in this table. Output only a JSON array of strings. | ["28.26", "29.45", "29.23", "30.03"] |
| 2 | HB_real_000916 | Aggregation & Comparison | Grouped Aggregation | Within the GPT-4o rows, what is the mean value of column Acc.? Output only the number as a string, rounded to two decimal places. | "28.19" |
| 3 | HB_real_001070 | Markup Grounding | Marked Value Listing | List green numbers in the GPT-4o rows only. Output only a JSON array of strings. | ["28.26", "29.45"] |
| 4 | HB_real_001762 | Aggregation & Comparison | Best Selection | Within group "GPT-4o", which row label in the leftmost row-label column is best on column Acc.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o / 1B/1.5B", "value": "29.45"} |
| 5 | HB_real_001763 | Consistency & Missingness | Column Missingness | Which row labels have missing entries marked with "-" in column TFLOPs? Output only a JSON array of row-label strings. | ["GPT-4o / Uniform", "GPT-4o / 1B/0.5B", "GPT-4o / 1B/1.5B"] |
| 6 | HB_real_001764 | Markup Grounding | Marked Value Listing | List green numbers in the InternVL2 rows only. Output only a JSON array of strings. | ["29.23", "30.03"] |
| 7 | HB_real_001765 | Markup Grounding | Marked Value Listing | List numbers whose text is colored in red in this table. Output only a JSON array of strings. | ["23.18", "23.85", "24.20", "24.26"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000213 | Aggregation & Comparison | Extremum Reasoning | For row MPNN, what is the minimum value across all metrics? Output only the number as a string. | "0.146" |
| 2 | HB_real_000487 | Aggregation & Comparison | Extremum Reasoning | For row MPNN, what is the maximum value across all metrics? Output only the number as a string. | "0.358" |
| 3 | HB_real_000492 | Aggregation & Comparison | Rank Selection | In column hom(F,G) column 1, who ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Subgraph GNN", "value": "0.011"} |
| 4 | HB_real_000836 | Local Relations | Neighbor Indexing | At position (r=1,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.358", "up": null, "down": "0.010", "left": "0.254", "right": "0.208"} |
| 5 | HB_real_000842 | Aggregation & Comparison | Grouped Aggregation | What is the mean value across the three hom(F,G) columns for row MPNN? Output the value rounded to three decimal places as a string. | "0.262" |
| 6 | HB_real_001770 | Aggregation & Comparison | Best Selection | Within group hom(F,G), which row label in the leftmost row-label column is best on column 1? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "MPNN", "value": "0.300"} |
| 7 | HB_real_001771 | Aggregation & Comparison | Argmax Reasoning | In column sub(F,G) column 1, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "MPNN", "value": "0.358"} |
| 8 | HB_real_001772 | Aggregation & Comparison | Pairwise Comparison | In column sub(F,G) column 2, compare Subgraph GNN and Local 2-GNN. Which value is larger? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Subgraph GNN", "value": "0.020"} |
| 9 | HB_real_001773 | Aggregation & Comparison | Extremum Reasoning | For item r-ℓGIN, what is the minimum value across all metrics? Output only the number as a string. | "0.0003" |
| 10 | HB_real_001774 | Aggregation & Comparison | Pairwise Comparison | For row r-ℓGIN, compare column hom(F,G) column 1 and column sub(F,G) column 1. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "hom(F,G) column 1", "value": "0.001"} |
| 11 | HB_real_001775 | Aggregation & Comparison | Argmax Reasoning | In column sub(F,G) column 3, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 1, "col": 9, "value": "0.205"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000219 | Markup Grounding | Marked Column Listing | From this table, list all underlined numbers in column ZINC12K. Output only a JSON array of strings. | ["0.079 ± 0.006"] |
| 2 | HB_real_000221 | Markup Grounding | Marked Column Listing | List all underlined numbers in column ZINC250K. Output only a JSON array of strings. | ["0.024 ± 0.003"] |
| 3 | HB_real_000428 | Consistency & Missingness | Column Missingness | In column ZINC250K, which row labels have missing entries marked as "-"? Output only a JSON array of strings. | ["GCN", "GAT", "GSN", "SUN", "GNNAK+", "PathNN"] |
| 4 | HB_real_000541 | Markup Grounding | Marked Value Listing | List all underlined numbers in the table. Output only a JSON array of strings. | ["0.079 ± 0.006", "0.024 ± 0.003"] |
| 5 | HB_real_000543 | Markup Grounding | Underline Presence | Is the number with value 0.024 ± 0.003 underlined? Answer JSON {"underlined":true/false}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 6 | HB_real_000546 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "2" |
| 7 | HB_real_000604 | Constrained Retrieval | Cell Retrieval | Find the underlined value in column ZINC12K. Output only the number as a string. | "0.079 ± 0.006" |
| 8 | HB_real_000787 | Local Relations | Marker-Anchor Neighbor | According to the table, find the right neighbor of the underlined cell with value 0.079 ± 0.006. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "0.022 ± 0.002" |
| 9 | HB_real_000893 | Markup Grounding | Marked Value Listing | List all numbers with green box or green filled highlight. Output only a JSON array of strings. | ["0.022 ± 0.002", "0.077 ± 0.002", "0.023 ± 0.001", "0.072 ± 0.002", "0.022 ± 0.001"] |
| 10 | HB_real_000894 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.022 ± 0.002, list all numbers highlighted with the same green filled color as the anchor. Output only a JSON array of strings. | ["0.022 ± 0.002"] |
| 11 | HB_real_001776 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column ZINC12K? Exclude N/A and —. Output only the number as a string. | "0.072 ± 0.002" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000180 | Aggregation & Comparison | Grouped Aggregation | Using the table only, take the cells highlighted in deep green. What is their max? Exclude N/A and —. Output only the number as a string. | "13.21" |
| 2 | HB_real_000182 | Markup Grounding | Marker Counting | How many underlined numeric cells are in the table? Output only the count as a string. | "12" |
| 3 | HB_real_000183 | Local Relations | Neighbor Value Tracing | The value 13.21 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "15.04", "right": null} |
| 4 | HB_real_000449 | Constrained Retrieval | Value Filtering | Among deep-green highlighted cells, list the values greater than 1.0. Output only a JSON array of strings. | ["13.21"] |
| 5 | HB_real_000548 | Markup Grounding | Marked Column Listing | By reading the table in the image, list all underlined numbers in column DRFWL. Output only a JSON array of strings. | ["0.153", "0.145", "0.156", "0.0901"] |
| 6 | HB_real_001067 | Constrained Retrieval | Color-Anchor Retrieval | According to the table image, for the number with value 0.217, list all numbers highlighted with the same deep-green color as the anchor. Output only a JSON array of strings. | ["0.244", "0.217", "0.00205", "0.00216", "0.00321", "13.21", "0.000127", "0.0418", "0.023", "0.0352", "0.0118", "0.0702"] |
| 7 | HB_real_001781 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in row ZVPE? Exclude N/A and —. Output only the number as a string. | "0.000127 ±0.000003" |
| 8 | HB_real_001782 | Markup Grounding | Underline Presence | Is the number with value 0.230 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 9 | HB_real_002259 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in deep green in this table? Output only the count as a string. | "12" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000204 | Markup Grounding | Marker Counting | According to the table, how many underlined numbers are in the table? Output only the count as a string. | "0" |
| 2 | HB_real_000364 | Consistency & Missingness | Column Missingness | Within group "VQA Benchmark", which row labels have missing entries marked with "-" in column Camera? Output only a JSON array of row-label strings. | ["NExT-QA (Xiao et al., 2021)", "EgoSchema (Mangalam et al., 2023)", "PerceptionTest (Patraucean et al., 2024)", "MVBench (Li et al., 2024d)", "Video-MME (Fu et al., 2024)", "MMBench-Video (Fang et al., 2024)", "VideoVista (Li et al., 2024e)", "TOMATO (Shangguan et al., 2024)"] |
| 3 | HB_real_000366 | Aggregation & Comparison | Best Selection | Within group "Multi-task Benchmark", which row label in the leftmost row-label column is best on column #Samp.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "E.T.Bench (Liu et al., 2024e)", "value": "7,289"} |
| 4 | HB_real_000729 | Markup Grounding | Marked Value Listing | How many cells in the row TUNA are visually marked with green check marks? Output only the count as a string. | "9" |
| 5 | HB_real_001785 | Aggregation & Comparison | Best Selection | Within group "VQA Benchmark", which row label in the leftmost row-label column is best on column #Videos? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "PerceptionTest (Patraucean et al., 2024)", "value": "11,620"} |
| 6 | HB_real_001786 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column #Samp.? Output only the number as a string. | "44,000" |
| 7 | HB_real_001787 | Aggregation & Comparison | Pairwise Comparison | In column #Videos, compare Video-MME (Fu et al., 2024) and MMBench-Video (Fang et al., 2024). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Video-MME (Fu et al., 2024)", "value": "900"} |
| 8 | HB_real_001788 | Constrained Retrieval | Cell Retrieval | What is the value at row TUNA, column Fine.? Output only the cell text. | "✓" |
| 9 | HB_real_001789 | Consistency & Missingness | Cell Missingness | Is the cell (row DREAM-1K (Wang et al., 2024a), column Global) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 10 | HB_real_001790 | Consistency & Missingness | Cell Missingness | Is the cell (row TUNA, column Global) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000051 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": false, "color": null} |
| 2 | HB_real_000058 | Markup Grounding | Bold Presence | Does the numeric cell at position (r=3,c=5) in the numeric part of the table contain any boldfaced number? Answer JSON {"contains_bold":true/false}. | {"contains_bold": true} |
| 3 | HB_real_000231 | Local Relations | Neighbor Indexing | At position (r=2,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "56.5/34.4/40.6", "up": "52.2/30.9/36.6", "down": "45.3/28.2/33.1", "left": "52.3/26.0/32.5", "right": "38.9/17.2/22.0"} |
| 4 | HB_real_000328 | Constrained Retrieval | Cell Retrieval | According to the image table, find the value at (r=1, c=1) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and left text columns. Output only the number as a string. | "49.4/22.6/28.9" |
| 5 | HB_real_000333 | Markup Grounding | Underline Presence | Is the numeric cell at position (r=2,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 6 | HB_real_000591 | Markup Grounding | Underline Presence | Does the numeric cell at position (r=6,c=2) in the numeric part of the table contain any underlined number? Answer JSON {"contains_underlined":true/false}. | {"contains_underlined": true} |
| 7 | HB_real_000701 | Markup Grounding | Bold Presence | Is the numeric cell at position (r=1,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 8 | HB_real_000972 | Markup Grounding | Marked Column Listing | In column Overall, list every underlined numeric value. Output only a JSON array of strings. | ["51.7", "77.1", "73.7", "41.1", "51.7", "73.7", "48.1", "57.4"] |
| 9 | HB_real_001791 | Constrained Retrieval | Cell Retrieval | What is the value at row GPT-4o, column Overall? Output only the cell text. | "77.7/48.2/58.5" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000218 | Markup Grounding | Marked Column Listing | Using the table only, list all underlined numbers in column Overall. Output only a JSON array of strings. | ["60.0", "60.3"] |
| 2 | HB_real_000222 | Aggregation & Comparison | Extremum Reasoning | In column Overall, what is the maximum value? Output only the number as a string. | "60.8" |
| 3 | HB_real_000539 | Aggregation & Comparison | Argmax Reasoning | In column Overall, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Gemini 1.5 Pro", "value": "60.8"} |
| 4 | HB_real_000540 | Markup Grounding | Marked Column Listing | List all bold numbers in column Overall. Output only a JSON array of strings. | ["60.7", "60.7", "60.8"] |
| 5 | HB_real_000545 | Constrained Retrieval | Cell Retrieval | What is the value at column Description (block Background Scene) for row GPT-4o? Output only the number as a string. | "81.5" |
| 6 | HB_real_000891 | Markup Grounding | Underline Presence | Is the number with value 61.9 underlined? Answer JSON {"underlined":true/false}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 7 | HB_real_000892 | Markup Grounding | Marker Counting | How many underlined numeric cells are in column Overall? Output only the count as a string. | "2" |
| 8 | HB_real_001797 | Aggregation & Comparison | Pairwise Comparison | For row GPT-4o, compare column Description and column Transition (block Background Scene). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Description", "value": "81.5"} |
| 9 | HB_real_001798 | Constrained Retrieval | Cell Retrieval | What is the Object Attribute Location value for row Qwen2-VL-72B? Output only the number as a string. | "66.7" |
| 10 | HB_real_001800 | Markup Grounding | Highlight Presence | Is the number with value 53.9 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000567 | Consistency & Missingness | Cell Missingness | From this table image, is the cell (row Random Choice, column L-3 (EM) (block Triple Granularity)) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 2 | HB_real_001801 | Aggregation & Comparison | Argmax Reasoning | In column L-3 (GPT) (block Subgraph Granularity), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "MLP RotateE", "value": "55.80"} |
| 3 | HB_real_001802 | Aggregation & Comparison | Best Selection | Within group "MLP", which row label in the leftmost row-label column is best on column L-2 (Acc) (block Subgraph Granularity)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "R-GCN", "value": "91.05"} |
| 4 | HB_real_001803 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column L-3 (EM) (block Triple Granularity)? Output only the number as a string. | "97.91" |
| 5 | HB_real_001805 | Constrained Retrieval | Value Filtering | In column L-2 (Acc) (block Triple Granularity), list all row labels in the primary row-label column with value >= 94. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Q-former TransE", "value": "94.42"}, {"item": "Q-former DistMult", "value": "94.11"}, {"item": "Q-former RotateE", "value": "94.02"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000038 | Aggregation & Comparison | Grouped Aggregation | Ignoring colored delta/gain text, among non-ZeroShot methods, which row has the maximum main score in column ViT-S/16 (6B)? Return the row index as a string (top-to-bottom among [Prototypical, Default Tip, XVal Tip, Plurality NN, Rank NN, Softmax NN]). | "5" |
| 2 | HB_real_000047 | Aggregation & Comparison | Grouped Aggregation | Ignoring colored delta/gain text, among non-ZeroShot methods, which row has the minimum main score in column ViT-S/16 (6B)? Return the row index as a string, counting [Prototypical, Default Tip, XVal Tip, Plurality NN, Rank NN, Softmax NN] from top to bottom. | "1" |
| 3 | HB_real_000369 | Constrained Retrieval | Cell Retrieval | Ignoring colored delta/gain text, using the table only, find the value at position (r=4,c=2) in the numeric part of the table. Output only the value as a string. | "67.8 ± 0.3" |
| 4 | HB_real_000378 | Local Relations | Neighbor Indexing | Ignoring colored delta/gain text, using the table only, at position (r=4,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "67.8 ± 0.3", "up": "63.6 ± 0.3", "down": "63.8 ± 0.2", "left": "64.7 ± 0.2", "right": "73.8 ± 0.2"} |
| 5 | HB_real_000409 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, for column ViT-S/16 (6B) among non-ZeroShot methods ordered as [Prototypical, Default Tip, XVal Tip, Plurality NN, Rank NN, Softmax NN], what is the main score at row 3? Output the full string. | "67.8 ± 0.3" |
| 6 | HB_real_000761 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, for column ViT-S/16 (6B) among non-ZeroShot methods ordered as [Prototypical, Default Tip, XVal Tip, Plurality NN, Rank NN, Softmax NN], what is the main score at row 5? Output the full string. | "68.1 ± 0.1" |
| 7 | HB_real_001078 | Markup Grounding | Marked Value Listing | Ignoring colored delta/gain text, list the main scores in the ZeroShot row from left to right. Output only a JSON array of strings. | ["46.9", "52.1", "60.3", "62.5", "64.1"] |
| 8 | HB_real_001806 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, for the Default Tip row, list the main scores across columns in order (ViT-S/16 1.5B, ViT-S/16 6B, ViT-B/16 6B, ViT-B/16 15B, ViT-L/16 8B). Output only a JSON array of strings. | ["60.3 ± 0.1", "63.6 ± 0.3", "69.5 ± 0.2", "70.2 ± 0.3", "73.2 ± 0.3"] |
| 9 | HB_real_001807 | Consistency & Missingness | Consistency Probing | Ignoring colored delta/gain text, in the ZeroShot row, list all columns whose main score is greater than 60. Return JSON array of {"item":"<column label>","value":"<cell value>"}. | [{"item": "ViT-B/16 (6B)", "value": "60.3"}, {"item": "ViT-B/16 (15B)", "value": "62.5"}, {"item": "ViT-L/16 (8B)", "value": "64.1"}] |
| 10 | HB_real_001808 | Aggregation & Comparison | Argmax Reasoning | Find the single largest main numeric score in the numeric part of the table, ignoring colored delta/gain text. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 4, "col": 5, "value": "77.0 ± 0.3"} |
| 11 | HB_real_001809 | Aggregation & Comparison | Grouped Aggregation | Ignoring colored delta/gain text, within row Softmax NN, what is the mean of the five main scores? Output only the number as a string, rounded to two decimal places. | "70.70" |
| 12 | HB_real_001810 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, in the ZeroShot row, what is the main score in column ViT-B/16 (15B)? Output only the number as a string. | "62.5" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000001 | Aggregation & Comparison | Argmax Reasoning | Based on the table, find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 12, "col": 4, "value": "94.4"} |
| 2 | HB_real_000005 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column DTD. Output JSON {"row":<int>,"col":<int>,"value":"<string>"}. The numeric part of the table excludes headers and left text columns. | {"row": 12, "col": 2, "value": "76.7"} |
| 3 | HB_real_000281 | Local Relations | Neighbor Value Tracing | The value 94.4 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "92.6", "right": "92.8"} |
| 4 | HB_real_000285 | Consistency & Missingness | Consistency Probing | Compare SigLixP against DMN across all dataset columns. List the columns where SigLixP has a higher value. Output only a JSON array of column labels. | ["IN-1K", "DTD", "Food101", "Pets", "Cars"] |
| 5 | HB_real_000286 | Consistency & Missingness | Cell Missingness | Is the cell at row TIP-X [84], column DTD missing (shown as "-")? Answer JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_real_000346 | Aggregation & Comparison | Argmax Reasoning | Based on the table image, find the location (row,col) and value of the maximum cell in column Cars. Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 12, "col": 5, "value": "92.8"} |
| 7 | HB_real_001814 | Markup Grounding | Bold Presence | Is the number at (r=12,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric columns Method/Train-free. | {"bold": true} |
| 8 | HB_real_001815 | Constrained Retrieval | Cell Retrieval | Find the value at (r=4, c=2) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and left text columns. Output only the number as a string. | "71.9" |
| 9 | HB_real_001816 | Consistency & Missingness | Column Missingness | Which numeric columns contain at least one missing entry marked as "-"? Output only a JSON array of column names. | ["DTD", "Food101", "Pets", "Cars"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000037 | Aggregation & Comparison | Grouped Aggregation | Ignoring colored delta/gain text, among non-ZeroShot methods, which row has the maximum main score in column ViT-S/16 (6B)? Return the row index as a string (top-to-bottom among [Prototypical, Default Tip, Softmax NN]). | "3" |
| 2 | HB_real_000046 | Aggregation & Comparison | Grouped Aggregation | Ignoring colored delta/gain text, among non-ZeroShot methods, which row has the minimum main score in column ViT-S/16 (6B)? Return the row index as a string, counting [Prototypical, Default Tip, Softmax NN] from top to bottom. | "1" |
| 3 | HB_real_000370 | Constrained Retrieval | Cell Retrieval | Ignoring colored delta/gain text, using the table only, find the value at position (r=4,c=5) in the numeric part of the table. Output only the value as a string. | "73.1 ± 0.1" |
| 4 | HB_real_000376 | Local Relations | Neighbor Indexing | Ignoring colored delta/gain text, using the table only, at position (r=4,c=5) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "73.1 ± 0.1", "up": "69.5 ± 0.2", "down": null, "left": "72.0 ± 0.2", "right": null} |
| 5 | HB_real_000407 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, for column ViT-S/16 (6B) among non-ZeroShot methods ordered as [Prototypical, Default Tip, Softmax NN], what is the main score at row 1? Output the full string. | "59.5 ± 0.3" |
| 6 | HB_real_000649 | Markup Grounding | Marked Value Listing | Ignoring colored delta/gain text, list the main scores in the ZeroShot row from left to right. Output only a JSON array of strings. | ["47.9", "52.1", "53.3", "60.8", "61.9"] |
| 7 | HB_real_000742 | Aggregation & Comparison | Argmax Reasoning | Find the single largest main numeric score in the numeric part of the table, ignoring colored delta/gain text. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 4, "col": 5, "value": "73.1 ± 0.1"} |
| 8 | HB_real_001076 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, in the ZeroShot row, what is the main score in column ViT-S/16 (15B)? Output only the number as a string. | "53.3" |
| 9 | HB_real_001817 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, for the Softmax NN row, list the main scores across columns in order (ViT-S/16 1.5B, ViT-S/16 6B, ViT-S/16 15B, ViT-B/16 6B, ViT-B/16 15B). Output only a JSON array of strings. | ["64.5 ± 0.2", "66.8 ± 0.1", "68.2 ± 0.2", "72.0 ± 0.2", "73.1 ± 0.1"] |
| 10 | HB_real_001818 | Consistency & Missingness | Consistency Probing | Ignoring colored delta/gain text, in the ZeroShot row, list all columns whose main score is greater than 60. Return JSON array of {"item":"<column label>","value":"<cell value>"}. | [{"item": "ViT-B/16 (6B)", "value": "60.8"}, {"item": "ViT-B/16 (15B)", "value": "61.9"}] |
| 11 | HB_real_001819 | Aggregation & Comparison | Grouped Aggregation | Ignoring colored delta/gain text, within the Softmax NN row, what is the mean of the five main scores? Output only the number as a string, rounded to two decimal places. | "68.92" |
| 12 | HB_real_001820 | Constrained Retrieval | Value Filtering | Ignoring colored delta/gain text, for column ViT-S/16 (6B) among non-ZeroShot methods ordered as [Prototypical, Default Tip, Softmax NN], what is the main score at row 3? Output the full string. | "66.8 ± 0.1" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001822 | Aggregation & Comparison | Best Selection | Within group "(a) Self-Attention Masking", which row label in the leftmost row-label column is best on column 16-Shot? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "SigLIxP", "value": "64.1 ± 0.5"} |
| 2 | HB_real_001823 | Aggregation & Comparison | Best Selection | Within group "(b) Relative weighting", which row label in the leftmost row-label column is best on column 16-Shot? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "α = 0.9", "value": "64.1 ± 0.5"} |
| 3 | HB_real_001824 | Aggregation & Comparison | Best Selection | Within group "(c) Contextualization Type", which row label in the leftmost row-label column is best on column 16-Shot? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Single-Stage", "value": "64.1 ± 0.5"} |
| 4 | HB_real_001825 | Aggregation & Comparison | Best Selection | Within group "(d) Uncoupled, learnable temperatures", which row label in the leftmost row-label column is best on column 16-Shot? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "τ1, τ2, τctx learnable", "value": "64.1 ± 0.5"} |
| 5 | HB_real_001826 | Aggregation & Comparison | Best Selection | Within group "(e) Back-propagation through memory buffer", which row label in the leftmost row-label column is best on column 16-Shot? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Full back-propagation", "value": "64.1 ± 0.5"} |
| 6 | HB_real_001827 | Aggregation & Comparison | Pairwise Comparison | For row τ1, τ2, τctx learnable (block (d) Uncoupled, learnable temperatures), compare column Zero-Shot and column 16-Shot. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "16-Shot", "value": "64.1 ± 0.5"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000053 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| 2 | HB_real_000056 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 3 | HB_real_000070 | Markup Grounding | Marked Value Listing | Please answer this question: return all numbers highlighted in light blue. Output only a JSON array of strings. | ["51.0", "60.1 ± 0.4", "51.0", "60.1 ± 0.4"] |
| 4 | HB_real_000133 | Markup Grounding | Underline Presence | Please answer this question: is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric columns. | {"underlined": false} |
| 5 | HB_real_000139 | Aggregation & Comparison | Grouped Aggregation | In subtable (b) Value Heads for M_V, what is the mean Zero-Shot value across all listed methods? Output only the number as a string, rounded to one decimal place. | "49.9" |
| 6 | HB_real_000504 | Constrained Retrieval | Value Filtering | In subtable (a) Context Batch Separation, what is the difference between SigLIP 16-Shot and SigLIP Zero-Shot? Output only the number as a string rounded to one decimal place. | "9.1" |
| 7 | HB_real_000593 | Constrained Retrieval | Value Filtering | Among light-blue highlighted cells, list the values whose numeric part is greater than 55. Output only a JSON array of strings, preserving reading order. | ["60.1 ± 0.4", "60.1 ± 0.4"] |
| 8 | HB_real_001000 | Markup Grounding | Marker Counting | From this highlighted table, how many numeric cells are highlighted in light blue in this table? Output only the count as a string. | "4" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000190 | Constrained Retrieval | Color-Anchor Retrieval | Using the table only, for the number at row 8, column 1 in the numeric part of the table, list all numbers highlighted with the same deep-pink color as the anchor. The numeric part of the table excludes header and left text columns. Output only a JSON array of strings. | ["0.40", "22.15", "0.342", "0.399"] |
| 2 | HB_real_000194 | Constrained Retrieval | Value Filtering | Among deep-pink highlighted cells, list the values that are greater than 1.0. Output only a JSON array of strings. | ["22.15"] |
| 3 | HB_real_000230 | Local Relations | Neighbor Indexing | Based on the table, at (r=6,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and left text columns. | {"target": "0.416", "up": "0.444", "down": "0.425", "left": "0.375", "right": "23.16"} |
| 4 | HB_real_000232 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column CD↓? Exclude N/A and —. Output only the number as a string. | "0.342" |
| 5 | HB_real_000459 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column JSD↓? Exclude N/A and —. Output only the number as a string. | "0.399" |
| 6 | HB_real_000460 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column EMD↓. Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and the left text column. | {"row": 8, "col": 3, "value": "23.26"} |
| 7 | HB_real_000462 | Consistency & Missingness | Column Missingness | In column EMD↓, which row labels have missing entries marked as "-"? Output only a JSON array of strings. | ["LMSCNet [30]", "LODE [16]", "MID [38]", "PVD [55]"] |
| 8 | HB_real_000822 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are > 23. Output JSON array of strings. | ["23.16", "23.14"] |
| 9 | HB_real_000958 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in the table? Count any highlighted color. Output only the count as a string. | "12" |
| 10 | HB_real_001833 | Constrained Retrieval | Cell Retrieval | Find the value at (r=6, c=3) in the numeric part of the table. Output only the number as a string. | "23.16" |
| 11 | HB_real_001837 | Consistency & Missingness | Cell Missingness | Is the cell (row LODE [16], column EMD↓) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000003 | Aggregation & Comparison | Argmax Reasoning | From this table, find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 5, "col": 4, "value": "29.43"} |
| 2 | HB_real_000004 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in the table? Output only the count as a string. | "12" |
| 3 | HB_real_000008 | Local Relations | Neighbor Value Tracing | The value 29.43 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "22.96", "right": null} |
| 4 | HB_real_000282 | Aggregation & Comparison | Extremum Reasoning | For row LiDiff (Refined) [28], what is the minimum value across CD, JSD, EMD, and Times (s)? Output only the number as a string. | "0.446" |
| 5 | HB_real_000283 | Constrained Retrieval | Value Filtering | Among red highlighted cells, list the values greater than 1. Output only a JSON array of strings. | ["21.98", "4.98", "5.14"] |
| 6 | HB_real_000652 | Markup Grounding | Underline Presence | From this table image, is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric columns. | {"underlined": false} |
| 7 | HB_real_001839 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column CD. Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 2, "col": 1, "value": "1.565"} |
| 8 | HB_real_001840 | Constrained Retrieval | Value Filtering | Among highlighted cells (any highlighted color), list the values that are > 23. Output JSON array of strings. | ["29.18"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000063 | Markup Grounding | Marked Value Listing | Read the table carefully and list numbers highlighted in deep pink. Output only a JSON array of strings. | ["22.15", "0.326", "0.379", "1.10"] |
| 2 | HB_real_000065 | Constrained Retrieval | Cell Retrieval | What is the JSD↓ value for ScoreLiDAR (4 Steps Refined)? Output only the cell value as a string. | "0.386" |
| 3 | HB_real_000187 | Aggregation & Comparison | Grouped Aggregation | From this table, take the cells highlighted in deep pink. What is their max? Exclude N/A and —. Output only the number as a string. | "22.15" |
| 4 | HB_real_000448 | Aggregation & Comparison | Extremum Reasoning | In column CD↓, what is the smallest numeric value? Lower values are better. Output only the number as a string. | "0.326" |
| 5 | HB_real_000455 | Aggregation & Comparison | Pairwise Comparison | For metric CD↓, which row has the smaller value: ScoreLiDAR (4 Steps Refined) or ScoreLiDAR (8 Steps Refined)? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "ScoreLiDAR (4 Steps Refined)", "value": "0.326"} |
| 6 | HB_real_000638 | Aggregation & Comparison | Extremum Reasoning | In column EMD↓, what is the largest numeric value, ignoring missing entries marked as "-"? Output only the number as a string. | "25.74" |
| 7 | HB_real_000774 | Local Relations | Neighbor Value Tracing | The value 22.15 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "0.444", "right": "30.38"} |
| 8 | HB_real_001001 | Markup Grounding | Marker Counting | Using visual markers in the table, how many numeric cells are highlighted in deep pink in this table? Output only the count as a string. | "4" |
| 9 | HB_real_001844 | Aggregation & Comparison | Best Selection | In metric Time (s)↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "ScoreLiDAR (1 Steps Refined)", "value": "1.10"} |
| 10 | HB_real_001845 | Consistency & Missingness | Column Missingness | In column EMD↓, which row labels have missing entries marked as "-"? Output only a JSON array of strings. | ["ScoreLiDAR (2 Steps Refined)", "ScoreLiDAR (1 Steps Refined)"] |
| 11 | HB_real_002262 | Constrained Retrieval | Value Filtering | Among deep-pink highlighted cells, list the values that are > 1.0. Output JSON array of strings. | ["22.15", "1.10"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001847 | Aggregation & Comparison | Argmax Reasoning | In column VQA Volume, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "RSVQA-HR [35]", "value": "1,066,316"} |
| 2 | HB_real_001848 | Constrained Retrieval | Value Filtering | In column VQA Volume, list all row labels in the primary row-label column with value >= 100000. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "RSVQA-HR [35]", "value": "1,066,316"}, {"item": "RSIVQA [78]", "value": "111,134"}, {"item": "VRSBench [28]", "value": "123,231"}] |
| 3 | HB_real_001849 | Constrained Retrieval | Cell Retrieval | What is the value at row XLRS-Bench, column Visual Grounding Annotation Method? Output only the cell text. | "✔" |
| 4 | HB_real_001850 | Consistency & Missingness | Cell Missingness | Is the cell (row RSVG [53], column VQA Volume) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000009 | Aggregation & Comparison | Grouped Aggregation | Based on the table, exclude cells that are boldfaced. On the remaining cells in column Avg., what is the max? Exclude N/A and —. Output only the number as a string. | "40.58" |
| 2 | HB_real_000016 | Markup Grounding | Marked Column Listing | In column Avg., list all boldfaced numeric values. Output only a JSON array of strings. | ["39.80", "41.10"] |
| 3 | HB_real_000019 | Markup Grounding | Marker Counting | How many numeric cells are boldfaced in the whole table? Output only the count as a string. | "18" |
| 4 | HB_real_000437 | Constrained Retrieval | Cell Retrieval | Within the zh group, what is the Avg. value for Qwen2-VL? Output only the number as a string. | "38.99" |
| 5 | HB_real_000438 | Constrained Retrieval | Cell Retrieval | Find the value at (r=3, c=12) in the numeric part of the table. Output only the number as a string. | "39.80" |
| 6 | HB_real_000443 | Constrained Retrieval | Cell Retrieval | Within the zh group, what is the Planning value for GPT-4o-mini? Output only the cell value as a string. | "41.86" |
| 7 | HB_real_000642 | Aggregation & Comparison | Grouped Aggregation | From this table image, take the cells that are boldfaced. What is their max? Exclude N/A and —. Output only the number as a string. | "76.57" |
| 8 | HB_real_001851 | Aggregation & Comparison | Best Selection | Within group "en", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CogVLM2", "value": "39.80"} |
| 9 | HB_real_001852 | Aggregation & Comparison | Best Selection | Within group "zh", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Qwen2-VL", "value": "41.10"} |
| 10 | HB_real_001853 | Consistency & Missingness | Cell Missingness | Is the cell (row GeoChat (en), column Spatiotemporal Reasoning) missing (N/A or — or -)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000113 | Aggregation & Comparison | Grouped Aggregation | According to the table, within "en", what is the mean value of column BLEU-1? Exclude N/A and —. Output only the number as a string, rounded to two decimal places. | "31.22" |
| 2 | HB_real_000114 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are boldfaced. On the remaining cells in column BLEU-1 within group en, what is the max? Exclude N/A and —. Output only the number as a string. | "38.29" |
| 3 | HB_real_000227 | Markup Grounding | Marked Column Listing | Please answer this question: list all bold numbers in column BLEU-4. Output only a JSON array of strings. | ["4.56", "0.98"] |
| 4 | HB_real_000273 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are boldfaced. On the remaining cells in column BLEU-1 within group zh, what is the max? Exclude N/A and —. Output only the number as a string. | "34.13" |
| 5 | HB_real_000279 | Markup Grounding | Marker Counting | How many bold numeric cells are in the table? Output only the count as a string. | "12" |
| 6 | HB_real_000612 | Markup Grounding | Marked Column Listing | In column BLEU-1, list every boldfaced numeric value. Output only a JSON array of strings. | ["41.12", "37.30"] |
| 7 | HB_real_000658 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=5,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 8 | HB_real_000697 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=1,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 9 | HB_real_000978 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 5, "col": 1, "value": "41.12"} |
| 10 | HB_real_001856 | Markup Grounding | Marked Column Listing | In column ROUGE_L, list every boldfaced numeric value. Output only a JSON array of strings. | ["21.30", "37.11"] |
| 11 | HB_real_001857 | Aggregation & Comparison | Grouped Aggregation | Take the cells that are boldfaced in column BLEU-1. What is their max? Exclude N/A and —. Output only the number as a string. | "41.12" |
| 12 | HB_real_001858 | Markup Grounding | Marked Column Listing | In column BLEU-3, list every boldfaced numeric value. Output only a JSON array of strings. | ["9.94", "2.47"] |
| 13 | HB_real_001859 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=13,c=6) in the numeric part of the table. Output only the value as a string. | "37.11" |
| 14 | HB_real_001860 | Aggregation & Comparison | Argmax Reasoning | In column BLEU-2, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 5, "col": 2, "value": "20.42"} |
| 15 | HB_real_001861 | Aggregation & Comparison | Argmax Reasoning | In column ROUGE_L, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 13, "col": 6, "value": "37.11"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000268 | Consistency & Missingness | Column Missingness | Which benchmark/method rows have missing entries marked with "-" in column GeoChat? Output only a JSON array of row-label strings. | ["VRSBench Acc@0.5", "VRSBench Acc@0.7"] |
| 2 | HB_real_000423 | Consistency & Missingness | Column Missingness | Which benchmark blocks have missing entries marked with "-" in column GeoChat? Output only a JSON array of benchmark names. | ["VRSBench"] |
| 3 | HB_real_001862 | Aggregation & Comparison | Best Selection | Within group "XLRS-Bench-EN", which row label in the leftmost row-label column is best on column Acc@0.5? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o*", "value": "0.46"} |
| 4 | HB_real_001863 | Aggregation & Comparison | Best Selection | Within group "XLRS-Bench-ZH", which row label in the leftmost row-label column is best on column Acc@0.5? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o*", "value": "0.45"} |
| 5 | HB_real_001864 | Aggregation & Comparison | Grouped Aggregation | Within XLRS-Bench-EN, what is the mean value of column GPT-4o* across Acc@0.5 and Acc@0.7? Output only the number as a string, rounded to three decimal places. | "0.255" |
| 6 | HB_real_001865 | Constrained Retrieval | Value Filtering | In column GPT-4o-mini, list all row labels in the primary row-label column with value >= 0.20. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "XLRS-Bench-ZH Acc@0.5", "value": "0.21"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000208 | Aggregation & Comparison | Extremum Reasoning | Based on the table, for item B-VGG16-SN (consider only ShoeV2 columns Top1 (%), Top10 (%), FLOPS (G)), what is the minimum value? Output only the number as a string. | "33.03" |
| 2 | HB_real_000482 | Aggregation & Comparison | Extremum Reasoning | For item B-VGG16-SN (consider only ShoeV2 columns Top1 (%), Top10 (%), FLOPS (G)), what is the maximum value? Output only the number as a string. | "78.51" |
| 3 | HB_real_000838 | Aggregation & Comparison | Rank Selection | In block ShoeV2, list the top 2 baseline rows by Top1 (%) value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "B-VGG16-SN", "value": "33.03"}, {"item": "B-VGG16-SN SketchyNetV2", "value": "32.77"}] |
| 4 | HB_real_001866 | Aggregation & Comparison | Argmax Reasoning | In column Top1 (%) within block ShoeV2, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows are counted within the baseline rows in that block. | {"row": 1, "col": 1, "value": "33.03"} |
| 5 | HB_real_001867 | Aggregation & Comparison | Pairwise Comparison | For row B-VGG16-SN (block ShoeV2), compare column Top1 (%) and column Top10 (%). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Top10 (%)", "value": "78.51"} |
| 6 | HB_real_001868 | Aggregation & Comparison | Rank Selection | In block ShoeV2, list the top 2 baseline rows by Top10 (%) value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "B-VGG16-SN", "value": "78.51"}, {"item": "B-VGG16-SN SketchyNetV2", "value": "78.21"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000002 | Aggregation & Comparison | Argmax Reasoning | Using the table only, find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 1, "col": 4, "value": "78.51"} |
| 2 | HB_real_000006 | Constrained Retrieval | Cell Retrieval | What is the ShoeV2 Top10 (%) value for row SketchyNetV2? Output only the number as a string. | "78.21" |
| 3 | HB_real_000007 | Aggregation & Comparison | Pairwise Comparison | In column ShoeV2 Top10 (%), compare B-VGG16-SN and SketchyNetV2. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "∅ B-VGG16-SN", "value": "78.51"} |
| 4 | HB_real_000274 | Aggregation & Comparison | Grouped Aggregation | By reading the table in the image, within "Group I-III", what is the mean value of column ShoeV2 Top1 (%)? Exclude N/A and —. Output only the number as a string, rounded to two decimal places. | "28.67" |
| 5 | HB_real_000287 | Constrained Retrieval | Value Filtering | In column ShoeV2 Top10 (%), list all row labels with value >= 78. Output a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "∅ B-VGG16-SN", "value": "78.51"}, {"item": "III w/o L_F-reward", "value": "78.39"}, {"item": "VII Offset s_v", "value": "78.40"}, {"item": "IX Decoder-Tf", "value": "78.02"}, {"item": "SketchyNetV2", "value": "78.21"}] |
| 6 | HB_real_000289 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Params (mil.)? Exclude N/A and —. Output only the number as a string. | "2.25" |
| 7 | HB_real_000343 | Aggregation & Comparison | Argmax Reasoning | Using the visual table content, find the location (row,col) and value of the maximum cell in column FLOPs (G). Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 1, "col": 1, "value": "40.18"} |
| 8 | HB_real_000666 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in the SketchyNetV2 row? Output only the count as a string. | "4" |
| 9 | HB_real_000669 | Local Relations | Neighbor Value Tracing | The value 78.51 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "33.03", "right": null} |
| 10 | HB_real_001869 | Aggregation & Comparison | Argmax Reasoning | In column ShoeV2 Top1 (%), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "∅ B-VGG16-SN", "value": "33.03"} |
| 11 | HB_real_001873 | Markup Grounding | Bold Presence | Is the number at (r=1,c=3) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric columns Type/Method. | {"bold": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000171 | Constrained Retrieval | Value Filtering | In the ARC-C column of the NQ + ARC-C block, considering only the two Ours rows in order (Ours(Mistral-7B), Ours(Yi-6B)), what is the green parenthesized value at row 2? Output the full string. | "+3.33" |
| 2 | HB_real_000174 | Constrained Retrieval | Value Filtering | In the NQ column of the NQ + ARC-C block, considering only the two Ours rows in order (Ours(Mistral-7B), Ours(Yi-6B)), what is the green parenthesized value at row 2? Output the full string. | "+0.73" |
| 3 | HB_real_000402 | Constrained Retrieval | Value Filtering | In the ARC-C column of the NQ + ARC-C block, considering only the two Ours rows in order (Ours(Mistral-7B), Ours(Yi-6B)), what is the green parenthesized value at row 1? Output the full string. | "+2.05" |
| 4 | HB_real_000405 | Aggregation & Comparison | Grouped Aggregation | Among the two Ours rows in the NQ + ARC-C block, which row has the largest green parenthesized value in column ARC-C? Return the row index as a string (top-to-bottom among Ours(Mistral-7B), Ours(Yi-6B)). | "2" |
| 5 | HB_real_000584 | Constrained Retrieval | Value Filtering | Considering only the two Ours rows in the NQ + ARC-C block in order (Ours(Mistral-7B), Ours(Yi-6B)), list the green parenthesized values in column ARC-C. Output only a JSON array of strings. | ["+2.05", "+3.33"] |
| 6 | HB_real_000754 | Aggregation & Comparison | Argmax Reasoning | In column ARC-C within block NQ + ARC-C, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows are counted within that block and columns within the block numeric part. | {"row": 6, "col": 2, "value": "77.86"} |
| 7 | HB_real_001875 | Consistency & Missingness | Consistency Probing | In the NQ column of the NQ + ARC-C block, considering only the two Ours rows, list all rows whose parenthesized value is green and positive. Return JSON array of {"item":"...","value":"..."}. | [{"item": "Ours(Mistral-7B)", "value": "+0.51"}, {"item": "Ours(Yi-6B)", "value": "+0.73"}] |
| 8 | HB_real_001876 | Constrained Retrieval | Value Filtering | In the NQ column of the NQ + ARC-C block, considering only the two Ours rows in order (Ours(Mistral-7B), Ours(Yi-6B)), what is the green parenthesized value at row 1? Output the full string. | "+0.51" |
| 9 | HB_real_001877 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 10 | HB_real_rebal_000047 | Aggregation & Comparison | Grouped Aggregation | In the GSM column of the left GSM/PIQA block, considering the five colored parenthesized values from top to bottom, which row has the largest positive green value? Return the row index as a string. | "3" |
| 11 | HB_real_rebal_000071 | Markup Grounding | Marked Value Listing | In the left GSM/PIQA block, list all red delta annotations from top to bottom. Output only a JSON array of strings. | ["-4.09", "-4.67", "-4.31"] |
| 12 | HB_real_rebal_000086 | Consistency & Missingness | Consistency Probing | In the right ARC-C column, list all rows with positive green delta annotations. Return a JSON array of {"item":"<row label>","value":"<delta>"}. | [{"item": "DeePen(Mistral-7B)", "value": "+1.01"}, {"item": "DeePen(Yi-6B)", "value": "+4.65"}, {"item": "Ours(Mistral-7B)", "value": "+2.05"}, {"item": "Ours(Yi-6B)", "value": "+3.33"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000062 | Markup Grounding | Marked Value Listing | From this table, list numbers highlighted in light green. Output only a JSON array of strings. | ["73.46", "67.24", "87.50", "62.13", "78.70", "31.78", "66.54"] |
| 2 | HB_real_000066 | Local Relations | Neighbor Value Tracing | The value 87.50 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "73.31", "right": "62.13"} |
| 3 | HB_real_000069 | Consistency & Missingness | Consistency Probing | Compare UniTE against OpenChat across all dataset and Avg. columns. List the columns where UniTE has a higher value. Output only a JSON array of column labels. | ["PIQA", "MMLU", "ARC-C", "TriviaQA", "NQ", "Avg."] |
| 4 | HB_real_000453 | Aggregation & Comparison | Extremum Reasoning | For the UniTE row, what is the minimum value across all dataset and Avg. columns? Output only the number as a string. | "31.78" |
| 5 | HB_real_001878 | Markup Grounding | Marked Column Listing | List all underlined numbers in column NQ. Output only a JSON array of strings. | ["24.25", "31.08"] |
| 6 | HB_real_001879 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "12" |
| 7 | HB_real_001883 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light green in this table? Output only the count as a string. | "7" |
| 8 | HB_real_002264 | Constrained Retrieval | Value Filtering | Among light-green highlighted cells, list the values that are > 70. Output JSON array of strings. | ["73.46", "87.50", "78.70"] |
| 9 | HB_real_002265 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light green. What is their max? Exclude N/A and —. Output only the number as a string. | "87.50" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000109 | Aggregation & Comparison | Grouped Aggregation | Within Two-model ensembling (LLaMA3+Qwen2), what is the mean value of column Avg.? Exclude OOM/missing entries. Output only the number as a string, rounded to two decimal places. | "79.04" |
| 2 | HB_real_000119 | Consistency & Missingness | Cell Missingness | Is the Avg. cell for DeepEn within Two-model ensembling missing/OOM? Answer JSON {"missing":true/false}. | {"missing": true} |
| 3 | HB_real_000275 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=7,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_green"} |
| 4 | HB_real_000280 | Constrained Retrieval | Color-Anchor Retrieval | In the same row as the number 84.17, list all numbers highlighted with the same light-green color as 84.17. Output only a JSON array of strings. | ["84.17", "85.53", "85.07", "69.78", "81.14"] |
| 5 | HB_real_000574 | Aggregation & Comparison | Grouped Aggregation | Among the light-green highlighted numbers in the UniTE row within Two-model ensembling, what is the maximum value? Output only the number as a string. | "85.53" |
| 6 | HB_real_000656 | Aggregation & Comparison | Grouped Aggregation | Within Three-model ensembling, what is the mean value of column Avg.? Exclude OOM/missing entries. Output only the number as a string, rounded to two decimal places. | "79.66" |
| 7 | HB_real_000663 | Markup Grounding | Marked Value Listing | List all light-green highlighted numbers in the UniTE row within Three-model ensembling. Output only a JSON array of strings. | ["84.99", "84.98", "85.39", "69.12", "81.12"] |
| 8 | HB_real_001884 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Avg.? Exclude OOM/missing entries. Output only the number as a string. | "81.14" |
| 9 | HB_real_001885 | Aggregation & Comparison | Grouped Aggregation | Within Three-model ensembling, what is the mean value of column GSM8K? Exclude OOM/missing entries. Output only the number as a string, rounded to two decimal places. | "84.15" |
| 10 | HB_real_001886 | Aggregation & Comparison | Best Selection | Within group Three-model ensembling, which row label in the leftmost row-label column is best on column GSM8K? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "UniTE", "value": "84.99"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000246 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the highlighted purple cell with value 83.44. The requested neighbor is numeric. Output only the number as a string. | "93.35" |
| 2 | HB_real_000640 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=13,c=2) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_purple"} |
| 3 | HB_real_000786 | Markup Grounding | Bold Presence | Is the number with value 83.44 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 4 | HB_real_000911 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are greater than 93. Output JSON array of strings. | ["93.21", "93.10", "93.35", "93.20"] |
| 5 | HB_real_001010 | Aggregation & Comparison | Grouped Aggregation | Among purple highlighted cells in column nnU-Net 2021 Dice (↑), what is the maximum value? Output only the number as a string. | "83.44" |
| 6 | HB_real_001042 | Markup Grounding | Highlight Presence | Is the number with value 83.44 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 7 | HB_real_001887 | Aggregation & Comparison | Best Selection | In metric nnU-Net 2021 Dice (↑), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "LeFusion-H+DiffMask (Ours), P+P’+N’’", "value": "83.44"} |
| 8 | HB_real_001888 | Constrained Retrieval | Value Filtering | Among highlighted cells in column SwinUNETR 2021 NSD (↑), list the values that are greater than 90. Output JSON array of strings. | ["90.98", "90.42", "91.67", "92.77", "93.10", "93.20"] |
| 9 | HB_real_001889 | Markup Grounding | Marked Value Listing | List all bold values in the final LeFusion-H+DiffMask (Ours), P+P’+N’’ row. Output only a JSON array of strings. | ["83.44", "93.35", "83.13", "93.20"] |
| 10 | HB_real_001890 | Aggregation & Comparison | Pairwise Comparison | For row LeFusion-H+DiffMask (Ours), P+P’+N’’, compare column nnU-Net 2021 Dice (↑) and column SwinUNETR 2021 Dice (↑). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "nnU-Net 2021 Dice (↑)", "value": "83.44"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000248 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the highlighted purple cell with value 69.95. The requested neighbor is numeric. Output only the number as a string. | "38.01" |
| 2 | HB_real_000631 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are greater than 70. Output JSON array of strings. | ["70.17", "70.34", "71.28"] |
| 3 | HB_real_000784 | Markup Grounding | Marked Value Listing | List the bold values in column nnU-Net 2021 MI Dice (↑). Output only a JSON array of strings. | ["69.95", "69.87", "70.34", "71.28"] |
| 4 | HB_real_001016 | Markup Grounding | Bold Presence | Is the value at position (r=7,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 5 | HB_real_001043 | Local Relations | Marker-Anchor Neighbor | Find the left neighbor of the highlighted purple cell with value 38.01. The requested neighbor is numeric. Output only the number as a string. | "69.95" |
| 6 | HB_real_001046 | Aggregation & Comparison | Grouped Aggregation | Among the purple highlighted cells in column nnU-Net 2021 MI Dice (↑), what is the maximum value? Output only the number as a string. | "71.28" |
| 7 | HB_real_001891 | Aggregation & Comparison | Best Selection | In metric nnU-Net 2021 MI Dice (↑), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "LeFusion-J-H+DiffMask (Ours), P+P’+N’’", "value": "71.28"} |
| 8 | HB_real_001892 | Constrained Retrieval | Value Filtering | Among highlighted cells in column SwinUNETR 2021 MI Dice (↑), list the values that are greater than 59. Output JSON array of strings. | ["59.61", "59.42", "59.56", "60.54", "59.30"] |
| 9 | HB_real_001893 | Aggregation & Comparison | Argmax Reasoning | In column nnU-Net 2021 MI Dice (↑), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 7, "col": 1, "value": "71.28"} |
| 10 | HB_real_001894 | Markup Grounding | Highlight Presence | Is the number with value 36.32 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": false} |
| 11 | HB_real_001895 | Markup Grounding | Highlight Presence | Is the number with value 69.95 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000558 | Constrained Retrieval | Value Filtering | From this table image, among highlighted cells (any highlighted color), list the values that are > 48. Output JSON array of strings. | ["49.04"] |
| 2 | HB_real_000910 | Constrained Retrieval | Value Filtering | Among highlighted cells in the Answer Correctness (↑) columns, list the values that are > 45. Output JSON array of strings. | ["45.43", "47.94", "46.83", "49.04", "47.74"] |
| 3 | HB_real_001015 | Aggregation & Comparison | Rank Selection | In column Mandatory (↑), list the top 2 rows by value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "GPT 4o", "value": "49.04"}, {"item": "GPT 4-Turbo", "value": "47.94"}] |
| 4 | HB_real_001041 | Local Relations | Marker-Anchor Neighbor | Using visual markers in the table, find the right neighbor of the highlighted cyan cell with value 49.04. The requested neighbor is guaranteed to be numeric. Output only the number as a string. | "47.74" |
| 5 | HB_real_001896 | Aggregation & Comparison | Best Selection | In metric Mandatory (↑), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT 4o", "value": "49.04"} |
| 6 | HB_real_001897 | Aggregation & Comparison | Extremum Reasoning | In row GPT 4o, what is the maximum numeric value across the table? Output only the number as a string. | "49.04" |
| 7 | HB_real_001898 | Local Relations | Marker-Anchor Neighbor | For the highlighted cyan cell with value 49.04, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "23.12", "right": "47.74"} |
| 8 | HB_real_001899 | Aggregation & Comparison | Pairwise Comparison | For row GPT 4o, compare column Mandatory (↑) and column Free (↑). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Mandatory (↑)", "value": "49.04"} |
| 9 | HB_real_001900 | Markup Grounding | Highlight Presence | Is the number with value 49.04 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000552 | Aggregation & Comparison | Extremum Reasoning | From this table image, what is the maximum value in column Avg.? Output only the number as a string. | "66.5" |
| 2 | HB_real_001901 | Aggregation & Comparison | Best Selection | Within group "CLIP (ViT-Large)", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "59.2"} |
| 3 | HB_real_001902 | Aggregation & Comparison | Best Selection | Within group "DINOv2 (ViT-Large)", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "66.5"} |
| 4 | HB_real_001903 | Aggregation & Comparison | Extremum Reasoning | In row Ours within group DINOv2 (ViT-Large), what is the minimum value across Citys, BDD, Map, and Avg.? Output only the number as a string. | "57.5" |
| 5 | HB_real_001904 | Aggregation & Comparison | Pairwise Comparison | In column Map (block DINOv2 (ViT-Large)), compare Ours and ChildTune [66]. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "68.0"} |
| 6 | HB_real_001905 | Markup Grounding | Highlight Presence | Is the number with value 66.5 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001906 | Aggregation & Comparison | Best Selection | Within group "DINOv2 (block Cityscapes→BDD100K)", which row label in the leftmost row-label column is best on column mIoU? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "67.7"} |
| 2 | HB_real_001907 | Aggregation & Comparison | Best Selection | Within group "EVA02 (block Cityscapes→ACDC)", which row label in the leftmost row-label column is best on column mIoU? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "72.9"} |
| 3 | HB_real_001908 | Markup Grounding | Highlight Presence | Is the number with value 77.6 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001909 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000025 | Aggregation & Comparison | Argmax Reasoning | From this table, find the location (row,col) and value of the maximum cell in column mIoU. Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and left text columns. | {"row": 6, "col": 7, "value": "66.3"} |
| 2 | HB_real_000026 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in this table? Output only the count as a string. | "14" |
| 3 | HB_real_000029 | Constrained Retrieval | Color-Anchor Retrieval | The value 56.9 appears exactly once. List all values highlighted with the same blue color as 56.9. Output only a JSON array of strings. | ["56.9", "60.0", "66.6", "53.2", "78.6", "82.2", "66.3"] |
| 4 | HB_real_000033 | Local Relations | Neighbor Value Tracing | The value 64.5 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "56.9", "right": "50.7"} |
| 5 | HB_real_000034 | Consistency & Missingness | Consistency Probing | Compare Ours against ChildTune across all metric columns. List the columns where Ours has a higher value. Output only a JSON array of column labels. | ["Foggy Zurich [52]", "Foggy Driving [52]", "Dark Zurich [53]", "Nighttime Driving [55]", "ACDC-Rain [54]", "ACDC-Snow [54]", "mIoU"] |
| 6 | HB_real_000318 | Markup Grounding | Marked Column Listing | According to the image table, in column mIoU, return the values that are bold. Output only a JSON array of strings. | ["66.3"] |
| 7 | HB_real_001912 | Aggregation & Comparison | Argmax Reasoning | In column mIoU, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours", "value": "66.3"} |
| 8 | HB_real_001913 | Aggregation & Comparison | Pairwise Comparison | For row Ours, compare column ACDC-Rain [54] and column ACDC-Snow [54]. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "ACDC-Snow [54]", "value": "82.2"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000169 | Constrained Retrieval | Value Filtering | Based on the table, for the delta column Cityscapes→ACDC gain (scope: DRFθ rows only, ordered as EVA02 then DINOv2), what is the value at row 2? Output the full string. | "+5.8" |
| 2 | HB_real_000387 | Constrained Retrieval | Value Filtering | Using the visual table content, considering only the two DRFθ rows in order (EVA02, DINOv2), list the parenthesized gain values in column Cityscapes→ACDC. Output only a JSON array of strings. | ["+5.3", "+5.8"] |
| 3 | HB_real_000403 | Aggregation & Comparison | Extremum Reasoning | In column Cityscapes→ACDC, what is the largest numeric entry shown in the table? Output the full cell string. | "77.5 (+5.8)" |
| 4 | HB_real_000589 | Constrained Retrieval | Value Filtering | Considering only the two DRFθ rows in order (EVA02, DINOv2), list the parenthesized gain values in column Cityscapes→BDD100K. Output only a JSON array of strings. | ["+3.7", "+4.0"] |
| 5 | HB_real_000755 | Markup Grounding | Marked Value Listing | List all numeric values highlighted with the blue background. Output only a JSON array of strings. | ["65.8 (+3.7)", "72.9 (+5.3)", "67.7 (+4.0)", "77.5 (+5.8)"] |
| 6 | HB_real_000756 | Constrained Retrieval | Value Filtering | For the delta column Cityscapes→ACDC gain, considering only the two DRFθ rows ordered as EVA02 then DINOv2, what is the value at row 1? Output the full string. | "+5.3" |
| 7 | HB_real_000976 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=18,c=2) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| 8 | HB_real_001915 | Aggregation & Comparison | Best Selection | Within group "EVA02 (Large)", which row label in the leftmost row-label column is best on column Cityscapes→BDD100K? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DRFθ", "value": "65.8 (+3.7)"} |
| 9 | HB_real_001917 | Markup Grounding | Marker Counting | How many numeric cells are highlighted with the blue background in this table? Output only the count as a string. | "4" |
| 10 | HB_real_001918 | Local Relations | Neighbor Indexing | At position (r=18,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "77.5 (+5.8)", "up": "76.1", "down": null, "left": "67.7 (+4.0)", "right": null} |
| 11 | HB_real_001919 | Markup Grounding | Highlight Presence | Is the number with value 77.5 (+5.8) highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000250 | Consistency & Missingness | Consistency Probing | In the AAN column, among the not gray-highlighted rows, list all items whose parenthesized value is positive. Return JSON array of {"item":"...","value":"..."}. | [{"item": "Depth 26 / Width 52 / GAP", "value": "+0.27"}] |
| 2 | HB_real_000382 | Constrained Retrieval | Value Filtering | Considering the not gray-highlighted rows in order (Depth 10/96/GAP, Depth 18/64/4×4 AvgPool, Depth 26/52/GAP), list all parenthesized values in column AAN. Output only a JSON array of strings. | ["-2.97", "-1.77", "+0.27"] |
| 3 | HB_real_000395 | Constrained Retrieval | Value Filtering | In column AAN, considering the not gray-highlighted rows in order (Depth 10/96/GAP, Depth 18/64/4×4 AvgPool, Depth 26/52/GAP), what is the parenthesized value at row 3? Output the full string. | "+0.27" |
| 4 | HB_real_000585 | Constrained Retrieval | Value Filtering | Considering the not gray-highlighted rows in order (Depth 10/96/GAP, Depth 18/64/4×4 AvgPool, Depth 26/52/GAP), list all parenthesized values in column FAF↓. Output only a JSON array of strings. | ["-2.60", "-1.59", "+0.26"] |
| 5 | HB_real_000904 | Constrained Retrieval | Value Filtering | Considering the not gray-highlighted rows in order (Depth 10/96/GAP, Depth 18/64/4×4 AvgPool, Depth 26/52/GAP), list all parenthesized values in column FRF↓. Output only a JSON array of strings. | ["-1.94", "-1.12", "+0.11"] |
| 6 | HB_real_001920 | Aggregation & Comparison | Best Selection | In metric AAN↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Depth 26 / Width 52 / GAP", "value": "86.68±0.70 (+0.27)"} |
| 7 | HB_real_001921 | Aggregation & Comparison | Best Selection | In metric FAF↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Depth 10 / Width 96 / GAP", "value": "33.16±1.28 (-2.60)"} |
| 8 | HB_real_001923 | Aggregation & Comparison | Grouped Aggregation | Among the three not gray-highlighted rows, which row has the largest parenthesized value in column AAN? Return the row index as a string (top-to-bottom among the not gray-highlighted rows). | "3" |
| 9 | HB_real_001924 | Aggregation & Comparison | Grouped Aggregation | Among the three not gray-highlighted rows, which row has the most negative parenthesized value in column FRF↓? Return the row index as a string (top-to-bottom among the not gray-highlighted rows). | "1" |
| 10 | HB_real_rebal_000040 | Consistency & Missingness | Consistency Probing | In the FAF↓ column, list rows with positive delta annotations. Return a JSON array of {"item":"<row label>","value":"<delta>"}. | [{"item": "Depth 26 / Width 52 / GAP", "value": "+0.26"}] |
| 11 | HB_real_rebal_000041 | Consistency & Missingness | Consistency Probing | In the FRF↓ column, list rows with positive delta annotations. Return a JSON array of {"item":"<row label>","value":"<delta>"}. | [{"item": "Depth 26 / Width 52 / GAP", "value": "+0.11"}] |
| 12 | HB_real_rebal_000042 | Aggregation & Comparison | Grouped Aggregation | Across the three not gray-highlighted rows, what is the mean #P (M) value? Output the value rounded to two decimal places as a string. | "11.35" |
| 13 | HB_real_rebal_000069 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000173 | Constrained Retrieval | Value Filtering | For the ImageNet100/10 FAF↓ delta column, considering only the five Improvement rows in order (iCaRL, WA, DER, Foster, MEMO), what is the value at row 3? Output the full string. | "↓2.26" |
| 2 | HB_real_000177 | Constrained Retrieval | Value Filtering | For the ImageNet100/10 FAF↓ delta column, considering only the five Improvement rows in order (iCaRL, WA, DER, Foster, MEMO), what is the value at row 1? Output the full string. | "↓7.69" |
| 3 | HB_real_000384 | Constrained Retrieval | Value Filtering | Considering only the five Improvement rows in order (iCaRL, WA, DER, Foster, MEMO), list all delta values in column FAF↓ within block ImageNet100/10. Output only a JSON array of strings. | ["↓7.69", "↓3.74", "↓2.26", "↓0.14", "↓5.34"] |
| 4 | HB_real_000404 | Aggregation & Comparison | Grouped Aggregation | Among the five Improvement rows, which row has the largest decrease in FAF↓ within block ImageNet100/10? Return the row index as a string (top-to-bottom among iCaRL, WA, DER, Foster, MEMO). | "1" |
| 5 | HB_real_000750 | Consistency & Missingness | Consistency Probing | In the CIFAR100/20 LA↑ improvement values (Improvement rows only), list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "iCaRL Improvement", "value": "+2.75"}, {"item": "WA Improvement", "value": "+8.24"}, {"item": "DER Improvement", "value": "+5.69"}, {"item": "Foster Improvement", "value": "+7.70"}, {"item": "MEMO Improvement", "value": "+10.29"}] |
| 6 | HB_real_000759 | Constrained Retrieval | Value Filtering | For the ImageNet100/10 FAF↓ delta column, considering only the five Improvement rows in order (iCaRL, WA, DER, Foster, MEMO), what is the value at row 5? Output the full string. | "↓5.34" |
| 7 | HB_real_001925 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column FAF↓ within block CIFAR100/20? Exclude missing entries. Output only the number as a string. | "19.05" |
| 8 | HB_real_001926 | Constrained Retrieval | Value Filtering | For the ImageNet100/10 FAF↓ delta column, considering only the five Improvement rows in order (iCaRL, WA, DER, Foster, MEMO), what is the value at row 4? Output the full string. | "↓0.14" |
| 9 | HB_real_001927 | Aggregation & Comparison | Grouped Aggregation | Among the five Improvement rows, which row has the smallest decrease in FAF↓ within block ImageNet100/10? Return the row index as a string (top-to-bottom among iCaRL, WA, DER, Foster, MEMO). | "4" |
| 10 | HB_real_001928 | Constrained Retrieval | Value Filtering | For the CIFAR100/20 LA↑ improvement values, considering only the five Improvement rows in order (iCaRL, WA, DER, Foster, MEMO), what is the value at row 2? Output the full string. | "+8.24" |
| 11 | HB_real_rebal_000046 | Aggregation & Comparison | Grouped Aggregation | Across the Improvement rows, in the CIFAR100/10 LA↑ column, which row has the largest positive delta? Return the row index as a string. | "5" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000143 | Constrained Retrieval | Cell Retrieval | Read the table carefully and find the value at (r=2, c=6) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and the two left text columns. Output only the number/string as shown. | "70.29 (-2.63)" |
| 2 | HB_real_000146 | Constrained Retrieval | Cell Retrieval | Find the value at (r=1, c=6) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and the two left text columns. Output only the number/string as shown. | "72.92" |
| 3 | HB_real_000148 | Constrained Retrieval | Cell Retrieval | Find the value at (r=5, c=5) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and the two left text columns. Output only the number/string as shown. | "69.92±0.12" |
| 4 | HB_real_000234 | Local Relations | Neighbor Indexing | Please answer this question: at (r=4,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and the two left text columns. | {"target": "74.30±0.08", "up": "74.64±0.28", "down": "74.25±0.35", "left": "71.38±0.25", "right": "72.65±0.11"} |
| 5 | HB_real_000236 | Local Relations | Neighbor Indexing | At (r=2,c=6) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and the two left text columns. | {"target": "70.29 (-2.63)", "up": "72.92", "down": "71.18 (-1.74)", "left": "72.11±0.27", "right": null} |
| 6 | HB_real_000329 | Markup Grounding | Marked Value Listing | List all boldfaced numeric values in the table. Output only a JSON array of strings. | ["70.21±0.19", "71.53±0.13", "75.26±0.20", "73.18±0.04", "74.44±0.07", "72.92"] |
| 7 | HB_real_000331 | Markup Grounding | Marker Counting | How many numeric cells are boldfaced in the table? Output only the count as a string. | "6" |
| 8 | HB_real_000465 | Aggregation & Comparison | Pairwise Comparison | For column Average, which row has the larger value: Sta-Net + Pla-Net or Sta-Net + Sta-Net? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Sta-Net + Pla-Net", "value": "72.92"} |
| 9 | HB_real_000580 | Markup Grounding | Marker Counting | How many underlined numeric cells are in the table? Output only the count as a string. | "0" |
| 10 | HB_real_000702 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column WA. Output JSON {"row":<int>,"col":<int>,"value":"<string>"}. The numeric part of the table excludes headers and the two left text columns. | {"row": 1, "col": 2, "value": "71.53±0.13"} |
| 11 | HB_real_000708 | Constrained Retrieval | Value Filtering | In row Sta-Net + None†, what is the difference between column MEMO and column Foster? Output only the number as a string rounded to two decimals. | "1.27" |
| 12 | HB_real_001933 | Aggregation & Comparison | Pairwise Comparison | For row Sta-Net + None†, compare column iCaRL and column MEMO. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "MEMO", "value": "72.11±0.27"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001934 | Aggregation & Comparison | Best Selection | Within group "Qwen 2.5 Coder 7B", which row label in the leftmost row-label column is best on column AVG? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CODEI/O++", "value": "57.7"} |
| 2 | HB_real_001935 | Aggregation & Comparison | Best Selection | Within group "Gemma2 27B", which row label in the leftmost row-label column is best on column AVG? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CODEI/O++", "value": "61.5"} |
| 3 | HB_real_001936 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000023 | Aggregation & Comparison | Argmax Reasoning | Based on the table, find the location (row,col) and value of the maximum cell in column AVG. Output JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and the left text label column. | {"row": 1, "col": 16, "value": "57.2"} |
| 2 | HB_real_000028 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part of the table excludes headers and the left text label column. | {"row": 5, "col": 3, "value": "87.4"} |
| 3 | HB_real_000325 | Constrained Retrieval | Cell Retrieval | Based on the table image, find the value at (r=6, c=16) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes the left text label column. Output only the number/string as shown. | "56.6" |
| 4 | HB_real_000332 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 5 | HB_real_000339 | Aggregation & Comparison | Extremum Reasoning | For row w/o wrong, what is the maximum value across all numeric metrics? Output only the number as a string. | "87.4" |
| 6 | HB_real_000345 | Aggregation & Comparison | Pairwise Comparison | For row CODEI/O, compare column GSM8K and column MMLU-STEM. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "GSM8K", "value": "86.4"} |
| 7 | HB_real_000615 | Aggregation & Comparison | Extremum Reasoning | Using visual evidence in the table, what is the maximum value in column AVG? Output only the number as a string. | "57.2" |
| 8 | HB_real_000704 | Aggregation & Comparison | Extremum Reasoning | For row CODEI/O, what is the minimum value across all numeric metrics? Output only the number as a string. | "10.7" |
| 9 | HB_real_000711 | Local Relations | Neighbor Value Tracing | The value 57.2 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "28.5", "right": null} |
| 10 | HB_real_000713 | Constrained Retrieval | Cell Retrieval | Find the value at (r=2, c=16) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes headers and the left text label column. Output only the number as a string. | "56.7" |
| 11 | HB_real_001942 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000027 | Local Relations | Neighbor Value Tracing | The value 43.3 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "71.9", "right": "77.3"} |
| 2 | HB_real_000030 | Consistency & Missingness | Consistency Probing | Compare row Q+Code / CoT against row Q / CoT across all numeric columns. List the columns where Q+Code / CoT has a higher value. Output only a JSON array of column labels. | ["Wino Grande", "MATH", "GPQA", "CRUX-I", "BBH-EN", "BBH-ZH", "AVG"] |
| 3 | HB_real_000031 | Aggregation & Comparison | Rank Selection | In column GPQA, which row has the second-highest value? Return JSON {"item":"<row label using Prompt / Response>","value":"<cell value>"}. | {"item": "Q / Code+CoT", "value": "42.2"} |
| 4 | HB_real_000137 | Markup Grounding | Underline Presence | Read the table carefully and is the number at (r=2,c=4) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the two left text columns Prompt/Response. | {"underlined": true} |
| 5 | HB_real_000294 | Markup Grounding | Bold Presence | Using the visual table content, is the number at (r=5,c=6) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the two left text columns Prompt/Response. | {"bold": true} |
| 6 | HB_real_001021 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": false, "color": null} |
| 7 | HB_real_001946 | Markup Grounding | Marked Column Listing | List all underlined numbers in column MATH. Output only a JSON array of strings. | ["70.4"] |
| 8 | HB_real_000885 | Markup Grounding | Marked Value Listing | In the Q / Code row, list all underlined numeric values. Output only a JSON array of strings. | ["73.1", "84.8", "20.8", "59.5", "62.4", "67.2", "68.3", "10.1", "40.3", "26.3", "54.9"] |
| 9 | HB_real_001882 | Markup Grounding | Marked Column Listing | In column DROP, list all underlined values. Output only a JSON array of strings. | ["73.1"] |
| 10 | HB_real_001743 | Markup Grounding | Underline Presence | At position (r=1,c=2) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Prompt and Response columns. | {"underlined": false} |
| 11 | HB_real_000985 | Markup Grounding | Marked Column Listing | In column Wino Grande, list all bold values. Output only a JSON array of strings. | ["67.9", "67.9"] |
| 12 | HB_real_001703 | Markup Grounding | Marker Counting | How many numeric values are boldfaced in the table? Output only the count as a string. | "19" |
| 13 | HB_real_001943 | Markup Grounding | Bold Presence | At position (r=1,c=2) in the numeric part of the table, is the value boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left Prompt and Response columns. | {"bold": false} |
| 14 | HB_real_000716 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. Among the remaining values in column MATH, what is the maximum value? Output only the number as a string. | "71.7" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000013 | Aggregation & Comparison | Grouped Aggregation | Please answer this question: exclude cells that are highlighted in light pink. On the remaining cells in column Reasoning Δ, what is the max? Exclude N/A and —. Output only the number as a string. | "8.5%" |
| 2 | HB_real_000014 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are highlighted in light pink. On the remaining cells in column Score Δ, what is the max? Exclude N/A and —. Output only the number as a string. | "15.5%" |
| 3 | HB_real_000021 | Aggregation & Comparison | Grouped Aggregation | Exclude cells that are highlighted in light pink. On the remaining cells in column Chat Δ, what is the max? Exclude N/A and —. Output only the number as a string. | "1.7%" |
| 4 | HB_real_000388 | Constrained Retrieval | Value Filtering | According to the image table, considering only the non-baseline rows in order (Llama-3.1-8B-Instruct, Hermes-3-Llama-3.1-8B, Llama-3.1-Tulu-3-8B-SFT, Llama-3.1-Tulu-3-8B-DPO, Llama-3.1-Tulu-3-8B), list all delta values in the Chat Δ column. Output only a JSON array of strings. | ["1.5%", "1.7%", "1.5%", "0.9%", "-0.6%"] |
| 5 | HB_real_000440 | Aggregation & Comparison | Grouped Aggregation | Among the five non-baseline rows, which row has the maximum Reasoning Δ value? Return the row index as a string, counting only the non-baseline rows from top to bottom. | "3" |
| 6 | HB_real_000445 | Markup Grounding | Highlight Presence | Is the value at position (r=3,c=8) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_pink"} |
| 7 | HB_real_000745 | Consistency & Missingness | Consistency Probing | Using the table only, in the Reasoning Δ values of non-baseline rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "Llama-3.1-8B-Instruct", "value": "7.1%"}, {"item": "Llama-3.1-Tulu-3-8B-SFT", "value": "8.5%"}, {"item": "Llama-3.1-Tulu-3-8B-DPO", "value": "1.3%"}] |
| 8 | HB_real_000802 | Local Relations | Neighbor Value Tracing | The value 8.5% appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "85.8", "right": "84.2"} |
| 9 | HB_real_001039 | Local Relations | Marker-Anchor Neighbor | Based on the image table, find the right neighbor of the highlighted cell with value -6.4%. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "81.1" |
| 10 | HB_real_001950 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=4,c=9) in the numeric part of the table. Output only the value as a string. | "84.2" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000172 | Constrained Retrieval | Value Filtering | Read the table carefully and for the delta column Avg. delta (scope: + TP (Ours) rows only, ordered as PromptEOL, MetaEOL, Pretended CoT, Knowledge), what is the value at row 3? Output the full string. | "0.68" |
| 2 | HB_real_000176 | Aggregation & Comparison | Grouped Aggregation | Among the four + TP (Ours) rows, which row has the largest orange number shown after the Avg. score? Return the row index as a string, counting [PromptEOL, MetaEOL, Pretended CoT, Knowledge] from top to bottom. | "1" |
| 3 | HB_real_000394 | Consistency & Missingness | Consistency Probing | According to the image table, in the STS12 delta values for the four "+ TP (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "PromptEOL + TP (Ours)", "value": "8.09"}, {"item": "MetaEOL + TP (Ours)", "value": "1.99"}, {"item": "Pretended CoT + TP (Ours)", "value": "1.07"}, {"item": "Knowledge + TP (Ours)", "value": "0.43"}] |
| 4 | HB_real_000406 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 5 | HB_real_000408 | Aggregation & Comparison | Grouped Aggregation | Among the four + TP (Ours) rows, which row has the minimum Avg. delta? Return the row index as a string, counting [PromptEOL, MetaEOL, Pretended CoT, Knowledge] from top to bottom. | "4" |
| 6 | HB_real_000587 | Constrained Retrieval | Value Filtering | For the Avg. delta column, considering only the four + TP (Ours) rows ordered as [PromptEOL, MetaEOL, Pretended CoT, Knowledge], what is the value at row 1? Output the full string. | "7.16" |
| 7 | HB_real_000758 | Constrained Retrieval | Value Filtering | For the STS-B delta column, considering only the four + TP (Ours) rows ordered as [PromptEOL, MetaEOL, Pretended CoT, Knowledge], what is the value at row 2? Output the full string. | "3.50" |
| 8 | HB_real_001955 | Constrained Retrieval | Value Filtering | Considering only the four "+ TP (Ours)" rows in order (PromptEOL, MetaEOL, Pretended CoT, Knowledge), list all delta values in the Avg. column. Output only a JSON array of strings. | ["7.16", "1.95", "0.68", "0.40"] |
| 9 | HB_real_001958 | Aggregation & Comparison | Grouped Aggregation | Within the four + TP (Ours) rows, what is the mean of the main Avg. scores before the orange delta annotations? Output only the number as a string, rounded to two decimal places. | "77.55" |
| 10 | HB_real_001959 | Markup Grounding | Highlight Presence | Is the value at position (r=7,c=8) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_gray"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000055 | Aggregation & Comparison | Grouped Aggregation | Among the five Pretended CoT + TP (Ours) rows, which row has the maximum Avg. delta? Return the row index as a string, counting only those Ours rows from top to bottom. | "4" |
| 2 | HB_real_000233 | Local Relations | Neighbor Indexing | Read the table carefully and at (r=4,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes headers and the two left text columns. | {"target": "77.27↑0.90", "up": "76.37", "down": "79.36", "left": "71.01↑1.08", "right": "80.07↑0.79"} |
| 3 | HB_real_000327 | Constrained Retrieval | Cell Retrieval | Using the visual table content, find the value at (r=8, c=8) in the numeric part of the table. The numeric part of the table includes numeric cells only and excludes the two left text columns Method/Backbone. Output only the number/string as shown. | "75.11↑2.17" |
| 4 | HB_real_000330 | Constrained Retrieval | Value Filtering | For the Avg. delta column, within the five Pretended CoT + TP (Ours) rows only, what is the value at row 4? Output the full string. | "2.17" |
| 5 | HB_real_000464 | Aggregation & Comparison | Grouped Aggregation | Within the five Pretended CoT + TP (Ours) rows, what is the mean of the main Avg. scores before the orange delta annotations? Output only the number as a string, rounded to two decimal places. | "76.31" |
| 6 | HB_real_000468 | Aggregation & Comparison | Argmax Reasoning | In column STS15, find the row and value of the maximum main score, ignoring orange delta annotations. Return JSON {"row":<int>,"col":<int>,"value":"<string>"}. The numeric part of the table excludes headers and left text columns. | {"row": 10, "col": 4, "value": "80.71"} |
| 7 | HB_real_000703 | Aggregation & Comparison | Argmax Reasoning | Find the single largest main score in the numeric part of the table, ignoring orange delta annotations. Return JSON {"row":<int>,"col":<int>,"value":"<string>"}. | {"row": 1, "col": 2, "value": "83.89"} |
| 8 | HB_real_000770 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 9 | HB_real_001963 | Aggregation & Comparison | Pairwise Comparison | For row Pretended CoT + TP (Ours), Backbone Qwen2-7B, compare column STS12 and column STS15. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "STS15", "value": "77.94↑3.5"} |
| 10 | HB_real_001964 | Consistency & Missingness | Consistency Probing | In the Avg. delta values for all five "Pretended CoT + TP (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "LLaMA2-7B", "value": "0.68"}, {"item": "LLaMA2-13B", "value": "1.28"}, {"item": "LLaMA3-8B", "value": "0.69"}, {"item": "Qwen2-7B", "value": "2.17"}, {"item": "Gemma2-9B", "value": "0.50"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000170 | Constrained Retrieval | Value Filtering | For the Avg. delta column, considering only the three "+ TP (Ours)" rows in order (PromptEOL, Pretended CoT, Knowledge), what is the value at row 2? Output the delta value as a string. | "0.64" |
| 2 | HB_real_000175 | Constrained Retrieval | Value Filtering | For the MR delta column, considering only the three "+ TP (Ours)" rows in order (PromptEOL, Pretended CoT, Knowledge), what is the value at row 3? Output the delta value as a string. | "0.55" |
| 3 | HB_real_000393 | Consistency & Missingness | Consistency Probing | In the SST2 delta values for the three "+ TP (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "PromptEOL + TP (Ours)", "value": "0.50"}, {"item": "Pretended CoT + TP (Ours)", "value": "0.66"}] |
| 4 | HB_real_000399 | Consistency & Missingness | Consistency Probing | In the MPQA delta values for the three "+ TP (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "PromptEOL + TP (Ours)", "value": "0.32"}, {"item": "Pretended CoT + TP (Ours)", "value": "0.05"}, {"item": "Knowledge + TP (Ours)", "value": "0.02"}] |
| 5 | HB_real_000752 | Consistency & Missingness | Consistency Probing | In the MRPC delta values for the three "+ TP (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "PromptEOL + TP (Ours)", "value": "0.93"}, {"item": "Pretended CoT + TP (Ours)", "value": "1.04"}, {"item": "Knowledge + TP (Ours)", "value": "2.15"}] |
| 6 | HB_real_000757 | Aggregation & Comparison | Grouped Aggregation | Among the three "+ TP (Ours)" rows, which row has the maximum Avg. delta? Return the row index as a string (top-to-bottom among PromptEOL, Pretended CoT, Knowledge). | "2" |
| 7 | HB_real_000977 | Aggregation & Comparison | Grouped Aggregation | Among the three "+ TP (Ours)" rows, which row has the minimum MPQA delta? Return the row index as a string (top-to-bottom among PromptEOL, Pretended CoT, Knowledge). | "3" |
| 8 | HB_real_001966 | Aggregation & Comparison | Grouped Aggregation | What is the mean Avg. score across the three "+ TP (Ours)" rows? Output the value rounded to two decimal places as a string. | "91.40" |
| 9 | HB_real_001967 | Consistency & Missingness | Consistency Probing | In the TREC delta values for the three "+ TP (Ours)" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "PromptEOL + TP (Ours)", "value": "0.60"}, {"item": "Pretended CoT + TP (Ours)", "value": "1.80"}, {"item": "Knowledge + TP (Ours)", "value": "0.40"}] |
| 10 | HB_real_001968 | Constrained Retrieval | Value Filtering | Considering only the three "+ TP (Ours)" rows in order (PromptEOL, Pretended CoT, Knowledge), list all delta values in the Avg. column. Output only a JSON array of strings. | ["0.48", "0.64", "0.46"] |
| 11 | HB_real_rebal_000024 | Aggregation & Comparison | Grouped Aggregation | Considering the three orange SUBJ delta annotations from top to bottom, which row has the largest delta? Return the row index as a string. | "1" |
| 12 | HB_real_rebal_000025 | Consistency & Missingness | Consistency Probing | For the SUBJ column, list all rows with positive orange delta annotations. Return a JSON array of {"item":"<row label>","value":"<delta>"}. | [{"item": "PromptEOL + TP (Ours)", "value": "0.26"}, {"item": "Pretended CoT + TP (Ours)", "value": "0.20"}, {"item": "Knowledge + TP (Ours)", "value": "0.10"}] |
| 13 | HB_real_rebal_000026 | Markup Grounding | Marked Value Listing | In the Knowledge + TP (Ours) row, list all orange delta annotations from left to right. Output only a JSON array of strings. | ["↑0.55", "↑0.29", "↑0.10", "↑0.02", "↓0.27", "↑0.40", "↑2.15", "↑0.46"] |
| 14 | HB_real_rebal_000027 | Constrained Retrieval | Color-Anchor Retrieval | In the first gray highlighted +TP row, list all entries from MR through Avg. including orange deltas. Output only a JSON array of strings. | ["90.90", "↑0.27", "93.35", "↑0.48", "96.58", "↑0.26", "91.51", "↑0.32", "95.50", "↑0.50", "96.00", "↑0.60", "76.12", "↑0.93", "91.42", "↑0.48"] |
| 15 | HB_real_rebal_000028 | Aggregation & Comparison | Grouped Aggregation | Across the three +TP rows, what is the mean MR value? Output the value rounded to two decimal places as a string. | "90.58" |
| 16 | HB_real_rebal_000029 | Local Relations | Neighbor Value Tracing | The value 91.42 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "↑0.93", "right": "↑0.48"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000129 | Markup Grounding | Underline Presence | Based on the table, is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric column Method. | {"underlined": false} |
| 2 | HB_real_000130 | Constrained Retrieval | Cell Retrieval | What is the PPL value for row Adam? Output only the number as a string. | "15.56" |
| 3 | HB_real_000131 | Markup Grounding | Marked Value Listing | List all red numeric delta values in the table. Output only a JSON array of strings. | ["+1.8%", "+4.3G", "+8.2%", "+4.1%", "+2.1G"] |
| 4 | HB_real_000132 | Markup Grounding | Marked Value Listing | List all green numeric delta values in the table. Output only a JSON array of strings. | ["+6.5%", "+6.5%", "+6.5%"] |
| 5 | HB_real_000386 | Constrained Retrieval | Value Filtering | From this table image, considering only the three "+SGG" rows in order (GPU, CPU, Hybrid), list all delta values in the Memory column. Output only a JSON array of strings. | ["+4.3G", "+0.0G", "+2.1G"] |
| 6 | HB_real_000506 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light blue in this table? Output only the count as a string. | "3" |
| 7 | HB_real_000508 | Aggregation & Comparison | Pairwise Comparison | In column Memory, compare Adam+SGG (GPU) and Adam+SGG (Hybrid). Which displayed delta value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Adam+SGG (GPU)", "value": "+4.3G"} |
| 8 | HB_real_000513 | Markup Grounding | Highlight Presence | Is the value at position (r=4,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| 9 | HB_real_000651 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 10 | HB_real_000847 | Constrained Retrieval | Color-Anchor Retrieval | Read the table carefully and for the number at row 4, column 1 in the numeric part of the table, list all numbers highlighted with the same light-blue color as the anchor. The numeric part of the table excludes header and left method-name column. Output only a JSON array of strings. | ["+6.5% (-1.00)", "+4.1% (+4h)", "+2.1G"] |
| 11 | HB_real_001971 | Consistency & Missingness | Consistency Probing | In the Training Time delta values for the three "+SGG" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "Adam+SGG (GPU)", "value": "+2h"}, {"item": "Adam+SGG (CPU)", "value": "+9h"}, {"item": "Adam+SGG (Hybrid)", "value": "+4h"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000265 | Constrained Retrieval | Value Filtering | Based on the table, list the 1B delta values from the four "△ Gains" rows in top-to-bottom order. Output only a JSON array of strings. | ["-1.26", "-0.33", "-0.25", "-4.48"] |
| 2 | HB_real_000383 | Constrained Retrieval | Value Filtering | List the 60M delta values from the four "△ Gains" rows in top-to-bottom order. Output only a JSON array of strings. | ["-3.75", "-1.22", "-1.37", "-4.37"] |
| 3 | HB_real_000400 | Constrained Retrieval | Value Filtering | For the 1B delta values from the four △ Gains rows, what is the value at row 3? Output the full string. | "-0.25" |
| 4 | HB_real_000748 | Constrained Retrieval | Value Filtering | List the 130M delta values from the four "△ Gains" rows in top-to-bottom order. Output only a JSON array of strings. | ["-2.90", "-0.46", "-0.42", "-10.30"] |
| 5 | HB_real_001974 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column 1B? Exclude N/A and —. Output only the number as a string. | "13.95" |
| 6 | HB_real_001975 | Aggregation & Comparison | Pairwise Comparison | In column 1B, compare APOLLO and APOLLO+SGG. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "APOLLO", "value": "14.20"} |
| 7 | HB_real_001976 | Constrained Retrieval | Value Filtering | List the 350M delta values from the four "△ Gains" rows in top-to-bottom order. Output only a JSON array of strings. | ["-1.52", "-0.36", "-0.31", "-7.72"] |
| 8 | HB_real_001977 | Aggregation & Comparison | Rank Selection | In column 1B, list the top 2 methods with the smallest values, sorted from lowest to highest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "APOLLO+SGG", "value": "13.95"}, {"item": "APOLLO", "value": "14.20"}] |
| 9 | HB_real_001978 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000390 | Constrained Retrieval | Value Filtering | Based on the table image, considering the six "+SGG" rows in order (full-rank AdamW/LAMB, rank-4 AdamW/LAMB, rank-8 AdamW/LAMB), list all delta values in column Average. Output only a JSON array of strings. | ["+1.00", "+0.45", "+1.27", "+0.49", "+0.92", "+0.50"] |
| 2 | HB_real_000398 | Consistency & Missingness | Consistency Probing | Using the visual table content, in the Average delta values for the six "+SGG" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "AdamW+SGG (Full-Rank SFT)", "value": "+1.00"}, {"item": "LAMB+SGG (Full-Rank SFT)", "value": "+0.45"}, {"item": "AdamW+SGG (Low-Rank SFT rank 4)", "value": "+1.27"}, {"item": "LAMB+SGG (Low-Rank SFT rank 4)", "value": "+0.49"}, {"item": "AdamW+SGG (Low-Rank SFT rank 8)", "value": "+0.92"}, {"item": "LAMB+SGG (Low-Rank SFT rank 8)", "value": "+0.50"}] |
| 3 | HB_real_000753 | Constrained Retrieval | Value Filtering | Considering the six "+SGG" rows in order (full-rank AdamW/LAMB, rank-4 AdamW/LAMB, rank-8 AdamW/LAMB), list all delta values in column CoLA. Output only a JSON array of strings. | ["+1.12", "+0.38", "+0.98", "+0.96", "+0.53", "+0.58"] |
| 4 | HB_real_001979 | Constrained Retrieval | Value Filtering | Considering the two rank-4 "+SGG" rows in order (AdamW+SGG, LAMB+SGG), list the Average delta values. Output only a JSON array of strings. | ["+1.27", "+0.49"] |
| 5 | HB_real_001980 | Consistency & Missingness | Consistency Probing | In the CoLA delta values for the six "+SGG" rows, list all items whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "AdamW+SGG (Full-Rank SFT)", "value": "+1.12"}, {"item": "LAMB+SGG (Full-Rank SFT)", "value": "+0.38"}, {"item": "AdamW+SGG (Low-Rank SFT rank 4)", "value": "+0.98"}, {"item": "LAMB+SGG (Low-Rank SFT rank 4)", "value": "+0.96"}, {"item": "AdamW+SGG (Low-Rank SFT rank 8)", "value": "+0.53"}, {"item": "LAMB+SGG (Low-Rank SFT rank 8)", "value": "+0.58"}] |
| 6 | HB_real_000737 | Aggregation & Comparison | Grouped Aggregation | Among the six +SGG rows, which row has the maximum Average delta? Return the row index as a string (top-to-bottom among full-rank AdamW/LAMB, rank-4 AdamW/LAMB, rank-8 AdamW/LAMB). | "3" |
| 7 | HB_real_000740 | Aggregation & Comparison | Grouped Aggregation | Among the six +SGG rows, which row has the maximum RTE delta? Return the row index as a string (top-to-bottom among full-rank AdamW/LAMB, rank-4 AdamW/LAMB, rank-8 AdamW/LAMB). | "4" |
| 8 | HB_real_000103 | Markup Grounding | Marked Value Listing | List all green RTE delta values in the six +SGG rows from top to bottom. Output only a JSON array of strings. | ["+1.45", "+1.08", "+1.81", "+0.80", "+1.45", "+1.80"] |
| 9 | HB_real_000284 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=6,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["62.36", "+0.98", "91.10", "+0.53", "92.12", "+1.05", "80.51", "+1.81", "95.06", "+2.17", "88.18", "+1.36", "92.62", "+0.44", "93.06", "+1.77", "86.88", "+1.27"] |
| 10 | HB_real_001794 | Aggregation & Comparison | Grouped Aggregation | Considering the six green MRPC delta annotations from top to bottom, which row has the largest delta? Return the row index as a string. | "1" |
| 11 | HB_real_002010 | Consistency & Missingness | Consistency Probing | For the MRPC column, list all optimizer rows with positive green delta annotations. Return a JSON array of {"item":"<row label>","value":"<delta>"}. | [{"item": "AdamW+SGG (Full-Rank SFT)", "value": "+1.35"}, {"item": "LAMB+SGG (Full-Rank SFT)", "value": "+0.74"}, {"item": "AdamW+SGG (Low-Rank SFT rank 4)", "value": "+1.05"}, {"item": "LAMB+SGG (Low-Rank SFT rank 4)", "value": "+0.30"}, {"item": "AdamW+SGG (Low-Rank SFT rank 8)", "value": "+0.22"}, {"item": "LAMB+SGG (Low-Rank SFT rank 8)", "value": "+0.25"}] |
| 12 | HB_real_001970 | Constrained Retrieval | Value Filtering | List the green STS-B delta annotations from top to bottom. Output only a JSON array of strings. | ["+0.30", "+0.31", "+0.53", "+0.57", "+0.30", "+0.12"] |
| 13 | HB_real_001795 | Markup Grounding | Marked Value Listing | In the AdamW+SGG row under Low-Rank SFT (rank 4), list all green delta annotations from left to right. Output only a JSON array of strings. | ["+0.98", "+0.53", "+1.05", "+1.81", "+2.17", "+1.36", "+0.44", "+1.77", "+1.27"] |
| 14 | HB_real_001813 | Constrained Retrieval | Color-Anchor Retrieval | At the AdamW+SGG row under Low-Rank SFT (rank 8), list all entries highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["62.36", "+0.53", "91.10", "+0.30", "92.12", "+0.22", "80.51", "+1.45", "95.06", "+1.60", "88.17", "+1.23", "92.65", "+0.40", "92.85", "+1.63", "86.85", "+0.92"] |
| 15 | HB_real_001846 | Markup Grounding | Marked Value Listing | In the LAMB+SGG row under Low-Rank SFT (rank 8), list all entries highlighted by the light-blue row background. Output only a JSON array of strings. | ["62.47", "+0.58", "90.90", "+0.12", "89.46", "+0.25", "76.53", "+1.80", "94.95", "+0.34", "87.85", "+0.24", "92.87", "+0.36", "91.78", "+0.36", "85.85", "+0.50"] |
| 16 | HB_real_001961 | Markup Grounding | Underline Presence | The value 62.36 appears exactly once in the AdamW+SGG row under Low-Rank SFT (rank 4). Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 17 | HB_real_001962 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000120 | Markup Grounding | Bold Presence | Based on the table, is the number at (r=6,c=2) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric column Method. | {"bold": true} |
| 2 | HB_real_000123 | Markup Grounding | Marked Column Listing | In column PIQA, return the values that are bold. Output only a JSON array of strings. | ["83.6", "84.8"] |
| 3 | HB_real_000298 | Aggregation & Comparison | Extremum Reasoning | For row DoRA+SGG, what is the maximum value across all main metrics? Output only the number as a string. | "84.8" |
| 4 | HB_real_000510 | Markup Grounding | Underline Presence | From this table image, is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric column Method. | {"underlined": false} |
| 5 | HB_real_000515 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 6 | HB_real_000675 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light blue in this table? Output only the count as a string. | "14" |
| 7 | HB_real_000677 | Consistency & Missingness | Consistency Probing | For the "△ Gains" row under LoRA+SGG, list all delta values from BoolQ to Avg. in order. Output only a JSON array of strings. | ["+1.4", "+2.9", "+1.4", "+2.1", "+3.7", "+4.2", "+2.9"] |
| 8 | HB_real_000851 | Constrained Retrieval | Color-Anchor Retrieval | Read the table carefully and for the number at row 8, column 1 in the numeric part of the table, list all numbers highlighted with the same light-blue color as the anchor. The numeric part of the table excludes header and left method-name column. Output only a JSON array of strings. | ["70.3", "83.6", "78.8", "80.9", "81.5", "79.0", "77.6", "71.4", "84.8", "79.5", "82.8", "83.8", "81.2", "79.6"] |
| 9 | HB_real_000979 | Markup Grounding | Marked Column Listing | In column Avg., return the values that are bold. Output only a JSON array of strings. | ["79.6"] |
| 10 | HB_real_001053 | Constrained Retrieval | Value Filtering | Among light-blue highlighted cells, list the values greater than 84. Output only a JSON array of strings. | ["84.8"] |
| 11 | HB_real_001981 | Constrained Retrieval | Value Filtering | For the "△ Gains" row under DoRA+SGG, list all delta values from BoolQ to Avg. in order. Output only a JSON array of strings. | ["+1.7", "+1.4", "+0.9", "+1.8", "+1.9", "+2.0", "+1.2"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000099 | Markup Grounding | Marked Value Listing | From this table, list all underlined numbers. Output only a JSON array of strings. | [] |
| 2 | HB_real_000104 | Markup Grounding | Underline Presence | Using the table only, is the value at position (r=5,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 3 | HB_real_000309 | Markup Grounding | Marked Value Listing | Based on the table image, list all bold numbers. Output only a JSON array of strings. | ["71.85", "72.02"] |
| 4 | HB_real_000348 | Markup Grounding | Marker Counting | From this table image, how many bold numbers are in the table? Output only the count as a string. | "2" |
| 5 | HB_real_000534 | Markup Grounding | Marker Counting | How many underlined numeric cells are in this table? Output only the count as a string. | "0" |
| 6 | HB_real_000578 | Aggregation & Comparison | Grouped Aggregation | Take the main LoRA scores in cells highlighted in light blue. What is their max? Exclude N/A and —. Output only the number as a string. | "72.02" |
| 7 | HB_real_000682 | Markup Grounding | Marked Value Listing | List all numeric cell strings highlighted in light blue. Output only a JSON array of strings. | ["70.82 +0.72", "70.76 +1.03", "71.85 +0.47", "72.02 +1.80", "71.32 +0.50", "71.28 +0.89"] |
| 8 | HB_real_000689 | Local Relations | Neighbor Indexing | Using the table only, at position (r=5,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "72.02 +1.80", "up": "70.76 +1.03", "down": "71.28 +0.89", "left": "71.85 +0.47", "right": null} |
| 9 | HB_real_000726 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=5,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 10 | HB_real_001985 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=4,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 11 | HB_real_001986 | Markup Grounding | Highlight Presence | Is the value at position (r=4,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000251 | Consistency & Missingness | Consistency Probing | In the first △ Gains row under AdamW+SGG, list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "GQA", "value": "+0.4"}, {"item": "VizWiz", "value": "+0.2"}, {"item": "SciVQA^I / VQA^T", "value": "+3.0"}, {"item": "MMB", "value": "+1.6"}, {"item": "MMB^CN", "value": "+1.8"}, {"item": "POPE", "value": "+0.4"}, {"item": "Avg.", "value": "+1.0"}] |
| 2 | HB_real_000380 | Constrained Retrieval | Value Filtering | For the first △ Gains row under AdamW+SGG, list all delta values from GQA to Avg. in order. Output only a JSON array of strings. | ["+0.4", "+0.2", "+3.0", "-0.8", "+1.6", "+1.8", "+0.4", "+1.0"] |
| 3 | HB_real_000396 | Constrained Retrieval | Value Filtering | For the Avg. delta column, considering the five △ Gains rows in order (AdamW+SGG, Adafactor+SGG, LAMB+SGG, LoRA+SGG, Q-LoRA+SGG), what is the value at row 4? Output the full string. | "+1.0" |
| 4 | HB_real_000746 | Consistency & Missingness | Consistency Probing | In the △ Gains row under Q-LoRA+SGG, list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "GQA", "value": "+0.8"}, {"item": "VizWiz", "value": "+0.6"}, {"item": "SciVQA^I / VQA^T", "value": "+0.3"}, {"item": "VQA^T", "value": "+0.5"}, {"item": "MMB", "value": "+0.1"}, {"item": "MMB^CN", "value": "+0.2"}, {"item": "POPE", "value": "+0.5"}, {"item": "Avg.", "value": "+0.6"}] |
| 5 | HB_real_000907 | Constrained Retrieval | Value Filtering | For the △ Gains row under LoRA+SGG, list all delta values from GQA to Avg. in order. Output only a JSON array of strings. | ["+0.4", "+2.2", "+1.5", "+0.4", "+0.6", "+0.5", "+0.2", "+1.0"] |
| 6 | HB_real_001991 | Aggregation & Comparison | Grouped Aggregation | Among the five △ Gains rows, which row has the maximum Avg. delta? Return the row index as a string (top-to-bottom among AdamW+SGG, Adafactor+SGG, LAMB+SGG, LoRA+SGG, Q-LoRA+SGG). | "1" |
| 7 | HB_real_001992 | Constrained Retrieval | Value Filtering | For the first △ Gains row under AdamW+SGG, what is the SciVQA^I / VQA^T delta value? Output the full string. | "+3.0" |
| 8 | HB_real_001326 | Aggregation & Comparison | Grouped Aggregation | Among the five △ Gains rows, which row has the maximum VizWiz delta? Return the row index as a string (top-to-bottom among AdamW+SGG, Adafactor+SGG, LAMB+SGG, LoRA+SGG, Q-LoRA+SGG). | "4" |
| 9 | HB_real_001330 | Aggregation & Comparison | Grouped Aggregation | Among the five △ Gains rows, which row has the maximum VQA^T delta? Return the row index as a string (top-to-bottom among AdamW+SGG, Adafactor+SGG, LAMB+SGG, LoRA+SGG, Q-LoRA+SGG). | "5" |
| 10 | HB_real_000337 | Markup Grounding | Marked Value Listing | List all green delta values in the Q-LoRA+SGG △ Gains row from GQA to Avg. Output only a JSON array of strings. | ["+0.8", "+0.6", "+0.3", "+0.5", "+0.1", "+0.2", "+0.5", "+0.6"] |
| 11 | HB_real_rebal_000043 | Aggregation & Comparison | Grouped Aggregation | Considering the five VizWiz green gain rows from top to bottom, which row has the largest gain? Return the row index as a string. | "2" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000112 | Aggregation & Comparison | Grouped Aggregation | Within Image-level (AUC), Task=Medical, what is the mean value of column AnomalyMoE? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "85.13" |
| 2 | HB_real_000117 | Markup Grounding | Marked Value Listing | Within Image-level (AUC), Task=Medical, list the bold values in column AnomalyMoE. Output only a JSON array of strings. | ["92.1", "71.1", "92.2"] |
| 3 | HB_real_000276 | Aggregation & Comparison | Grouped Aggregation | Within Pixel-level (AUC), Task=Medical, what is the mean value of column AnomalyMoE? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "96.83" |
| 4 | HB_real_000278 | Aggregation & Comparison | Extremum Reasoning | Within Image-level (AUC), for row BrainMRI, what is the maximum value across methods? Output only the number as a string. | "92.1" |
| 5 | HB_real_000657 | Markup Grounding | Bold Presence | Is the value at position (r=4,c=7) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 6 | HB_real_000662 | Aggregation & Comparison | Rank Selection | Within Image-level (AUC), in column AnomalyMoE, which dataset ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "VisA", "value": "98.1"} |
| 7 | HB_real_002004 | Aggregation & Comparison | Argmax Reasoning | In column AnomalyMoE within block Image-level (AUC), which row label has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "MVTec-AD", "value": "99.5"} |
| 8 | HB_real_002005 | Aggregation & Comparison | Best Selection | Within Image-level (AUC), which row label is best on Dinomaly? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "MVTec-AD", "value": "98.8"} |
| 9 | HB_real_002006 | Aggregation & Comparison | Grouped Aggregation | Within Image-level (AUC), Task=Industrial, what is the mean value of column AnomalyMoE? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "98.80" |
| 10 | HB_real_002007 | Consistency & Missingness | Column Missingness | In column Recontrast within block Image-level (AUC), list the row labels whose entry is missing. Output only a JSON array of strings. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000154 | Markup Grounding | Marked Value Listing | Read the table carefully and list all bold numbers in the dataset-value area (excluding Patch/Component/Global indicator columns). Output only a JSON array of strings. | ["(99.5, 97.7)", "(98.1, 99.0)", "(87.5, 82.2)", "(92.1, 97.3)", "(92.2, 96.0)", "(87.2, 98.9)", "(97.1, 98.4)"] |
| 2 | HB_real_000157 | Markup Grounding | Marked Column Listing | In column MVTec-AD, list every boldfaced dataset-value cell. Output only a JSON array of strings. | ["(99.5, 97.7)"] |
| 3 | HB_real_000306 | Markup Grounding | Marked Column Listing | In column BrainMRI, list every boldfaced dataset-value cell. Output only a JSON array of strings. | ["(92.1, 97.3)"] |
| 4 | HB_real_000312 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=4,c=1) in the dataset-value part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 5 | HB_real_000352 | Markup Grounding | Marker Counting | Based on the table image, how many bold dataset-value cells are in the table? Exclude Patch/Component/Global indicator columns. Output only the count as a string. | "7" |
| 6 | HB_real_000575 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=1,c=1) in the dataset-value part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 7 | HB_real_000684 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=4,c=6) in the dataset-value part of the table. Output only the value as a string. | "(87.2, 98.9)" |
| 8 | HB_real_000721 | Local Relations | Neighbor Indexing | Using the table only, at position (r=4,c=6) in the dataset-value part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "(87.2, 98.9)", "up": "(86.3, 97.3)", "down": null, "left": "(92.2, 96.0)", "right": "(97.1, 98.4)"} |
| 9 | HB_real_000947 | Markup Grounding | Marked Column Listing | In column MVTec 3D, list every boldfaced dataset-value cell. Output only a JSON array of strings. | ["(87.2, 98.9)"] |
| 10 | HB_real_002025 | Markup Grounding | Marker Counting | How many underlined dataset-value cells are in the table? Exclude Patch/Component/Global indicator columns. Output only the count as a string. | "0" |
| 11 | HB_real_002027 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the boldfaced cell with value (92.1, 97.3). The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "(92.2, 96.0)" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000121 | Markup Grounding | Bold Presence | Using the table only, is the number at (r=13,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric column Models. | {"bold": true} |
| 2 | HB_real_000134 | Markup Grounding | Underline Presence | According to the table, is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric column Models. | {"underlined": false} |
| 3 | HB_real_000293 | Markup Grounding | Bold Presence | Is the number at (r=1,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric column Models. | {"bold": false} |
| 4 | HB_real_000302 | Markup Grounding | Marked Column Listing | In column Coarse-gr. R2 G.A, return the values that are bold. Output only a JSON array of strings. | ["76.1"] |
| 5 | HB_real_000507 | Constrained Retrieval | Cell Retrieval | What is the Coarse-gr. R2 G.A value for row Qwen + T? Output only the number as a string. | "76.1" |
| 6 | HB_real_000514 | Aggregation & Comparison | Pairwise Comparison | In column Coarse-gr. R2 G.A, compare Qwen + T and LLaMa_L + T. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Qwen + T", "value": "76.1"} |
| 7 | HB_real_000608 | Local Relations | Neighbor Value Tracing | The value 76.1 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "70.4", "right": null} |
| 8 | HB_real_000654 | Constrained Retrieval | Value Filtering | In column Coarse-gr. R2 G.A, list all row labels with value >= 75. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "LLaMa + T", "value": "75.0"}, {"item": "LLaMa_L + T", "value": "75.4"}, {"item": "Qwen + T", "value": "76.1"}] |
| 9 | HB_real_000673 | Markup Grounding | Highlight Presence | Is the value at position (r=11,c=8) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_pink"} |
| 10 | HB_real_001993 | Markup Grounding | Highlight Presence | Is the number with value 76.1 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000097 | Markup Grounding | Marked Value Listing | Based on the table, return every value that is underlined. Output only a JSON array of strings. | [] |
| 2 | HB_real_000102 | Aggregation & Comparison | Extremum Reasoning | In column Coarse-gr. G.A, what is the largest numeric value? Output only the number as a string. | "91.1" |
| 3 | HB_real_000105 | Constrained Retrieval | Cell Retrieval | What is the Coarse-gr. S.A value for SciTülu + RTE? Output only the cell value as a string. | "88.8" |
| 4 | HB_real_000296 | Markup Grounding | Bold Presence | By reading the table in the image, is the number at (r=11,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric column Models. | {"bold": true} |
| 5 | HB_real_000529 | Consistency & Missingness | Column Missingness | In column Fine-gr. G.A, which row labels have missing entries marked as "-"? Output only a JSON array of strings. | ["RANDOM", "MAJORITY"] |
| 6 | HB_real_000535 | Aggregation & Comparison | Pairwise Comparison | For row Qwen + TE, compare Fine-gr. S.A and Fine-gr. G.A. Which numeric value is larger? Return JSON {"item":"<column name>","value":"<cell value>"}. | {"item": "Fine-gr. G.A", "value": "56.4"} |
| 7 | HB_real_000674 | Markup Grounding | Marked Value Listing | List all boldfaced numeric values in the table. Output only a JSON array of strings. | ["31.1", "56.4", "88.8", "91.1"] |
| 8 | HB_real_000734 | Markup Grounding | Marker Counting | From this table, how many underlined numbers are in the table? Output only the count as a string. | "0" |
| 9 | HB_real_000884 | Markup Grounding | Bold Presence | Is the value at position (r=1,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": false} |
| 10 | HB_real_000888 | Markup Grounding | Highlight Presence | Is the value at position (r=11,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_pink"} |
| 11 | HB_real_001998 | Aggregation & Comparison | Argmax Reasoning | In column Fine-gr. G.A, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Qwen + TE", "value": "56.4"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000122 | Markup Grounding | Bold Presence | From this table, is the number at (r=15,c=6) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. The numeric part of the table excludes the left non-numeric column Models. | {"bold": true} |
| 2 | HB_real_000125 | Markup Grounding | Marker Counting | How many boldfaced numeric values are in the table? Output only the count as a string. | "6" |
| 3 | HB_real_000126 | Aggregation & Comparison | Extremum Reasoning | In column Coarse-gr. R3, what is the largest numeric value? Output only the number as a string. | "91.2" |
| 4 | HB_real_000135 | Markup Grounding | Underline Presence | Using the table only, is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left non-numeric column Models. | {"underlined": false} |
| 5 | HB_real_000292 | Aggregation & Comparison | Pairwise Comparison | For row SciTülu it + T, compare Coarse-gr. R2 and Coarse-gr. R3. Which numeric value is larger? Return JSON {"item":"<column name>","value":"<cell value>"}. | {"item": "Coarse-gr. R3", "value": "91.2"} |
| 6 | HB_real_000299 | Markup Grounding | Marked Value Listing | List all boldfaced numeric values in the table. Output only a JSON array of strings. | ["45.9", "48.6", "59.4", "75.4", "76.3", "91.2"] |
| 7 | HB_real_000511 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 8 | HB_real_000512 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Return JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": false, "color": null} |
| 9 | HB_real_000606 | Constrained Retrieval | Value Filtering | In column Coarse-gr. R3, list all numeric values greater than 90. Output only a JSON array of strings. | ["91.2"] |
| 10 | HB_real_000678 | Local Relations | Neighbor Value Tracing | The value 91.2 appears exactly once in the numeric part of the table. What are its immediate left and right numeric neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "68.4", "right": null} |
| 11 | HB_real_002001 | Aggregation & Comparison | Argmax Reasoning | In column Fine-gr. R3, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Qwen it + T", "value": "59.4"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000100 | Markup Grounding | Marked Value Listing | Read the table carefully and list all underlined numbers. Output only a JSON array of strings. | [] |
| 2 | HB_real_000101 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 3 | HB_real_000392 | Consistency & Missingness | Consistency Probing | From this table image, for row Knowledge + CP-NS (Ours), list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "STS12", "value": "↑1.56"}, {"item": "STS13", "value": "↑0.61"}, {"item": "STS15", "value": "↑0.54"}, {"item": "STS-B", "value": "↑0.46"}, {"item": "SICK-R", "value": "↑0.08"}, {"item": "Avg.", "value": "↑0.42"}] |
| 4 | HB_real_000401 | Aggregation & Comparison | Grouped Aggregation | Among the row Knowledge + CP-NS (Ours) deltas from STS12 to Avg., which column position has the largest positive delta? Return the 1-based column index as a string. | "1" |
| 5 | HB_real_000531 | Aggregation & Comparison | Extremum Reasoning | For row CK + CP-NS (Ours), what is the maximum value across the main metric values? Ignore red delta annotations. Output only the number as a string. | "84.21" |
| 6 | HB_real_000538 | Constrained Retrieval | Value Filtering | In column Avg., list all row labels with value >= 78.60. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "CK + CP-NS (Ours)", "value": "78.68"}, {"item": "CK + CP-NR (Ours)", "value": "78.60"}] |
| 7 | HB_real_000883 | Markup Grounding | Underline Presence | Is the number at (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. The numeric part of the table excludes the left Method and Params columns. | {"underlined": false} |
| 8 | HB_real_000886 | Markup Grounding | Marked Value Listing | List all main metric cells highlighted in light green for row Knowledge + CP-NS (Ours). Output only a JSON array of strings. | ["67.16 ↑1.56", "83.43 ↑0.61", "74.23 ↓0.25", "81.29 ↑0.54", "80.03 ↓0.10", "80.80 ↑0.46", "75.97 ↑0.08", "77.56 ↑0.42"] |
| 9 | HB_real_002008 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Avg.? Output only the number as a string. | "78.68" |
| 10 | HB_real_002009 | Constrained Retrieval | Value Filtering | For row PromptEOL + CP-NS (Ours), list all delta values from STS12 to Avg. in order. Output only a JSON array of strings. | ["↑4.53", "↑5.14", "↑5.39", "↑6.46", "↑3.67", "↑7.05", "↑4.40", "↑5.24"] |
| 11 | HB_real_002012 | Markup Grounding | Highlight Presence | Is the number with value ↑5.24 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000389 | Constrained Retrieval | Value Filtering | By reading the table in the image, for row Pretended CoT + CP-NS (Ours), backbone LLaMA2-7B, list all delta values from STS12 to Avg. in order. Output only a JSON array of strings. | ["↓0.75", "↑0.37", "↑0.30", "↑1.92", "↓0.05", "↑1.61", "↑0.74", "↑0.59"] |
| 2 | HB_real_000906 | Constrained Retrieval | Value Filtering | For row Pretended CoT + CP-NR (Ours), backbone LLaMA2-7B, list all delta values from STS12 to Avg. in order. Output only a JSON array of strings. | ["↓0.68", "↑0.42", "↑0.41", "↑1.98", "↓0.03", "↑1.56", "↑0.75", "↑0.63"] |
| 3 | HB_real_002013 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Avg.? Output only the number as a string. | "77.49↑0.63" |
| 4 | HB_real_002014 | Consistency & Missingness | Consistency Probing | For row Pretended CoT + CP-NR (Ours), backbone LLaMA3.1-8B, list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "STS12", "value": "↑2.03"}, {"item": "STS13", "value": "↑1.17"}, {"item": "STS14", "value": "↑1.11"}, {"item": "STS15", "value": "↑1.04"}, {"item": "STS16", "value": "↑1.18"}, {"item": "STS-B", "value": "↑1.14"}, {"item": "SICK-R", "value": "↑0.43"}, {"item": "Avg.", "value": "↑1.16"}] |
| 5 | HB_real_002015 | Constrained Retrieval | Value Filtering | For row Pretended CoT + CP-NS (Ours), backbone LLaMA3.1-8B, list all delta values from STS12 to Avg. in order. Output only a JSON array of strings. | ["↑1.54", "↑1.26", "↑1.25", "↑1.28", "↑1.14", "↑1.21", "↑0.36", "↑1.15"] |
| 6 | HB_real_002016 | Consistency & Missingness | Consistency Probing | For row Pretended CoT + CP-NS (Ours), backbone LLaMA2-7B, list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "STS13", "value": "↑0.37"}, {"item": "STS14", "value": "↑0.30"}, {"item": "STS15", "value": "↑1.92"}, {"item": "STS-B", "value": "↑1.61"}, {"item": "SICK-R", "value": "↑0.74"}, {"item": "Avg.", "value": "↑0.59"}] |
| 7 | HB_real_000743 | Aggregation & Comparison | Grouped Aggregation | Among the six Ours rows, which row has the maximum Avg. delta? Return the row index as a string (top-to-bottom among the displayed Ours rows). | "6" |
| 8 | HB_real_000991 | Aggregation & Comparison | Grouped Aggregation | Among the six Ours rows, which row has the maximum SICK-R delta? Return the row index as a string (top-to-bottom among the displayed Ours rows). | "3" |
| 9 | HB_real_000303 | Markup Grounding | Marked Value Listing | List all red down-arrow delta values in the first two Ours rows. Output only a JSON array of strings. | ["↓0.75", "↓0.05", "↓0.68", "↓0.03"] |
| 10 | HB_real_002258 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 77.45, list all numeric values highlighted with the same light-green background as the anchor row. Output only a JSON array of strings. | ["66.70", "84.26", "74.44", "81.39", "80.71", "80.56", "74.07", "77.45", "66.77", "84.31", "74.55", "81.45", "80.73", "80.51", "74.08", "77.49"] |
| 11 | HB_real_001854 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light green in the LLaMA2-7B Ours rows. What is their maximum numeric value? Output only the number as a string. | "84.31" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000391 | Consistency & Missingness | Consistency Probing | Based on the table image, for row PromptEOL + CP-NS (Ours), list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "CR", "value": "↑0.24"}, {"item": "SUBJ", "value": "↑0.65"}, {"item": "SST2", "value": "↑0.94"}, {"item": "TREC", "value": "↑1.60"}, {"item": "MRPC", "value": "↑2.32"}, {"item": "Avg.", "value": "↑0.79"}] |
| 2 | HB_real_000397 | Consistency & Missingness | Consistency Probing | For row PromptEOL + CP-NR (Ours), list all metrics whose delta is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "SUBJ", "value": "↑0.42"}, {"item": "SST2", "value": "↑0.83"}, {"item": "TREC", "value": "↑1.00"}, {"item": "MRPC", "value": "↑0.81"}, {"item": "Avg.", "value": "↑0.35"}] |
| 3 | HB_real_000744 | Constrained Retrieval | Value Filtering | For row PromptEOL + CP-NS (Ours), list all delta values from MR to Avg. in order. Output only a JSON array of strings. | ["↓0.14", "↑0.24", "↑0.65", "↓0.09", "↑0.94", "↑1.60", "↑2.32", "↑0.79"] |
| 4 | HB_real_002017 | Aggregation & Comparison | Pairwise Comparison | In column TREC, compare PromptEOL + CP-NS (Ours) and PromptEOL + CP-NR (Ours). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "PromptEOL + CP-NS (Ours)", "value": "97.00↑1.60"} |
| 5 | HB_real_002018 | Constrained Retrieval | Value Filtering | For row PromptEOL + CP-NR (Ours), list all delta values from MR to Avg. in order. Output only a JSON array of strings. | ["↓0.30", "↓0.23", "↑0.42", "↓0.08", "↑0.83", "↑1.00", "↑0.81", "↑0.35"] |
| 6 | HB_real_002020 | Aggregation & Comparison | Argmax Reasoning | Within the "Without fine-tuning" section, in column Avg., find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows are counted within that section and columns within the numeric part of the table. | {"row": 2, "col": 9, "value": "91.73↑0.79"} |
| 7 | HB_real_000992 | Aggregation & Comparison | Grouped Aggregation | Among the two PromptEOL Ours rows in the Without fine-tuning section, which row has the maximum MPQA delta? Return the row index as a string (top-to-bottom among CP-NS and CP-NR). | "2" |
| 8 | HB_real_000322 | Markup Grounding | Marked Value Listing | List all red down-arrow delta values in the two PromptEOL Ours rows. Output only a JSON array of strings. | ["↓0.14", "↓0.09", "↓0.30", "↓0.23", "↓0.08"] |
| 9 | HB_real_000324 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=5,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-green row background as the anchor row. Output only a JSON array of strings. | ["90.49", "93.11", "96.97", "91.10", "95.94", "97.00", "77.51", "91.73", "90.33", "92.64", "96.74", "91.11", "95.83", "96.40", "76.00", "91.29"] |
| 10 | HB_real_002250 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light green in the two PromptEOL Ours rows. What is their maximum numeric value? Output only the number as a string. | "97.00" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002021 | Aggregation & Comparison | Extremum Reasoning | For sentence "It is the first named storm to develop in the Caribbean in December." and method Knowledge+CP-NS, what is the maximum probability among the listed tokens? Output only the number as a string. | "0.1092" |
| 2 | HB_real_002022 | Aggregation & Comparison | Extremum Reasoning | For sentence "The rule - When in doubt throw it out!" and method Knowledge, what is the maximum probability among the listed tokens? Output only the number as a string. | "0.0401" |
| 3 | HB_real_002023 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002028 | Aggregation & Comparison | Best Selection | In metric RM↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "GPT-4o AR-MCTS", "value": "31.8"} |
| 2 | HB_real_002029 | Aggregation & Comparison | Best Selection | In metric IK↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "InternVL2-8B Self-Correction", "value": "62.7"} |
| 3 | HB_real_002030 | Constrained Retrieval | Value Filtering | In column RM↓, list all row labels in the primary row-label column with value <= 35. Here ↓ means smaller is better. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "GPT-4o Self-Consistency", "value": "32.8"}, {"item": "GPT-4o Self-Correction", "value": "34.2"}, {"item": "GPT-4o AR-MCTS", "value": "31.8"}] |
| 4 | HB_real_002031 | Aggregation & Comparison | Pairwise Comparison | For row GPT-4o AR-MCTS, compare column S3↑ and column AVG↑. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "S3↑", "value": "56.4"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002032 | Aggregation & Comparison | Pairwise Comparison | In column Mathematics, compare Zero-shot and AR-MCTS for model GPT-4o. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "AR-MCTS (GPT-4o)", "value": "62.5"} |
| 2 | HB_real_002033 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 3 | HB_real_000136 | Markup Grounding | Bold Presence | Is the value at position (r=3,c=1) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 4 | HB_real_000686 | Markup Grounding | Marked Value Listing | List all bold numeric values in the AR-MCTS row for GPT-4o. Output only a JSON array of strings. | ["52.2", "62.5", "33.3", "21.2", "42.9", "50.0", "80.0", "73.1", "100.0"] |
| 5 | HB_real_000505 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 6 | HB_real_000800 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. On the remaining cells in column Mathematics, what is the maximum value? Output only the number as a string. | "50.0" |
| 7 | HB_real_000452 | Aggregation & Comparison | Grouped Aggregation | Take the cells in the light-gray AR-MCTS rows. What is the maximum numeric value among them? Output only the number as a string. | "100.0" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002034 | Aggregation & Comparison | Pairwise Comparison | In block Few-shot ECE↓, compare rows LLM Prob. and LLM Prob.+PS. Which value is larger? Here ↓ means smaller is better, but this question asks numeric larger. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LLM Prob.", "value": "0.235"} |
| 2 | HB_real_002035 | Aggregation & Comparison | Pairwise Comparison | For row LLM Prob.+PS, compare column Zero-shot ECE↓ and column Few-shot ECE↓. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Zero-shot ECE↓", "value": "0.276"} |
| 3 | HB_real_002036 | Constrained Retrieval | Value Filtering | For row LLM Prob.+PS, among AUC columns (Verb., Zero-shot, CoT, Few-shot), list all row labels in the primary row-label column with value >= 0.67. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Verb. AUC↑", "value": "0.690"}, {"item": "Zero-shot AUC↑", "value": "0.713"}, {"item": "CoT AUC↑", "value": "0.672"}] |
| 4 | HB_real_002037 | Consistency & Missingness | Consistency Probing | List the leftmost row label for each row that has entries marked as missing using "-" or "N/A". Include repeated labels for distinct rows. If none, output an empty JSON array []. | ["Verbalized %", "Verbalized %", "Verbalized %", "Verbalized %", "Verbalized %"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002038 | Aggregation & Comparison | Best Selection | Within group "Popular / GPT-3.5-Turbo", which row label in the leftmost row-label column is best on column MDS? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Self-reflection", "value": "4.18"} |
| 2 | HB_real_002039 | Aggregation & Comparison | Best Selection | Within group "General / Llama3.1-8B", which row label in the leftmost row-label column is best on column MDS? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Self-reflection", "value": "3.91"} |
| 3 | HB_real_002040 | Aggregation & Comparison | Pairwise Comparison | For row Self-reflection (group General / GPT-3.5-Turbo), compare column T_Abs and column MDS. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "MDS", "value": "3.93"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000192 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=1,c=1) in the numeric part of the table, list all numbers highlighted with the same light-cyan color as the anchor. The numeric part excludes the #Can. and Ref. columns. Output only a JSON array of strings. | ["0.60", "0.61", "0.61", "0.63", "0.61"] |
| 2 | HB_real_000197 | Aggregation & Comparison | Argmax Reasoning | In column P_All, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 5, "col": 4, "value": "0.77"} |
| 3 | HB_real_000568 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 3.86, list all numbers highlighted with the same light-peach color as the anchor. Output only a JSON array of strings. | ["3.86", "3.87", "3.93", "3.89", "3.95"] |
| 4 | HB_real_000845 | Markup Grounding | Marker Counting | How many highlighted numeric cells are in the table? Output only the count as a string. | "25" |
| 5 | HB_real_000848 | Markup Grounding | Highlight Presence | Is the value at position (r=3,c=5) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 6 | HB_real_000960 | Local Relations | Neighbor Indexing | At position (r=3,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.75", "up": "0.74", "down": "0.74", "left": "0.73", "right": "0.70"} |
| 7 | HB_real_001055 | Aggregation & Comparison | Grouped Aggregation | Among the light-cyan highlighted T_All cells, what is the maximum value? Output only the number as a string. | "0.63" |
| 8 | HB_real_001057 | Aggregation & Comparison | Grouped Aggregation | Excluding the light-peach MDS column, what is the maximum highlighted value in the table? Output only the number as a string. | "0.77" |
| 9 | HB_real_002042 | Aggregation & Comparison | Grouped Aggregation | What is the mean value in column T_All across all five C rows? Output the value rounded to two decimal places as a string. | "0.61" |
| 10 | HB_real_002043 | Aggregation & Comparison | Pairwise Comparison | For row C=5, compare column B_All and column P_All. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "P_All", "value": "0.75"} |
| 11 | HB_real_002044 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_cyan"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000191 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=1,c=2) in the numeric part of the table, list all numbers highlighted with the same light-purple color as the anchor. The numeric part excludes the Warmup column. Output only a JSON array of strings. | ["0.73", "0.72", "0.70", "0.70"] |
| 2 | HB_real_000195 | Aggregation & Comparison | Argmax Reasoning | In column P_All, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 2, "col": 3, "value": "0.77"} |
| 3 | HB_real_000844 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 3.93, list all numbers highlighted with the same light-peach color as the anchor. Output only a JSON array of strings. | ["3.93", "4.03", "3.85", "3.94"] |
| 4 | HB_real_000961 | Markup Grounding | Marker Counting | How many highlighted numeric cells are in the table? Output only the count as a string. | "20" |
| 5 | HB_real_001054 | Markup Grounding | Highlight Presence | Is the value at position (r=2,c=5) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 6 | HB_real_001056 | Local Relations | Neighbor Indexing | At position (r=2,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.72", "up": "0.73", "down": "0.70", "left": "0.66", "right": "0.77"} |
| 7 | HB_real_002045 | Aggregation & Comparison | Argmax Reasoning | In column MDS, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "W=2", "value": "4.03"} |
| 8 | HB_real_002046 | Aggregation & Comparison | Grouped Aggregation | What is the mean value in column T_All across all four W rows? Output the value rounded to two decimal places as a string. | "0.64" |
| 9 | HB_real_002047 | Aggregation & Comparison | Grouped Aggregation | Excluding the light-peach MDS column, what is the maximum highlighted value in the table? Output only the number as a string. | "0.77" |
| 10 | HB_real_002049 | Aggregation & Comparison | Pairwise Comparison | For row W=2, compare column T_All and column R_All. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "R_All", "value": "0.73"} |
| 11 | HB_real_002050 | Markup Grounding | Highlight Presence | Is the value at position (r=2,c=3) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_purple"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000635 | Constrained Retrieval | Value Filtering | Using visual evidence in the table, in column Speech Acts (↓), list all row labels in the primary row-label column with value <= 0.4. Here ↓ means smaller is better. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Cantonese d_bi", "value": "0.373"}, {"item": "Urdu d_bi", "value": "0.311"}] |
| 2 | HB_real_002051 | Aggregation & Comparison | Best Selection | In metric Number Agreement (↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Japanese d_bi", "value": "0.010"} |
| 3 | HB_real_002052 | Aggregation & Comparison | Best Selection | In metric Modal Verbs Expressions (↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Mandarin d_bi", "value": "0.023"} |
| 4 | HB_real_002053 | Constrained Retrieval | Value Filtering | In column Reference Word (↓), list all row labels in the primary row-label column with value <= 0.2. Here ↓ means smaller is better. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Cantonese d_bi", "value": "0.138"}, {"item": "Japanese d_bi", "value": "0.183"}, {"item": "Malay d_bi", "value": "0.164"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002054 | Aggregation & Comparison | Best Selection | In metric DeepSeek V3 (↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Korean d_bi", "value": "0.021"} |
| 2 | HB_real_002055 | Aggregation & Comparison | Best Selection | In metric LLaMA 70B (↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Mandarin d_mono", "value": "0.008"} |
| 3 | HB_real_002056 | Aggregation & Comparison | Best Selection | In metric GPT 4o (↓), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Korean d_mono", "value": "0.017"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002057 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column S>T Rate (block zh⇒en)? Output only the number as a string. | "34.4%" |
| 2 | HB_real_002058 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column P (block zh⇒th)? Exclude N/A and —. Output only the number as a string. | "0.198" |
| 3 | HB_real_002059 | Aggregation & Comparison | Pairwise Comparison | For row Qwen2.5-14B-Instruct SSPO, compare column CR (zh⇒en) and column CR (zh⇒th). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "CR (zh⇒th)", "value": "36.1%"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000141 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=1,c=1) in the numeric/text data part of the table. The numeric/text data part excludes the two left text columns Challenger and Competitors. Output only the string. | "27:50:23" |
| 2 | HB_real_000144 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=1,c=2) in the numeric/text data part of the table. The numeric/text data part excludes the two left text columns Challenger and Competitors. Output only the string. | "16:67:17" |
| 3 | HB_real_000149 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=8,c=6) in the numeric/text data part of the table. The numeric/text data part excludes the two left text columns Challenger and Competitors. Output only the string. | "33:38:28" |
| 4 | HB_real_000326 | Local Relations | Neighbor Indexing | At position (r=1,c=1) in the numeric/text data part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "27:50:23", "up": null, "down": "26:50:24", "left": null, "right": "16:67:17"} |
| 5 | HB_real_000334 | Markup Grounding | Highlight Presence | Is the value at position (r=1,c=1) in the numeric/text data part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 6 | HB_real_000579 | Markup Grounding | Marker Counting | How many highlighted data cells are in the table? Output only the count as a string. | "64" |
| 7 | HB_real_000700 | Constrained Retrieval | Cell Retrieval | Find the value at position (r=4,c=8) in the numeric/text data part of the table. The numeric/text data part excludes the two left text columns Challenger and Competitors. Output only the string. | "27:51:22" |
| 8 | HB_real_002060 | Local Relations | Neighbor Indexing | At position (r=8,c=6) in the numeric/text data part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "33:38:28", "up": "27:51:22", "down": null, "left": "27:49:24", "right": "25:51:24"} |
| 9 | HB_real_002061 | Markup Grounding | Highlight Presence | Is the value at position (r=8,c=6) in the numeric/text data part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_blue"} |
| 10 | HB_real_002062 | Consistency & Missingness | Cell Missingness | Is the cell in row GPT-4o within block SSPO (Qwen2.5-14B), column Naturalness (es⇒zh), missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 11 | HB_real_002063 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000224 | Markup Grounding | Marked Column Listing | Using the table only, in column Accuracy_triple, return the values that are bold. Output only a JSON array of strings. | ["48.65"] |
| 2 | HB_real_000225 | Markup Grounding | Marked Column Listing | In column BARTScore, return the values that are bold. Output only a JSON array of strings. | ["-1.6299*"] |
| 3 | HB_real_000228 | Markup Grounding | Marked Column Listing | In column Spearman, return the values that are bold. Output only a JSON array of strings. | ["0.6314*"] |
| 4 | HB_real_000315 | Aggregation & Comparison | Extremum Reasoning | In column Accuracy_triple, what is the maximum value? Output only the number as a string. | "48.65" |
| 5 | HB_real_000323 | Aggregation & Comparison | Argmax Reasoning | In column Accuracy_2-tuple, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours (AutoMedEval)", "value": "74.61"} |
| 6 | HB_real_000609 | Constrained Retrieval | Cell Retrieval | What is the Pearson value for row Ours (AutoMedEval)? Output only the number/string as shown. | "0.6854*" |
| 7 | HB_real_000984 | Constrained Retrieval | Value Filtering | In column Accuracy_triple, list all row labels with value >= 44. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Ours (AutoMedEval)", "value": "48.65"}, {"item": "w/o Iterative Knowledge Introspection", "value": "44.61*"}] |
| 8 | HB_real_000986 | Markup Grounding | Marker Counting | How many bold numbers are in the table? Output only the count as a string. | "6" |
| 9 | HB_real_002064 | Aggregation & Comparison | Argmax Reasoning | In column Accuracy_triple, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. Exclude group header rows. | {"row": 10, "col": 6, "value": "48.65"} |
| 10 | HB_real_002065 | Aggregation & Comparison | Best Selection | Within group "Open-sourced models", which row label in the leftmost row-label column is best on column Pearson? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Ours (AutoMedEval)", "value": "0.6854*"} |
| 11 | HB_real_002066 | Markup Grounding | Bold Presence | Using the table only, is the value at position (r=10,c=6) in the numeric part of the table boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002068 | Markup Grounding | Highlight Presence | Is the number with value 70.6 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 2 | HB_real_000509 | Markup Grounding | Underline Presence | Is the value at position (r=14,c=10) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": true} |
| 3 | HB_real_000549 | Markup Grounding | Marked Column Listing | List all underlined values in column VideoChat. Output only a JSON array of strings. | ["3.17"] |
| 4 | HB_real_000854 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 70.6, list all numeric values highlighted with the same light-green background as the anchor in the Video-XL row. Output only a JSON array of strings. | ["64.9", "4.50", "45.5", "4.21", "55.5", "61.0", "61.6", "70.6", "50.7", "3.17", "55.3"] |
| 5 | HB_real_000456 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light green in the Video-XL row. What is their maximum numeric value? Output only the number as a string. | "70.6" |
| 6 | HB_real_001799 | Markup Grounding | Marked Column Listing | In column VideoMME W sub, list all underlined values. Output only a JSON array of strings. | ["54.9"] |
| 7 | HB_real_000710 | Markup Grounding | Underline Presence | The value 4.06 appears exactly once in the numeric part of the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 8 | HB_real_001754 | Markup Grounding | Marked Value Listing | List the bold numeric values in the Video-XL row. Output only a JSON array of strings. | ["64.9", "4.50", "45.5", "4.21", "55.5", "61.0", "61.6", "70.6", "50.7", "55.3"] |
| 9 | HB_real_002252 | Constrained Retrieval | Value Filtering | In the light-green highlighted Video-XL row, list the numeric values greater than 60. Output only a JSON array of strings. | ["64.9", "61.0", "61.6", "70.6"] |
| 10 | HB_real_001954 | Aggregation & Comparison | Grouped Aggregation | Exclude underlined cells. Among the remaining values in column VideoChat., what is the maximum value? Output only the number as a string. | "4.06" |
| 11 | HB_real_001617 | Markup Grounding | Marker Counting | How many bold numeric values are in the Video-XL row? Output only the count as a string. | "10" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002069 | Aggregation & Comparison | Argmax Reasoning | In column MMB, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Upper-bound", "value": "71.6"} |
| 2 | HB_real_002070 | Constrained Retrieval | Value Filtering | In column MME, list all row labels in the primary row-label column with value >= 1500. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Video-XL", "value": "1510.2"}, {"item": "Upper-bound", "value": "1533.7"}] |
| 3 | HB_real_002071 | Markup Grounding | Highlight Presence | Is the number with value 1510.2 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_002072 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 5 | HB_real_002073 | Aggregation & Comparison | Pairwise Comparison | For row Video-XL, compare column MLVU and column VideoMME. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "VideoMME", "value": "52.0"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000189 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=4,c=1) in the numeric part of the table, list all numbers highlighted with the same light-green color as the anchor. The numeric part excludes the Train and Test columns. Output only a JSON array of strings. | ["41.6", "52.3", "1520.0", "71.3"] |
| 2 | HB_real_000196 | Aggregation & Comparison | Argmax Reasoning | In column MME, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 4, "col": 3, "value": "1520.0"} |
| 3 | HB_real_000846 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light green? Output only the count as a string. | "4" |
| 4 | HB_real_000849 | Local Relations | Neighbor Indexing | At position (r=4,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "1520.0", "up": "1515.5", "down": null, "left": "52.3", "right": "71.3"} |
| 5 | HB_real_001051 | Aggregation & Comparison | Grouped Aggregation | Among the light-green highlighted cells, what is the maximum value? Output only the number as a string. | "1520.0" |
| 6 | HB_real_002074 | Aggregation & Comparison | Argmax Reasoning | In column MMB, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Train ✓ / Test ✓", "value": "71.3"} |
| 7 | HB_real_002075 | Aggregation & Comparison | Pairwise Comparison | In column MME, compare Train ✓ / Test ✗ and Train ✓ / Test ✓. Which value is larger? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Train ✓ / Test ✓", "value": "1520.0"} |
| 8 | HB_real_002076 | Markup Grounding | Highlight Presence | Is the value at position (r=4,c=4) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_green"} |
| 9 | HB_real_002077 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002078 | Aggregation & Comparison | Argmax Reasoning | In column TR, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Video=100k, Single Image=700k, Multi Image=40k", "value": "82.1"} |
| 2 | HB_real_002079 | Aggregation & Comparison | Argmax Reasoning | In column AO, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Video=100k, Single Image=700k, Multi Image=40k", "value": "55.4"} |
| 3 | HB_real_002080 | Constrained Retrieval | Value Filtering | In column NQA, list all row labels in the primary row-label column with value >= 70. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Video=100k, Single Image=700k, Multi Image=-", "value": "70.0"}, {"item": "Video=100k, Single Image=700k, Multi Image=20k", "value": "70.3"}, {"item": "Video=100k, Single Image=700k, Multi Image=40k", "value": "70.1"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002081 | Aggregation & Comparison | Argmax Reasoning | In column # Dim., which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "HEIM [34]", "value": "12"} |
| 2 | HB_real_002082 | Aggregation & Comparison | Argmax Reasoning | In column # Pmts. / Edits, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "HEIM [34]", "value": "500k"} |
| 3 | HB_real_002083 | Markup Grounding | Highlight Presence | Is the number with value 40.2k highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_002084 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002085 | Aggregation & Comparison | Best Selection | Within group "Text-to-image", which row label in the leftmost row-label column is best on column Bias? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DALL-E 3 [55]", "value": "0.91"} |
| 2 | HB_real_002086 | Aggregation & Comparison | Best Selection | Within group "Image-editing", which row label in the leftmost row-label column is best on column Avg (rightmost column under Robustness->Bias)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "HQEdit (w/ DTM)", "value": "0.98"} |
| 3 | HB_real_002087 | Aggregation & Comparison | Best Selection | Within group "Subject-driven Generation", which row label in the leftmost row-label column is best on column Bias? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "OmniGen [67]", "value": "0.96"} |
| 4 | HB_real_002088 | Markup Grounding | Highlight Presence | Is the number with value 0.98 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 5 | HB_real_002089 | Constrained Retrieval | Value Filtering | In column Bias, list all row labels in the primary row-label column with value >= 0.95. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "HQEdit [26]", "value": "0.97"}, {"item": "HQEdit (w/ DTM)", "value": "0.98"}, {"item": "OmniGen [67] (Subject-driven Generation)", "value": "0.96"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000210 | Aggregation & Comparison | Extremum Reasoning | From this table, for item SeaS, what is the minimum value across all numeric metrics in this row? Exclude N/A and —. Output only the number as a string. | "0.02" |
| 2 | HB_real_000488 | Aggregation & Comparison | Extremum Reasoning | For item SeaS, what is the maximum value across all numeric metrics in this row? Exclude N/A and —. Output only the number as a string. | "1.95" |
| 3 | HB_real_000835 | Aggregation & Comparison | Rank Selection | In column IS↑ (block MVTec 3D AD), list the top 2 methods by value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "SeaS", "value": "1.95"}, {"item": "DFMGAN[11]", "value": "1.80"}] |
| 4 | HB_real_002104 | Aggregation & Comparison | Best Selection | In metric KID↓ (block MVTec AD), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "SeaS", "value": "0.04"} |
| 5 | HB_real_002105 | Aggregation & Comparison | Best Selection | In metric IS↑ (block MVTec 3D AD), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "SeaS", "value": "1.95"} |
| 6 | HB_real_002106 | Consistency & Missingness | Column Missingness | List all missing entries in column KID↓ (block VisA) (N/A or —). Output only a JSON array of strings. | ["Crop&Paste[23]", "SDGAN[28]", "Defect-GAN[42]", "AnomalyDiffusion[17]"] |
| 7 | HB_real_002107 | Aggregation & Comparison | Rank Selection | In column KID↓ (block VisA), list the top 2 methods with the smallest values, sorted from lowest to highest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. Exclude missing values. | [{"item": "SeaS", "value": "0.02"}, {"item": "DFMGAN[11]", "value": "0.24"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000110 | Aggregation & Comparison | Grouped Aggregation | Within row DRAEM + SeaS, what is the mean value of AP across the three Image-level blocks (MVTec AD, VisA, MVTec 3D AD)? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "93.34" |
| 2 | HB_real_000118 | Aggregation & Comparison | Grouped Aggregation | Within row DRAEM + SeaS, what is the mean value of AUROC across the three Image-level blocks (MVTec AD, VisA, MVTec 3D AD)? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "90.74" |
| 3 | HB_real_000271 | Markup Grounding | Marked Value Listing | List all bold values in row DRAEM + SeaS. Output only a JSON array of strings. | ["98.64", "99.40", "97.89", "98.11", "76.55", "72.70", "88.12", "87.04", "83.04", "98.45", "49.05", "48.62", "35.00", "85.45", "93.58", "90.85", "95.43", "20.09", "26.10", "17.07"] |
| 4 | HB_real_000655 | Aggregation & Comparison | Extremum Reasoning | For row DRAEM + SeaS, what is the maximum value across all visible metrics? Output only the number as a string. | "99.40" |
| 5 | HB_real_002115 | Aggregation & Comparison | Best Selection | In metric AP↑ within block VisA Image-level, which row label is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "GLASS + SeaS", "value": "97.39"} |
| 6 | HB_real_002116 | Aggregation & Comparison | Pairwise Comparison | For Average(+ SeaS), compare AP and IoU in block MVTec 3D AD Pixel-level. Which metric value is larger? Return JSON {"item":"<metric name>","value":"<string>"}. | {"item": "AP", "value": "34.32"} |
| 7 | HB_real_002117 | Markup Grounding | Marker Counting | How many bold numeric cells are in row DRAEM + SeaS? Output only the count as a string. | "20" |
| 8 | HB_real_002118 | Aggregation & Comparison | Argmax Reasoning | In column AP within block MVTec AD Image-level, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 2, "col": 2, "value": "99.40"} |
| 9 | HB_real_002119 | Consistency & Missingness | Column Missingness | In column AUROC within block VisA Pixel-level, list the row labels whose entry is missing. Output only a JSON array of strings. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000090 | Aggregation & Comparison | Extremum Reasoning | For item PatchCore + SeaS in block MVTec AD Image-level, what is the maximum value across metrics AUROC, AP, and F1-max? Output only the number as a string. | "99.48" |
| 2 | HB_real_000474 | Aggregation & Comparison | Extremum Reasoning | For item PatchCore + SeaS in block MVTec AD Image-level, what is the minimum value across metrics AUROC, AP, and F1-max? Output only the number as a string. | "98.22" |
| 3 | HB_real_000481 | Markup Grounding | Marked Value Listing | List all bold values for PatchCore + SeaS in block MVTec AD Image-level. Output only a JSON array of strings. | ["98.64", "99.48", "98.22"] |
| 4 | HB_real_000633 | Constrained Retrieval | Value Filtering | In column AUROC within block MVTec 3D AD Image-level, list all row labels with value >= 85. Return a JSON array of {"item":"<row label>","value":"<cell value>"} objects. | [{"item": "MambaAD [15]", "value": "85.92"}, {"item": "MambaAD + SeaS", "value": "88.67"}] |
| 5 | HB_real_000826 | Markup Grounding | Marker Counting | How many bold numeric cells are in block MVTec AD Image-level for row PatchCore + SeaS? Output only the count as a string. | "3" |
| 6 | HB_real_002142 | Aggregation & Comparison | Best Selection | In metric IoU↑ within block MVTec AD Pixel-level, which row label is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "PatchCore + SeaS", "value": "55.43"} |
| 7 | HB_real_002143 | Aggregation & Comparison | Rank Selection | Within block MVTec AD Image-level, who ranks 2nd in column AP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "HVQ-Trans + SeaS", "value": "98.48"} |
| 8 | HB_real_002144 | Aggregation & Comparison | Rank Selection | Within block MVTec AD Image-level, who ranks 2nd in column F1-max? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "PatchCore [32]", "value": "98.18"} |
| 9 | HB_real_002145 | Consistency & Missingness | Cell Missingness | Is the cell at row HVQ-Trans [26], column AP, block MVTec AD Image-level missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000111 | Aggregation & Comparison | Grouped Aggregation | Within group Average (DFMGAN/AnomalyDiffusion/SeaS), what is the mean value of column AUROC in block VisA Image-level? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "77.82" |
| 2 | HB_real_000115 | Aggregation & Comparison | Grouped Aggregation | Within group Average (DFMGAN/AnomalyDiffusion/SeaS), what is the mean value of column AP in block VisA Image-level? Exclude missing entries. Output only the number as a string, rounded to two decimal places. | "77.88" |
| 3 | HB_real_000272 | Markup Grounding | Marked Value Listing | List all bold values in row Average SeaS. Output only a JSON array of strings. | ["96.72", "98.41", "95.97", "97.72", "73.59", "69.86", "57.66", "86.34", "86.75", "80.69", "95.32", "47.38", "49.29", "29.40", "78.38", "90.52", "87.27", "91.32", "34.93", "39.87", "35.37"] |
| 4 | HB_real_000659 | Aggregation & Comparison | Extremum Reasoning | For row Average SeaS, what is the maximum value across all visible metrics? Output only the number as a string. | "98.41" |
| 5 | HB_real_002150 | Aggregation & Comparison | Best Selection | In metric IoU↑ within block MVTec AD Pixel-level, which row label is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "UPerNet [38] + SeaS", "value": "61.24"} |
| 6 | HB_real_002151 | Aggregation & Comparison | Best Selection | Within LFD [45], which row label is best on AP within block MVTec 3D AD Pixel-level? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "SeaS", "value": "40.25"} |
| 7 | HB_real_002152 | Markup Grounding | Marker Counting | How many bold numeric cells are in row Average SeaS? Output only the count as a string. | "21" |
| 8 | HB_real_002153 | Markup Grounding | Bold Presence | Is the number with value 57.66 boldfaced? Answer JSON {"bold":true/false}. The value appears exactly once in the numeric part of the table. | {"bold": true} |
| 9 | HB_real_002154 | Consistency & Missingness | Column Missingness | In column IoU within block VisA Pixel-level, list the row labels whose entry is missing. Output only a JSON array of strings. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000267 | Constrained Retrieval | Value Filtering | From this table, considering the four non-reference rows in block Seqlen=32 and in order (Absolute+Local Relative, Absolute+Global Deltas [7, 28], Sequence Canonicalization [48], Absolute), list all % Diff values next to MPJPE. Output only a JSON array of strings. | ["2.4%", "4.9%", "17.9%", "23.2%"] |
| 2 | HB_real_000385 | Constrained Retrieval | Value Filtering | Considering the four non-reference rows in block Seqlen=32 and in order (Absolute+Local Relative, Absolute+Global Deltas [7, 28], Sequence Canonicalization [48], Absolute), list all % Diff values next to PA-MPJPE. Output only a JSON array of strings. | ["3.4%", "7.7%", "17.1%", "28.4%"] |
| 3 | HB_real_000905 | Constrained Retrieval | Value Filtering | Considering the four non-reference rows in block Seqlen=128 and in order (Absolute+Local Relative, Absolute+Global Deltas [7, 28], Sequence Canonicalization [48], Absolute), list all % Diff values next to MPJPE. Output only a JSON array of strings. | ["4.0%", "6.4%", "11.9%", "23.9%"] |
| 4 | HB_real_002090 | Aggregation & Comparison | Best Selection | In metric PA-MPJPE↓ (block Seqlen=128), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "EgoAllo (Eq. 4)", "value": "101.1±1.3"} |
| 5 | HB_real_002092 | Constrained Retrieval | Value Filtering | Considering the five rows in block Seqlen=128 from top to bottom, list all GND↑ values. Output only a JSON array of strings. | ["1.00±0.00", "1.00±0.00", "0.99±0.00", "0.88±0.02", "0.96±0.01"] |
| 6 | HB_real_002093 | Markup Grounding | Highlight Presence | Is the number with value 119.7±1.3 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 7 | HB_real_001652 | Aggregation & Comparison | Grouped Aggregation | In block Seqlen=32, among the four non-reference rows, which row has the largest % Diff next to MPJPE? Return the row index as a string. | "4" |
| 8 | HB_real_000781 | Markup Grounding | Marked Value Listing | List all values highlighted in light pink in the Seqlen=128 block. Output only a JSON array of strings. | ["127.4±1.3", "6.4%", "136.8±1.6", "11.9%", "152.4±1.6", "23.9%"] |
| 9 | HB_real_rebal_000044 | Aggregation & Comparison | Grouped Aggregation | For SeqLen=128, considering the PA-MPJPE % Diff column from top to bottom, which row has the largest percentage? Return the row index as a string. | "5" |
| 10 | HB_real_rebal_000045 | Aggregation & Comparison | Grouped Aggregation | Within the two red-highlighted SeqLen=32 rows, considering the MPJPE % Diff values from top to bottom, which row has the larger percentage? Return the row index as a string. | "2" |
| 11 | HB_real_rebal_000050 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 12 | HB_real_rebal_000051 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Conditioning and Seqlen columns. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000560 | Constrained Retrieval | Value Filtering | From this table image, in column GND↑ (group Aria Digital Twins, Seq=128), list all row labels in the primary row-label column with value >= 0.90. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "EgoAllo", "value": "0.94±0.01"}, {"item": "NoShape", "value": "0.96±0.01"}] |
| 2 | HB_real_002094 | Aggregation & Comparison | Best Selection | Within group "AMASS (Seq=32)", which row label in the leftmost row-label column is best on column MPJPE↓? Here ↓ means smaller is better. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "EgoAllo", "value": "129.8±1.1"} |
| 3 | HB_real_002095 | Aggregation & Comparison | Best Selection | Within group "RICH (Seq=128)", which row label in the leftmost row-label column is best on column PA-MPJPE↓? Here ↓ means smaller is better. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "EgoAllo", "value": "160.1±5.9"} |
| 4 | HB_real_002096 | Constrained Retrieval | Value Filtering | In column GND↑ (group RICH, Seq=128), list all row labels in the primary row-label column with value >= 0.90. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "EgoAllo", "value": "0.96±0.02"}] |
| 5 | HB_real_002097 | Consistency & Missingness | Cell Missingness | Is the cell (row VAE+Opt, column T_head↓, block Aria Digital Twins, Seq=32) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_real_002098 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000636 | Constrained Retrieval | Value Filtering | According to the table image, in column MPJPE↓, list all row labels in the primary row-label column with value <= 140. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "EgoAllo-Mono", "value": "131.45±0.39"}, {"item": "EgoAllo-Wrist3D", "value": "60.08±0.26"}] |
| 2 | HB_real_002099 | Aggregation & Comparison | Best Selection | In metric MPJPE↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "EgoAllo-Wrist3D", "value": "60.08±0.26"} |
| 3 | HB_real_002100 | Aggregation & Comparison | Best Selection | In metric PA-MPJPE↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "HaMeR", "value": "13.04±1.89"} |
| 4 | HB_real_002101 | Aggregation & Comparison | Pairwise Comparison | In column MPJPE↓, compare EgoAllo-NoReproj and EgoAllo-Mono. Which numeric value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "EgoAllo-NoReproj", "value": "143.20±0.42"} |
| 5 | HB_real_002102 | Aggregation & Comparison | Pairwise Comparison | For row EgoAllo-Wrist3D, compare column MPJPE↓ and column PA-MPJPE↓. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "MPJPE↓", "value": "60.08±0.26"} |
| 6 | HB_real_002103 | Constrained Retrieval | Value Filtering | In column PA-MPJPE↓, list all row labels in the primary row-label column with value <= 15. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "HaMeR", "value": "13.04±1.89"}, {"item": "EgoAllo-NoReproj", "value": "14.75±0.42"}, {"item": "EgoAllo-Mono", "value": "14.71±0.39"}, {"item": "EgoAllo-Wrist3D", "value": "14.38±0.26"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002108 | Aggregation & Comparison | Best Selection | Within group "PASCAL-5^i / 1-shot", which row label in the leftmost row-label column is best on column Mean? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DSV-LFS", "value": "74.96"} |
| 2 | HB_real_002109 | Aggregation & Comparison | Best Selection | Within group "PASCAL-5^i / 5-shot", which row label in the leftmost row-label column is best on column Mean? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LLaFS [68]", "value": "75.6"} |
| 3 | HB_real_002110 | Aggregation & Comparison | Best Selection | Within group "COCO-20^i / 1-shot", which row label in the leftmost row-label column is best on column Mean? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DSV-LFS", "value": "71.33"} |
| 4 | HB_real_002111 | Aggregation & Comparison | Best Selection | Within group "COCO-20^i / 5-shot", which row label in the leftmost row-label column is best on column Mean? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DSV-LFS", "value": "71.90"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002114 | Markup Grounding | Highlight Presence | Is the number with value 74.86 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 2 | HB_real_001030 | Markup Grounding | Marked Value Listing | List all values in the light-green highlighted DSV-LFS row. Output only a JSON array of strings. | ["74.86", "85.23", "82.23", "80.37"] |
| 3 | HB_real_000999 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in light green? Output only the count as a string. | "4" |
| 4 | HB_real_002260 | Constrained Retrieval | Value Filtering | In the light-green highlighted DSV-LFS row, list the numeric values greater than 82. Output only a JSON array of strings. | ["85.23", "82.23"] |
| 5 | HB_real_000712 | Aggregation & Comparison | Grouped Aggregation | Exclude gray highlighted cells. Among the remaining values in column mIoU, what is the maximum value? Output only the number as a string. | "79.6" |
| 6 | HB_real_001812 | Aggregation & Comparison | Argmax Reasoning | In the numeric part of the table, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and numeric columns left-to-right. | {"row": 10, "col": 2, "value": "85.23"} |
| 7 | HB_real_000982 | Markup Grounding | Bold Presence | At position (r=10,c=3) in the numeric part of the table, is the value boldfaced? Answer JSON {"bold":true/false}. The numeric part excludes the left Methods column. | {"bold": true} |
| 8 | HB_real_001997 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | ["71.0", "72.3", "71.1", "79.8", "79.6"] |
| 9 | HB_real_001910 | Markup Grounding | Underline Presence | The value 74.86 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 10 | HB_real_001911 | Constrained Retrieval | Color-Anchor Retrieval | In the DSV-LFS row, list the Fold-0 through Fold-3 values highlighted with the same light-green color as the anchor at Fold-0. Output only a JSON array of strings. | ["74.86", "85.23", "82.23", "80.37"] |
| 11 | HB_real_001914 | Constrained Retrieval | Color-Anchor Retrieval | For the value 80.67, list the mIoU-column values highlighted with the same gray color as the anchor. Output only a JSON array of strings. | ["80.67"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002123 | Markup Grounding | Highlight Presence | Is the number with value 0.940 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 2 | HB_real_001035 | Markup Grounding | Marked Value Listing | In the light-blue highlighted Effort (Ours) row, list the Cross-dataset Evaluation values only. Output only a JSON array of strings. | ["0.956", "0.965", "0.843", "0.909", "0.977", "0.848", "0.921", "0.917"] |
| 3 | HB_real_000971 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=13,c=1) in the numeric/text data part of the table, list all entries highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. The data part excludes the left Methods column. | ["0.19M", "0.956", "0.965", "0.843", "0.909", "0.977", "0.848", "0.921", "0.917", "0.962", "0.873", "0.953", "0.983", "0.926", "0.957", "0.936", "0.926", "0.940"] |
| 4 | HB_real_000973 | Constrained Retrieval | Color-Anchor Retrieval | For the value 0.983, list the Cross-method Evaluation values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["0.962", "0.873", "0.953", "0.983", "0.926", "0.957", "0.936", "0.926", "0.940"] |
| 5 | HB_real_001006 | Markup Grounding | Marker Counting | How many highlighted data entries are in the light-blue Effort (Ours) row, excluding the left Effort (Ours) row label and including Trainable Param. plus both evaluation sections? Output only the count as a string. | "18" |
| 6 | HB_real_000974 | Constrained Retrieval | Value Filtering | In the light-blue highlighted Effort (Ours) row, list the evaluation values greater than 0.95. Output only a JSON array of strings. | ["0.956", "0.965", "0.977", "0.962", "0.953", "0.983", "0.957"] |
| 7 | HB_real_000975 | Aggregation & Comparison | Grouped Aggregation | Exclude the light-blue highlighted Effort (Ours) row. Among the remaining values in the Cross-method Evaluation Avg. column, what is the maximum value? Output only the number as a string. | "0.867" |
| 8 | HB_real_001871 | Aggregation & Comparison | Argmax Reasoning | Within the Cross-method Evaluation section, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and section columns left-to-right. | {"row": 13, "col": 4, "value": "0.983"} |
| 9 | HB_real_001333 | Local Relations | Neighbor Indexing | Within the Cross-method Evaluation section, at position (r=13,c=5), return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.926", "up": "0.747", "down": null, "left": "0.983", "right": "0.957"} |
| 10 | HB_real_002268 | Local Relations | Neighbor Value Tracing | The value 0.983 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "0.953", "right": "0.926"} |
| 11 | HB_real_001932 | Constrained Retrieval | Cell Retrieval | Within the Cross-method Evaluation section, find the value at position (r=13,c=9). Output only the number as a string. | "0.940" |
| 12 | HB_real_001880 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002124 | Aggregation & Comparison | Argmax Reasoning | In column Dalle, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Effort (Ours)", "value": "99.96"} |
| 2 | HB_real_002125 | Aggregation & Comparison | Pairwise Comparison | For row Effort (Ours), compare column Dalle and column Guided. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Dalle", "value": "99.96"} |
| 3 | HB_real_000340 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=10,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["100.0", "100.0", "99.99", "99.77", "100.0", "100.0", "98.95", "97.53", "97.53", "100.0", "100.0", "95.39", "99.99", "99.89", "100.0", "99.87", "99.92", "99.98", "99.96", "99.41"] |
| 4 | HB_real_002253 | Aggregation & Comparison | Grouped Aggregation | Take the cells in the light-blue Effort (Ours) row excluding the value 100.0. What is their maximum numeric value? Output only the number as a string. | "99.99" |
| 5 | HB_real_000934 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 1, "col": "Pro-GAN", "value": "100.0"} |
| 6 | HB_real_000605 | Markup Grounding | Bold Presence | Is the value at position (r=10,c=19) in the numeric part of the table highlighted with the light-blue Effort (Ours) row background? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 7 | HB_real_000944 | Markup Grounding | Marker Counting | How many numeric cells are highlighted in the light-blue Effort (Ours) row? Output only the count as a string. | "20" |
| 8 | HB_real_000954 | Aggregation & Comparison | Grouped Aggregation | Exclude the light-blue Effort (Ours) row. On the remaining cells in column Dalle, what is the maximum value? Output only the number as a string. | "99.82" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002126 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column mAcc? Output only the number as a string. | "95.19" |
| 2 | HB_real_002127 | Aggregation & Comparison | Argmax Reasoning | In column Dalle, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "FatFormer† (Liu et al., 2024)", "value": "98.75"} |
| 3 | HB_real_002128 | Aggregation & Comparison | Pairwise Comparison | For row Effort (Ours), compare column Guided and column e4s. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "e4s", "value": "96.80"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002129 | Aggregation & Comparison | Pairwise Comparison | In column CDF-v2, compare row (SVD=on, L_ksv=on, L_orth=off) and row (SVD=on, L_ksv=off, L_orth=on). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "SVD=on, L_ksv=off, L_orth=on", "value": "0.945"} |
| 2 | HB_real_002130 | Aggregation & Comparison | Pairwise Comparison | For row (SVD=on, L_ksv=on, L_orth=on), compare column CDF-v2 and column SimSwap. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "CDF-v2", "value": "0.956"} |
| 3 | HB_real_002131 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000266 | Constrained Retrieval | Value Filtering | Using the table only, considering the three "+ Ours" rows in order (BEIT-v2, SigLIP, CLIP), list the Avg. gains over their corresponding base rows. Output only a JSON array of strings. | ["+0.034", "+0.072", "+0.082"] |
| 2 | HB_real_000381 | Constrained Retrieval | Value Filtering | Considering the three "+ Ours" rows in order (BEIT-v2, SigLIP, CLIP), list the CDF-v2 gains over their corresponding base rows. Output only a JSON array of strings. | ["+0.039", "+0.018", "+0.099"] |
| 3 | HB_real_000749 | Consistency & Missingness | Consistency Probing | According to the table, for the Avg. gains over base rows (scope: + Ours rows), list all items whose gain is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "BEIT-v2 + Ours", "value": "+0.034"}, {"item": "SigLIP + Ours", "value": "+0.072"}, {"item": "CLIP + Ours", "value": "+0.082"}] |
| 4 | HB_real_000908 | Constrained Retrieval | Value Filtering | Considering the three "+ Ours" rows in order (BEIT-v2, SigLIP, CLIP), list the SimSwap gains over their corresponding base rows. Output only a JSON array of strings. | ["+0.029", "+0.065", "+0.066"] |
| 5 | HB_real_002132 | Aggregation & Comparison | Best Selection | Within group "BEIT-v2 (Peng et al., 2022)", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "+ Ours", "value": "0.872"} |
| 6 | HB_real_002133 | Aggregation & Comparison | Best Selection | Within group "CLIP (Radford et al., 2021)", which row label in the leftmost row-label column is best on column Avg.? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "+ Ours", "value": "0.941"} |
| 7 | HB_real_002134 | Aggregation & Comparison | Pairwise Comparison | In column CDF-v2, compare CLIP (Radford et al., 2021) and its + Ours row. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "+ Ours (for CLIP)", "value": "0.956"} |
| 8 | HB_real_002135 | Aggregation & Comparison | Rank Selection | In column Avg., list the top 2 rows by value, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "CLIP + Ours", "value": "0.941"}, {"item": "BEIT-v2 + Ours", "value": "0.872"}] |
| 9 | HB_real_002136 | Consistency & Missingness | Consistency Probing | For the CDF-v2 gains over base rows (scope: + Ours rows), list all items whose gain is positive (>0). Return JSON array of {"item":"...","value":"..."}. | [{"item": "BEIT-v2 + Ours", "value": "+0.039"}, {"item": "SigLIP + Ours", "value": "+0.018"}, {"item": "CLIP + Ours", "value": "+0.099"}] |
| 10 | HB_real_002137 | Constrained Retrieval | Value Filtering | In column Avg., list all row labels in the primary row-label column with value >= 0.90. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "CLIP + Ours", "value": "0.941"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000202 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 2 | HB_real_000207 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=12,c=3) in the numeric part of the table. Output only the value as a string. | "95.19" |
| 3 | HB_real_000363 | Markup Grounding | Marked Column Listing | In column mAcc, list all underlined numbers. If none, output an empty JSON array []. | [] |
| 4 | HB_real_000365 | Markup Grounding | Underline Presence | Is the value at position (r=12,c=3) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 5 | HB_real_000733 | Consistency & Missingness | Column Missingness | In column n − r, list the row labels whose entry is missing. Output only a JSON array of strings. | ["UniFD (Linear-Prob)", "Baseline (FFT)", "LoRA Variant-1", "LoRA Variant-2", "LoRA Variant-3", "LoRA Variant-4", "LoRA Variant-5"] |
| 6 | HB_real_000952 | Consistency & Missingness | Column Missingness | In column r, list the row labels whose entry is missing. Output only a JSON array of strings. | ["UniFD (Linear-Prob)", "Baseline (FFT)", "Ours Variant-1", "Ours Variant-2", "Ours Variant-3", "Ours Variant-4", "Ours Variant-5"] |
| 7 | HB_real_002138 | Aggregation & Comparison | Argmax Reasoning | Which method has the maximum value in column mAcc? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours Variant-5", "value": "95.19"} |
| 8 | HB_real_002139 | Aggregation & Comparison | Best Selection | Within group LoRA, which variant is best on column mAcc? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Variant-4", "value": "93.53"} |
| 9 | HB_real_002140 | Constrained Retrieval | Value Filtering | In column mAcc, list all row labels with value >= 94.00. Return a JSON array of {"item":"<row label>","value":"<cell value>"} objects. | [{"item": "Ours Variant-3", "value": "94.45"}, {"item": "Ours Variant-4", "value": "94.37"}, {"item": "Ours Variant-5", "value": "95.19"}] |
| 10 | HB_real_002141 | Aggregation & Comparison | Rank Selection | In column mAcc, which method ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours Variant-3", "value": "94.45"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000493 | Aggregation & Comparison | Rank Selection | Based on the table image, within block 13B, list the top-2 methods by column Math, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "TR", "value": "2.21"}, {"item": "Medusa", "value": "1.98"}] |
| 2 | HB_real_000503 | Aggregation & Comparison | Rank Selection | Within block 7B, list the top-2 methods by column Math, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "TR", "value": "2.40"}, {"item": "Medusa", "value": "1.87"}] |
| 3 | HB_real_000863 | Aggregation & Comparison | Rank Selection | Within block 33B, list the top-2 methods by column Math, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "TR", "value": "2.24"}, {"item": "Medusa", "value": "2.09"}] |
| 4 | HB_real_002146 | Aggregation & Comparison | Argmax Reasoning | Which item has the maximum value in column SpecBench Ts/s? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "TR (7B)", "value": "110.06"} |
| 5 | HB_real_002147 | Consistency & Missingness | Cell Missingness | Is the cell (row Medusa (33B), column MAT (block MBPP)) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_real_002148 | Markup Grounding | Bold Presence | Is the number with value 3.08 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 7 | HB_real_002149 | Aggregation & Comparison | Rank Selection | Within block 13B, list the top-2 methods by column MBPP Speed, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. Exclude missing values. | [{"item": "TR", "value": "2.26"}, {"item": "REST", "value": "1.70"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002155 | Aggregation & Comparison | Best Selection | In metric PI (MMLU) R0 (block Random), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Deepseek-V3", "value": "3.97"} |
| 2 | HB_real_002156 | Aggregation & Comparison | Best Selection | In metric PI (MMLU) R0 (block Star), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Deepseek-V3", "value": "6.68"} |
| 3 | HB_real_002157 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column TA (InjecAgent) R0? Output only the number as a string. | "0.00" |
| 4 | HB_real_002158 | Constrained Retrieval | Value Filtering | In column PI (GSM8k) R0, list all row labels in the primary row-label column with value < 7.00. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "Chain / Claude-3.5-haiku", "value": "6.67"}, {"item": "Star / Claude-3.5-haiku", "value": "6.25"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000200 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 2 | HB_real_000361 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=4,c=4) in the numeric part of the table. Output only the value as a string. | "50.77/11.54" |
| 3 | HB_real_000735 | Markup Grounding | Underline Presence | Is the value at position (r=4,c=4) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 4 | HB_real_000951 | Markup Grounding | Marker Counting | How many GS numbers are shown after the slash in the Rounds numeric data cells? Count each displayed GS number, including repeated values. Output only the count as a string. | "15" |
| 5 | HB_real_002159 | Constrained Retrieval | Cell Retrieval | What is the value at column R2/R2+GS for row 65? Output only the full cell string. | "44.62/9.62" |
| 6 | HB_real_002160 | Constrained Retrieval | Cell Retrieval | What is the value at column R1/R1+GS for row 80? Output only the full cell string. | "5.29/1.25" |
| 7 | HB_real_002161 | Constrained Retrieval | Cell Retrieval | What is the value at column R3/R3+GS for row 35? Output only the full cell string. | "26.66/3.04" |
| 8 | HB_real_002162 | Markup Grounding | Bold Presence | Is the number with value 11.54 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 9 | HB_real_002163 | Markup Grounding | Marked Value Listing | List all bold numbers in row 65. Output only a JSON array of strings. | ["8.46", "9.62", "11.54"] |
| 10 | HB_real_002164 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000269 | Markup Grounding | Underline Presence | Based on the table, is the number with value 54.68 underlined? Answer JSON {"underlined":true/false}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 2 | HB_real_002165 | Aggregation & Comparison | Best Selection | Within group LoRA Rank=16, which method is best on column IconQA Avg? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LoRASculpt", "value": "70.13"} |
| 3 | HB_real_002166 | Aggregation & Comparison | Argmax Reasoning | Which item has the maximum value in column IconQA Avg? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LoRASculpt (Rank=32)", "value": "70.68"} |
| 4 | HB_real_002167 | Markup Grounding | Bold Presence | Is the number with value 121.26 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 5 | HB_real_002168 | Markup Grounding | Marked Value Listing | In the IconQA Target column, list all underlined numeric values. Output only a JSON array of strings. | ["84.66", "84.82", "81.79"] |
| 6 | HB_real_002169 | Markup Grounding | Highlight Presence | Is the number with value 70.04 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 7 | HB_real_001821 | Markup Grounding | Marked Value Listing | List all red numeric increase annotations in the table. Output only a JSON array of strings. | ["↑0.66", "↑0.83", "↑4.31", "↑1.01", "↑8.00", "↑3.45"] |
| 8 | HB_real_000747 | Consistency & Missingness | Consistency Probing | For the LoRASculpt rows, list all Avg. cells that show a positive red delta annotation. Return JSON array of {"item":"...","value":"..."}. | [{"item": "LoRA Rank=16 IconQA Avg", "value": "↑0.66"}, {"item": "LoRA Rank=16 COCO-Caption Avg", "value": "↑0.83"}, {"item": "LoRA Rank=32 IconQA Avg", "value": "↑4.31"}, {"item": "LoRA Rank=32 COCO-Caption Avg", "value": "↑1.01"}, {"item": "LoRA Rank=64 IconQA Avg", "value": "↑8.00"}, {"item": "LoRA Rank=64 COCO-Caption Avg", "value": "↑3.45"}] |
| 9 | HB_real_001989 | Constrained Retrieval | Color-Anchor Retrieval | In the light-blue LoRASculpt row under LoRA Rank=16, list the black main numeric values only, excluding red increase annotations. Output only a JSON array of strings. | ["54.07", "57.00", "57.66", "52.24", "55.24", "85.02", "70.13", "50.86", "59.10", "57.64", "51.10", "54.68", "121.26", "87.97"] |
| 10 | HB_real_000967 | Markup Grounding | Marked Column Listing | List all underlined values in column Source under COCO-Caption. Output only a JSON array of strings. | ["54.68", "54.25", "51.77"] |
| 11 | HB_real_001629 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light blue in LoRASculpt rows. What is the maximum numeric value? Output only the number as a string. | "121.26" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002170 | Aggregation & Comparison | Pairwise Comparison | For row Ours Ours, compare column Source and column Target. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Target", "value": "85.34"} |
| 2 | HB_real_002171 | Markup Grounding | Bold Presence | Is the number with value 72.45 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 3 | HB_real_000850 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=6,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-blue background as the anchor. Output only a JSON array of strings. | ["56.91", "64.15", "60.93", "56.22", "59.55", "85.34", "72.45"] |
| 4 | HB_real_000687 | Markup Grounding | Marked Value Listing | List all bold numeric values in row Ours / Ours. Output only a JSON array of strings. | ["59.55", "85.34", "72.45"] |
| 5 | HB_real_000719 | Markup Grounding | Marker Counting | How many bold numeric cells are in row Ours / Ours? Output only the count as a string. | "3" |
| 6 | HB_real_000290 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 6, "col": "Target", "value": "85.34"} |
| 7 | HB_real_000801 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. On the remaining cells in column Target, what is the maximum value? Output only the number as a string. | "84.74" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002172 | Markup Grounding | Highlight Presence | Is the number with value 87.30 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 2 | HB_real_002173 | Markup Grounding | Bold Presence | Is the number with value 120.35 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 3 | HB_real_001049 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=4,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-blue background as the anchor. Output only a JSON array of strings. | ["56.10", "85.26", "70.68", "54.25", "120.35", "87.30"] |
| 4 | HB_real_000720 | Markup Grounding | Marker Counting | How many bold numeric cells are in the light-blue Ours/Ours row? Output only the count as a string. | "6" |
| 5 | HB_real_000291 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 4, "col": "COCO-Caption Target", "value": "120.35"} |
| 6 | HB_real_000803 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. On the remaining cells in column COCO-Caption Target, what is the maximum value? Output only the number as a string. | "120.11" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000201 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. | "0" |
| 2 | HB_real_000206 | Markup Grounding | Marked Value Listing | List all underlined numbers in the table. Output only a JSON array of strings. | [] |
| 3 | HB_real_000362 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 4 | HB_real_000727 | Aggregation & Comparison | Rank Selection | Who ranks 2nd best in column FID-50k↓? Lower values are better. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "+ Rotary Pos Embed [36]", "value": "7.13"} |
| 5 | HB_real_002174 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column FID-50k↓? Output only the number as a string. Note: ↓ means smaller is better. | "4.29" |
| 6 | HB_real_002175 | Aggregation & Comparison | Best Selection | In metric FID-50k↓ within block Architecture Improvements, which row label is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "+ patch size=1 & VA-VAE (Sec. 3)", "value": "4.29"} |
| 7 | HB_real_002176 | Constrained Retrieval | Cell Retrieval | What is the value at column FID-50k↓ for row + Rotary Pos Embed [36]? Output only the number as a string. | "7.13" |
| 8 | HB_real_002177 | Aggregation & Comparison | Argmax Reasoning | In column FID-50k↓, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 1, "col": 1, "value": "19.50"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002178 | Aggregation & Comparison | Best Selection | In metric LightningDiT-B (block f16d32), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "LDM+VF loss (DINOv2)", "value": "15.82"} |
| 2 | HB_real_002179 | Aggregation & Comparison | Pairwise Comparison | In column rFID↓ (block f16d64), compare LDM+VF loss (MAE) vs LDM+VF loss (DINOv2). Which one is better under ↓ (smaller is better)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "LDM+VF loss (DINOv2)", "value": "0.14"} |
| 3 | HB_real_002180 | Aggregation & Comparison | Pairwise Comparison | For row LDM (block f16d32), compare column rFID↓ and column LPIPS↓. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "rFID↓", "value": "0.26"} |
| 4 | HB_real_002181 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column rFID↓? Output only the number as a string. Note: ↓ means smaller is better. | "0.14" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002185 | Markup Grounding | Highlight Presence | Is the number with value 5.14 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 2 | HB_real_001036 | Markup Grounding | Marked Value Listing | In the purple highlighted LightningDiT row with Training Epochs 800, list all highlighted data entries from Training Epochs through Generation w/ CFG Rec. Output only a JSON array of strings. | ["800", "675M", "2.17", "4.36", "205.6", "0.77", "0.65", "1.35", "4.15", "295.3", "0.79", "0.65"] |
| 3 | HB_real_001019 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=13,c=4) in the numeric/text data part of the table, list all entries highlighted with the same purple color as the anchor row. Output only a JSON array of strings. The data part excludes the left Method and Reconstruction columns. | ["64", "675M", "5.14", "4.22", "130.2", "0.76", "0.62", "2.11", "4.16", "252.3", "0.81", "0.58"] |
| 4 | HB_real_001083 | Markup Grounding | Marker Counting | How many data entries are highlighted in purple across the two LightningDiT lines, excluding the left Method and Reconstruction columns? Output only the count as a string. | "24" |
| 5 | HB_real_001020 | Constrained Retrieval | Value Filtering | In the purple highlighted LightningDiT rows, list the numeric values less than 2.0. Output only a JSON array of strings. | ["0.76", "0.62", "0.81", "0.58", "0.77", "0.65", "1.35", "0.79", "0.65"] |
| 6 | HB_real_001022 | Markup Grounding | Marked Value Listing | In the LightningDiT rows, list all bold numeric values. Output only a JSON array of strings. | ["0.28", "2.17", "205.6", "0.77", "1.35", "4.15", "0.65"] |
| 7 | HB_real_001988 | Markup Grounding | Marker Counting | How many bold numeric values are in the LightningDiT rows? Output only the count as a string. | "7" |
| 8 | HB_real_001994 | Markup Grounding | Bold Presence | At position (r=14,c=10) in the numeric/text data part of the table, is the value boldfaced? Answer JSON {"bold":true/false}. The data part excludes the left Method and Reconstruction columns. | {"bold": true} |
| 9 | HB_real_001023 | Aggregation & Comparison | Grouped Aggregation | Exclude purple highlighted cells. Among the remaining values in column gFID under Generation w/ CFG, what is the minimum value? Output only the number as a string. | "1.42" |
| 10 | HB_real_001024 | Aggregation & Comparison | Argmax Reasoning | Within the Generation w/ CFG section, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and section columns left-to-right. | {"row": 3, "col": 3, "value": "365.4"} |
| 11 | HB_real_001025 | Local Relations | Neighbor Value Tracing | The value 205.6 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "4.36", "right": "0.77"} |
| 12 | HB_real_001953 | Markup Grounding | Underline Presence | The value 2.17 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 13 | HB_real_001956 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 14 | HB_real_001960 | Consistency & Missingness | Column Missingness | In column sFID under Generation w/o CFG, list the method rows whose entry is shown as missing using "-". Output only a JSON array of strings. | ["MaskGIT", "MagViT-v2", "REPA"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002186 | Aggregation & Comparison | Best Selection | Within group f16d32, which VF Loss setting is best on column gFID (DiT-B)↓? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DINOv2", "value": "15.82"} |
| 2 | HB_real_002187 | Aggregation & Comparison | Best Selection | Within group f16d64, which VF Loss setting is best on column gini coefficient↓? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "DINOv2", "value": "0.141"} |
| 3 | HB_real_002188 | Aggregation & Comparison | Pairwise Comparison | For row f16d32 + DINOv2, compare column density cv↓ and column gini coefficient↓. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "density cv↓", "value": "0.178"} |
| 4 | HB_real_002189 | Markup Grounding | Bold Presence | Is the number with value 23.58 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 5 | HB_real_rebal_000015 | Markup Grounding | Marked Value Listing | List all bold numeric values in the table. Output only a JSON array of strings. | ["0.178", "0.096", "0.998", "15.82", "23.58", "0.251", "0.141", "0.996"] |
| 6 | HB_real_rebal_000016 | Markup Grounding | Marked Column Listing | In column density cv↓, list all bold values. Output only a JSON array of strings. | ["0.178", "0.251"] |
| 7 | HB_real_rebal_000017 | Markup Grounding | Marker Counting | How many bold numeric values are in the table? Output only the count as a string. | "8" |
| 8 | HB_real_rebal_000019 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 9 | HB_real_rebal_000020 | Markup Grounding | Underline Presence | The value 0.263 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 10 | HB_real_rebal_000023 | Aggregation & Comparison | Argmax Reasoning | In the numeric part of the table, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. The numeric part excludes Tokenizer and VF Loss. | {"row": 4, "col": 4, "value": "36.83"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000619 | Aggregation & Comparison | Pairwise Comparison | Using visual evidence in the table, in row scan65, compare MonoSDF vs Ours. This metric is Chamfer Distance (lower is better). Which one is better? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "MonoSDF", "value": "1.38"} |
| 2 | HB_real_002190 | Aggregation & Comparison | Best Selection | In metric Chamfer Distance↓ (group Average), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Ours", "value": "1.42"} |
| 3 | HB_real_002191 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Ours? Output only the number as a string. | "0.82" |
| 4 | HB_real_002192 | Aggregation & Comparison | Argmax Reasoning | Which row has the maximum value in column Ours? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "scan37", "value": "3.14"} |
| 5 | HB_real_002193 | Markup Grounding | Highlight Presence | Is the number with value 1.31 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 6 | HB_real_002194 | Aggregation & Comparison | Extremum Reasoning | In row scan65, what is the minimum Chamfer Distance value across all methods? Output only the number as a string. | "1.38" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000243 | Aggregation & Comparison | Rank Selection | Within block RANSAC rows, list the top-2 settings by column F-Score ↑, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "N_step=10, N_Ensemble=5", "value": "0.853"}, {"item": "N_step=10, N_Ensemble=1", "value": "0.850"}] |
| 2 | HB_real_000495 | Aggregation & Comparison | Rank Selection | List the top-2 settings by fastest Inference time ↓, sorted from lowest to highest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "N_step=1, N_Ensemble=1, No Alignment", "value": "0.829"}, {"item": "N_step=1, N_Ensemble=1, Least Square", "value": "0.836"}] |
| 3 | HB_real_000597 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column F-Score ↑ among all settings? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "N_step=10, N_Ensemble=1, RANSAC", "value": "0.850"} |
| 4 | HB_real_000866 | Aggregation & Comparison | Extremum Reasoning | For row N_step=1, N_Ensemble=1, RANSAC, what is the maximum value across Inference time and F-Score? Output only the number as a string. | "0.840" |
| 5 | HB_real_002195 | Aggregation & Comparison | Best Selection | In metric Inference time ↓, which row label is optimal? Lower values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "N_step=1, N_Ensemble=1, No Alignment", "value": "0.829"} |
| 6 | HB_real_002196 | Aggregation & Comparison | Best Selection | In metric F-Score ↑, which row label is optimal? Higher values indicate better performance. Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "N_step=10, N_Ensemble=5, RANSAC", "value": "0.853"} |
| 7 | HB_real_002197 | Aggregation & Comparison | Pairwise Comparison | For row N_step=1, N_Ensemble=1, Least Square, compare column Inference time ↓ and column F-Score ↑. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Inference time ↓", "value": "0.836"} |
| 8 | HB_real_002198 | Aggregation & Comparison | Argmax Reasoning | In column F-Score ↑, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 1, "col": 2, "value": "0.853"} |
| 9 | HB_real_002199 | Constrained Retrieval | Value Filtering | In column Inference time ↓, list all settings with value <= 1.000. Return a JSON array of {"item":"<row label>","value":"<cell value>"} objects. | [{"item": "N_step=1, N_Ensemble=1, RANSAC", "value": "0.840"}, {"item": "N_step=1, N_Ensemble=1, Least Square", "value": "0.836"}, {"item": "N_step=1, N_Ensemble=1, No Alignment", "value": "0.829"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000242 | Aggregation & Comparison | Rank Selection | According to the table, within block Ours rows, list the top-2 methods by column Full(1464), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "ConformalSAM(Ours)", "value": "83.85"}, {"item": "ConformalSAM(AllSpark)(Ours)", "value": "83.44"}] |
| 2 | HB_real_000498 | Aggregation & Comparison | Rank Selection | Within block Ours rows, list the top-2 methods by column 1/2(732), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "ConformalSAM(Ours)", "value": "83.52"}, {"item": "ConformalSAM(AllSpark)(Ours)", "value": "82.69"}] |
| 3 | HB_real_000617 | Aggregation & Comparison | Pairwise Comparison | Based on the table image, in column 1/4(366), compare ConformalSAM(AllSpark)(Ours) vs ConformalSAM(Ours). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "ConformalSAM(Ours)", "value": "81.84"} |
| 4 | HB_real_000864 | Aggregation & Comparison | Rank Selection | Within block Ours rows, list the top-2 methods by column 1/4(366), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "ConformalSAM(Ours)", "value": "81.84"}, {"item": "ConformalSAM(AllSpark)(Ours)", "value": "81.33"}] |
| 5 | HB_real_002200 | Aggregation & Comparison | Argmax Reasoning | Which method has the maximum value in column 1/2(732)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "ConformalSAM(Ours)", "value": "83.52"} |
| 6 | HB_real_002201 | Aggregation & Comparison | Pairwise Comparison | In column 1/8(183), compare ConformalSAM(AllSpark)(Ours) vs ConformalSAM(Ours). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "ConformalSAM(Ours)", "value": "82.22"} |
| 7 | HB_real_002202 | Constrained Retrieval | Value Filtering | In column 1/16(92), list all methods with value >= 80.00. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "ConformalSAM(AllSpark)(Ours)", "value": "80.69"}, {"item": "ConformalSAM(Ours)", "value": "81.21"}] |
| 8 | HB_real_002203 | Aggregation & Comparison | Rank Selection | Within block Ours rows, list the top-2 methods by column 1/16(92), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "ConformalSAM(Ours)", "value": "81.21"}, {"item": "ConformalSAM(AllSpark)(Ours)", "value": "80.69"}] |
| 9 | HB_real_002204 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | ["CPS", "LogicDiag", "UniMatch"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000239 | Aggregation & Comparison | Rank Selection | Within block Left block Ours rows, list the top-2 methods by column 1/8(1323), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "ConformalSAM", "value": "80.93"}, {"item": "ConformalSAM(AllSpark)", "value": "80.41"}] |
| 2 | HB_real_000496 | Markup Grounding | Marked Value Listing | Return all numbers whose text is colored red in this table. Output only a JSON array of strings. | ["80.24", "80.93", "81.04", "81.43", "83.40", "83.75", "82.14", "81.98"] |
| 3 | HB_real_000502 | Markup Grounding | Marked Value Listing | Return all numbers whose text is colored blue in this table. Output only a JSON array of strings. | ["80.42", "81.14", "80.09", "80.41", "82.34", "83.31", "81.36", "81.13"] |
| 4 | HB_real_000595 | Aggregation & Comparison | Rank Selection | Within block Right block proposed rows, who ranks 2nd in column 1/8(1323)? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "ConformalSAM(AllSpark)†", "value": "83.31"} |
| 5 | HB_real_000865 | Aggregation & Comparison | Argmax Reasoning | In column 1/2(5291) within the right block, which method has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "ConformalSAM†", "value": "81.98"} |
| 6 | HB_real_002205 | Aggregation & Comparison | Pairwise Comparison | For row ConformalSAM† in the right block, compare column 1/8(1323) and column 1/2(5291). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "1/8(1323)", "value": "83.75"} |
| 7 | HB_real_002206 | Aggregation & Comparison | Rank Selection | Within block Right block proposed rows, list the top-2 methods by column 1/8(1323), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "ConformalSAM†", "value": "83.75"}, {"item": "ConformalSAM(AllSpark)†", "value": "83.31"}] |
| 8 | HB_real_002207 | Consistency & Missingness | Column Missingness | In column 1/2(5291) in the left block, list the row labels whose entry is missing. Output only a JSON array of strings. | ["AugSeg", "UniMatch", "CorrMatch", "RankMatch"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000256 | Markup Grounding | Marked Value Listing | List numbers whose text is colored blue in this table. Output only a JSON array of strings. | ["33.2", "24.10", "27.26", "31.40", "36.25"] |
| 2 | HB_real_000572 | Aggregation & Comparison | Rank Selection | In column 1/128(158), which method ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "ConformalSAM(AllSpark)", "value": "24.10"} |
| 3 | HB_real_000650 | Markup Grounding | Marked Value Listing | Return all numbers whose text is colored red in this table. Output only a JSON array of strings. | ["38.0", "26.21", "30.02", "33.33", "34.64"] |
| 4 | HB_real_000919 | Aggregation & Comparison | Rank Selection | In column 1/8(2526), list the top-2 methods sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "AEL", "value": "38.0"}, {"item": "ConformalSAM", "value": "36.25"}] |
| 5 | HB_real_001069 | Aggregation & Comparison | Extremum Reasoning | For row ConformalSAM, what is the maximum value across all ratio columns? Output only the number as a string. | "36.25" |
| 6 | HB_real_002208 | Aggregation & Comparison | Argmax Reasoning | In column 1/16(1263), which method has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "ConformalSAM", "value": "34.64"} |
| 7 | HB_real_002209 | Aggregation & Comparison | Argmax Reasoning | In column 1/8(2526), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 3, "col": 5, "value": "38.0"} |
| 8 | HB_real_002210 | Aggregation & Comparison | Pairwise Comparison | For row ConformalSAM, compare column 1/16(1263) and column 1/8(2526). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "1/8(2526)", "value": "36.25"} |
| 9 | HB_real_002211 | Aggregation & Comparison | Rank Selection | In column 1/32(632), which method ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "ConformalSAM(AllSpark)", "value": "31.40"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000253 | Markup Grounding | Marked Value Listing | Return all numbers whose text is colored red in this table. Output only a JSON array of strings. | ["81.21", "82.22", "81.84", "83.52"] |
| 2 | HB_real_000571 | Aggregation & Comparison | Rank Selection | In column 1/16(92), which setting ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "SEEM+CP (✓,✓,blank)", "value": "78.09"} |
| 3 | HB_real_000648 | Markup Grounding | Marked Value Listing | Return all numbers whose text is colored blue in this table. Output only a JSON array of strings. | ["78.09", "78.23", "78.32", "79.10"] |
| 4 | HB_real_000920 | Aggregation & Comparison | Rank Selection | In column 1/2(732), list the top-2 settings sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "SEEM+CP+SR (✓,✓,✓)", "value": "83.52"}, {"item": "SEEM+CP (✓,✓,blank)", "value": "79.10"}] |
| 5 | HB_real_001073 | Aggregation & Comparison | Extremum Reasoning | For row SEEM+CP+SR (✓,✓,✓), what is the maximum value across the four ratio columns? Output only the number as a string. | "83.52" |
| 6 | HB_real_002212 | Aggregation & Comparison | Argmax Reasoning | In column 1/4(366), which setting has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "SEEM+CP+SR (✓,✓,✓)", "value": "81.84"} |
| 7 | HB_real_002213 | Aggregation & Comparison | Argmax Reasoning | In column 1/2(732), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 5, "col": 4, "value": "83.52"} |
| 8 | HB_real_002214 | Aggregation & Comparison | Pairwise Comparison | For row SEEM+SR (✓,blank,✓), compare column 1/8(183) and column 1/4(366). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "1/8(183)", "value": "70.67"} |
| 9 | HB_real_002216 | Aggregation & Comparison | Pairwise Comparison | For row SEEM+CP+SR (✓,✓,✓), compare column 1/8(183) and column 1/2(732). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "1/2(732)", "value": "83.52"} |
| 10 | HB_real_002217 | Aggregation & Comparison | Rank Selection | In column 1/4(366), which setting ranks 2nd? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "SEEM+CP (✓,✓,blank)", "value": "78.32"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000923 | Markup Grounding | Marked Value Listing | Please answer this question: list numbers whose text is colored in red in this table. Output only a JSON array of strings. | ["78.09", "79.10"] |
| 2 | HB_real_001074 | Aggregation & Comparison | Rank Selection | In column 1/2(732), list the top 2 rows by value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "Pixel α=0.05", "value": "79.10"}, {"item": "K-Means α=0.1", "value": "78.23"}] |
| 3 | HB_real_002218 | Aggregation & Comparison | Extremum Reasoning | In row Pixel α=0.05, what is the maximum numeric value across the two PASCAL VOC 2012 columns? Output only the number as a string. | "79.10" |
| 4 | HB_real_002219 | Aggregation & Comparison | Argmax Reasoning | In column 1/2(732), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 2, "col": 2, "value": "79.10"} |
| 5 | HB_real_002220 | Aggregation & Comparison | Best Selection | Within group Pixel, which error-rate setting is best on column 1/2(732)? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "α=0.05", "value": "79.10"} |
| 6 | HB_real_002221 | Aggregation & Comparison | Pairwise Comparison | In column 1/2(732), compare Pixel α=0.05 vs Image α=0.1. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "Pixel α=0.05", "value": "79.10"} |
| 7 | HB_real_002222 | Markup Grounding | Marked Value Listing | List numbers whose text is colored in red in the 1/16(92) column. Output only a JSON array of strings. | ["78.09"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000075 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column Var? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "BTRM Qwen 2 7B", "value": "15.9"} |
| 2 | HB_real_000418 | Aggregation & Comparison | Rank Selection | Who ranks 1st in column Var? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "Tulu 2 13B DPO", "value": "25.0"} |
| 3 | HB_real_000422 | Aggregation & Comparison | Rank Selection | Who ranks 3rd in column Var? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "URM LlaMa 3.1 8B", "value": "11.8"} |
| 4 | HB_real_000792 | Aggregation & Comparison | Extremum Reasoning | For row BTRM Qwen 2 7B, what is the maximum value across language columns? Exclude Var. Output only the number as a string. | "75.5" |
| 5 | HB_real_002223 | Aggregation & Comparison | Rank Selection | List the top-3 items by column Avg, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. Exclude missing entries. | [{"item": "GPT-4 Turbo", "value": "83.5"}, {"item": "GPT-4o", "value": "81.1"}, {"item": "Gemma 2 9B", "value": "76.6"}] |
| 6 | HB_real_002224 | Aggregation & Comparison | Extremum Reasoning | For row Tulu 2 13B DPO, what is the minimum value across language columns? Exclude Var. Output only the number as a string. | "57.5" |
| 7 | HB_real_002225 | Aggregation & Comparison | Pairwise Comparison | For GPT-4 Turbo, compare ar and zh. Which metric value is larger? Return JSON {"item":"<metric name>","value":"<string>"}. | {"item": "ar", "value": "83.7"} |
| 8 | HB_real_002226 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Avg? Exclude missing entries. Output only the number as a string. | "68.1" |
| 9 | HB_real_002227 | Consistency & Missingness | Cell Missingness | Is the cell at row Tulu 2 13B DPO, column he missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000215 | Aggregation & Comparison | Extremum Reasoning | Using the table only, for item Llama 3.1 70B, what is the minimum value across all metrics in this row? Exclude N/A and —. Output only the number as a string. | "-11.62" |
| 2 | HB_real_000485 | Aggregation & Comparison | Rank Selection | List the top-2 items by column Chat, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. Exclude N/A and —. | [{"item": "Gemma 2 9B", "value": "-0.58"}, {"item": "Command R+", "value": "-0.76"}] |
| 3 | HB_real_000834 | Aggregation & Comparison | Rank Selection | List the top-2 items by column Chat-Hard, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. Exclude N/A and —. | [{"item": "Tulu 2 13B DPO", "value": "-2.34"}, {"item": "Aya Expanse 32B", "value": "-2.44"}] |
| 4 | HB_real_002228 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=5,c=2) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_red"} |
| 5 | HB_real_002229 | Aggregation & Comparison | Rank Selection | List the top-2 items by column Safety Reasoning, sorted from highest to lowest. Return a JSON array of {"item","value"}. Exclude N/A and —. | [{"item": "Tulu 2 13B DPO", "value": "1.04"}, {"item": "GPT-4 Turbo", "value": "0.84"}] |
| 6 | HB_real_002230 | Aggregation & Comparison | Extremum Reasoning | For item Llama 3.1 70B, what is the maximum value across all metrics in this row? Exclude N/A and —. Output only the number as a string. | "-1.82" |
| 7 | HB_real_002231 | Consistency & Missingness | Cell Missingness | Is the cell (row GPT-4o, column Safety Reasoning) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000432 | Consistency & Missingness | Column Missingness | By reading the table in the image, list all missing entries in column de→en (block Translation-Hard) (N/A or —). Output only a JSON array of strings. | [] |
| 2 | HB_real_002232 | Aggregation & Comparison | Argmax Reasoning | In column Avg, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4o", "value": "82.5"} |
| 3 | HB_real_002233 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column zh→en (block Translation-Hard)? Exclude N/A and —. Output only the number as a string. | "52.0" |
| 4 | HB_real_002234 | Constrained Retrieval | Value Filtering | In column Avg, list all row labels in the primary row-label column with value >= 80. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. | [{"item": "GPT-4o", "value": "82.5"}, {"item": "GPT-4 Turbo", "value": "82.2"}, {"item": "Aya Expanse 32B", "value": "81.6"}, {"item": "Eurus RM 7B", "value": "80.0"}] |
| 5 | HB_real_002235 | Aggregation & Comparison | Pairwise Comparison | In column en→zh (block Translation-Hard), compare GPT-4 Turbo and GPT-4o. Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "GPT-4 Turbo", "value": "84.0"} |
| 6 | HB_real_002236 | Aggregation & Comparison | Extremum Reasoning | For row GPT-4o, what is the minimum value across all translation columns? Output only the number as a string. | "61.0" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000244 | Local Relations | Marker-Anchor Neighbor | Based on the table, find the right neighbor of the cell with value 0.8537. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "0.1551" |
| 2 | HB_real_000788 | Aggregation & Comparison | Extremum Reasoning | In row Ours-Large within block S22 - 4K, what is the minimum value across PSNR, SSIM, and LPIPS? Output only the number as a string. | "0.1290" |
| 3 | HB_real_001079 | Markup Grounding | Highlight Presence | Using the table only, is the S22 - 4K LPIPS↓ cell for row Ours-Large highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "red"} |
| 4 | HB_real_001080 | Local Relations | Neighbor Value Tracing | For the cell with value 0.8537, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.8537", "up": "0.8620", "down": "0.8662", "left": "29.2304", "right": "0.1551"} |
| 5 | HB_real_001081 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.8537, return its up and down neighbors as JSON {"up":"...","down":"..."}. | {"up": "0.8620", "down": "0.8662"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000264 | Markup Grounding | Marker Counting | How many highlighted numeric data cells are in this table (light pink, light yellow, or light orange)? Exclude headers and left identifier columns. Output only the count as a string. | "9" |
| 2 | HB_real_000555 | Markup Grounding | Marked Value Listing | List all light-pink highlighted numeric values in the table. Output only a JSON array of strings. | ["0.236", "0.768", "1.070"] |
| 3 | HB_real_000623 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.356, list all numbers highlighted with the same light-orange color as the anchor. Output only a JSON array of strings. | ["1.096", "0.356", "0.810"] |
| 4 | HB_real_000900 | Markup Grounding | Highlight Presence | Is the value at position (r=7,c=1) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false}. | {"highlighted": true} |
| 5 | HB_real_000903 | Constrained Retrieval | Value Filtering | Among highlighted cells of any color, list the values that are less than 0.4. Output JSON array of strings. | ["0.236", "0.360", "0.356"] |
| 6 | HB_real_001005 | Aggregation & Comparison | Grouped Aggregation | Among the light-orange highlighted cells, what is the minimum value? Output only the number as a string. | "0.356" |
| 7 | HB_real_001091 | Constrained Retrieval | Cell Retrieval | What is the value at column SI↓ for row Ours? Output only the number as a string. Note: SI↓ means lower is better. | "0.810" |
| 8 | HB_real_001092 | Local Relations | Neighbor Value Tracing | For the cell with value 0.356, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.356", "up": "0.387", "down": "0.365", "left": "1.109", "right": "0.822"} |
| 9 | HB_real_001093 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=5,c=2) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_orange"} |
| 10 | HB_real_001094 | Consistency & Missingness | Cell Missingness | Is the cell at row DDC, column Tex Res missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 11 | HB_real_001095 | Markup Grounding | Marker Counting | How many light-yellow highlighted numeric data cells are in the table? Output only the count as a string. | "3" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001096 | Constrained Retrieval | Cell Retrieval | What is the value at column Average rank for row Ours? Output only the number as a string. | "1.55" |
| 2 | HB_real_001097 | Constrained Retrieval | Cell Retrieval | What is the value at column 3DMatch (block Indoor) for row Ours? Output only the number as a string. | "95.58" |
| 3 | HB_real_001098 | Local Relations | Neighbor Value Tracing | For the cell with value 1.55, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "1.55", "up": "3.00", "down": null, "left": "99.67", "right": null} |
| 4 | HB_real_001099 | Aggregation & Comparison | Argmax Reasoning | In column Average rank, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 7, "col": 12, "value": "15.73"} |
| 5 | HB_real_001100 | Consistency & Missingness | Cell Missingness | Is the cell (row Predator [31], column ScanNet++F (block Indoor)) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_real_001101 | Local Relations | Neighbor Value Tracing | For the cell with value 95.58, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "95.58", "up": "93.38", "down": null, "left": null, "right": "74.18"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001102 | Constrained Retrieval | Cell Retrieval | What is the value at column RTE [cm]↓ for row Ours? Output only the number as a string. Note: RTE [cm]↓ means lower is better. | "7.74" |
| 2 | HB_real_001103 | Local Relations | Neighbor Value Tracing | For the cell with value 4.90, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "4.90", "up": "7.46", "down": "7.74", "left": null, "right": "0.23"} |
| 3 | HB_real_001104 | Aggregation & Comparison | Best Selection | In metric RRE [°]↓, which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "G-ICP [65]", "value": "0.22"} |
| 4 | HB_real_001105 | Local Relations | Neighbor Value Tracing | For the cell with value 0.22, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "0.22", "up": null, "down": "0.38", "left": "8.56", "right": "37.95"} |
| 5 | HB_real_001106 | Consistency & Missingness | Cell Missingness | Is the cell (row Ours, column Succ. rate [%]↑) missing (N/A or —)? Answer JSON {"missing":true/false}. Note: Succ. rate [%]↑ means larger is better. | {"missing": false} |
| 6 | HB_real_001107 | Markup Grounding | Highlight Presence | Is the number with value 0.22 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001108 | Constrained Retrieval | Cell Retrieval | What is the value at column RTE [cm]↓ for the row with Local=✓, Middle=✓, and an empty Global cell? Output only the number as a string. Note: ↓ means lower is better. | "5.73" |
| 2 | HB_real_001109 | Local Relations | Neighbor Value Tracing | For the cell with value 94.62, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "94.62", "up": "94.02", "down": "95.58", "left": "1.81", "right": "2.33"} |
| 3 | HB_real_001110 | Markup Grounding | Highlight Presence | Is the number with value 5.73 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001111 | Local Relations | Neighbor Value Tracing | For the cell with value 5.73, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "5.73", "up": "6.06", "down": "5.77", "left": null, "right": "1.81"} |
| 5 | HB_real_001112 | Consistency & Missingness | Cell Missingness | Is the cell (row Local=✗, Middle=✓, Global=✗, column RRE [°]↓) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_real_rebal_000085 | Markup Grounding | Underline Presence | The value 31.9 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000520 | Markup Grounding | Underline Presence | By reading the table in the image, is the number with value 15.28 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 2 | HB_real_000601 | Aggregation & Comparison | Rank Selection | List the top 2 methods by ΔP_task↑, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Rep-MTL (EW)", "value": "+1.70±0.29"}, {"item": "DB-MTL [30]", "value": "+1.15±0.16"}] |
| 3 | HB_real_000924 | Markup Grounding | Marked Value Listing | List the red positive ΔP_task↑ values in the table. Output only a JSON array of strings. | ["+0.30±0.30", "+0.05±0.29", "+0.06±0.11", "+1.15±0.16", "+1.70±0.29"] |
| 4 | HB_real_001113 | Constrained Retrieval | Cell Retrieval | What is the value at column ΔP_metric↑ (block Table-wide summary) for row Rep-MTL (EW)? Output only the number as a string. Note: ↑ means larger is better. | "+0.95" |
| 5 | HB_real_001114 | Local Relations | Neighbor Value Tracing | For the cell with value 0.1542, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.1542", "up": "0.1557", "down": null, "left": "0.3750", "right": "21.91"} |
| 6 | HB_real_001115 | Aggregation & Comparison | Extremum Reasoning | In row Rep-MTL (EW), what is the maximum numeric value among the performance columns before ΔP_task↑? Output only the number as a string. | "76.04" |
| 7 | HB_real_001116 | Consistency & Missingness | Cell Missingness | Is the cell (row Rep-MTL (EW), column Rel. Err↓ (block Depth Estimation)) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 8 | HB_real_001117 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=1,c=6) in the numeric part of the table. Output only the value as a string. | "15.28" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001119 | Local Relations | Neighbor Value Tracing | For the cell with value 0.01265, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.01265", "up": "0.01340", "down": "0.01270", "left": "91.52", "right": "45.92"} |
| 2 | HB_real_001120 | Aggregation & Comparison | Best Selection | In metric Abs. Err↓ (block Depth Estimation), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Nash-MTL [49]", "value": "0.01265"} |
| 3 | HB_real_001121 | Local Relations | Neighbor Value Tracing | For the cell with value 69.72, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "69.72", "up": "69.17", "down": null, "left": null, "right": "91.85"} |
| 4 | HB_real_001122 | Markup Grounding | Bold Presence | Is the number with value 69.72 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 5 | HB_real_001123 | Markup Grounding | Underline Presence | Is the number with value 43.42 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 6 | HB_real_001780 | Markup Grounding | Marked Column Listing | In column Abs. Err↓ under Depth Estimation, list all underlined values. Include repeated values each time they appear. Output only a JSON array of strings. | ["0.01270", "0.01270", "0.01270", "0.01270"] |
| 7 | HB_real_000881 | Markup Grounding | Underline Presence | At position (r=2,c=3) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Method column. | {"underlined": false} |
| 8 | HB_real_000583 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.62, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["69.72", "91.85", "0.01270", "43.42", "+0.62"] |
| 9 | HB_real_000611 | Markup Grounding | Marked Value Listing | List the positive red numeric values in the Δp_task↑ column. Output only a JSON array of strings. | ["+0.20", "+0.62"] |
| 10 | HB_real_001702 | Markup Grounding | Marked Value Listing | List all bold numeric values in the table. Output only a JSON array of strings. | ["0.01265", "43.17", "69.72", "91.85", "+0.62"] |
| 11 | HB_real_000945 | Markup Grounding | Marker Counting | How many numeric values are boldfaced in the table? Output only the count as a string. | "5" |
| 12 | HB_real_000629 | Constrained Retrieval | Value Filtering | Among values in the light-blue highlighted row, list the numeric values greater than 50. Output only a JSON array of strings. | ["69.72", "91.85"] |
| 13 | HB_real_001747 | Aggregation & Comparison | Grouped Aggregation | Exclude boldfaced cells. Among the remaining cells in column mIoU↑ under Segmentation, what is the maximum value? Output only the number as a string. | "69.62" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000565 | Local Relations | Marker-Anchor Neighbor | From this table image, find the right neighbor of the cell with value 79.14. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "+0.41" |
| 2 | HB_real_000790 | Local Relations | Marker-Anchor Neighbor | For the highlighted cell with value 79.14, return its up and left neighbors as JSON {"up":"...","left":"..."}. | {"up": "78.95", "left": "80.04"} |
| 3 | HB_real_001124 | Constrained Retrieval | Cell Retrieval | What is the value at column Avg.↑ for row Rep-MTL (EW)? Output only the number as a string. Note: ↑ means larger is better. | "79.14" |
| 4 | HB_real_001125 | Local Relations | Neighbor Value Tracing | For the cell with value 79.14, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "79.14", "up": "78.95", "down": null, "left": "80.04", "right": "+0.41"} |
| 5 | HB_real_001126 | Aggregation & Comparison | Extremum Reasoning | In row Rep-MTL (EW), what is the minimum numeric value among Artistic, Clipart, Product, Real, and Avg.? Output only the number as a string. | "67.40" |
| 6 | HB_real_001127 | Consistency & Missingness | Consistency Probing | List rows whose Δp_task↑ value is positive. Return JSON array of {"item":"...","value":"..."}. | [{"item": "DB-MTL [30]", "value": "+0.17±0.44"}, {"item": "Rep-MTL (EW)", "value": "+0.41±0.58"}] |
| 7 | HB_real_001128 | Aggregation & Comparison | Pairwise Comparison | For row Rep-MTL (EW), compare column Artistic and column Product. Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "Product", "value": "90.37"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000245 | Local Relations | Marker-Anchor Neighbor | Using the table only, find the right neighbor of the cell with value +1.06. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "+0.21" |
| 2 | HB_real_000785 | Constrained Retrieval | Value Filtering | List the NYUv2 Δp_task↑ values from top to bottom. Output only a JSON array of strings. | ["−1.78±0.45", "+1.06±0.27", "+0.23±0.29", "+1.70±0.29"] |
| 3 | HB_real_001130 | Local Relations | Neighbor Value Tracing | For the cell with value +1.06, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "+1.06", "up": "-1.78", "down": "+0.23", "left": null, "right": "+0.21"} |
| 4 | HB_real_001131 | Local Relations | Marker-Anchor Neighbor | For the cell with value +1.06, return its up and down neighbors as JSON {"up":"...","down":"..."}. | {"up": "−1.78", "down": "+0.23"} |
| 5 | HB_real_000767 | Aggregation & Comparison | Grouped Aggregation | In the NYUv2 Δp_task↑ column, which row has the largest delta value? Return the row index as a string, counted from top to bottom. | "4" |
| 6 | HB_real_001957 | Consistency & Missingness | Consistency Probing | In the Cityscapes Δp_task↑ column, list the row settings with positive delta values. Output only a JSON array of strings. | ["CSA=✓, TSR=×", "CSA=✓, TSR=✓"] |
| 7 | HB_real_001951 | Constrained Retrieval | Value Filtering | List the Cityscapes Δp_task↑ values from top to bottom. Output only a JSON array of strings. | ["−2.05±0.56", "+0.21±0.68", "−0.34±0.24", "+0.62±0.53"] |
| 8 | HB_real_001033 | Markup Grounding | Marked Value Listing | List the delta values in the gray-highlighted rows, excluding uncertainty terms. Output only a JSON array of strings. | ["+1.06", "+0.21", "+1.70", "+0.62"] |
| 9 | HB_real_000768 | Constrained Retrieval | Value Filtering | Among gray-highlighted delta values, list the values greater than 1.0, excluding uncertainty terms. Output only a JSON array of strings. | ["+1.06", "+1.70"] |
| 10 | HB_real_001755 | Markup Grounding | Marked Value Listing | List all bold delta values, excluding uncertainty terms. Output only a JSON array of strings. | ["+1.70", "+0.62"] |
| 11 | HB_real_001982 | Markup Grounding | Bold Presence | At position (r=2,c=2) in the numeric part of the table, is the delta value boldfaced? Answer JSON {"bold":true/false}. The numeric part contains the two Δp_task columns only. | {"bold": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002237 | Aggregation & Comparison | Extremum Reasoning | In column Avg., what is the minimum value? Treat γ/ρ as row-header tags (not numeric data cells). Output only the number as a string. | "0.557" |
| 2 | HB_real_002238 | Aggregation & Comparison | Pairwise Comparison | For SentPT-M² (γ row), compare CoNLL-2014 EW and CoNLL-2014 TS. Which value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "CoNLL-2014 EW", "value": "0.949"} |
| 3 | HB_real_002239 | Aggregation & Comparison | Pairwise Comparison | In column Avg., compare SentCLEME2.0-sim-dep (Ours, ρ row) and SentCLEME2.0-sim-ind (Ours, ρ row). Which value is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "SentCLEME2.0-sim-dep (Ours, ρ)", "value": "0.915"} |
| 4 | HB_real_002240 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_002241 | Aggregation & Comparison | Argmax Reasoning | In column Avg., which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CLEME2.0-sim-ind (Ours)", "value": "0.941"} |
| 2 | HB_real_002242 | Aggregation & Comparison | Best Selection | Within group "CLEME family (non-Sent block)", which row label in the leftmost row-label column is best on SEEDA-E ρ? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "CLEME2.0-sim-ind (Ours)", "value": "0.981"} |
| 3 | HB_real_002243 | Consistency & Missingness | Cell Missingness | Is the cell (row GLEU, column SEEDA-E γ) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000257 | Markup Grounding | Marked Value Listing | Please answer this question: list text spans whose text is colored in blue in this figure. Output only a JSON array of strings. | ["suffer", "to inform", "diagnosed out"] |
| 2 | HB_real_000921 | Markup Grounding | Marked Value Listing | From this table, list text spans whose text is colored in orange in this figure. Output only a JSON array of strings. | ["the results", "their"] |
| 3 | HB_real_001075 | Constrained Retrieval | Cell Retrieval | In the second example block, what is the value of Wrong in the summary line "Hit = ..., Wrong = ..., Under = ..., Over = ..."? Output only the number as a string. | "0.90" |
| 4 | HB_real_002244 | Markup Grounding | Marked Value Listing | In the first example block, list text spans whose text is colored in blue. Output only a JSON array of strings. | ["suffer", "to inform"] |
| 5 | HB_real_002245 | Markup Grounding | Marked Value Listing | In the second example block, list text spans whose text is colored in blue. Output only a JSON array of strings. | ["diagnosed out"] |
| 6 | HB_real_002246 | Constrained Retrieval | Cell Retrieval | In the first example block, what is the value of Wrong in the summary line "Hit = ..., Wrong = ..., Under = ..., Over = ..."? Output only the number as a string. | "0.79" |
| 7 | HB_real_002247 | Aggregation & Comparison | Pairwise Comparison | In the first example block summary line, compare Wrong and Under. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. | {"item": "Wrong", "value": "0.79"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000249 | Local Relations | Marker-Anchor Neighbor | According to the table, find the right neighbor of the cell with value 1.36. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "69.05" |
| 2 | HB_real_000639 | Aggregation & Comparison | Rank Selection | Within CIFAR-10 / Evasion, list the top 2 Poisoning DOV rows by ACC(↑), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Narcissus", "value": "94.37 ± 0.18"}, {"item": "Badnets", "value": "93.46 ± 0.20"}] |
| 3 | HB_real_000783 | Aggregation & Comparison | Rank Selection | Within Tiny-Imagenet / Evasion, list the top 2 Poisoning DOV rows by ACC(↑), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Badnets", "value": "66.48 ± 0.14"}, {"item": "Label-Consistent", "value": "66.46 ± 0.10"}] |
| 4 | HB_real_001133 | Constrained Retrieval | Cell Retrieval | What is the value at column VSR(↓) (block CIFAR-10 / Evasion, section Poisoning DOV) for row Badnets? Output only the number as a string. Note: ↓ means smaller is better. | "1.36" |
| 5 | HB_real_001134 | Local Relations | Neighbor Value Tracing | For the cell with value 1.36, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "1.36", "up": null, "down": "1.74", "left": "93.46", "right": "69.05"} |
| 6 | HB_real_001135 | Aggregation & Comparison | Extremum Reasoning | In row Badnets, what is the minimum VSR value across the two VSR(↓) columns? Output only the number as a string. | "1.36 ± 0.07" |
| 7 | HB_real_001136 | Aggregation & Comparison | Argmax Reasoning | Within CIFAR-10 / Evasion, in column ACC(↑), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows are counted within the Poisoning DOV section and columns within that block. | {"row": 4, "col": 3, "value": "94.37 ± 0.18"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000247 | Local Relations | Marker-Anchor Neighbor | Read the table carefully and find the right neighbor of the cell with value 0.0532. The requested neighbor is guaranteed to be numeric. Output only the number/string as a single cell value. | "86.88" |
| 2 | HB_real_000564 | Aggregation & Comparison | Rank Selection | In block Dataset Inference, list the top 2 methods by ACC(↑), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Escaping DOV (Ours)", "value": "93.97 ± 0.49"}, {"item": "BCU", "value": "93.09 ± 0.18"}] |
| 3 | HB_real_000782 | Aggregation & Comparison | Rank Selection | In block Isotope, list the top 2 methods by ACC(↑), sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Escaping DOV (Ours)", "value": "93.99 ± 0.49"}, {"item": "BCU", "value": "92.77 ± 0.36"}] |
| 4 | HB_real_001137 | Constrained Retrieval | Cell Retrieval | What is the value at column VSR(↓) (block Narcissus) for row Escaping DOV (Ours)? Output only the number as a string. Note: ↓ means smaller is better. | "4.59" |
| 5 | HB_real_001138 | Local Relations | Neighbor Value Tracing | For the cell with value 0.0532, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.0532", "up": null, "down": "0.0264", "left": "86.78", "right": "86.88"} |
| 6 | HB_real_001139 | Aggregation & Comparison | Extremum Reasoning | In row Escaping DOV (Ours), what is the maximum ACC value across the four ACC(↑) columns? Output only the number as a string. | "94.37" |
| 7 | HB_real_001140 | Aggregation & Comparison | Argmax Reasoning | Within block Dataset Inference, in column ACC(↑), find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows are counted in the numeric part of the table. | {"row": 10, "col": 7, "value": "93.97 ± 0.49"} |
| 8 | HB_real_001141 | Consistency & Missingness | Cell Missingness | Is the cell (row Escaping DOV (Ours), column p-value(↑) (block Dataset Inference)) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001142 | Constrained Retrieval | Cell Retrieval | What is the value at column p-value (block Vanilla, section Non-poisoning DOV) for row Dataset Inference? Output only the number as a string. | "0.0937" |
| 2 | HB_real_001143 | Aggregation & Comparison | Best Selection | In metric p-value(↑) (block Evasion, section Non-poisoning DOV), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Isotope", "value": "0.7013"} |
| 3 | HB_real_001144 | Aggregation & Comparison | Best Selection | In metric ACC(↑) (block Evasion, section Non-poisoning DOV), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Dataset Inference", "value": "92.78"} |
| 4 | HB_real_001145 | Aggregation & Comparison | Pairwise Comparison | For row Narcissus, compare column ACC (block Vanilla, section Poisoning DOV) and column ACC(↑) (block Evasion, section Poisoning DOV). Which numeric value is larger? Return JSON {"item":"...","value":"..."}. | {"item": "ACC (Vanilla)", "value": "94.04"} |
| 5 | HB_real_001146 | Consistency & Missingness | Cell Missingness | Is the cell (row Isotope, column p-value(↑) (block Evasion, section Non-poisoning DOV)) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000077 | Aggregation & Comparison | Rank Selection | Within block DA methods (with unlabeled target data), who ranks 2nd in column mAP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours (Diff. Guided, R101)", "value": "51.3"} |
| 2 | HB_real_000080 | Aggregation & Comparison | Rank Selection | Within block DA methods (with unlabeled target data), who ranks 1st in column mAP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "CAT [36] (CVPR’24)", "value": "52.5"} |
| 3 | HB_real_000413 | Aggregation & Comparison | Rank Selection | Within the upper methods block (without target data), who ranks 1st in column mAP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours (Diff. Detector, SD-1.5)", "value": "49.3"} |
| 4 | HB_real_000421 | Markup Grounding | Marked Value Listing | List all bold numbers in row Ours (Diff. Detector, SD-1.5). Output only a JSON array of strings. | ["41.2", "41.7", "72.7", "62.8", "48.7", "40.5", "49.3"] |
| 5 | HB_real_001147 | Constrained Retrieval | Cell Retrieval | What is the value at column mAP within block DA methods (with unlabeled target data) for row Ours (Diff. Guided, R101)? Output only the number as a string. | "51.3" |
| 6 | HB_real_001148 | Markup Grounding | Marked Value Listing | List all light-yellow highlighted numbers in row Ours (Diff. Detector, SD-1.5). Output only a JSON array of strings. | ["41.2", "41.7", "72.7", "37.2", "62.8", "48.7", "40.5", "49.3"] |
| 7 | HB_real_001149 | Markup Grounding | Bold Presence | Is the number with value 75.2 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000072 | Aggregation & Comparison | Rank Selection | Within the upper methods block, who ranks 2nd in column mAP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "GDD (R101) [24] (CVPR’25)", "value": "52.5"} |
| 2 | HB_real_000082 | Aggregation & Comparison | Rank Selection | Within the upper methods block, who ranks 1st in column mAP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours (Diff. Guided, R101)", "value": "54.1"} |
| 3 | HB_real_000412 | Aggregation & Comparison | Rank Selection | Within the DA methods block, who ranks 1st in column mAP? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours (Diff. Guided, R101)", "value": "56.6"} |
| 4 | HB_real_000419 | Markup Grounding | Marked Value Listing | List all bold numbers in row Ours (Diff. Guided, R101) within the upper methods block. Output only a JSON array of strings. | ["62.7", "68.2", "52.9", "62.0", "54.1"] |
| 5 | HB_real_001150 | Constrained Retrieval | Cell Retrieval | What is the value at column mAP within block DA methods for row Ours (Diff. Guided, R101)? Output only the number as a string. | "56.6" |
| 6 | HB_real_001151 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column mAP within the upper methods block? Output only the number as a string. | "54.1" |
| 7 | HB_real_001152 | Markup Grounding | Marked Value Listing | List all light-yellow highlighted numbers in row Ours (Diff. Guided, R101) within the upper methods block. Output only a JSON array of strings. | ["55.3", "62.7", "68.2", "45.5", "52.9", "62.0", "48.4", "37.4", "54.1"] |
| 8 | HB_real_001153 | Markup Grounding | Highlight Presence | Is the number with value 54.1 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 9 | HB_real_001154 | Markup Grounding | Bold Presence | Is the number with value 62.7 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001156 | Local Relations | Neighbor Value Tracing | For the cell with value 55.2, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "55.2", "up": "51.9", "down": "30.0", "left": "64.1", "right": "69.7"} |
| 2 | HB_real_001157 | Markup Grounding | Bold Presence | Is the number with value 69.7 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 3 | HB_real_001158 | Local Relations | Neighbor Value Tracing | For the cell with value 69.7, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "69.7", "up": "68.4", "down": "56.6", "left": "55.2", "right": null} |
| 4 | HB_real_rebal_000030 | Markup Grounding | Marked Column Listing | In column Com., list all bold numeric values. Output only a JSON array of strings. | ["55.2"] |
| 5 | HB_real_rebal_000031 | Markup Grounding | Marked Value Listing | In the yellow highlighted Ours (Diff. Detector, SD-1.5) row, list all bold numeric values. Output only a JSON array of strings. | ["64.1", "55.2", "69.7"] |
| 6 | HB_real_rebal_000032 | Markup Grounding | Marker Counting | How many bold numeric values are in the yellow highlighted row? Output only the count as a string. | "3" |
| 7 | HB_real_rebal_000034 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 8 | HB_real_rebal_000035 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Methods column. | {"underlined": false} |
| 9 | HB_real_rebal_000037 | Constrained Retrieval | Color-Anchor Retrieval | In the yellow highlighted row, list the Cli. and Com. values highlighted with the same yellow color. Output only a JSON array of strings. | ["64.1", "55.2"] |
| 10 | HB_real_rebal_000038 | Aggregation & Comparison | Grouped Aggregation | Take the yellow highlighted numeric cells. What is their minimum value? Output only the number as a string. | "55.2" |
| 11 | HB_real_rebal_000039 | Local Relations | Neighbor Value Tracing | The value 64.1 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": null, "right": "55.2"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000084 | Aggregation & Comparison | Extremum Reasoning | For item Ours (Diff. Guided, R101), what is the maximum value across all metrics? Output only the number as a string. | "68.0" |
| 2 | HB_real_000093 | Aggregation & Comparison | Extremum Reasoning | For item Ours (Diff. Guided, R101), what is the minimum value across all metrics? Output only the number as a string. | "50.5" |
| 3 | HB_real_000471 | Markup Grounding | Highlight Presence | Using the table only, is the value at position (r=15,c=3) in the numeric part of the table highlighted? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. | {"highlighted": true, "color": "light_yellow"} |
| 4 | HB_real_000477 | Markup Grounding | Marked Value Listing | List all light-yellow highlighted numbers in row Ours (Diff. Guided, R101). Output only a JSON array of strings. | ["58.2", "50.5", "68.0"] |
| 5 | HB_real_000827 | Markup Grounding | Marker Counting | How many bold numeric cells are in the table? Output only the count as a string. | "3" |
| 6 | HB_real_000832 | Consistency & Missingness | Column Missingness | In column Com., list the row labels whose entry is missing. Output only a JSON array of strings. | ["UMT", "SADA", "AT", "TIA", "LODS", "CIGAR", "CMT", "UMGA", "CAT"] |
| 7 | HB_real_001159 | Constrained Retrieval | Cell Retrieval | Using the table only, find the value at position (r=15,c=1) in the numeric part of the table. Output only the value as a string. | "58.2" |
| 8 | HB_real_001160 | Aggregation & Comparison | Argmax Reasoning | Find the single largest numeric cell in the numeric part of the table. Return JSON {"row":<int>,"col":<int>,"value":<string>}. | {"row": 15, "col": 3, "value": "68.0"} |
| 9 | HB_real_001161 | Markup Grounding | Highlight Presence | Is the number with value 58.2 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 10 | HB_real_001162 | Markup Grounding | Bold Presence | Is the number with value 68.0 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000209 | Aggregation & Comparison | Extremum Reasoning | Using the table only, for item Ours (Diff. Detector, SD-1.5), what is the minimum value across all metrics? Output only the number as a string. | "31.3" |
| 2 | HB_real_000484 | Aggregation & Comparison | Rank Selection | In column Average, list the top 2 methods by value, sorted from highest to lowest. Return a JSON array of {"item":"<row label>","value":"<cell value>"}. | [{"item": "Ours (Diff. Detector, SD-1.5)", "value": "45.0"}, {"item": "GDD (SD-1.5)", "value": "40.2"}] |
| 3 | HB_real_000837 | Aggregation & Comparison | Extremum Reasoning | In row Ours (Diff. Detector, SD-1.5), what is the maximum numeric value across all metrics? Output only the number as a string. | "51.6" |
| 4 | HB_real_001163 | Constrained Retrieval | Cell Retrieval | What is the value at column Average for row Ours (Diff. Detector, SD-1.5)? Output only the number as a string. | "45.0" |
| 5 | HB_real_001164 | Markup Grounding | Highlight Presence | Is the number with value 45.0 highlighted (background-colored)? Answer JSON {"highlighted":true/false,"color":"<color name or null>"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "light_yellow"} |
| 6 | HB_real_001165 | Local Relations | Marker-Anchor Neighbor | For the highlighted cell with value 51.6, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "31.3", "right": "45.0"} |
| 7 | HB_real_001166 | Markup Grounding | Highlight Presence | Is the number with value 51.6 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 8 | HB_real_001167 | Markup Grounding | Bold Presence | Is the number with value 48.4 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001168 | Constrained Retrieval | Cell Retrieval | What is the value at column mPC↑ for row Ours (Diff. Detector, SD-1.5)? Output only the number as a string. Note: ↑ means larger is better. | "27.4" |
| 2 | HB_real_001169 | Aggregation & Comparison | Best Selection | In metric mPC↑, which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Ours (Diff. Detector, SD-1.5)", "value": "27.4"} |
| 3 | HB_real_001170 | Markup Grounding | Highlight Presence | Is the number with value 27.4 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001171 | Markup Grounding | Bold Presence | Is the number with value 26.6 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 5 | HB_real_001172 | Consistency & Missingness | Cell Missingness | Is the cell (row Ours (Diff. Guided, R50), column Clean) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000086 | Aggregation & Comparison | Extremum Reasoning | For item Ours (SD-1.5) in block 100% COCO, Faster R-CNN, what is the maximum value across all metrics? Output only the number as a string. | "86.6" |
| 2 | HB_real_000091 | Aggregation & Comparison | Extremum Reasoning | For item Ours (SD-1.5) in block 100% COCO, Faster R-CNN, what is the minimum value across all metrics? Exclude missing entries. Output only the number as a string. | "21.7" |
| 3 | HB_real_000476 | Markup Grounding | Marked Value Listing | List all bold numbers in row Ours (SD-1.5) within block 100% COCO, Faster R-CNN. Output only a JSON array of strings. | ["67.0", "86.6", "54.1", "51.2", "45.8", "64.6", "51.9", "66.6", "41.7", "39.2", "21.7", "40.7", "51.3"] |
| 4 | HB_real_000479 | Markup Grounding | Marked Value Listing | List all light-yellow highlighted numbers in row Ours (SD-1.5) within block 100% COCO, Faster R-CNN. Output only a JSON array of strings. | ["67.0", "86.6", "54.1", "51.2", "45.8", "64.6", "51.9", "66.6", "41.7", "39.2", "21.7", "40.7", "51.3", "164"] |
| 5 | HB_real_000824 | Markup Grounding | Marker Counting | How many bold numeric cells are in row Ours (SD-1.5) within block 100% COCO, Faster R-CNN? Output only the count as a string. | "13" |
| 6 | HB_real_001173 | Constrained Retrieval | Cell Retrieval | What is the value at column Avg. on 11 in block 100% COCO, Faster R-CNN for row Ours (SD-1.5)? Output only the number as a string. | "51.3" |
| 7 | HB_real_001174 | Aggregation & Comparison | Argmax Reasoning | In column VOC within block 100% COCO, Faster R-CNN, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows are counted within that block and columns within the numeric part of the block. | {"row": 6, "col": 2, "value": "86.6"} |
| 8 | HB_real_001175 | Aggregation & Comparison | Argmax Reasoning | In column Avg. on 11 within block 100% COCO, Faster R-CNN, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Ours (SD-1.5)", "value": "51.3"} |
| 9 | HB_real_001176 | Markup Grounding | Bold Presence | Is the number with value 86.6 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 10 | HB_real_001177 | Consistency & Missingness | Cell Missingness | Is the cell at row ResNet-R50 [25], column Inf. time (ms), block 1% COCO, Faster R-CNN missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000550 | Markup Grounding | Bold Presence | Based on the table image, is the number with value 49.3+4.6 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 2 | HB_real_000998 | Aggregation & Comparison | Extremum Reasoning | For the setting row where all five switches are ✓, what is the minimum numeric value across City., VOC, Foggy., BDD., and Cli.? Output only the number as a string. | "61.1" |
| 3 | HB_real_001178 | Constrained Retrieval | Cell Retrieval | What is the value at column Cli. (block Cross-Domain) for the setting row with Domain Aug.=✓, Feature Coll.=✓, Feature Fusion.=✓, Aux. Branch=✓, Consist. Loss=✓? Output only the number as a string. | "64.1+4.3" |
| 4 | HB_real_001179 | Markup Grounding | Highlight Presence | Is the number with value 62.4 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 5 | HB_real_001180 | Local Relations | Marker-Anchor Neighbor | For the highlighted cell with value 62.4, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": null, "right": "87.9"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000089 | Aggregation & Comparison | Extremum Reasoning | For item Corvid-o1-8B, what is the maximum value across all metrics? Exclude missing entries. Output only the number as a string. | "93.8" |
| 2 | HB_real_000092 | Markup Grounding | Marked Value Listing | List all bold numbers in row Corvid-o1-8B. Output only a JSON array of strings. | ["65.2", "59.7", "59.8", "30.1", "33.2"] |
| 3 | HB_real_000475 | Markup Grounding | Marked Value Listing | List all underlined numbers in row Corvid-o1-8B. Output only a JSON array of strings. | ["72.0", "40.1"] |
| 4 | HB_real_000478 | Aggregation & Comparison | Rank Selection | In column MMStar, which MLLM ranks 1st? Return JSON {"item":"<row label>","value":"<cell value>"}. Exclude missing entries. | {"item": "Corvid-o1-8B", "value": "65.2"} |
| 5 | HB_real_000830 | Aggregation & Comparison | Extremum Reasoning | For item Corvid-o1-8B, what is the minimum value across all metrics? Exclude missing entries. Output only the number as a string. | "30.1" |
| 6 | HB_real_001181 | Constrained Retrieval | Cell Retrieval | What is the value at column DynaMath for row Corvid-o1-8B? Output only the number as a string. | "33.2" |
| 7 | HB_real_001182 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column MMStar? Exclude missing entries. Output only the number as a string. | "65.2" |
| 8 | HB_real_001183 | Aggregation & Comparison | Rank Selection | In column WeMath, list the top-2 MLLMs sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. Exclude missing entries. | [{"item": "Corvid-o1-8B", "value": "59.8"}, {"item": "Corvid-base-8B", "value": "54.0"}] |
| 9 | HB_real_001184 | Markup Grounding | Highlight Presence | Is the number with value 65.2 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 10 | HB_real_001185 | Markup Grounding | Bold Presence | Is the number with value 59.8 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| 11 | HB_real_001186 | Consistency & Missingness | Cell Missingness | Is the cell at row Molmo-7B-D [25], column WeMath missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000237 | Aggregation & Comparison | Rank Selection | Within block base LLM: Llama3-8B-Instruct, list the top-2 items by column BLINK, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Corvid-o1-8B", "value": "57.9"}, {"item": "Corvid-base-8B", "value": "55.7"}] |
| 2 | HB_real_000497 | Aggregation & Comparison | Extremum Reasoning | For row Corvid-o1-8B within block base LLM: Llama3-8B-Instruct, what is the maximum value across all metrics? Output only the number as a string. | "77.0" |
| 3 | HB_real_000501 | Markup Grounding | Marked Value Listing | List all bold numbers in the table. Output only a JSON array of strings. | ["77.2", "66.6", "77.0", "57.9"] |
| 4 | HB_real_000862 | Markup Grounding | Marked Value Listing | List all underlined numbers in row Corvid-o1-8B within block base LLM: Llama3-8B-Instruct. Output only a JSON array of strings. | ["77.0", "65.9", "76.4"] |
| 5 | HB_real_000870 | Aggregation & Comparison | Rank Selection | Within block base LLM: Llama3-8B-Instruct, who ranks 2nd in column RWQA? Return JSON {"item":"<row label>","value":"<cell value>"}. | {"item": "Corvid-base-8B", "value": "77.0"} |
| 6 | HB_real_001187 | Constrained Retrieval | Cell Retrieval | What is the value at column RWQA for row Corvid-o1-8B within block base LLM: Llama3-8B-Instruct? Output only the number as a string. | "76.4" |
| 7 | HB_real_001188 | Aggregation & Comparison | Rank Selection | Within block base LLM: Llama3-8B-Instruct, list the top-2 items by column SEED-IMG, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Corvid-o1-8B", "value": "77.0"}, {"item": "Corvid-base-8B", "value": "76.4"}] |
| 8 | HB_real_001189 | Markup Grounding | Highlight Presence | Is the number with value 77.0 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric part of the table. | {"highlighted": true} |
| 9 | HB_real_001190 | Markup Grounding | Bold Presence | Is the number with value 66.6 boldfaced? Answer JSON {"bold":true/false}. | {"bold": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001191 | Constrained Retrieval | Cell Retrieval | What is the value at column Average for row GateMixer? Output only the number as a string. | "55.6" |
| 2 | HB_real_001192 | Local Relations | Neighbor Value Tracing | For the cell with value 93.2, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "93.2", "up": "92.2", "down": "91.7", "left": "57.4", "right": "82.8"} |
| 3 | HB_real_001193 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column WeMath? Output only the number as a string. | "60.0" |
| 4 | HB_real_001194 | Markup Grounding | Bold Presence | Is the number with value 93.2 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 5 | HB_real_001195 | Local Relations | Neighbor Value Tracing | For the cell with value 55.6, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "55.6", "up": "53.7", "down": "55.0", "left": null, "right": "62.4"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000241 | Aggregation & Comparison | Rank Selection | Please answer this question: list the top-2 items by column Average, sorted from highest to lowest. Return a JSON array of {"item":"...","value":"..."}. | [{"item": "Corvid-1M", "value": "55.6"}, {"item": "Ovis-1M†", "value": "53.8"}] |
| 2 | HB_real_000500 | Aggregation & Comparison | Extremum Reasoning | In row Corvid-1M, what is the maximum numeric value across the table? Output only the number as a string. | "93.2" |
| 3 | HB_real_000867 | Markup Grounding | Marker Counting | How many boldfaced numbers are in the table? Output only the count as a string. | "10" |
| 4 | HB_real_001196 | Constrained Retrieval | Cell Retrieval | What is the value at column Average for row Corvid-1M? Output only the number as a string. | "55.6" |
| 5 | HB_real_001197 | Local Relations | Neighbor Value Tracing | For the cell with value 55.6, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "55.6", "up": "53.8", "down": "51.2", "left": null, "right": "62.4"} |
| 6 | HB_real_001198 | Local Relations | Marker-Anchor Neighbor | For the highlighted cell with value 55.6, return its down and right neighbors as JSON {"down":"...","right":"..."}. | {"down": "51.2", "right": "62.4"} |
| 7 | HB_real_001199 | Markup Grounding | Highlight Presence | Is the number with value 55.6 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 8 | HB_real_001200 | Markup Grounding | Bold Presence | Is the number with value 93.2 boldfaced? Answer JSON {"bold":true/false}. Example: {"bold":true}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 9 | HB_real_001201 | Aggregation & Comparison | Argmax Reasoning | In column Average, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 4, "col": 1, "value": "55.6"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001202 | Constrained Retrieval | Cell Retrieval | What is the value at column AUC of ADD-S for row HCCE f(h3)? Output only the number as a string. | "81.11" |
| 2 | HB_real_001203 | Local Relations | Neighbor Value Tracing | For the cell with value 67.00, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "67.00", "up": "63.50", "down": "65.23", "left": "61.35", "right": "76.35"} |
| 3 | HB_real_001204 | Aggregation & Comparison | Argmax Reasoning | In column AUC of ADD(-S), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "HCCE f(h3)", "value": "77.67"} |
| 4 | HB_real_001205 | Markup Grounding | Highlight Presence | Is the number with value 61.95 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 5 | HB_real_001206 | Local Relations | Neighbor Value Tracing | For the cell with value 81.11, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "81.11", "up": "78.68", "down": null, "left": "77.67", "right": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001207 | Constrained Retrieval | Cell Retrieval | For method HCCE+f(h0), what is the value of 5% Diameter? Output only the number as a string. | "42.23" |
| 2 | HB_real_001208 | Local Relations | Neighbor Value Tracing | For the cell with value 42.23, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "42.23", "up": "37.72", "down": "43.85", "left": "11.97", "right": "60.25"} |
| 3 | HB_real_001209 | Local Relations | Neighbor Value Tracing | For the cell with value 43.85, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the chart. | {"target": "43.85", "up": "42.23", "down": null, "left": "13.00", "right": "60.76"} |
| 4 | HB_real_001210 | Consistency & Missingness | Cell Missingness | Is the cell (row HBCE+f, column 2% Diameter) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001211 | Constrained Retrieval | Cell Retrieval | What is the value at column bfu for row IC-BIN [43] / obj02? Output only the number as a string. | "71.5" |
| 2 | HB_real_001212 | Local Relations | Neighbor Value Tracing | For the cell with value 57.1, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "57.1", "up": "82.1", "down": "83.9", "left": "56.6", "right": "56.0"} |
| 3 | HB_real_001213 | Local Relations | Neighbor Value Tracing | For the cell with value 71.5, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The value appears exactly once in the numeric part of the table. | {"target": "71.5", "up": "63.7", "down": null, "left": "70.7", "right": null} |
| 4 | HB_real_001469 | Constrained Retrieval | Color-Anchor Retrieval | In column f, list all numbers highlighted with the same darker blue shade as 84.9. Output only a JSON array of strings. | ["84.9", "94.8", "93.1", "78.7", "95.0", "85.4"] |
| 5 | HB_real_000776 | Markup Grounding | Marked Value Listing | List all numbers highlighted in dark pink in the bfu column. Output only a JSON array of strings. | ["85.8", "95.5", "87.7", "94.5", "82.4", "57.1", "86.1", "82.3", "95.8", "86.8", "93.3", "63.7", "71.5", "83.3"] |
| 6 | HB_real_000809 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light pink in column bfu. What is their maximum numeric value? Output only the number as a string. | "95.8" |
| 7 | HB_real_000670 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 9, "col": "bfu", "value": "95.8"} |
| 8 | HB_real_000461 | Local Relations | Neighbor Indexing | At position (r=1,c=1) in the numeric part of the table, return the target and its four neighbors as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. Use null if a neighbor does not exist. | {"target": "84.9", "up": null, "down": "94.8", "left": null, "right": "85.7"} |
| 9 | HB_real_001058 | Constrained Retrieval | Color-Anchor Retrieval | In column bfu, list all numbers highlighted in dark pink. Output only a JSON array of strings. | ["85.8", "95.5", "87.7", "94.5", "82.4", "57.1", "86.1", "82.3", "95.8", "86.8", "93.3", "63.7", "71.5", "83.3"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001215 | Aggregation & Comparison | Argmax Reasoning | In column Avg. (block ViT-S), which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. | {"item": "IN-lbGen(ours)", "value": "74.4"} |
| 2 | HB_real_001216 | Markup Grounding | Underline Presence | Is the number with value 56.8 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 3 | HB_real_001649 | Markup Grounding | Marked Value Listing | List all red numeric values in the two IN-lbGen(ours) rows. Output only a JSON array of strings. | ["62.1", "58.5", "72.8", "95.0", "86.4", "64.3", "73.2", "62.6", "58.0", "71.2", "86.2", "68.5", "66.0", "74.4"] |
| 4 | HB_real_001050 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=5,c=1) in the numeric part of the table, list all numeric values highlighted with the same light-gray row background as the anchor row. Output only a JSON array of strings. | ["46.1", "62.1", "58.5", "72.8", "95.0", "86.4", "87.2", "65.3", "64.3", "73.2"] |
| 5 | HB_real_000653 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=7) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 6 | HB_real_000895 | Markup Grounding | Marked Column Listing | List all underlined values in column Avg. Output only a JSON array of strings. | ["71.8", "72.3"] |
| 7 | HB_real_000882 | Markup Grounding | Marked Value Listing | In the ResNet50 block, list all underlined numeric values in the IN-RealFake row. Output only a JSON array of strings. | ["59.9", "54.1", "70.6", "83.7", "90.0", "71.8"] |
| 8 | HB_real_001784 | Markup Grounding | Marked Column Listing | List all underlined values in column Aircraft across both backbones. Output only a JSON array of strings. | ["59.9", "59.4"] |
| 9 | HB_real_000694 | Constrained Retrieval | Color-Anchor Retrieval | For the value 74.4, list all numeric values highlighted with the same light-gray row background as the anchor row. Output only a JSON array of strings. | ["46.3", "62.6", "58.0", "71.2", "94.1", "86.2", "88.6", "68.5", "66.0", "74.4"] |
| 10 | HB_real_001009 | Constrained Retrieval | Value Filtering | In the bottom light-gray highlighted row, list the numeric values greater than 70. Output only a JSON array of strings. | ["71.2", "94.1", "86.2", "88.6", "74.4"] |
| 11 | HB_real_000695 | Markup Grounding | Marked Value Listing | In the ViT-S block, list the red numeric values in the IN-GenRobust row. Output only a JSON array of strings. | ["94.4"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000521 | Markup Grounding | Underline Presence | Based on the table image, is the number with value 35.09 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 2 | HB_real_001218 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column 0.5× (block COCO (AP^bbox))? Output only the number as a string. | "35.24" |
| 3 | HB_real_001220 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["25.51", "35.09", "29.25", "38.05", "27.64", "32.28"] |
| 4 | HB_real_001761 | Markup Grounding | Marked Value Listing | List all red numeric values in the two IN-lbGen(ours) rows. Output only a JSON array of strings. | ["35.24", "30.68", "25.64", "38.61", "33.57", "27.82"] |
| 5 | HB_real_001052 | Constrained Retrieval | Color-Anchor Retrieval | For the value at position (r=5,c=1) in the numeric part of the COCO block, list all numeric values highlighted with the same light-gray row background as the anchor row. Output only a JSON array of strings. | ["39.26", "35.24", "30.68", "25.64"] |
| 6 | HB_real_000872 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 7 | HB_real_000896 | Markup Grounding | Marked Column Listing | List all underlined values in column 0.2× of the COCO block. Output only a JSON array of strings. | ["29.25", "30.68"] |
| 8 | HB_real_001636 | Markup Grounding | Marked Column Listing | In column 0.1× across both COCO and ADE20K blocks, list all underlined values. Output only a JSON array of strings. | ["25.51", "27.64"] |
| 9 | HB_real_001699 | Markup Grounding | Marked Value Listing | In the COCO block, list all red numeric values. Output only a JSON array of strings. | ["39.32", "35.24", "30.68", "25.64"] |
| 10 | HB_real_001744 | Constrained Retrieval | Color-Anchor Retrieval | For the value 41.50, list all values highlighted with the same light-gray row background within the ADE20K block. Output only a JSON array of strings. | ["41.50", "38.61", "33.57", "27.82"] |
| 11 | HB_real_001831 | Markup Grounding | Marked Value Listing | In the bottom light-gray highlighted row within ADE20K, list the 0.2× and 0.1× values. Output only a JSON array of strings. | ["33.57", "27.82"] |
| 12 | HB_real_001779 | Aggregation & Comparison | Grouped Aggregation | Exclude red text values. In the COCO 1.0× column, what is the maximum remaining value? Output only the number as a string. | "39.26" |
| 13 | HB_real_001793 | Aggregation & Comparison | Grouped Aggregation | In the bottom IN-lbGen(ours) row of the ADE20K block, what is the mean of the red values? Output the value rounded to two decimal places as a string. | "33.33" |
| 14 | HB_real_001965 | Consistency & Missingness | Consistency Probing | Which leftmost row labels have entries marked as missing using "-" or "N/A"? If none, output an empty JSON array []. | [] |
| 15 | HB_real_002003 | Local Relations | Neighbor Value Tracing | The value 38.61 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "41.50", "right": "33.57"} |
| 16 | HB_real_002011 | Aggregation & Comparison | Argmax Reasoning | Within the ADE20K block, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and columns left-to-right within that block. | {"row": 1, "col": 1, "value": "42.44"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000525 | Markup Grounding | Underline Presence | By reading the table in the image, is the number with value 61.8 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 2 | HB_real_000600 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["6.8", "60.1", "61.8", "6.7", "60.7", "63.8"] |
| 3 | HB_real_001222 | Aggregation & Comparison | Best Selection | In metric CB_avg.(↑) (block ResNet50), which row label in the leftmost row-label column is optimal? Higher values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "IN-lbGen(ours)", "value": "64.7"} |
| 4 | HB_real_001223 | Aggregation & Comparison | Best Selection | In metric BG_Gap(↓) (block ViT-S), which row label in the leftmost row-label column is optimal? Lower values indicate better performance. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "IN-lbGen(ours)", "value": "6.1"} |
| 5 | HB_real_001224 | Markup Grounding | Highlight Presence | Is the number with value 64.7 highlighted (emphasized in red text)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 6 | HB_real_000717 | Markup Grounding | Marked Column Listing | In column CBavg.(↑), list all underlined values. Output only a JSON array of strings. | ["60.1", "61.8"] |
| 7 | HB_real_001832 | Markup Grounding | Underline Presence | At position (r=1,c=2) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Backbone and Pre-Trained Data columns. | {"underlined": false} |
| 8 | HB_real_000718 | Markup Grounding | Marked Value Listing | List all red numeric values in the table. Output only a JSON array of strings. | ["56.1", "64.7", "6.4", "57.2", "66.0", "6.1"] |
| 9 | HB_real_001032 | Markup Grounding | Marked Value Listing | List the numeric values in the bottom light-gray highlighted row. Output only a JSON array of strings. | ["57.2", "66.0", "6.1"] |
| 10 | HB_real_000728 | Constrained Retrieval | Color-Anchor Retrieval | For the value 56.1, list all numeric values highlighted with the same light-gray row background as the anchor row. Output only a JSON array of strings. | ["56.1", "64.7", "6.4"] |
| 11 | HB_real_000732 | Aggregation & Comparison | Grouped Aggregation | Exclude underlined cells. Among the remaining values in column BG_Gap(↓), what is the minimum value? Output only the number as a string. | "6.1" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000216 | Aggregation & Comparison | Extremum Reasoning | From this table, for item IN100-lbGen (L_en=✓, L_in=✓, L_q=✓, CLIP Size=Large), what is the minimum value across all dataset metrics? Output only the number as a string. | "34.3" |
| 2 | HB_real_000841 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["92.4", "68.8", "72.1", "42.3", "32.6", "58.4", "49.5", "39.2", "56.7"] |
| 3 | HB_real_001227 | Aggregation & Comparison | Argmax Reasoning | In column Cars196, find the row and value of the maximum cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Rows and columns are counted in the numeric part of the table. | {"row": 2, "col": 3, "value": "34.3"} |
| 4 | HB_real_001228 | Aggregation & Comparison | Extremum Reasoning | For item IN100-lbGen (L_en=✓, L_in=✓, L_q=✓, CLIP Size=Base), what is the minimum value across all dataset metrics? Output only the number as a string. | "32.6" |
| 5 | HB_real_001229 | Markup Grounding | Underline Presence | Is the number with value 72.1 underlined? Answer JSON {"underlined":true/false}. Example: {"underlined":true}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 6 | HB_real_000962 | Markup Grounding | Marked Column Listing | In column Avg., list all underlined values. Output only a JSON array of strings. | ["56.7"] |
| 7 | HB_real_000963 | Markup Grounding | Underline Presence | The value 88.3 appears exactly once in the numeric part of the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 8 | HB_real_001996 | Markup Grounding | Marker Counting | How many underlined numeric values are in the table? Output only the count as a string. | "9" |
| 9 | HB_real_000964 | Markup Grounding | Marked Value Listing | List all red numeric values in the table. Output only a JSON array of strings. | ["44.8", "34.3", "59.1", "92.7", "71.6", "72.0", "50.8", "42.3", "58.5"] |
| 10 | HB_real_000970 | Aggregation & Comparison | Grouped Aggregation | Exclude red text values. Among the remaining values in column Avg., what is the maximum value? Output only the number as a string. | "56.7" |
| 11 | HB_real_rebal_000068 | Markup Grounding | Marked Column Listing | In column Aircraft, list all underlined values. Output only a JSON array of strings. | ["42.3"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001230 | Constrained Retrieval | Cell Retrieval | What is the value at column C_b (block CIFAR-10) for row BadNets [9]? Output only the number as a string. | "0.9313" |
| 2 | HB_real_001231 | Local Relations | Neighbor Value Tracing | For the cell with value 0.8144, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.8144", "up": "0.8196", "down": "0.7471", "left": "75.77", "right": "0.2022"} |
| 3 | HB_real_001232 | Markup Grounding | Highlight Presence | Is the number with value 47.69* highlighted (text-color emphasized)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001233 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.8144, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "75.77", "right": "0.2022"} |
| 5 | HB_real_001999 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 6 | HB_real_001941 | Markup Grounding | Underline Presence | The value 0.1076 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 7 | HB_real_001947 | Markup Grounding | Marker Counting | How many underlined numeric values are in the table? Output only the count as a string. | "0" |
| 8 | HB_real_001948 | Markup Grounding | Marked Value Listing | List all red numeric values in the table. Output only a JSON array of strings. | ["0.1076", "0.0361", "0.1167", "0.0872", "0.0888", "0.2022", "0.0675", "0.0065", "0.2301", "8.21", "0.1430", "47.69*"] |
| 9 | HB_real_001949 | Consistency & Missingness | Column Missingness | In the CIFAR-10 WSR (%) column, list row labels whose entry is shown as missing using "-". Output only a JSON array of strings. | ["Clean Model"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001235 | Local Relations | Neighbor Value Tracing | For the cell with value 0.8157, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.8157", "up": "0.0764", "down": "0.0039", "left": "89.03", "right": "0.9837"} |
| 2 | HB_real_001236 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.8157, return its up and down neighbors as JSON {"up":"...","down":"..."}. | {"up": "0.0764", "down": "0.0039"} |
| 3 | HB_real_001237 | Consistency & Missingness | Cell Missingness | Is the cell (row IFT=✗, GM=✗, AS=✗, column Other) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_real_002026 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 5 | HB_real_002266 | Markup Grounding | Underline Presence | The value 88.82 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 6 | HB_real_rebal_000002 | Constrained Retrieval | Color-Anchor Retrieval | In the two pink highlighted rows, list the metric values from WSR (%) through Other. Output only a JSON array of strings. | ["88.82", "88.69", "0.0764", "0.9998", "0.0001", "88.96", "89.03", "0.8157", "0.9837", "0.5794"] |
| 7 | HB_real_rebal_000003 | Constrained Retrieval | Color-Anchor Retrieval | For the value 0.0039, list metric values in the same light-green highlighted row from WSR (%) through Other. Output only a JSON array of strings. | ["86.83", "89.00", "0.0039", "0.9837", "0.8655"] |
| 8 | HB_real_rebal_000004 | Markup Grounding | Marked Value Listing | In the bottom light-green highlighted row, list the metric values from WSR (%) through Other. Output only a JSON array of strings. | ["82.47", "87.79", "0.0017", "1.0000", "1.0000"] |
| 9 | HB_real_rebal_000005 | Aggregation & Comparison | Grouped Aggregation | Take the pink highlighted rows. What is the maximum BA (%) value? Output only the number as a string. | "89.03" |
| 10 | HB_real_rebal_000006 | Aggregation & Comparison | Grouped Aggregation | Across the light-green highlighted rows, what is the mean Benign value? Output the value rounded to four decimal places as a string. | "0.9946" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001239 | Local Relations | Neighbor Value Tracing | For the cell with value 0.3266, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.3266", "up": "0.1539", "down": "0.3616", "left": "80.72", "right": "1.0000"} |
| 2 | HB_real_001240 | Markup Grounding | Highlight Presence | Is the number with value 85.06 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 3 | HB_real_001241 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.3266, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "80.72", "right": "1.0000"} |
| 4 | HB_real_rebal_000007 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 5 | HB_real_rebal_000008 | Markup Grounding | Underline Presence | The value 85.06 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 6 | HB_real_rebal_000010 | Constrained Retrieval | Color-Anchor Retrieval | In the first light-green row, list the metric values from WSR (%) through Other. Output only a JSON array of strings. | ["85.06", "88.20", "0.0071", "1.0000", "0.5772"] |
| 7 | HB_real_rebal_000011 | Constrained Retrieval | Color-Anchor Retrieval | For the value 0.2253, list metric values in the same light-green row from WSR (%) through Other. Output only a JSON array of strings. | ["88.06", "92.26", "0.2253", "0.9602", "0.8639"] |
| 8 | HB_real_rebal_000012 | Aggregation & Comparison | Grouped Aggregation | Across all rows, what is the mean BA (%) value? Output the value rounded to two decimal places as a string. | "87.99" |
| 9 | HB_real_rebal_000013 | Local Relations | Neighbor Value Tracing | The value 0.2253 appears exactly once in the table. What are its immediate left and right neighbors in the same row? Return JSON {"left":"<value or null>","right":"<value or null>"}. | {"left": "92.26", "right": "0.9602"} |
| 10 | HB_real_rebal_000014 | Aggregation & Comparison | Argmax Reasoning | Within the metric columns WSR (%) through Other, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and metric columns left-to-right. | {"row": 1, "col": 4, "value": "1.0000"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001242 | Constrained Retrieval | Cell Retrieval | What is the BA (%) value when R = 3? Output only the number as a string. Note: R is the left header column (row label), not a numeric metric column. | "88.04" |
| 2 | HB_real_001243 | Local Relations | Neighbor Value Tracing | For the cell with value 0.0079, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.0079", "up": "0.0017", "down": "0.4439", "left": "88.04", "right": "0.9870"} |
| 3 | HB_real_001244 | Markup Grounding | Highlight Presence | Is the number with value 72.47 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001245 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.0079, return its up and down neighbors as JSON {"up":"...","down":"..."}. | {"up": "0.0017", "down": "0.4439"} |
| 5 | HB_real_001246 | Consistency & Missingness | Cell Missingness | Is the cell (row R=1, column Benign) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000060 | Markup Grounding | Marked Value Listing | Based on the table, list numbers highlighted in light pink, excluding the left header column γ (γ values are row labels, not data values). Output only a JSON array of strings. | ["79.71", "87.10", "0.6583", "1.0000", "0.2552"] |
| 2 | HB_real_000067 | Markup Grounding | Marker Counting | How many underlined numeric cells are in the table? Output only the count as a string. | "0" |
| 3 | HB_real_000556 | Markup Grounding | Marker Counting | From this table image, how many numeric cells are highlighted in light pink in this table? Output only the count as a string. | "5" |
| 4 | HB_real_000637 | Consistency & Missingness | Consistency Probing | Compare row γ = 0.20 against row γ = 0.10 across WSR (%), BA (%), Watermark, Benign, and Other. List the columns where γ = 0.20 has a higher value. Output only a JSON array of column labels. | ["Watermark"] |
| 5 | HB_real_000988 | Markup Grounding | Marker Counting | From the table image, how many highlighted numeric data cells are in the table (light green or light pink)? Exclude the left header column γ (treat γ values as row labels, not data values). Output only the count as a string. Example: "7". Keep the answer strictly in the requested format. | "20" |
| 6 | HB_real_001085 | Constrained Retrieval | Cell Retrieval | What is the BA (%) value when γ = 0.05? Output only the number as a string. Example: "88.50". | "88.50" |
| 7 | HB_real_001087 | Local Relations | Neighbor Value Tracing | At the cell with value 0.6583, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes the left header column γ. Use null if a neighbor does not exist. | {"target": "0.6583", "up": "0.0017", "down": null, "left": "87.10", "right": "1.0000"} |
| 8 | HB_real_001088 | Local Relations | Neighbor Value Tracing | At the cell with value 88.50, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. The numeric part of the table excludes the left header column γ. Use null if a neighbor does not exist. | {"target": "88.50", "up": "88.86", "down": "87.79", "left": "81.11", "right": "0.0018"} |
| 9 | HB_real_002267 | Constrained Retrieval | Value Filtering | Among light-pink highlighted cells, list the values that are > 1.0. Output JSON array of strings. | ["79.71", "87.10"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001248 | Local Relations | Neighbor Value Tracing | For the cell with value 74.6, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "74.6", "up": "36.7", "down": "80.7", "left": "71.5", "right": "82.2"} |
| 2 | HB_real_001249 | Markup Grounding | Underline Presence | Is the number with value 85.1 underlined? Answer JSON {"underlined":true/false}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 3 | HB_real_001250 | Local Relations | Marker-Anchor Neighbor | For the cell with value 74.6, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "71.5", "right": "82.2"} |
| 4 | HB_real_000887 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["85.1", "89.8", "99.3", "70.6", "88.2", "71.5", "74.6", "79.7", "63.1"] |
| 5 | HB_real_000771 | Markup Grounding | Marked Column Listing | In column Average, list all underlined values. Output only a JSON array of strings. | ["79.7"] |
| 6 | HB_real_000889 | Markup Grounding | Underline Presence | The value 70.4 appears exactly once in the numeric part of the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 7 | HB_real_001987 | Markup Grounding | Marked Value Listing | List all bold numeric values in the table. Output only a JSON array of strings. | ["78.0", "63.5", "97.1", "100.0", "78.0", "74.2", "95.9", "72.3", "80.7", "87.1", "83.2"] |
| 8 | HB_real_001753 | Markup Grounding | Marker Counting | How many numeric values are boldfaced in the table? Output only the count as a string. | "11" |
| 9 | HB_real_001034 | Markup Grounding | Marked Value Listing | List all values in the light-blue highlighted CPath-CLIP row. Output only a JSON array of strings. | ["97.1", "100.0", "78.0", "74.2", "95.9", "72.3", "80.7", "87.1", "63.1", "83.2"] |
| 10 | HB_real_000890 | Constrained Retrieval | Value Filtering | In the light-blue highlighted CPath-CLIP row, list the numeric values greater than 90. Output only a JSON array of strings. | ["97.1", "100.0", "95.9"] |
| 11 | HB_real_000953 | Aggregation & Comparison | Grouped Aggregation | Exclude light-blue highlighted cells. Among the remaining values in column Average, what is the maximum value? Output only the number as a string. | "79.7" |
| 12 | HB_real_001838 | Aggregation & Comparison | Argmax Reasoning | In the numeric part of the table, find the largest numeric cell. Return JSON {"row":<int>,"col":<int>,"value":<string>}. Count rows top-to-bottom and numeric columns left-to-right. | {"row": 7, "col": 2, "value": "100.0"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000516 | Markup Grounding | Underline Presence | Based on the table image, is the number with value 53.9 underlined? Answer JSON {"underlined":true/false}. The value appears exactly once in the numeric data grid. | {"underlined": true} |
| 2 | HB_real_001252 | Constrained Retrieval | Cell Retrieval | What is the value at column PathCLS ALL for row CPath-Omni? Output only the number as a string. | "79.0" |
| 3 | HB_real_001253 | Local Relations | Neighbor Value Tracing | For the cell with value 60.7, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "60.7", "up": "45.3", "down": "70.6", "left": "60.8", "right": "63.5"} |
| 4 | HB_real_001254 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | ["53.9", "49.8", "59.4", "53.5", "58.7", "53.9", "60.4", "53.6", "51.4", "36.2", "34.7", "52.8"] |
| 5 | HB_real_001255 | Local Relations | Marker-Anchor Neighbor | For the cell with value 60.7, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "60.8", "right": "63.5"} |
| 6 | HB_real_001256 | Aggregation & Comparison | Extremum Reasoning | In row CPath-Omni, what is the minimum value across all dataset metrics? Output only the number as a string. | "65.9" |
| 7 | HB_real_001843 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 79.0, list all numeric values highlighted with the same light-blue row background as the anchor row. Output only a JSON array of strings. | ["72.4", "72.2", "74.0", "69.9", "76.6", "71.8", "69.8", "70.6", "65.9", "70.6", "75.7", "79.0"] |
| 8 | HB_real_000877 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| 9 | HB_real_000968 | Markup Grounding | Marked Column Listing | List all underlined values in column PathCLS Tiny. Output only a JSON array of strings. | ["36.2"] |
| 10 | HB_real_001701 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light pink in the Expert performance row. What is their maximum numeric value? Output only the number as a string. | "78.9" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001258 | Local Relations | Neighbor Value Tracing | For the cell with value 12.8, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "12.8", "up": "8.0", "down": "25.6", "left": "0.1", "right": "20.5"} |
| 2 | HB_real_001259 | Markup Grounding | Bold Presence | Is the number with value 67.3 boldfaced? Answer JSON {"bold":true/false}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 3 | HB_real_001260 | Local Relations | Marker-Anchor Neighbor | For the cell with value 12.8, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "0.1", "right": "20.5"} |
| 4 | HB_real_rebal_000072 | Markup Grounding | Marked Value Listing | Based on the table, return every underlined value. Output only a JSON array of strings. | [] |
| 5 | HB_real_rebal_000073 | Markup Grounding | Underline Presence | At position (r=1,c=1) in the numeric part of the table, is the value underlined? Answer JSON {"underlined":true/false}. The numeric part excludes the left Model column. | {"underlined": false} |
| 6 | HB_real_rebal_000075 | Markup Grounding | Marked Value Listing | In the CPath-Omni row, list all bold numeric values. Output only a JSON array of strings. | ["33.7", "20.1", "12.9", "8.7", "25.6", "67.3", "70.8"] |
| 7 | HB_real_rebal_000076 | Markup Grounding | Marked Column Listing | In column BLEU_1, list all bold values. Output only a JSON array of strings. | ["33.7"] |
| 8 | HB_real_rebal_000077 | Markup Grounding | Marker Counting | How many bold numeric values are in the CPath-Omni row? Output only the count as a string. | "7" |
| 9 | HB_real_rebal_000079 | Constrained Retrieval | Color-Anchor Retrieval | In the light-blue CPath-Omni row, list the Report Generation values highlighted with the same color. Output only a JSON array of strings. | ["33.7", "20.1", "12.9", "8.7", "25.6"] |
| 10 | HB_real_rebal_000080 | Constrained Retrieval | Color-Anchor Retrieval | For the value 67.3, list the VQA values highlighted with the same light-blue row background. Output only a JSON array of strings. | ["67.3", "70.8"] |
| 11 | HB_real_rebal_000081 | Markup Grounding | Marker Counting | How many Report Generation numeric cells are highlighted in the light-blue CPath-Omni row? Output only the count as a string. | "5" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001261 | Constrained Retrieval | Cell Retrieval | In panel (b) Attention Alignment, what is the value in column V for row +CoV? Output only the number as a string. | "54.3" |
| 2 | HB_real_001262 | Local Relations | Neighbor Value Tracing | For the cell with value 52.2, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "52.2", "up": "54.3", "down": null, "left": "67.2", "right": null} |
| 3 | HB_real_001263 | Local Relations | Marker-Anchor Neighbor | For the cell with value 52.2, return its up and left neighbors as JSON {"up":"...","left":"..."}. | {"up": "54.3", "left": "67.2"} |
| 4 | HB_real_001264 | Consistency & Missingness | Cell Missingness | Is the cell in panel (a) Module at row MDM=✗, DAA=✗ and column G missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_real_rebal_000082 | Markup Grounding | Underline Presence | The value 54.3 appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001265 | Constrained Retrieval | Cell Retrieval | What is the value at column CV-Bench3D (block Vision-Centric) for row MODA-34B? Output only the number as a string. | "81.3" |
| 2 | HB_real_001266 | Local Relations | Neighbor Value Tracing | For the cell with value 74.3, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "74.3", "up": "73.3", "down": "79.8", "left": "72.0", "right": "65.2"} |
| 3 | HB_real_001267 | Local Relations | Marker-Anchor Neighbor | For the cell with value 74.3, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "72.0", "right": "65.2"} |
| 4 | HB_real_001835 | Constrained Retrieval | Color-Anchor Retrieval | In the light-blue MODA-8B row under Base LLM: Llama-3-Ins-8B, list all numeric values in that highlighted row. Output only a JSON array of strings. | ["72.1", "1535.9", "73.8", "74.9", "63.0", "61.5", "80.4", "43.1", "48.8", "73.6", "72.0", "74.3", "65.2", "70.4", "78.1", "66.0", "52.6", "64.1", "73.5", "73.8"] |
| 5 | HB_real_000778 | Markup Grounding | Marked Value Listing | List all numeric values highlighted in light cyan in the MODA-34B row. Output only a JSON array of strings. | ["76.7", "1639.2", "82.3", "75.8", "66.2", "69.5", "88.1", "52.5", "54.0", "83.4", "74.7", "79.8", "62.7", "78.3", "78.2", "69.9", "53.8", "68.5", "75.8", "81.3"] |
| 6 | HB_real_000812 | Aggregation & Comparison | Grouped Aggregation | Take the cells highlighted in light cyan in the MODA-34B row. What is their maximum numeric value? Output only the number as a string. | "1639.2" |
| 7 | HB_real_000671 | Aggregation & Comparison | Argmax Reasoning | Considering only numeric cells, find the single largest numeric cell in the table. Return JSON {"row":<1-based data row index>,"col":"<column label>","value":"<cell value>"}. | {"row": 17, "col": "MME^P", "value": "1689.3"} |
| 8 | HB_real_000467 | Local Relations | Neighbor Indexing | At position (r=14,c=15) in the numeric part of the table, return the target and its four neighbors as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. Use null if a neighbor does not exist. | {"target": "79.8", "up": "74.3", "down": "76.1", "left": "74.7", "right": "62.7"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_000566 | Aggregation & Comparison | Grouped Aggregation | Within the GPT/Gemini/Claude/Qwen group, what is the mean value of column Response Accuracy? Output only the number as a string, rounded to three decimal places. | "1.070" |
| 2 | HB_real_001268 | Constrained Retrieval | Cell Retrieval | What is the value at column Personality Consistency for row MODA-8B (In-Test)? Output only the number as a string. | "1.013" |
| 3 | HB_real_001269 | Local Relations | Neighbor Value Tracing | For the cell with value 1.127, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "1.127", "up": "0.999", "down": "0.884", "left": "1.157", "right": "1.070"} |
| 4 | HB_real_001270 | Markup Grounding | Highlight Presence | Is the number with value 1.013 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 5 | HB_real_001271 | Local Relations | Marker-Anchor Neighbor | For the highlighted cell with value 1.013, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "0.992", "right": "0.996"} |
| 6 | HB_real_001272 | Aggregation & Comparison | Extremum Reasoning | In row MODA-8B (In-Test), what is the minimum value across all Cognition metrics? Output only the number as a string. | "0.996" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001273 | Constrained Retrieval | Cell Retrieval | What is the value at column HFM (F1) for row MODA-8B in the Emotion-Specialized block? Output only the number as a string. | "0.881" |
| 2 | HB_real_001274 | Local Relations | Neighbor Value Tracing | For the cell with value 0.739, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.739", "up": "0.732", "down": "0.697", "left": "0.755", "right": "0.738"} |
| 3 | HB_real_001275 | Markup Grounding | Highlight Presence | Is the number with value 0.588 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001276 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.739, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "0.755", "right": "0.738"} |
| 5 | HB_real_rebal_000097 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.881, list all numbers highlighted with the same green color as the anchor. Output only a JSON array of strings. | ["0.881"] |
| 6 | HB_real_rebal_000098 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | [] |
| 7 | HB_real_rebal_000099 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001277 | Constrained Retrieval | Cell Retrieval | What is the value at column LPIPS↓ (block Average) for row Ours? Output only the number as a string. Note: LPIPS↓ means lower is better. | "0.0040" |
| 2 | HB_real_001278 | Local Relations | Neighbor Value Tracing | For the cell with value 40.82, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "40.82", "up": "41.24", "down": "40.52", "left": "0.0015", "right": "0.9992"} |
| 3 | HB_real_001279 | Markup Grounding | Highlight Presence | Is the number with value 44.29 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001280 | Local Relations | Marker-Anchor Neighbor | For the cell with value 40.82, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "0.0015", "right": "0.9992"} |
| 5 | HB_real_001281 | Consistency & Missingness | Cell Missingness | Is the cell (row D-NeRF, column MS-SSIM↑, block Mutant) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_real_rebal_000100 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | [] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001282 | Constrained Retrieval | Cell Retrieval | What is the value at column CC_NN for dataset B? Output the full cell text as a string. | "90.0(1.2)" |
| 2 | HB_real_001283 | Local Relations | Neighbor Value Tracing | For the cell with value 89.9(1.6), return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "89.9(1.6)", "up": "88.4(1.5)", "down": "93.4(1.4)", "left": "82.3(1.4)", "right": "89.9(1.6)"} |
| 3 | HB_real_001284 | Markup Grounding | Bold Presence | Is the number with value 94.5(1.4) boldfaced? Answer JSON {"bold":true/false}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 4 | HB_real_001285 | Local Relations | Marker-Anchor Neighbor | For the cell with value 89.9(1.6), return its up and down neighbors as JSON {"up":"...","down":"..."}. | {"up": "88.4(1.5)", "down": "93.4(1.4)"} |
| 5 | HB_real_rebal_000083 | Markup Grounding | Underline Presence | The value 90.0(1.2) appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001286 | Constrained Retrieval | Cell Retrieval | What is the value at column ENN for dataset E? Output only the number as a string. | "0.634" |
| 2 | HB_real_001287 | Local Relations | Neighbor Value Tracing | For the cell with value 0.403, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "0.403", "up": "0.032", "down": "0.018", "left": "0.023", "right": "0.197"} |
| 3 | HB_real_001288 | Markup Grounding | Highlight Presence | Is the number with value 0.692 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001289 | Local Relations | Marker-Anchor Neighbor | For the cell with value 0.403, return its up and down neighbors as JSON {"up":"...","down":"..."}. | {"up": "0.032", "down": "0.018"} |
| 5 | HB_real_rebal_000084 | Markup Grounding | Underline Presence | The value 88.8(1.8) appears exactly once in the table. Is it underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001290 | Constrained Retrieval | Cell Retrieval | What is the value at column SBC for dataset I? Output only the number as a string. | "2.861" |
| 2 | HB_real_001291 | Local Relations | Neighbor Value Tracing | For the cell with value 1.867, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "1.867", "up": "1.521", "down": "1.612", "left": "1.619", "right": "1.728"} |
| 3 | HB_real_001292 | Markup Grounding | Highlight Presence | Is the number with value 0.800 highlighted (background-colored)? Answer JSON {"highlighted":true/false}. The value appears exactly once in the numeric data grid. | {"highlighted": true} |
| 4 | HB_real_001293 | Local Relations | Marker-Anchor Neighbor | For the cell with value 1.867, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "1.619", "right": "1.728"} |
| 5 | HB_real_001294 | Consistency & Missingness | Cell Missingness | Is the cell (dataset C, column TL) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_real_rebal_000094 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 0.800, list all numbers highlighted with the same deep-red color as the anchor. Output only a JSON array of strings. | ["0.800"] |
| 7 | HB_real_rebal_000095 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | [] |
| 8 | HB_real_rebal_000096 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_real_001296 | Local Relations | Neighbor Value Tracing | For the cell with value 1.557, return the target and its four neighbors (up, down, left, right) as JSON {"target":"...","up":"...","down":"...","left":"...","right":"..."}. | {"target": "1.557", "up": "1.333", "down": "1.586", "left": "1.338", "right": "1.554"} |
| 2 | HB_real_001297 | Markup Grounding | Bold Presence | Is the number with value 2.468 boldfaced? Answer JSON {"bold":true/false}. The value appears exactly once in the numeric data grid. | {"bold": true} |
| 3 | HB_real_001298 | Local Relations | Marker-Anchor Neighbor | For the cell with value 1.557, return its left and right neighbors as JSON {"left":"...","right":"..."}. | {"left": "1.338", "right": "1.554"} |
| 4 | HB_real_001299 | Consistency & Missingness | Cell Missingness | Is the cell (size Large, dataset L, column CNN, block Macro-F1) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_real_rebal_000087 | Markup Grounding | Marked Column Listing | In column NoUnder (block Macro-F1), list all boldfaced numeric values. Output only a JSON array of strings. | ["91.2(1.0)", "93.9(1.4)", "95.2(0.7)", "91.7(1.8)", "83.9(5.4)", "88.8(3.7)", "76.3(6.0)", "96.7(0.9)"] |
| 6 | HB_real_rebal_000088 | Markup Grounding | Marked Column Listing | In column E2SC_US (block Macro-F1), list all boldfaced numeric values. Output only a JSON array of strings. | ["92.1(1.4)", "93.4(4.6)", "92.4(0.5)"] |
| 7 | HB_real_rebal_000089 | Markup Grounding | Marked Value Listing | List all boldfaced numeric values in row K. Output only a JSON array of strings. | ["96.7(0.9)", "0.029", "1.923", "1.923"] |
| 8 | HB_real_rebal_000090 | Markup Grounding | Marker Counting | How many boldfaced numeric cells are in the whole table? Output only the count as a string. | "40" |
| 9 | HB_real_rebal_000091 | Constrained Retrieval | Color-Anchor Retrieval | At position (r=2,c=8) in the numeric part of the table, list all numbers highlighted with the same red color as the anchor. The numeric part starts at column NoUnder in block Macro-F1 and proceeds left to right across blocks. Output only a JSON array of strings. | ["0.030"] |
| 10 | HB_real_rebal_000092 | Markup Grounding | Marked Value Listing | List all underlined numeric values in the table. Output only a JSON array of strings. | [] |
| 11 | HB_real_rebal_000093 | Markup Grounding | Underline Presence | Is the value at position (r=1,c=1) in the numeric part of the table underlined? Answer JSON {"underlined":true/false}. | {"underlined": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000001 | Aggregation & Comparison | Extremum Reasoning | For item Approach-Y, what is the maximum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.01" |
| 2 | HB_syn_000002 | Local Relations | Neighbor Indexing | At (r=9,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "58.61", "up": "55.04", "down": "90.99", "left": "80.48", "right": "16.26"} |
| 3 | HB_syn_000003 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 2, column 8, list all numbers highlighted with the same color (purple, hex #B07CFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["74.19", "22.96", "80.92", "86.30", "26.01", "50.06", "7.96", "24.68", "25.27", "−1.95 ±0.46"] |
| 4 | HB_syn_000004 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Val↑ under Grp-C. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["87.68"] |
| 5 | HB_syn_000005 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-Z, ΔDelta↑ under Cfg-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000006 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number at (r=9,c=4) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000007 | Aggregation & Comparison | Pairwise Comparison | Compare Method-D vs Baseline-Z on Score. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-D", "value": "44.09"} |
| 2 | HB_syn_000008 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the text-colored in green (hex #2E7D32) with value +0.42 ±0.41 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "97.61" |
| 3 | HB_syn_000009 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 9.17, list all numbers highlighted with the same color (red, hex #FF5A5C). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["9.17", "93.59", "−0.33 ±0.34", "22.40"] |
| 4 | HB_syn_000010 | Consistency & Missingness | Column Missingness | Return the items whose Score is missing (N/A or —). Output JSON array of strings. | ["Variant-F"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000011 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Score under Sect-A. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-D", "value": "83.31"} |
| 2 | HB_syn_000012 | Local Relations | Neighbor Value Tracing | Locate value 34.23 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "34.23", "up": "14.37", "down": "65.44", "left": "72.52", "right": "57.44"} |
| 3 | HB_syn_000013 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "24" |
| 4 | HB_syn_000014 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-C, S-D under Cfg-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000015 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are <= 89.37. Output JSON array of strings. | ["87.62", "68.23", "40.21", "57.44", "68.34", "83.31", "62.40", "32.21", "89.37", "70.99", "4.33", "78.87", "79.18"] |
| 6 | HB_syn_000016 | Markup Grounding | Highlight Presence | Check whether value 91.36 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_pink(#FFD1E7), mid_pink(#FC88BC), dark_pink(#F74B98). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "mid_pink", "hex": "#FC88BC"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000017 | Constrained Retrieval | Cell Retrieval | Return the value where Approach-E intersects Rate under Sect-C. Output only the number as a string. Example: "12.34". | "31.89" |
| 2 | HB_syn_000018 | Aggregation & Comparison | Extremum Reasoning | Return the max of column Rate. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.11" |
| 3 | HB_syn_000019 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Val under Sect-B. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 4, "col": 2, "value": "96.30"} |
| 4 | HB_syn_000020 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the highlighted in orange (hex #F9B163) with value −1.97 ±0.50 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "49.36" |
| 5 | HB_syn_000021 | Markup Grounding | Marked Value Listing | Find all orange-highlighted values (hex #F9B163) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["53.57", "−1.97 ±0.50", "−2.04 ±0.68", "82.07", "54.03", "4.45", "97.17", "2.07", "−1.52 ±0.55"] |
| 6 | HB_syn_000022 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "5" |
| 7 | HB_syn_000023 | Consistency & Missingness | Column Missingness | In Val under Sect-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Method-B [26]"] |
| 8 | HB_syn_000024 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are <= 53.57. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["53.57", "4.45", "6.81", "22.73", "53.53", "2.07"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000025 | Aggregation & Comparison | Extremum Reasoning | In row System-D, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "11.45" |
| 2 | HB_syn_000026 | Local Relations | Neighbor Indexing | At (r=11,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "66.91", "up": "42.43", "down": "0.74", "left": "26.19", "right": "35.08"} |
| 3 | HB_syn_000027 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 57.22, list all numbers highlighted with the same color (purple, hex #B07CFF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["20.25", "10.37", "53.91", "33.24", "26.19", "57.22"] |
| 4 | HB_syn_000028 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["20.91", "87.05", "93.40", "85.08", "10.99"] |
| 5 | HB_syn_000029 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-D, S-A↓ under Zone-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000030 | Aggregation & Comparison | Rank Selection | In Eff, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-Z [46]", "value": "83.33"}, {"item": "System-E", "value": "71.94"}, {"item": "Baseline-A", "value": "69.51"}] |
| 2 | HB_syn_000031 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in light_cyan_blue (hex #C8ECFF) with value 69.51 number and return the number immediately to its right. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "38.41" |
| 3 | HB_syn_000032 | Markup Grounding | Marked Value Listing | Which values are highlighted in light_cyan_blue (#C8ECFF)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["69.51", "55.05"] |
| 4 | HB_syn_000033 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=4,c=2) as an anchor, list all cells sharing its highlight color (dark_cyan_blue/#55BFFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["83.33", "38.41"] |
| 5 | HB_syn_000034 | Consistency & Missingness | Column Missingness | List all missing entries in column Eff (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-X", "Approach-B"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000035 | Aggregation & Comparison | Extremum Reasoning | Return the min of column MetB. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "34.14" |
| 2 | HB_syn_000036 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column MetDB? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-D", "value": "91.76"} |
| 3 | HB_syn_000037 | Local Relations | Neighbor Value Tracing | Locate value 16.18 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "16.18", "up": "23.38", "down": "13.27", "left": "7.57", "right": "84.01"} |
| 4 | HB_syn_000038 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 35.15. Return all values sharing its highlight color (light_purple/#E6D1FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["23.38", "37.03", "7.57", "84.01", "35.15", "19.33", "66.30", "33.91", "66.00", "1.34"] |
| 5 | HB_syn_000039 | Markup Grounding | Marked Column Listing | In MetA↓, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["32.55"] |
| 6 | HB_syn_000040 | Consistency & Missingness | Column Missingness | Return the items whose Qual is missing (N/A or —). Output JSON array of strings. | ["Baseline-D", "Method-D", "Approach-E"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000041 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in Err. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "94.13" |
| 2 | HB_syn_000042 | Local Relations | Neighbor Indexing | At (r=5,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "31.59", "up": "37.86", "down": "94.13", "left": "36.68", "right": "4.93"} |
| 3 | HB_syn_000043 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-Z, Stat↓ under Set-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_syn_000044 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are >= 67.56. Output JSON array of strings. | ["82.88", "67.56", "90.53", "94.13", "81.79", "91.32", "97.38"] |
| 5 | HB_syn_000045 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number with value 97.38 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "light purple", "hex": "#E593FC"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000046 | Aggregation & Comparison | Extremum Reasoning | For item Baseline-F [56], what is the minimum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "1.74" |
| 2 | HB_syn_000047 | Local Relations | Neighbor Value Tracing | Locate value 5.86 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "5.86", "up": "89.07", "down": "22.99", "left": "84.24", "right": "11.03"} |
| 3 | HB_syn_000048 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=7,c=3) as an anchor, list all cells sharing its highlight color (mid_pink/#FC88BC). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["87.84", "97.52", "71.97", "3.34", "90.02", "81.37", "15.36"] |
| 4 | HB_syn_000049 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["87.84", "97.52", "71.97", "3.34", "90.02", "81.37", "15.36"] |
| 5 | HB_syn_000050 | Consistency & Missingness | Column Missingness | In MetD↑ under Grp-C, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Baseline-Y"] |
| 6 | HB_syn_000051 | Markup Grounding | Highlight Presence | Check whether (r=2,c=1) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_pink(#FFD1E7), mid_pink(#FC88BC), dark_pink(#F74B98). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000052 | Constrained Retrieval | Cell Retrieval | For item Baseline-E, what is the value in column MetDB? Output only the number as a string. Example: "12.34". | "84.69" |
| 2 | HB_syn_000053 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Eff. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-F [53]", "value": "94.64"} |
| 3 | HB_syn_000054 | Aggregation & Comparison | Pairwise Comparison | On metric Eff, which is higher: Approach-F [53] or Model-E? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-F [53]", "value": "94.64"} |
| 4 | HB_syn_000055 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the highlighted in light_green (hex #D7F6D1) with value 54.05 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "73.02" |
| 5 | HB_syn_000056 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "24" |
| 6 | HB_syn_000057 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "8" |
| 7 | HB_syn_000058 | Consistency & Missingness | Cell Missingness | Check whether Method-F at MetD↑ is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 8 | HB_syn_000059 | Constrained Retrieval | Value Filtering | Filter MetD↑ by >= 34.73. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-Y", "value": "74.30"}, {"item": "Approach-F [53]", "value": "46.68"}, {"item": "System-D [17]", "value": "94.27"}, {"item": "Method-Y", "value": "34.73"}, {"item": "Baseline-E", "value": "86.97"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000060 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in Rate. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "11.77" |
| 2 | HB_syn_000061 | Local Relations | Neighbor Value Tracing | At the cell with value 23.32, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "23.32", "up": "85.78", "down": "—", "left": "85.01", "right": "18.57"} |
| 3 | HB_syn_000062 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 92.05, list all numbers highlighted with the same color (mid_orange, hex #F1B17D). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["12.90", "73.91", "90.61", "92.05", "79.81", "86.51"] |
| 4 | HB_syn_000063 | Consistency & Missingness | Cell Missingness | Check whether Baseline-C at Val is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000064 | Markup Grounding | Highlight Presence | Check whether value 98.50 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "dark_orange", "hex": "#E47B30"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000065 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column Err under Blk-B? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-X", "value": "77.66"} |
| 2 | HB_syn_000066 | Local Relations | Neighbor Indexing | At (r=5,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "8.06", "up": "7.09", "down": "37.60", "left": "95.01", "right": "5.74"} |
| 3 | HB_syn_000067 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=9,c=9). Return all values with the same highlight color (red, #FF5A5C). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["2.97", "3.58", "19.72", "26.02", "16.97", "36.56"] |
| 4 | HB_syn_000068 | Consistency & Missingness | Column Missingness | Return the items whose MetD under Cfg-C is missing (N/A or —). Output JSON array of strings. | ["Model-E", "System-X"] |
| 5 | HB_syn_000069 | Markup Grounding | Highlight Presence | Check whether (r=8,c=8) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "red", "hex": "#FF5A5C"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000070 | Aggregation & Comparison | Extremum Reasoning | In row Approach-F [52], what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "92.17" |
| 2 | HB_syn_000071 | Local Relations | Neighbor Value Tracing | Locate value 89.82 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "89.82", "up": "57.02", "down": "11.76", "left": "17.21", "right": "83.82"} |
| 3 | HB_syn_000072 | Markup Grounding | Marked Value Listing | Which values are highlighted in blue (#6DA1FD)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["61.83", "31.38", "4.74", "11.76", "68.30", "72.99", "20.36", "99.04", "−2.47 ±0.45", "24.54"] |
| 4 | HB_syn_000073 | Consistency & Missingness | Column Missingness | List all missing entries in column MetD↑ under Cfg-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Baseline-Z"] |
| 5 | HB_syn_000074 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are >= 57.02. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["61.83", "57.02", "68.30", "72.99", "75.45", "99.04", "74.92"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000075 | Aggregation & Comparison | Pairwise Comparison | In row Baseline-X, is Qual under Grp-A higher or S-A under Grp-A higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "Qual under Grp-A", "value": "23.10"} |
| 2 | HB_syn_000076 | Local Relations | Marker-Anchor Neighbor | Locate the underlined with value 67.05 number and return the number immediately above it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "61.47" |
| 3 | HB_syn_000077 | Markup Grounding | Marked Column Listing | Which entries in column MetD↓ under Set-B are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["33.54"] |
| 4 | HB_syn_000078 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "7" |
| 5 | HB_syn_000079 | Consistency & Missingness | Cell Missingness | Check whether Model-B at Qual under Grp-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000080 | Constrained Retrieval | Value Filtering | In column MetD↓ under Set-B, list all row labels in the leftmost row-label column with value < 46.46. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Variant-E", "value": "46.18"}, {"item": "Baseline-A [51]", "value": "39.11"}, {"item": "Model-B", "value": "44.95"}, {"item": "System-Z", "value": "33.54"}, {"item": "System-B", "value": "41.49"}, {"item": "Model-D", "value": "38.49"}, {"item": "Variant-F", "value": "4.90"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000081 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column S-C under Sect-C. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 12, "col": 8, "value": "98.97"} |
| 2 | HB_syn_000082 | Local Relations | Neighbor Indexing | At (r=9,c=7) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "12.92", "up": "71.37", "down": "43.66", "left": "75.91", "right": "20.08"} |
| 3 | HB_syn_000083 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 8.35. Return all values sharing its highlight color (dark_cyan_blue/#55BFFF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["99.75", "22.81", "8.35", "5.48", "71.78", "95.60", "14.49", "98.97"] |
| 4 | HB_syn_000084 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "24" |
| 5 | HB_syn_000085 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-X, Score↓ under Sect-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000086 | Constrained Retrieval | Cell Retrieval | Locate row Baseline-A [3] and column Val↓ under Sect-B. What is the cell value? Output only the number as a string. Example: "12.34". | "47.49" |
| 2 | HB_syn_000087 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Variant-D, what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "3.43" |
| 3 | HB_syn_000088 | Aggregation & Comparison | Extremum Reasoning | In row Baseline-Z, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "78.06" |
| 4 | HB_syn_000089 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the highlighted in light_green (hex #D7F6D1) with value 7.07 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "73.49" |
| 5 | HB_syn_000090 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=7,c=1). Return all values with the same highlight color (mid_green, #AEEA9F). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["76.94", "12.04", "9.25", "3.02", "68.04"] |
| 6 | HB_syn_000091 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["76.94", "12.04", "9.25", "3.02", "68.04"] |
| 7 | HB_syn_000092 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Val↓ under Sect-B. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["12.04"] |
| 8 | HB_syn_000093 | Constrained Retrieval | Value Filtering | In column MetE↓ under Slot-A, list all row labels in the leftmost row-label column with value < 16.48. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Variant-C", "value": "1.33"}, {"item": "Approach-Z", "value": "15.37"}, {"item": "Variant-D", "value": "7.07"}, {"item": "Baseline-F", "value": "3.02"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000094 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Val under Grp-A in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Approach-B [40]", "value": "93.32"}, {"item": "System-F", "value": "74.83"}, {"item": "Method-Z", "value": "67.06"}] |
| 2 | HB_syn_000095 | Local Relations | Neighbor Value Tracing | At the cell with value 16.86, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "16.86", "up": "64.40", "down": "42.80", "left": "49.17", "right": "92.01"} |
| 3 | HB_syn_000096 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["92.01", "3.12", "93.32", "98.62", "78.22", "6.58", "87.24", "3.35", "92.34"] |
| 4 | HB_syn_000097 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "27" |
| 5 | HB_syn_000098 | Consistency & Missingness | Column Missingness | Return the items whose Idx↑ under Slot-C is missing (N/A or —). Output JSON array of strings. | ["System-F"] |
| 6 | HB_syn_000099 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those < 78.22 as a JSON array of strings. | ["72.60", "10.27", "74.83", "14.74", "15.89", "7.15", "67.06", "3.12", "10.33", "71.65", "14.90", "6.58", "3.35"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000100 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column MetA? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "76.12" |
| 2 | HB_syn_000101 | Aggregation & Comparison | Argmax Reasoning | In column EffB, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-Y", "value": "96.33"} |
| 3 | HB_syn_000102 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column EffB. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 5, "col": 8, "value": "96.33"} |
| 4 | HB_syn_000103 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the text-colored in red (hex #C62828) with value −2.06 ±0.60 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "29.73" |
| 5 | HB_syn_000104 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 80.48, list all numbers highlighted with the same color (blue, hex #6FA0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["35.99", "47.55", "27.89", "80.48"] |
| 6 | HB_syn_000105 | Consistency & Missingness | Cell Missingness | Check whether Method-A at S-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 7 | HB_syn_000106 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number at (r=4,c=1) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| 8 | HB_syn_000107 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number with value 56.45 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000108 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in Score. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "1.04" |
| 2 | HB_syn_000109 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=10,c=3), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "72.93", "up": "10.37", "down": "15.16", "left": "1.86", "right": "42.32"} |
| 3 | HB_syn_000110 | Markup Grounding | Marked Value Listing | Which values are highlighted in light_orange (#FFE6C7)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["65.85", "6.12"] |
| 4 | HB_syn_000111 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 1, column 2, list all numbers highlighted with the same color (light_orange, hex #FFE6C7). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["65.85", "6.12"] |
| 5 | HB_syn_000112 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "6" |
| 6 | HB_syn_000113 | Consistency & Missingness | Column Missingness | In Score under Blk-C, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["System-C [63]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000114 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Variant-B, what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "4.91" |
| 2 | HB_syn_000115 | Aggregation & Comparison | Extremum Reasoning | In row Variant-Y [68], what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "88.30" |
| 3 | HB_syn_000116 | Local Relations | Neighbor Value Tracing | Locate value 20.86 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "20.86", "up": "97.50", "down": "82.14", "left": "61.82", "right": "22.48"} |
| 4 | HB_syn_000117 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-Y [68], MetA) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000118 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are > 61.84. Output JSON array of strings. | ["96.14", "68.03", "66.10", "97.50", "98.77", "89.12", "88.30", "74.11", "82.14", "76.48", "71.68", "82.75", "84.76", "81.78", "73.64", "93.84"] |
| 6 | HB_syn_000119 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_red(#FFD6D6), mid_red(#F86D7B), dark_red(#FC5255). Is the number at (r=1,c=9) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000120 | Constrained Retrieval | Cell Retrieval | Give the number at (Model-B, Idx↑ under Slot-C). Output only the number as a string. Example: "12.34". | "30.46" |
| 2 | HB_syn_000121 | Aggregation & Comparison | Pairwise Comparison | Between Baseline-Y and System-E for Idx under Zone-A, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Baseline-Y", "value": "92.29"} |
| 3 | HB_syn_000122 | Local Relations | Marker-Anchor Neighbor | Find the down neighbor of the highlighted in light_blue (hex #E6F2FF) with value 36.23 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "60.13" |
| 4 | HB_syn_000123 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["84.33", "10.15", "73.98", "84.75", "86.22", "93.79", "93.74", "89.88"] |
| 5 | HB_syn_000124 | Consistency & Missingness | Column Missingness | Return the items whose Rate↑ under Slot-D is missing (N/A or —). Output JSON array of strings. | ["Model-Z [34]", "System-X"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000125 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Score under Set-A in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Model-B", "value": "67.99"}, {"item": "Approach-A", "value": "65.25"}, {"item": "Baseline-Y", "value": "58.85"}] |
| 2 | HB_syn_000126 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Stat under Set-C and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Approach-A", "value": "81.58"} |
| 3 | HB_syn_000127 | Local Relations | Marker-Anchor Neighbor | Locate the underlined with value 65.25 number and return the number immediately above it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "57.12" |
| 4 | HB_syn_000128 | Consistency & Missingness | Column Missingness | In Val under Sect-D, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Approach-A", "System-X"] |
| 5 | HB_syn_000129 | Constrained Retrieval | Value Filtering | Filter Val↑ under Slot-B by >= 37.26. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Variant-X", "value": "89.43"}, {"item": "System-A [72]", "value": "37.26"}, {"item": "Model-Z", "value": "82.87"}, {"item": "Variant-B", "value": "95.65"}, {"item": "System-X", "value": "49.56"}] |
| 6 | HB_syn_000130 | Markup Grounding | Highlight Presence | Check whether value 70.04 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000131 | Aggregation & Comparison | Argmax Reasoning | In Eff under Sect-C, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 6, "col": 8, "value": "99.81"} |
| 2 | HB_syn_000132 | Local Relations | Neighbor Indexing | At (r=5,c=7) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "25.75", "up": "65.90", "down": "49.55", "left": "37.91", "right": "78.58"} |
| 3 | HB_syn_000133 | Markup Grounding | Marked Value Listing | Which values are highlighted in light_blue (#E6F2FF)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["39.21", "78.58", "65.19", "24.54", "35.91"] |
| 4 | HB_syn_000134 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 49.55. Return all values sharing its highlight color (orange/#F9B163). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["15.56", "46.26", "49.55", "89.81", "79.18", "77.95", "10.30", "58.13", "40.33"] |
| 5 | HB_syn_000135 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-A [77], Cost↑ under Part-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000136 | Aggregation & Comparison | Extremum Reasoning | Return the max of column S-D. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "78.75" |
| 2 | HB_syn_000137 | Local Relations | Neighbor Value Tracing | At the cell with value 70.03, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "70.03", "up": "45.52", "down": "85.59", "left": "91.01", "right": "36.39"} |
| 3 | HB_syn_000138 | Markup Grounding | Marked Column Listing | In Rate↑ under Grp-A, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["90.26"] |
| 4 | HB_syn_000139 | Consistency & Missingness | Cell Missingness | Check whether Variant-F [38] at S-D under Part-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000140 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are <= 85.59. Output JSON array of strings. | ["78.73", "78.55", "81.13", "75.94", "81.74", "35.97", "78.75", "30.94", "72.90", "28.32", "85.45", "85.47", "85.59"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000141 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column MetE? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "9.28" |
| 2 | HB_syn_000142 | Aggregation & Comparison | Argmax Reasoning | In column Eff under Zone-C, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-C", "value": "98.42"} |
| 3 | HB_syn_000143 | Aggregation & Comparison | Pairwise Comparison | Between Approach-C and Baseline-E [94] for Eff under Zone-C, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Approach-C", "value": "98.42"} |
| 4 | HB_syn_000144 | Local Relations | Marker-Anchor Neighbor | Locate the boldfaced with value 86.79 number and return the number immediately above it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "10.16" |
| 5 | HB_syn_000145 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 8, column 5, list all numbers highlighted with the same color (orange, hex #FF8C3A). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["74.53", "0.96", "74.44"] |
| 6 | HB_syn_000146 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "7" |
| 7 | HB_syn_000147 | Consistency & Missingness | Column Missingness | Return the items whose MetE under Zone-C is missing (N/A or —). Output JSON array of strings. | ["Baseline-E [94]", "Method-Z"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000148 | Constrained Retrieval | Cell Retrieval | For item System-E, what is the value in column Val↑ under Blk-A? Output only the number as a string. Example: "12.34". | "39.42" |
| 2 | HB_syn_000149 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Method-Z, what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "71.81" |
| 3 | HB_syn_000150 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=8,c=9), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "58.13", "up": "66.64", "down": "33.61", "left": "10.42", "right": "90.88"} |
| 4 | HB_syn_000151 | Markup Grounding | Marked Value Listing | Which values are highlighted in mid_pink (#FC88BC)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["75.87", "18.54", "71.51", "82.64", "10.69", "76.33", "12.52", "12.50", "85.82", "95.86", "91.10", "93.52"] |
| 5 | HB_syn_000152 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "12" |
| 6 | HB_syn_000153 | Consistency & Missingness | Column Missingness | Return the items whose Val↑ under Blk-A is missing (N/A or —). Output JSON array of strings. | ["System-Y"] |
| 7 | HB_syn_000154 | Constrained Retrieval | Value Filtering | In column Val↑ under Blk-A, list all row labels in the leftmost row-label column with value < 59.45. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-Y", "value": "49.83"}, {"item": "System-E", "value": "39.42"}, {"item": "Variant-Z", "value": "5.34"}, {"item": "Method-Z", "value": "5.46"}, {"item": "Approach-E [61]", "value": "27.75"}] |
| 8 | HB_syn_000155 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_pink(#FFD1E7), mid_pink(#FC88BC), dark_pink(#F74B98). Is the number with value 12.50 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "mid_pink", "hex": "#FC88BC"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000156 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Err under Cfg-C and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Method-E", "value": "49.40"} |
| 2 | HB_syn_000157 | Local Relations | Neighbor Value Tracing | Locate value 17.46 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "17.46", "up": "63.42", "down": "61.46", "left": "1.46", "right": "35.54"} |
| 3 | HB_syn_000158 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 58.31, list all numbers highlighted with the same color (pink, hex #FFB7D6). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["58.31", "58.89"] |
| 4 | HB_syn_000159 | Markup Grounding | Marked Column Listing | Which entries in column Eff↑ under Blk-B are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["62.08"] |
| 5 | HB_syn_000160 | Consistency & Missingness | Cell Missingness | Is the cell (System-Y, S-A↑ under Cfg-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000161 | Aggregation & Comparison | Extremum Reasoning | In row Variant-B, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "12.44" |
| 2 | HB_syn_000162 | Local Relations | Neighbor Indexing | At (r=4,c=6) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "21.34", "up": "55.73", "down": "—", "left": "36.60", "right": "28.54"} |
| 3 | HB_syn_000163 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 9, column 10, list all numbers highlighted with the same color (light_cyan_blue, hex #C8ECFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["16.51", "54.55", "34.24", "31.53", "58.09", "39.27", "54.11", "66.74", "28.83", "25.81"] |
| 4 | HB_syn_000164 | Consistency & Missingness | Column Missingness | Return the items whose Rank↑ under Set-C is missing (N/A or —). Output JSON array of strings. | ["Baseline-B"] |
| 5 | HB_syn_000165 | Markup Grounding | Highlight Presence | Check whether (r=2,c=8) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_cyan_blue(#C8ECFF), mid_cyan_blue(#95D9FF), dark_cyan_blue(#55BFFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000166 | Aggregation & Comparison | Pairwise Comparison | In row System-Y, is S-A under Sect-A higher or Err under Sect-A higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "S-A under Sect-A", "value": "91.12"} |
| 2 | HB_syn_000167 | Local Relations | Neighbor Value Tracing | At the cell with value 58.09, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "58.09", "up": "0.87", "down": "24.86", "left": "40.55", "right": "58.30"} |
| 3 | HB_syn_000168 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "24" |
| 4 | HB_syn_000169 | Consistency & Missingness | Cell Missingness | Is the cell (System-Y, S-D↑ under Cfg-B) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000170 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those <= 81.66 as a JSON array of strings. | ["23.50", "81.66", "65.44", "78.40", "73.29", "67.97", "79.09", "70.08", "9.35", "72.24", "12.42", "73.63", "72.29"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000171 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in MetC in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Baseline-F", "value": "96.69"}, {"item": "Variant-A", "value": "89.24"}, {"item": "Model-B", "value": "44.97"}] |
| 2 | HB_syn_000172 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column MetC? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Variant-A", "value": "89.24"} |
| 3 | HB_syn_000173 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (down) to the highlighted in mid_orange (hex #F1B17D) with value 84.96 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "87.85" |
| 4 | HB_syn_000174 | Consistency & Missingness | Cell Missingness | Check whether Baseline-Y at MetD↑ is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000175 | Constrained Retrieval | Value Filtering | Filter MetC by < 89.24. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-Y", "value": "35.20"}, {"item": "Model-B", "value": "44.97"}] |
| 6 | HB_syn_000176 | Markup Grounding | Highlight Presence | Check whether (r=3,c=3) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000177 | Constrained Retrieval | Cell Retrieval | Locate row Baseline-D and column MetA↓ under Blk-B. What is the cell value? Output only the number as a string. Example: "12.34". | "38.29" |
| 2 | HB_syn_000178 | Aggregation & Comparison | Extremum Reasoning | Return the min of column MetE. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "24.48" |
| 3 | HB_syn_000179 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in light_green (hex #D7F6D1) with value 26.33 number and return the number immediately below it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "42.70" |
| 4 | HB_syn_000180 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["64.37", "75.57", "17.02", "15.56", "9.05", "13.40", "17.86", "88.98", "92.39"] |
| 5 | HB_syn_000181 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "9" |
| 6 | HB_syn_000182 | Consistency & Missingness | Column Missingness | In MetA↑ under Slot-D, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Approach-Y", "Model-C"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000183 | Aggregation & Comparison | Pairwise Comparison | Within row Method-Y, choose the larger between S-C under Zone-D and MetA under Zone-D, output {item,value}. Exclude N/A and —. | {"item": "S-C under Zone-D", "value": "79.90"} |
| 2 | HB_syn_000184 | Local Relations | Neighbor Indexing | At (r=9,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "58.91", "up": "50.04", "down": "23.54", "left": "73.94", "right": "57.45"} |
| 3 | HB_syn_000185 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 95.48, list all numbers highlighted with the same color (light_teal, hex #D3FAEE). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["80.30", "73.83", "11.77", "95.48", "79.90", "75.98", "23.10", "63.63", "76.33", "12.30"] |
| 4 | HB_syn_000186 | Markup Grounding | Marked Column Listing | In Stat under Blk-B, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["95.97"] |
| 5 | HB_syn_000187 | Consistency & Missingness | Cell Missingness | Check whether Model-Y at MetA under Zone-D is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000188 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Rank? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "98.84" |
| 2 | HB_syn_000189 | Aggregation & Comparison | Rank Selection | List the top-3 items by column Score under Zone-A, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-D", "value": "98.58"}, {"item": "Baseline-Z", "value": "93.62"}, {"item": "Baseline-Y", "value": "74.05"}] |
| 3 | HB_syn_000190 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the highlighted in orange (hex #FF8C3A) with value −0.43 ±0.38 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "8.20" |
| 4 | HB_syn_000191 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "15" |
| 5 | HB_syn_000192 | Consistency & Missingness | Column Missingness | In Score under Zone-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Baseline-D"] |
| 6 | HB_syn_000193 | Constrained Retrieval | Value Filtering | Take all highlighted standalone numeric values (any color) and return those > 34.72. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["37.27", "44.09", "57.15", "85.82", "93.29"] |
| 7 | HB_syn_000194 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number with value 16.98 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000195 | Aggregation & Comparison | Argmax Reasoning | In column MetD under Zone-A, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-B [73]", "value": "82.39"} |
| 2 | HB_syn_000196 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=5,c=5), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "37.08", "up": "83.30", "down": "57.24", "left": "34.59", "right": "57.80"} |
| 3 | HB_syn_000197 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=4,c=6) as an anchor, list all cells sharing its highlight color (red/#FF5A5C). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["29.73", "59.15", "98.48", "95.07"] |
| 4 | HB_syn_000198 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["59.15", "78.60", "21.16", "85.86", "66.03", "87.30", "90.19"] |
| 5 | HB_syn_000199 | Consistency & Missingness | Column Missingness | Return the items whose Cost under Zone-A is missing (N/A or —). Output JSON array of strings. | ["Baseline-E"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000200 | Constrained Retrieval | Cell Retrieval | In the MetC↓ under Set-C column, read the entry for Variant-Y [59]. Output only the number as a string. Example: "12.34". | "71.99" |
| 2 | HB_syn_000201 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Method-F [2], what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "7.37" |
| 3 | HB_syn_000202 | Local Relations | Neighbor Value Tracing | At the cell with value 26.89, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "26.89", "up": "32.37", "down": "42.40", "left": "21.87", "right": "23.99"} |
| 4 | HB_syn_000203 | Markup Grounding | Marked Value Listing | Return all numbers highlighted in dark_cyan_blue (hex #55BFFF) as a JSON array. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["4.01", "16.02", "91.12", "7.01", "99.85", "91.18", "92.94", "7.91"] |
| 5 | HB_syn_000204 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "8" |
| 6 | HB_syn_000205 | Consistency & Missingness | Cell Missingness | Check whether Approach-A at MetC↑ under Slot-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000206 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest Score under Cfg-B? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-F", "value": "94.19"} |
| 2 | HB_syn_000207 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Score under Cfg-B. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 10, "col": 3, "value": "94.19"} |
| 3 | HB_syn_000208 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Method-Z [20], what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "97.37" |
| 4 | HB_syn_000209 | Local Relations | Marker-Anchor Neighbor | Locate the text-colored in green (hex #2E7D32) with value +0.13 ±0.75 number and return the number immediately to its left. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "86.18" |
| 5 | HB_syn_000210 | Constrained Retrieval | Value Filtering | Filter Score↑ under Blk-A by < 73.80. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-Y", "value": "2.37"}, {"item": "System-E", "value": "28.58"}, {"item": "Method-Z [20]", "value": "29.46"}, {"item": "Variant-A", "value": "3.89"}, {"item": "System-F", "value": "29.86"}] |
| 6 | HB_syn_000211 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). Is the number at (r=5,c=5) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000212 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in Cost. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.68" |
| 2 | HB_syn_000213 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Cost? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "2.61" |
| 3 | HB_syn_000214 | Aggregation & Comparison | Rank Selection | List the top-3 items by column Cost, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-C [3]", "value": "99.68"}, {"item": "System-Z [81]", "value": "97.61"}, {"item": "Baseline-A [18]", "value": "96.75"}] |
| 4 | HB_syn_000215 | Aggregation & Comparison | Pairwise Comparison | Compare Method-B [55] vs Method-A on Cost. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-B [55]", "value": "68.25"} |
| 5 | HB_syn_000216 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in light purple (hex #E593FC) with value 80.67 number and return the number immediately to its right. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "32.41" |
| 6 | HB_syn_000217 | Markup Grounding | Marked Value Listing | List numbers highlighted in blue (hex #6FA0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["89.09"] |
| 7 | HB_syn_000218 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 89.09. Return all values sharing its highlight color (blue/#6FA0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["89.09"] |
| 8 | HB_syn_000219 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number with value 89.09 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "blue", "hex": "#6FA0FF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000220 | Aggregation & Comparison | Argmax Reasoning | In S-D under Sect-A, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 9, "col": 1, "value": "95.35"} |
| 2 | HB_syn_000221 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=11,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "77.22", "up": "3.16", "down": "24.13", "left": "20.05", "right": "48.13"} |
| 3 | HB_syn_000222 | Markup Grounding | Marked Column Listing | Which entries in column MetE↑ under Sect-C are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["95.02"] |
| 4 | HB_syn_000223 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "18" |
| 5 | HB_syn_000224 | Consistency & Missingness | Cell Missingness | Check whether Method-Z [31] at Val↑ under Sect-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_syn_000225 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those < 91.31 as a JSON array of strings. | ["1.23", "8.22", "11.52", "9.86", "90.11", "3.16", "87.59", "12.30", "91.02"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000226 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column MetA under Set-C? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-Y [33]", "value": "77.37"} |
| 2 | HB_syn_000227 | Aggregation & Comparison | Pairwise Comparison | Compare System-B vs Model-Z [50] on MetA under Set-C. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-B", "value": "68.41"} |
| 3 | HB_syn_000228 | Local Relations | Neighbor Value Tracing | Locate value 70.93 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "70.93", "up": "93.97", "down": "74.79", "left": "93.85", "right": "94.14"} |
| 4 | HB_syn_000229 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=6,c=2) as an anchor, list all cells sharing its highlight color (dark_green/#72D46A). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["1.25", "2.39", "93.97", "3.59", "2.46", "97.55", "94.07", "4.51", "98.07"] |
| 5 | HB_syn_000230 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "9" |
| 6 | HB_syn_000231 | Consistency & Missingness | Column Missingness | In MetD↑ under Cfg-D, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["System-Z [35]", "Model-C"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000232 | Constrained Retrieval | Cell Retrieval | In the MetA↑ column, read the entry for System-E [79]. Output only the number as a string. Example: "12.34". | "78.27" |
| 2 | HB_syn_000233 | Aggregation & Comparison | Extremum Reasoning | Across all columns for System-D, what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "8.78" |
| 3 | HB_syn_000234 | Local Relations | Neighbor Value Tracing | Locate value 77.53 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "77.53", "up": "42.72", "down": "55.72", "left": "85.27", "right": "97.71"} |
| 4 | HB_syn_000235 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["91.14", "2.57", "81.24", "98.66"] |
| 5 | HB_syn_000236 | Consistency & Missingness | Column Missingness | In MetE↑, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-C", "Method-A", "Baseline-E"] |
| 6 | HB_syn_000237 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_green(#D7F6D1), mid_green(#AEEA9F), dark_green(#72D46A). Is the number with value 15.74 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000238 | Aggregation & Comparison | Pairwise Comparison | Compare Variant-C vs Variant-F on Err. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Variant-F", "value": "37.31"} |
| 2 | HB_syn_000239 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=4,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "2.88", "up": "37.22", "down": "37.31", "left": "2.75", "right": "45.95"} |
| 3 | HB_syn_000240 | Markup Grounding | Marked Column Listing | In Err, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["74.16"] |
| 4 | HB_syn_000241 | Consistency & Missingness | Cell Missingness | Check whether Variant-E at Qual↑ is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000242 | Constrained Retrieval | Value Filtering | In column Qual↑, list all row labels in the leftmost row-label column with value < 45.95. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-Y [36]", "value": "31.09"}, {"item": "Variant-C", "value": "5.02"}, {"item": "Method-Z", "value": "36.66"}, {"item": "Model-X", "value": "31.15"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000243 | Aggregation & Comparison | Pairwise Comparison | In row System-A, is Eff under Blk-D higher or MetB under Blk-D higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "MetB under Blk-D", "value": "38.58"} |
| 2 | HB_syn_000244 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=5,c=3). Return all values with the same highlight color (red, #FF5A5C). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["81.05", "12.08", "45.22", "34.16", "73.97"] |
| 3 | HB_syn_000245 | Consistency & Missingness | Column Missingness | List all missing entries in column Val↑ under Grp-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-A", "Baseline-Z"] |
| 4 | HB_syn_000246 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number at (r=1,c=8) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000247 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in S-C. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.76" |
| 2 | HB_syn_000248 | Aggregation & Comparison | Extremum Reasoning | In row Method-Z, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "94.16" |
| 3 | HB_syn_000249 | Local Relations | Neighbor Value Tracing | Locate value 50.87 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "50.87", "up": "30.80", "down": "10.58", "left": "34.55", "right": "89.46"} |
| 4 | HB_syn_000250 | Markup Grounding | Marked Value Listing | Return all numbers highlighted in red (hex #FF5A5C) as a JSON array. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["56.16"] |
| 5 | HB_syn_000251 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 39.52, list all numbers highlighted with the same color (yellow, hex #FFE9A3). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["5.94", "53.42", "1.48", "11.18", "92.54", "94.02", "39.52"] |
| 6 | HB_syn_000252 | Consistency & Missingness | Cell Missingness | Check whether System-Y at Acc↓ under Part-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000253 | Aggregation & Comparison | Argmax Reasoning | In Rate under Part-C, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 1, "col": 7, "value": "97.32"} |
| 2 | HB_syn_000254 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Rate under Part-C and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "System-Y", "value": "90.35"} |
| 3 | HB_syn_000255 | Local Relations | Neighbor Indexing | At (r=10,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "97.35", "up": "53.35", "down": "57.90", "left": "52.15", "right": "86.89"} |
| 4 | HB_syn_000256 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "8" |
| 5 | HB_syn_000257 | Consistency & Missingness | Column Missingness | List all missing entries in column Val under Part-C (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Variant-C"] |
| 6 | HB_syn_000258 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are >= 62.78. Output JSON array of strings. | ["62.78", "65.22", "93.38", "97.35", "86.89"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000259 | Aggregation & Comparison | Extremum Reasoning | Return the min of column Rate. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "0.44" |
| 2 | HB_syn_000260 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (down) to the underlined with value 2.07 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "49.39" |
| 3 | HB_syn_000261 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["4.22", "75.53", "2.07", "93.49", "89.33"] |
| 4 | HB_syn_000262 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "15" |
| 5 | HB_syn_000263 | Consistency & Missingness | Cell Missingness | Check whether Variant-Z at MetB↑ under Grp-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000264 | Constrained Retrieval | Value Filtering | In column Val↓ under Set-B, list all row labels in the leftmost row-label column with value > 32.18. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-A [48]", "value": "64.80"}, {"item": "Method-D", "value": "89.73"}, {"item": "System-X", "value": "49.80"}, {"item": "System-Z", "value": "65.42"}, {"item": "Method-X", "value": "76.15"}, {"item": "Baseline-Y", "value": "54.33"}, {"item": "Method-B", "value": "91.81"}, {"item": "Baseline-X [76]", "value": "33.02"}, {"item": "Method-Z", "value": "64.87"}, {"item": "Baseline-Z [9]", "value": "51.61"}, {"item": "Variant-B", "value": "60.34"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000265 | Constrained Retrieval | Cell Retrieval | Return the value where System-D intersects MetC↑ under Cfg-C. Output only the number as a string. Example: "12.34". | "98.01" |
| 2 | HB_syn_000266 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest Eff under Part-B? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Model-E [69]", "value": "96.67"} |
| 3 | HB_syn_000267 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Eff under Part-B. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 8, "col": 4, "value": "96.67"} |
| 4 | HB_syn_000268 | Local Relations | Marker-Anchor Neighbor | Find the left neighbor of the highlighted in light purple (hex #E593FC) with value 65.90 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "32.13" |
| 5 | HB_syn_000269 | Consistency & Missingness | Column Missingness | In Rate↑ under Cfg-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Method-X [94]"] |
| 6 | HB_syn_000270 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number at (r=1,c=4) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "light purple", "hex": "#E593FC"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000271 | Aggregation & Comparison | Pairwise Comparison | For System-D [69], compare MetD and MetB. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "MetB", "value": "91.19"} |
| 2 | HB_syn_000272 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=3,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "33.33", "up": "61.27", "down": "41.36", "left": "60.36", "right": "10.93"} |
| 3 | HB_syn_000273 | Markup Grounding | Marked Value Listing | Which values are highlighted in pink (#FFB7D6)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["68.51", "14.15", "57.86", "35.55", "7.82"] |
| 4 | HB_syn_000274 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=19,c=1). Return all values with the same highlight color (pink, #FFB7D6). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["68.51", "14.15", "57.86", "35.55", "7.82"] |
| 5 | HB_syn_000275 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-B [30], MetB) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000276 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Baseline-C, what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "74.99" |
| 2 | HB_syn_000277 | Aggregation & Comparison | Extremum Reasoning | In row System-E [50], what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "14.89" |
| 3 | HB_syn_000278 | Local Relations | Neighbor Value Tracing | At the cell with value 43.25, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "43.25", "up": "62.33", "down": "—", "left": "43.63", "right": "81.27"} |
| 4 | HB_syn_000279 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 66.43, list all numbers highlighted with the same color (mid_teal, hex #83EDC4). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["66.43", "71.20", "69.79", "8.29", "97.30", "79.93"] |
| 5 | HB_syn_000280 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Qual↑. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["66.43"] |
| 6 | HB_syn_000281 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-F, S-D↑) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 7 | HB_syn_000282 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_teal(#D3FAEE), mid_teal(#83EDC4), dark_teal(#22D28E). Is the number with value 69.79 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "mid_teal", "hex": "#83EDC4"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000283 | Aggregation & Comparison | Pairwise Comparison | On metric Idx under Grp-B, which is higher: Approach-A or Approach-C? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-A", "value": "28.42"} |
| 2 | HB_syn_000284 | Local Relations | Neighbor Value Tracing | At the cell with value 55.57, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "55.57", "up": "5.33", "down": "4.78", "left": "71.16", "right": "8.80"} |
| 3 | HB_syn_000285 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "27" |
| 4 | HB_syn_000286 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "9" |
| 5 | HB_syn_000287 | Consistency & Missingness | Column Missingness | List all missing entries in column S-D↑ under Sect-C (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-B [46]"] |
| 6 | HB_syn_000288 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those < 70.73 as a JSON array of strings. | ["7.81", "4.93", "65.52", "69.40", "0.44", "8.80", "15.53", "4.59", "12.38", "65.29", "2.51", "17.10", "69.59"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000289 | Constrained Retrieval | Cell Retrieval | Give the number at (Baseline-F, S-C under Sect-B). Output only the number as a string. Example: "12.34". | "58.94" |
| 2 | HB_syn_000290 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Val under Cfg-C. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-E [37]", "value": "98.94"} |
| 3 | HB_syn_000291 | Aggregation & Comparison | Rank Selection | In Rank under Sect-B, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-E", "value": "97.43"}, {"item": "Approach-E", "value": "93.70"}, {"item": "Variant-C", "value": "92.26"}] |
| 4 | HB_syn_000292 | Local Relations | Marker-Anchor Neighbor | Find the up neighbor of the boldfaced with value 97.43 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "84.94" |
| 5 | HB_syn_000293 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["87.22", "84.48", "67.30", "94.68", "78.64", "93.70", "80.94", "79.57", "+1.93 ±0.64", "15.22"] |
| 6 | HB_syn_000294 | Consistency & Missingness | Column Missingness | Return the items whose Val under Cfg-C is missing (N/A or —). Output JSON array of strings. | ["Baseline-F", "System-E"] |
| 7 | HB_syn_000295 | Markup Grounding | Highlight Presence | Check whether value 66.08 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "blue", "hex": "#6FA0FF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000296 | Aggregation & Comparison | Extremum Reasoning | In column Val, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.07" |
| 2 | HB_syn_000297 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=9,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "36.77", "up": "2.73", "down": "79.78", "left": "57.34", "right": "57.57"} |
| 3 | HB_syn_000298 | Markup Grounding | Marked Column Listing | In MetC↑ under Blk-B, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["87.08"] |
| 4 | HB_syn_000299 | Consistency & Missingness | Cell Missingness | Check whether Model-D at Val under Cfg-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000300 | Constrained Retrieval | Value Filtering | Filter Rate↑ under Cfg-A by <= 43.19. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-B [37]", "value": "43.19"}, {"item": "Approach-A", "value": "37.78"}, {"item": "Model-Y [92]", "value": "29.79"}, {"item": "Variant-E", "value": "10.63"}, {"item": "Baseline-D", "value": "12.98"}, {"item": "Model-D", "value": "42.12"}, {"item": "Variant-C", "value": "18.65"}, {"item": "Method-A", "value": "42.01"}] |
| 6 | HB_syn_000301 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_red(#FFD6D6), mid_red(#F86D7B), dark_red(#FC5255). Is the number at (r=5,c=1) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000302 | Aggregation & Comparison | Extremum Reasoning | Find the highest metric value for Variant-D [72]. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "96.03" |
| 2 | HB_syn_000303 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Approach-Y [20]. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "10.85" |
| 3 | HB_syn_000304 | Local Relations | Marker-Anchor Neighbor | Find the up neighbor of the highlighted in orange (hex #FF8C3A) with value 42.12 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "14.08" |
| 4 | HB_syn_000305 | Markup Grounding | Marked Value Listing | Find all red-highlighted values (hex #FF5A5C) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["16.26", "72.45", "95.02", "65.20", "9.14"] |
| 5 | HB_syn_000306 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 16.26. Return all values sharing its highlight color (red/#FF5A5C). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["16.26", "72.45", "95.02", "65.20", "9.14"] |
| 6 | HB_syn_000307 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "7" |
| 7 | HB_syn_000308 | Constrained Retrieval | Value Filtering | In column Rank↓, list all row labels in the leftmost row-label column with value >= 65.58. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-Z", "value": "66.27"}, {"item": "Approach-C", "value": "99.77"}, {"item": "Variant-E", "value": "65.58"}, {"item": "Variant-C", "value": "86.87"}, {"item": "Model-Z [10]", "value": "78.98"}, {"item": "Variant-Z", "value": "69.57"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000309 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in MetC. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "3.24" |
| 2 | HB_syn_000310 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 7, column 4, list all numbers highlighted with the same color (dark_cyan_blue, hex #55BFFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["99.30", "0.43", "0.10", "1.32", "58.17", "6.34", "5.48"] |
| 3 | HB_syn_000311 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["19.20", "54.45", "9.82", "10.35", "95.04", "0.81", "5.15"] |
| 4 | HB_syn_000312 | Consistency & Missingness | Column Missingness | List all missing entries in column MetC↑ under Grp-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Approach-E [48]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000313 | Aggregation & Comparison | Rank Selection | In Eff under Grp-B, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-B", "value": "94.89"}, {"item": "System-Z", "value": "86.82"}, {"item": "Method-A [83]", "value": "48.07"}] |
| 2 | HB_syn_000314 | Local Relations | Neighbor Indexing | At (r=3,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "38.68", "up": "94.89", "down": "14.51", "left": "75.47", "right": "62.31"} |
| 3 | HB_syn_000315 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "9" |
| 4 | HB_syn_000316 | Consistency & Missingness | Cell Missingness | Is the cell (System-Z, MetE under Part-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000317 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those <= 74.16 as a JSON array of strings. | ["13.59", "74.16", "62.31", "67.84", "10.06", "48.07", "68.86", "15.14", "15.82", "0.89", "71.86", "66.37", "66.18", "0.35"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000318 | Constrained Retrieval | Cell Retrieval | For item System-Y, what is the value in column Err under Slot-C? Output only the number as a string. Example: "12.34". | "1.90" |
| 2 | HB_syn_000319 | Aggregation & Comparison | Pairwise Comparison | Within row Model-B [40], choose the larger between Rate under Set-B and MetB under Set-B, output {item,value}. Exclude N/A and —. | {"item": "MetB under Set-B", "value": "77.56"} |
| 3 | HB_syn_000320 | Local Relations | Neighbor Value Tracing | Locate value 88.39 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "88.39", "up": "68.65", "down": "3.14", "left": "28.43", "right": "40.43"} |
| 4 | HB_syn_000321 | Markup Grounding | Marked Column Listing | In Err under Slot-C, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["91.53"] |
| 5 | HB_syn_000322 | Consistency & Missingness | Cell Missingness | Check whether Model-B [40] at Score under Slot-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_syn_000323 | Markup Grounding | Highlight Presence | Check whether (r=5,c=15) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "mid_orange", "hex": "#F1B17D"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000324 | Aggregation & Comparison | Rank Selection | In S-D under Slot-C, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-B [96]", "value": "92.10"}, {"item": "Approach-Y", "value": "62.41"}, {"item": "Approach-X", "value": "60.90"}] |
| 2 | HB_syn_000325 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in S-D under Slot-C and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Approach-Y", "value": "62.41"} |
| 3 | HB_syn_000326 | Local Relations | Neighbor Value Tracing | Locate value 61.70 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "61.70", "up": "34.38", "down": "51.40", "left": "27.24", "right": "46.39"} |
| 4 | HB_syn_000327 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 35.61, list all numbers highlighted with the same color (blue, hex #6FA0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["35.61", "11.07", "+1.33 ±0.09"] |
| 5 | HB_syn_000328 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "9" |
| 6 | HB_syn_000329 | Consistency & Missingness | Column Missingness | In S-C↑ under Grp-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Approach-B [96]", "Baseline-E"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000330 | Aggregation & Comparison | Extremum Reasoning | Find the highest metric value for System-B [45]. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "89.35" |
| 2 | HB_syn_000331 | Local Relations | Neighbor Indexing | At (r=4,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "86.98", "up": "49.55", "down": "85.20", "left": "29.71", "right": "13.07"} |
| 3 | HB_syn_000332 | Consistency & Missingness | Cell Missingness | Is the cell (System-F, S-D↑) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_syn_000333 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are >= 84.44. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["84.44", "95.93", "94.70"] |
| 5 | HB_syn_000334 | Markup Grounding | Highlight Presence | Check whether value 22.89 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "red", "hex": "#FF5A5C"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000335 | Aggregation & Comparison | Pairwise Comparison | For Baseline-Z [96], compare S-B under Zone-B and MetA under Zone-A. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "MetA under Zone-A", "value": "28.27"} |
| 2 | HB_syn_000336 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=6,c=10). Return all values with the same highlight color (orange, #FF8C3A). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["36.50", "36.94", "75.73", "28.27", "94.59"] |
| 3 | HB_syn_000337 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "15" |
| 4 | HB_syn_000338 | Consistency & Missingness | Column Missingness | List all missing entries in column Val↑ under Part-C (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-Y"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000339 | Constrained Retrieval | Cell Retrieval | What is the MetE↑ under Blk-B value for Approach-D [33]? Output only the number as a string. Example: "12.34". | "1.42" |
| 2 | HB_syn_000340 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in MetC. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "93.41" |
| 3 | HB_syn_000341 | Aggregation & Comparison | Extremum Reasoning | Return the min of column MetC. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "2.69" |
| 4 | HB_syn_000342 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the highlighted in mid_cyan_blue (hex #95D9FF) with value 5.71 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "93.29" |
| 5 | HB_syn_000343 | Markup Grounding | Marked Value Listing | List numbers highlighted in dark_cyan_blue (hex #55BFFF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["0.30", "93.41", "14.95", "91.70", "4.86", "94.79", "95.53", "99.35"] |
| 6 | HB_syn_000344 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["91.07", "92.57", "37.24", "93.12", "89.19", "5.71", "86.17", "3.84"] |
| 7 | HB_syn_000345 | Consistency & Missingness | Column Missingness | List all missing entries in column Val↑ under Blk-B (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Variant-Y [8]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000346 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in S-C under Slot-D and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Baseline-C", "value": "81.36"} |
| 2 | HB_syn_000347 | Markup Grounding | Marked Column Listing | Which entries in column MetA under Part-B are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["79.93"] |
| 3 | HB_syn_000348 | Consistency & Missingness | Cell Missingness | Check whether Baseline-C at S-D↑ under Slot-D is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_syn_000349 | Constrained Retrieval | Value Filtering | In column MetA under Part-B, list all row labels in the leftmost row-label column with value < 74.82. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-Y [25]", "value": "64.49"}, {"item": "Baseline-C", "value": "74.18"}, {"item": "Method-C", "value": "48.68"}, {"item": "Model-F", "value": "34.18"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000350 | Aggregation & Comparison | Argmax Reasoning | In column Idx under Sect-A, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-C", "value": "87.90"} |
| 2 | HB_syn_000351 | Aggregation & Comparison | Argmax Reasoning | In Idx under Sect-A, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 6, "col": 2, "value": "87.90"} |
| 3 | HB_syn_000352 | Aggregation & Comparison | Pairwise Comparison | Between Model-D [26] and Variant-C for S-D under Sect-A, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Variant-C", "value": "27.85"} |
| 4 | HB_syn_000353 | Local Relations | Marker-Anchor Neighbor | Find the down neighbor of the highlighted in orange (hex #FF8C3A) with value 76.39 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "22.36" |
| 5 | HB_syn_000354 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 10, column 2, list all numbers highlighted with the same color (orange, hex #FF8C3A). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["12.24", "27.85", "36.68", "76.39", "17.60", "79.91", "79.22"] |
| 6 | HB_syn_000355 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "7" |
| 7 | HB_syn_000356 | Consistency & Missingness | Column Missingness | List all missing entries in column MetC↑ under Cfg-B (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Baseline-X", "Model-D [26]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000357 | Aggregation & Comparison | Extremum Reasoning | In column Idx, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "6.49" |
| 2 | HB_syn_000358 | Local Relations | Neighbor Value Tracing | At the cell with value 63.06, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "63.06", "up": "68.69", "down": "83.68", "left": "44.88", "right": "51.02"} |
| 3 | HB_syn_000359 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 99.50. Return all values sharing its highlight color (light_blue/#E6F2FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["86.43", "63.06", "94.25", "99.50", "84.22", "3.91", "86.83", "35.11"] |
| 4 | HB_syn_000360 | Consistency & Missingness | Cell Missingness | Check whether System-E at Rate↓ under Blk-D is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000361 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). Is the number at (r=10,c=7) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000362 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Baseline-Y, what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "51.53" |
| 2 | HB_syn_000363 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Baseline-Y, what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "36.17" |
| 3 | HB_syn_000364 | Local Relations | Neighbor Indexing | At (r=11,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "55.19", "up": "43.22", "down": "31.74", "left": "N/A", "right": "48.34"} |
| 4 | HB_syn_000365 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-Y, MetB↓) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000366 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those < 14.66 as a JSON array of strings. | ["9.06", "13.93", "7.64", "5.93"] |
| 6 | HB_syn_000367 | Markup Grounding | Highlight Presence | Check whether value 31.74 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000368 | Constrained Retrieval | Cell Retrieval | For item Method-X, what is the value in column Cost under Slot-B? Output only the number as a string. Example: "12.34". | "15.26" |
| 2 | HB_syn_000369 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Cost under Slot-B. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-E", "value": "88.53"} |
| 3 | HB_syn_000370 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Cost under Slot-B. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 6, "col": 5, "value": "88.53"} |
| 4 | HB_syn_000371 | Local Relations | Marker-Anchor Neighbor | Find the left neighbor of the boldfaced with value 10.28 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "15.98" |
| 5 | HB_syn_000372 | Markup Grounding | Marked Value Listing | Find all red-highlighted values (hex #FF5A5C) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["47.67", "10.28", "88.53", "61.87"] |
| 6 | HB_syn_000373 | Consistency & Missingness | Column Missingness | In MetC↑ under Slot-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Method-Y", "Baseline-C [29]", "Baseline-D [36]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000374 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column MetE? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "76.86" |
| 2 | HB_syn_000375 | Local Relations | Neighbor Indexing | At (r=2,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "45.36", "up": "N/A", "down": "97.74", "left": "9.76", "right": "56.41"} |
| 3 | HB_syn_000376 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["82.91", "8.23", "75.90", "82.04", "91.08", "73.72", "83.60", "84.02", "94.77", "92.59"] |
| 4 | HB_syn_000377 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-Z, Qual↑ under Grp-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000378 | Constrained Retrieval | Value Filtering | Filter Score↑ under Grp-C by < 59.36. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-Y", "value": "16.04"}, {"item": "Variant-E [17]", "value": "24.46"}, {"item": "Baseline-B", "value": "29.71"}, {"item": "System-X", "value": "39.85"}, {"item": "System-Y", "value": "8.50"}, {"item": "Variant-X", "value": "13.16"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000379 | Aggregation & Comparison | Pairwise Comparison | Within row Variant-F, choose the larger between MetE under Grp-B and MetD under Grp-B, output {item,value}. Exclude N/A and —. | {"item": "MetD under Grp-B", "value": "81.50"} |
| 2 | HB_syn_000380 | Local Relations | Neighbor Value Tracing | Locate value 60.57 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "60.57", "up": "60.92", "down": "79.34", "left": "2.46", "right": "59.56"} |
| 3 | HB_syn_000381 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 81.98. Return all values sharing its highlight color (mid_yellow/#F7DA67). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["79.60", "87.46", "70.64", "96.68", "66.57", "92.91", "81.98", "84.26", "10.31"] |
| 4 | HB_syn_000382 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "27" |
| 5 | HB_syn_000383 | Consistency & Missingness | Column Missingness | In S-D under Sect-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-A"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000384 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Variant-E. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "8.98" |
| 2 | HB_syn_000385 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the highlighted in mid_blue (hex #8DB3FF) with value 87.59 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "95.64" |
| 3 | HB_syn_000386 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=11,c=2). Return all values with the same highlight color (light_blue, #CFE0FF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["80.81", "13.55"] |
| 4 | HB_syn_000387 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-Z [97], Stat↓) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000388 | Markup Grounding | Highlight Presence | Check whether (r=3,c=2) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_blue(#CFE0FF), mid_blue(#8DB3FF), dark_blue(#4F8AFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "mid_blue", "hex": "#8DB3FF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000389 | Constrained Retrieval | Cell Retrieval | For item Method-A, what is the value in column MetA↑ under Cfg-A? Output only the number as a string. Example: "12.34". | "44.75" |
| 2 | HB_syn_000390 | Aggregation & Comparison | Extremum Reasoning | Find the highest metric value for Variant-A. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "86.57" |
| 3 | HB_syn_000391 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (right) to the highlighted in light_orange (hex #FFE6C7) cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "64.48" |
| 4 | HB_syn_000392 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Stat↓ under Zone-B. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["40.43"] |
| 5 | HB_syn_000393 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "2" |
| 6 | HB_syn_000394 | Consistency & Missingness | Column Missingness | List all missing entries in column MetA↑ under Cfg-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Model-B [1]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000395 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Method-E. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "18.77" |
| 2 | HB_syn_000396 | Local Relations | Neighbor Value Tracing | At the cell with value 12.52, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "12.52", "up": "95.13", "down": "98.83", "left": "75.46", "right": "31.38"} |
| 3 | HB_syn_000397 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-E, Val↑ under Grp-B) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_syn_000398 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are <= 52.37. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["12.86", "21.30", "46.57", "12.81", "37.37", "52.37", "36.08"] |
| 5 | HB_syn_000399 | Markup Grounding | Highlight Presence | Check whether value 0.11 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000400 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in S-A in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Model-D", "value": "96.16"}, {"item": "Baseline-Z", "value": "94.02"}, {"item": "System-X [94]", "value": "84.27"}] |
| 2 | HB_syn_000401 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=7,c=7), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "22.76", "up": "63.56", "down": "33.09", "left": "36.51", "right": "—"} |
| 3 | HB_syn_000402 | Markup Grounding | Marked Value Listing | List numbers highlighted in mid_purple (hex #B681F3). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["94.02", "97.61", "91.18", "97.94", "90.25", "3.21", "99.32", "9.24"] |
| 4 | HB_syn_000403 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["94.02", "97.61", "91.18", "97.94", "90.25", "3.21", "99.32", "9.24"] |
| 5 | HB_syn_000404 | Consistency & Missingness | Column Missingness | In Err↑, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-C", "Approach-A", "Method-B", "Model-Y"] |
| 6 | HB_syn_000405 | Constrained Retrieval | Value Filtering | In column S-CB↓, list all row labels in the leftmost row-label column with value < 48.54. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-C", "value": "34.56"}, {"item": "Variant-Y [28]", "value": "28.63"}, {"item": "Baseline-Y [4]", "value": "28.79"}, {"item": "Variant-E [76]", "value": "37.63"}, {"item": "Baseline-A", "value": "28.37"}, {"item": "System-X [94]", "value": "37.14"}, {"item": "Model-D", "value": "6.60"}, {"item": "System-D [39]", "value": "3.21"}, {"item": "Method-X", "value": "35.23"}, {"item": "Baseline-D [44]", "value": "0.13"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000406 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Eff? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "13.42" |
| 2 | HB_syn_000407 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=5,c=5), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "39.70", "up": "87.65", "down": "13.66", "left": "41.28", "right": "69.09"} |
| 3 | HB_syn_000408 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 6, column 9, list all numbers highlighted with the same color (mid_blue, hex #8DB3FF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["61.86", "10.20", "14.03", "79.84", "87.65", "67.19", "24.78", "46.39", "71.91", "21.87", "51.28", "85.44"] |
| 4 | HB_syn_000409 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "36" |
| 5 | HB_syn_000410 | Consistency & Missingness | Column Missingness | Return the items whose F1↑ under Cfg-D is missing (N/A or —). Output JSON array of strings. | ["Model-D", "Baseline-C [97]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000411 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in S-D and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Variant-A", "value": "46.23"} |
| 2 | HB_syn_000412 | Local Relations | Neighbor Value Tracing | At the cell with value 46.63, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "46.63", "up": "91.48", "down": "64.54", "left": "48.91", "right": "95.84"} |
| 3 | HB_syn_000413 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 89.60. Return all values sharing its highlight color (orange/#F9B163). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["57.01", "89.60"] |
| 4 | HB_syn_000414 | Markup Grounding | Marked Column Listing | Which entries in column Err↓ are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["10.19"] |
| 5 | HB_syn_000415 | Consistency & Missingness | Cell Missingness | Check whether Model-Y [13] at Err↓ is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000416 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Approach-C [89], what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "91.33" |
| 2 | HB_syn_000417 | Aggregation & Comparison | Extremum Reasoning | For item Approach-D, what is the minimum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "13.31" |
| 3 | HB_syn_000418 | Local Relations | Neighbor Value Tracing | At the cell with value 32.92, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "32.92", "up": "70.00", "down": "26.73", "left": "86.67", "right": "89.29"} |
| 4 | HB_syn_000419 | Markup Grounding | Marked Value Listing | Return all numbers highlighted in light_blue (hex #E6F2FF) as a JSON array. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["53.86", "5.18", "30.08", "84.33", "89.66", "41.07"] |
| 5 | HB_syn_000420 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are >= 37.92. Output JSON array of strings. | ["86.35", "96.11", "79.87", "53.86", "84.33", "89.66", "37.92", "41.07"] |
| 6 | HB_syn_000421 | Markup Grounding | Highlight Presence | Check whether value 1.21 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "blue", "hex": "#6DA1FD"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000422 | Aggregation & Comparison | Pairwise Comparison | Compare Approach-D [92] vs Model-E on MetD under Zone-B. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-D [92]", "value": "74.90"} |
| 2 | HB_syn_000423 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in light_orange (hex #FCE0C7) with value 71.46 number and return the number immediately to its left. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "82.68" |
| 3 | HB_syn_000424 | Consistency & Missingness | Column Missingness | In MetB under Zone-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Variant-B"] |
| 4 | HB_syn_000425 | Constrained Retrieval | Value Filtering | In column MetD under Zone-B, list all row labels in the leftmost row-label column with value >= 43.00. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-B [33]", "value": "70.50"}, {"item": "Baseline-F [69]", "value": "43.00"}, {"item": "Model-E", "value": "71.46"}, {"item": "Approach-D [92]", "value": "74.90"}, {"item": "Variant-B", "value": "68.98"}, {"item": "System-F [4]", "value": "77.03"}] |
| 5 | HB_syn_000426 | Markup Grounding | Highlight Presence | Check whether (r=4,c=4) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000427 | Constrained Retrieval | Cell Retrieval | In the MetB under Blk-A column, read the entry for Model-B [4]. Output only the number as a string. Example: "12.34". | "88.33" |
| 2 | HB_syn_000428 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest MetD under Zone-B? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-C", "value": "96.97"} |
| 3 | HB_syn_000429 | Local Relations | Neighbor Indexing | At (r=2,c=3) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "6.03", "up": "68.62", "down": "N/A", "left": "98.32", "right": "85.95"} |
| 4 | HB_syn_000430 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["94.67", "98.32", "21.76", "99.85", "96.97", "89.16", "89.82", "81.78"] |
| 5 | HB_syn_000431 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "24" |
| 6 | HB_syn_000432 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "8" |
| 7 | HB_syn_000433 | Consistency & Missingness | Cell Missingness | Check whether Variant-X at MetD under Zone-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000434 | Aggregation & Comparison | Extremum Reasoning | In row System-D [56], what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "95.35" |
| 2 | HB_syn_000435 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=9,c=3), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "46.81", "up": "31.79", "down": "80.79", "left": "75.11", "right": "—"} |
| 3 | HB_syn_000436 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=11,c=2) as an anchor, list all cells sharing its highlight color (purple/#B07CFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["89.47", "23.30", "46.83", "6.22", "16.75"] |
| 4 | HB_syn_000437 | Markup Grounding | Marked Column Listing | In Score↓ under Cfg-A, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["6.22"] |
| 5 | HB_syn_000438 | Consistency & Missingness | Cell Missingness | Check whether Approach-C at MetE↓ under Cfg-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000439 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Variant-A. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "11.80" |
| 2 | HB_syn_000440 | Local Relations | Neighbor Value Tracing | At the cell with value 22.52, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "22.52", "up": "60.93", "down": "56.24", "left": "—", "right": "29.96"} |
| 3 | HB_syn_000441 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 24.23, list all numbers highlighted with the same color (light_blue, hex #CFE0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["95.45", "81.48", "25.55", "24.23"] |
| 4 | HB_syn_000442 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "12" |
| 5 | HB_syn_000443 | Consistency & Missingness | Column Missingness | Return the items whose Val↑ is missing (N/A or —). Output JSON array of strings. | ["Variant-A", "Model-F", "Variant-C", "System-A", "System-X"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000444 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Approach-F. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "84.21" |
| 2 | HB_syn_000445 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=7,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "43.47", "up": "99.55", "down": "62.30", "left": "72.73", "right": "1.96"} |
| 3 | HB_syn_000446 | Consistency & Missingness | Cell Missingness | Check whether System-Y at MetA↑ is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_syn_000447 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are < 80.24. Output JSON array of strings. | ["1.96", "62.30", "8.90", "2.29"] |
| 5 | HB_syn_000448 | Markup Grounding | Highlight Presence | Check whether (r=3,c=1) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_yellow(#FEF4B8), mid_yellow(#F7DA67), dark_yellow(#EFBC23). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "light_yellow", "hex": "#FEF4B8"} |
| 6 | HB_syn_000449 | Markup Grounding | Highlight Presence | Check whether value 99.55 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_yellow(#FEF4B8), mid_yellow(#F7DA67), dark_yellow(#EFBC23). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "dark_yellow", "hex": "#EFBC23"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000450 | Aggregation & Comparison | Extremum Reasoning | Find the highest metric value for Approach-B [11]. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "49.74" |
| 2 | HB_syn_000451 | Local Relations | Marker-Anchor Neighbor | Locate the underlined with value 5.87 number and return the number immediately below it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "81.98" |
| 3 | HB_syn_000452 | Markup Grounding | Marked Value Listing | Which values are highlighted in mid_orange (#F1B17D)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["5.87", "0.77", "4.96", "96.45", "2.43", "7.02"] |
| 4 | HB_syn_000453 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["5.87", "0.77", "4.96", "96.45", "2.43", "7.02"] |
| 5 | HB_syn_000454 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "6" |
| 6 | HB_syn_000455 | Consistency & Missingness | Column Missingness | List all missing entries in column MetD↑ (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Variant-A [82]"] |
| 7 | HB_syn_000456 | Constrained Retrieval | Value Filtering | Filter S-D↓ by < 44.52. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-F", "value": "19.58"}, {"item": "Variant-E", "value": "44.16"}, {"item": "Approach-D", "value": "4.96"}, {"item": "Approach-B [11]", "value": "8.01"}, {"item": "Model-F", "value": "4.63"}, {"item": "Method-B", "value": "29.66"}, {"item": "Model-E", "value": "41.53"}, {"item": "System-A", "value": "30.66"}, {"item": "Approach-Z", "value": "9.79"}, {"item": "Baseline-X", "value": "13.11"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000457 | Constrained Retrieval | Cell Retrieval | For item Approach-X [91], what is the value in column MetB under Cfg-B? Output only the number as a string. Example: "12.34". | "43.80" |
| 2 | HB_syn_000458 | Aggregation & Comparison | Pairwise Comparison | For Model-E, compare Cost under Sect-A and Rank under Cfg-B. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "Cost under Sect-A", "value": "63.84"} |
| 3 | HB_syn_000459 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (up) to the boldfaced with value 89.86 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "54.67" |
| 4 | HB_syn_000460 | Consistency & Missingness | Cell Missingness | Check whether Method-Z at Rank under Cfg-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000461 | Markup Grounding | Highlight Presence | Check whether value 56.97 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_cyan_blue(#C8ECFF), mid_cyan_blue(#95D9FF), dark_cyan_blue(#55BFFF). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "dark_cyan_blue", "hex": "#55BFFF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000462 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column MetE. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 13, "col": 1, "value": "93.49"} |
| 2 | HB_syn_000463 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column MetE? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-A", "value": "89.56"} |
| 3 | HB_syn_000464 | Local Relations | Neighbor Value Tracing | Locate value 4.15 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "4.15", "up": "74.75", "down": "9.05", "left": "72.84", "right": "54.87"} |
| 4 | HB_syn_000465 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 86.67. Return all values sharing its highlight color (dark_teal/#22D28E). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["86.67", "93.49", "97.07", "96.60"] |
| 5 | HB_syn_000466 | Consistency & Missingness | Column Missingness | List all missing entries in column MetE (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-A", "Approach-Z [93]", "Approach-X"] |
| 6 | HB_syn_000467 | Markup Grounding | Highlight Presence | Check whether (r=13,c=1) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_teal(#D3FAEE), mid_teal(#83EDC4), dark_teal(#22D28E). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "dark_teal", "hex": "#22D28E"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000468 | Aggregation & Comparison | Extremum Reasoning | For item Approach-E [33], what is the maximum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "84.07" |
| 2 | HB_syn_000469 | Local Relations | Neighbor Value Tracing | At the cell with value 24.67, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "24.67", "up": "64.77", "down": "33.83", "left": "N/A", "right": "36.99"} |
| 3 | HB_syn_000470 | Markup Grounding | Marked Value Listing | List numbers highlighted in dark_red (hex #FC5255). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["1.25", "99.08", "6.77", "95.77", "2.11"] |
| 4 | HB_syn_000471 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=18,c=3) as an anchor, list all cells sharing its highlight color (dark_red/#FC5255). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["1.25", "99.08", "6.77", "95.77", "2.11"] |
| 5 | HB_syn_000472 | Consistency & Missingness | Cell Missingness | Check whether Variant-E at S-B↑ under Set-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000473 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column MetE? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "88.21" |
| 2 | HB_syn_000474 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Eff. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-A [82]", "value": "94.62"} |
| 3 | HB_syn_000475 | Aggregation & Comparison | Rank Selection | In Eff, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-A [82]", "value": "94.62"}, {"item": "Approach-C", "value": "94.49"}, {"item": "System-A", "value": "58.52"}] |
| 4 | HB_syn_000476 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Eff and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Approach-C", "value": "94.49"} |
| 5 | HB_syn_000477 | Aggregation & Comparison | Pairwise Comparison | Between Method-X and Method-Y [98] for MetE, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Method-Y [98]", "value": "80.16"} |
| 6 | HB_syn_000478 | Local Relations | Marker-Anchor Neighbor | Locate the underlined with value 94.49 number and return the number immediately to its left. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "36.63" |
| 7 | HB_syn_000479 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "4" |
| 8 | HB_syn_000480 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are < 54.27. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["10.52", "17.82"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000481 | Aggregation & Comparison | Extremum Reasoning | In column S-D, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "3.34" |
| 2 | HB_syn_000482 | Local Relations | Neighbor Indexing | At (r=18,c=5) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "98.02", "up": "9.78", "down": "77.99", "left": "80.80", "right": "71.42"} |
| 3 | HB_syn_000483 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "18" |
| 4 | HB_syn_000484 | Consistency & Missingness | Column Missingness | Return the items whose MetE↑ under Sect-C is missing (N/A or —). Output JSON array of strings. | ["System-Y", "Model-B", "Baseline-Z", "Baseline-F"] |
| 5 | HB_syn_000485 | Constrained Retrieval | Value Filtering | In column MetE↑ under Sect-C, list all row labels in the leftmost row-label column with value < 55.47. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-Y", "value": "32.53"}, {"item": "Method-C", "value": "35.78"}, {"item": "Model-F", "value": "17.65"}, {"item": "Method-B", "value": "16.67"}, {"item": "Method-E", "value": "4.76"}, {"item": "Baseline-E", "value": "36.83"}, {"item": "Approach-C", "value": "47.27"}, {"item": "Approach-X [9]", "value": "39.91"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000486 | Constrained Retrieval | Cell Retrieval | In the Idx↑ under Cfg-C column, read the entry for Model-A. Output only the number as a string. Example: "12.34". | "20.70" |
| 2 | HB_syn_000487 | Aggregation & Comparison | Extremum Reasoning | Return the max of column S-A. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "83.48" |
| 3 | HB_syn_000488 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Qual under Part-D. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["94.56"] |
| 4 | HB_syn_000489 | Consistency & Missingness | Column Missingness | Return the items whose MetD↑ under Zone-B is missing (N/A or —). Output JSON array of strings. | ["Approach-F [69]", "System-B"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000490 | Aggregation & Comparison | Argmax Reasoning | In S-D under Zone-B, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 9, "col": 3, "value": "89.11"} |
| 2 | HB_syn_000491 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 72.28, list all numbers highlighted with the same color (mid_yellow, hex #F7DA67). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["85.57", "72.28", "69.44", "17.82", "27.61", "96.79"] |
| 3 | HB_syn_000492 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["99.65", "3.13", "87.45", "13.71", "89.11", "76.31"] |
| 4 | HB_syn_000493 | Consistency & Missingness | Cell Missingness | Check whether Model-Y [34] at Eff under Grp-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000494 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Eff under Slot-A in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Model-Y", "value": "94.78"}, {"item": "Approach-B", "value": "81.73"}, {"item": "Approach-X [88]", "value": "61.14"}] |
| 2 | HB_syn_000495 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Eff under Slot-A and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Approach-B", "value": "81.73"} |
| 3 | HB_syn_000496 | Local Relations | Neighbor Indexing | At (r=7,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "96.07", "up": "2.38", "down": "63.04", "left": "59.75", "right": "—"} |
| 4 | HB_syn_000497 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=1,c=1) as an anchor, list all cells sharing its highlight color (pink/#FFB7D6). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["67.55", "38.61", "+0.74 ±0.49", "73.58", "96.53"] |
| 5 | HB_syn_000498 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "8" |
| 6 | HB_syn_000499 | Consistency & Missingness | Column Missingness | List all missing entries in column S-C↑ under Grp-C (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Variant-C", "Variant-B", "Baseline-C [16]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000500 | Aggregation & Comparison | Pairwise Comparison | Between Variant-B and Baseline-Y for Qual under Set-A, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Baseline-Y", "value": "84.38"} |
| 2 | HB_syn_000501 | Aggregation & Comparison | Pairwise Comparison | Within row Model-A [77], choose the larger between Idx under Blk-B and Qual under Set-A, output {item,value}. Exclude N/A and —. | {"item": "Idx under Blk-B", "value": "34.35"} |
| 3 | HB_syn_000502 | Local Relations | Neighbor Value Tracing | Locate value 79.65 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "79.65", "up": "25.19", "down": "19.91", "left": "83.43", "right": "23.01"} |
| 4 | HB_syn_000503 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-X, Idx under Blk-B) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000504 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those > 54.47 as a JSON array of strings. | ["92.82", "87.33", "76.16", "84.38", "65.77", "83.43", "57.90", "66.29"] |
| 6 | HB_syn_000505 | Markup Grounding | Highlight Presence | Check whether (r=2,c=5) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_teal(#D3FAEE), mid_teal(#83EDC4), dark_teal(#22D28E). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000506 | Constrained Retrieval | Cell Retrieval | For item Variant-Y, what is the value in column MetC under Grp-D? Output only the number as a string. Example: "12.34". | "12.51" |
| 2 | HB_syn_000507 | Aggregation & Comparison | Extremum Reasoning | Return the max of column Rank. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "97.44" |
| 3 | HB_syn_000508 | Local Relations | Marker-Anchor Neighbor | Locate the underlined with value 93.07 number and return the number immediately below it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "53.39" |
| 4 | HB_syn_000509 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-C [36], MetC under Grp-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000510 | Markup Grounding | Highlight Presence | Check whether value 77.02 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "dark_orange", "hex": "#E47B30"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000511 | Aggregation & Comparison | Extremum Reasoning | In column Eff, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "9.69" |
| 2 | HB_syn_000512 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (up) to the highlighted in light_orange (hex #FFE6C7) with value 23.66 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "51.15" |
| 3 | HB_syn_000513 | Markup Grounding | Marked Column Listing | Which entries in column Cost↑ are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["96.67"] |
| 4 | HB_syn_000514 | Consistency & Missingness | Column Missingness | Return the items whose Eff is missing (N/A or —). Output JSON array of strings. | ["Approach-B [27]", "Approach-A [30]"] |
| 5 | HB_syn_000515 | Constrained Retrieval | Value Filtering | Filter Err↑ by < 48.41. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Variant-B", "value": "28.56"}, {"item": "Method-C", "value": "26.82"}, {"item": "Variant-D", "value": "5.06"}, {"item": "Method-D", "value": "42.57"}, {"item": "Approach-Z", "value": "45.35"}, {"item": "Approach-A [30]", "value": "28.25"}, {"item": "Variant-Z [25]", "value": "4.37"}, {"item": "Model-C", "value": "0.83"}, {"item": "Baseline-Z", "value": "33.45"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000516 | Aggregation & Comparison | Extremum Reasoning | In row Model-Z, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "0.43" |
| 2 | HB_syn_000517 | Local Relations | Neighbor Value Tracing | Locate value 96.73 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "96.73", "up": "93.13", "down": "26.21", "left": "93.35", "right": "85.46"} |
| 3 | HB_syn_000518 | Markup Grounding | Marked Value Listing | Find all dark_green-highlighted values (hex #72D46A) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["0.43", "6.27", "2.00"] |
| 4 | HB_syn_000519 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 5.65. Return all values sharing its highlight color (light_green/#D7F6D1). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["6.65", "5.65", "11.38"] |
| 5 | HB_syn_000520 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["6.32", "6.36", "3.16"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000521 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column MetA. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 3, "col": 9, "value": "65.19"} |
| 2 | HB_syn_000522 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=2,c=11), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "67.73", "up": "31.46", "down": "36.03", "left": "41.65", "right": "86.04"} |
| 3 | HB_syn_000523 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=6,c=12) as an anchor, list all cells sharing its highlight color (light_orange/#FFE6C7). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["87.66", "68.02"] |
| 4 | HB_syn_000524 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "10" |
| 5 | HB_syn_000525 | Consistency & Missingness | Column Missingness | In Err↑, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-Y", "Model-X [76]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000526 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest Score under Sect-D? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Model-F", "value": "97.65"} |
| 2 | HB_syn_000527 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=12,c=6), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "88.22", "up": "32.24", "down": "83.28", "left": "80.99", "right": "96.89"} |
| 3 | HB_syn_000528 | Markup Grounding | Marked Value Listing | Find all dark_blue-highlighted values (hex #4F8AFF) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["94.16", "11.68", "95.23", "97.65", "93.38", "10.62", "1.47", "96.89", "98.13"] |
| 4 | HB_syn_000529 | Consistency & Missingness | Cell Missingness | Is the cell (Model-Y, S-A↑ under Slot-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000530 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are < 78.16. Output JSON array of strings. | ["11.68", "72.26", "2.21", "11.86", "24.00", "73.94", "20.62", "10.62", "45.05", "1.47", "32.24", "38.81", "36.20"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000531 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Idx under Sect-D in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Approach-F", "value": "99.14"}, {"item": "System-D", "value": "96.59"}, {"item": "Model-C", "value": "92.22"}] |
| 2 | HB_syn_000532 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column Idx under Sect-D? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-D", "value": "96.59"} |
| 3 | HB_syn_000533 | Local Relations | Marker-Anchor Neighbor | Locate the boldfaced with value 3.72 number and return the number immediately to its right. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "27.09" |
| 4 | HB_syn_000534 | Markup Grounding | Marked Column Listing | In ΔDiff↑ under Sect-D, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["+1.56 ±0.59"] |
| 5 | HB_syn_000535 | Consistency & Missingness | Cell Missingness | Check whether Variant-D at S-A↑ under Grp-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000536 | Constrained Retrieval | Value Filtering | In column Idx under Sect-D, list all row labels in the leftmost row-label column with value > 53.54. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-C", "value": "92.22"}, {"item": "Approach-F", "value": "99.14"}, {"item": "Variant-Y", "value": "54.94"}, {"item": "System-D", "value": "96.59"}, {"item": "Method-Y", "value": "71.31"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000537 | Constrained Retrieval | Cell Retrieval | Give the number at (Model-B [7], MetD↑ under Cfg-A). Output only the number as a string. Example: "12.34". | "71.06" |
| 2 | HB_syn_000538 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Model-B [7], what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "71.06" |
| 3 | HB_syn_000539 | Aggregation & Comparison | Extremum Reasoning | Find the highest metric value for Approach-X. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "2.06" |
| 4 | HB_syn_000540 | Local Relations | Marker-Anchor Neighbor | Find the up neighbor of the highlighted in light_orange (hex #FFE6C7) cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "71.06" |
| 5 | HB_syn_000541 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["66.40", "+1.13 ±0.17"] |
| 6 | HB_syn_000542 | Consistency & Missingness | Column Missingness | In MetD↑ under Cfg-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Approach-A"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000543 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in S-C. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "96.26" |
| 2 | HB_syn_000544 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 54.55. Return all values sharing its highlight color (light_blue/#CFE0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["75.98", "51.60", "22.01", "78.53", "30.88", "67.33", "54.55", "65.08", "54.73", "62.81"] |
| 3 | HB_syn_000545 | Consistency & Missingness | Cell Missingness | Check whether Approach-D at S-D↑ under Blk-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_syn_000546 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_blue(#CFE0FF), mid_blue(#8DB3FF), dark_blue(#4F8AFF). Is the number at (r=6,c=6) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "mid_blue", "hex": "#8DB3FF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000547 | Aggregation & Comparison | Argmax Reasoning | In Val under Cfg-A, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 11, "col": 2, "value": "77.49"} |
| 2 | HB_syn_000548 | Aggregation & Comparison | Pairwise Comparison | Compare Variant-B [7] vs Approach-X on Val under Cfg-A. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-X", "value": "71.99"} |
| 3 | HB_syn_000549 | Local Relations | Neighbor Value Tracing | At the cell with value 59.48, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "59.48", "up": "93.03", "down": "0.53", "left": "90.93", "right": "8.01"} |
| 4 | HB_syn_000550 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 18, column 3, list all numbers highlighted with the same color (light_purple, hex #E6D1FF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["10.66", "91.40", "7.22", "94.62", "18.39", "72.58"] |
| 5 | HB_syn_000551 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "18" |
| 6 | HB_syn_000552 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "6" |
| 7 | HB_syn_000553 | Consistency & Missingness | Column Missingness | Return the items whose Val under Cfg-A is missing (N/A or —). Output JSON array of strings. | ["Approach-A"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000554 | Aggregation & Comparison | Rank Selection | List the top-3 items by column MetD, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-Z [24]", "value": "93.53"}, {"item": "Baseline-Z", "value": "87.05"}, {"item": "Model-E [84]", "value": "77.71"}] |
| 2 | HB_syn_000555 | Local Relations | Neighbor Value Tracing | At the cell with value 35.18, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "35.18", "up": "14.05", "down": "27.76", "left": "9.60", "right": "0.55"} |
| 3 | HB_syn_000556 | Consistency & Missingness | Cell Missingness | Check whether Model-X at MetD is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_syn_000557 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those <= 77.71 as a JSON array of strings. | ["6.41", "33.80", "0.55", "77.71"] |
| 5 | HB_syn_000558 | Markup Grounding | Highlight Presence | Check whether value 91.17 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "light_orange", "hex": "#FFE6C7"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000559 | Constrained Retrieval | Cell Retrieval | For item Baseline-X, what is the value in column Cost↓ under Blk-D? Output only the number as a string. Example: "12.34". | "41.42" |
| 2 | HB_syn_000560 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Stat under Blk-D and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Approach-E", "value": "89.77"} |
| 3 | HB_syn_000561 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=2,c=9), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "69.34", "up": "90.08", "down": "46.28", "left": "10.92", "right": "35.14"} |
| 4 | HB_syn_000562 | Markup Grounding | Marked Column Listing | List all underlined numbers in column S-D↑ under Blk-D. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["70.89"] |
| 5 | HB_syn_000563 | Consistency & Missingness | Column Missingness | List all missing entries in column Stat under Blk-D (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-E"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000564 | Aggregation & Comparison | Argmax Reasoning | In column Err under Cfg-A, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-F", "value": "94.87"} |
| 2 | HB_syn_000565 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (right) to the underlined with value 82.57 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "6.52" |
| 3 | HB_syn_000566 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["7.22", "88.13", "83.38", "5.50", "92.15", "6.06", "82.57"] |
| 4 | HB_syn_000567 | Consistency & Missingness | Column Missingness | In Score under Sect-C, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["System-D", "Model-A"] |
| 5 | HB_syn_000568 | Constrained Retrieval | Value Filtering | In column Stat↓ under Blk-B, list all row labels in the leftmost row-label column with value <= 45.88. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-Z", "value": "7.22"}, {"item": "Approach-F", "value": "27.41"}, {"item": "System-D", "value": "13.91"}, {"item": "System-C", "value": "35.07"}, {"item": "Method-D [47]", "value": "1.40"}, {"item": "Model-C", "value": "35.01"}, {"item": "Method-F", "value": "45.88"}, {"item": "Method-B", "value": "7.66"}, {"item": "Baseline-Z", "value": "24.83"}, {"item": "System-Z [35]", "value": "36.01"}, {"item": "Method-Z [48]", "value": "22.41"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000569 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in S-A. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "13.22" |
| 2 | HB_syn_000570 | Aggregation & Comparison | Pairwise Comparison | Compare Baseline-Y vs Baseline-Z on S-A. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-Y", "value": "69.54"} |
| 3 | HB_syn_000571 | Local Relations | Neighbor Value Tracing | Locate value 73.97 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "73.97", "up": "96.60", "down": "60.05", "left": "N/A", "right": "44.53"} |
| 4 | HB_syn_000572 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 3.37, list all numbers highlighted with the same color (dark_cyan_blue, hex #55BFFF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["98.94", "10.84", "97.92", "89.27", "3.37", "7.37", "93.67"] |
| 5 | HB_syn_000573 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "7" |
| 6 | HB_syn_000574 | Consistency & Missingness | Cell Missingness | Check whether Model-C [93] at Idx is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000575 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest Score under Sect-B? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-A [31]", "value": "96.90"} |
| 2 | HB_syn_000576 | Local Relations | Neighbor Indexing | At (r=3,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "49.98", "up": "73.37", "down": "98.35", "left": "99.42", "right": "68.13"} |
| 3 | HB_syn_000577 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=2,c=2) as an anchor, list all cells sharing its highlight color (red/#FF5A5C). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["73.37", "61.09"] |
| 4 | HB_syn_000578 | Consistency & Missingness | Column Missingness | Return the items whose Score under Sect-B is missing (N/A or —). Output JSON array of strings. | ["Model-A"] |
| 5 | HB_syn_000579 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number at (r=4,c=1) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000580 | Aggregation & Comparison | Extremum Reasoning | In column Idx, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "98.09" |
| 2 | HB_syn_000581 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Idx under Slot-B. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 15, "col": 3, "value": "98.09"} |
| 3 | HB_syn_000582 | Aggregation & Comparison | Pairwise Comparison | On metric Idx under Slot-B, which is higher: Baseline-E or Model-X? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-E", "value": "74.51"} |
| 4 | HB_syn_000583 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=15,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "22.32", "up": "7.31", "down": "41.51", "left": "56.37", "right": "98.09"} |
| 5 | HB_syn_000584 | Markup Grounding | Marked Value Listing | List numbers highlighted in light_blue (hex #E6F2FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["7.37", "26.44", "84.51", "+1.86 ±0.09", "88.68", "−1.74 ±0.24"] |
| 6 | HB_syn_000585 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are > 32.21. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["84.51", "88.68"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000586 | Constrained Retrieval | Cell Retrieval | Return the value where System-Z intersects Score↓. Output only the number as a string. Example: "12.34". | "7.61" |
| 2 | HB_syn_000587 | Aggregation & Comparison | Pairwise Comparison | In row System-D [42], is MetB higher or Rate higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "MetB", "value": "90.53"} |
| 3 | HB_syn_000588 | Local Relations | Marker-Anchor Neighbor | Find the down neighbor of the highlighted in light_cyan_blue (hex #C8ECFF) with value 69.27 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "90.53" |
| 4 | HB_syn_000589 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-C [83], Rate) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 5 | HB_syn_000590 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_cyan_blue(#C8ECFF), mid_cyan_blue(#95D9FF), dark_cyan_blue(#55BFFF). Is the number with value 8.45 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000591 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Err? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "0.50" |
| 2 | HB_syn_000592 | Local Relations | Neighbor Value Tracing | Locate value 54.23 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "54.23", "up": "54.65", "down": "40.26", "left": "25.16", "right": "71.04"} |
| 3 | HB_syn_000593 | Markup Grounding | Marked Column Listing | Which entries in column MetD↑ are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["90.73"] |
| 4 | HB_syn_000594 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "9" |
| 5 | HB_syn_000595 | Consistency & Missingness | Column Missingness | Return the items whose MetD↑ is missing (N/A or —). Output JSON array of strings. | ["Method-C"] |
| 6 | HB_syn_000596 | Constrained Retrieval | Value Filtering | Filter MetA↓ by > 42.77. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-E", "value": "58.72"}, {"item": "Variant-D [91]", "value": "85.11"}, {"item": "System-F [58]", "value": "97.31"}, {"item": "Model-X [40]", "value": "83.88"}, {"item": "Variant-Z", "value": "88.19"}, {"item": "Method-C", "value": "61.87"}, {"item": "Model-D", "value": "57.45"}, {"item": "Model-B", "value": "75.71"}, {"item": "Variant-E", "value": "75.37"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000597 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Stat? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "5.19" |
| 2 | HB_syn_000598 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=8,c=7). Return all values with the same highlight color (light_blue, #CFE0FF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["11.08", "86.71", "14.30", "88.05", "78.93", "11.19", "89.77", "89.43"] |
| 3 | HB_syn_000599 | Consistency & Missingness | Cell Missingness | Check whether Approach-D at Cost↑ under Cfg-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_syn_000600 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_blue(#CFE0FF), mid_blue(#8DB3FF), dark_blue(#4F8AFF). Is the number with value 95.21 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "dark_blue", "hex": "#4F8AFF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000601 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column MetB? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "93.89" |
| 2 | HB_syn_000602 | Aggregation & Comparison | Argmax Reasoning | In column MetB under Sect-C, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-Z [34]", "value": "93.89"} |
| 3 | HB_syn_000603 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column MetB under Sect-C. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 4, "col": 6, "value": "93.89"} |
| 4 | HB_syn_000604 | Aggregation & Comparison | Pairwise Comparison | Between Variant-Z and Model-Z for Rate under Sect-C, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Model-Z", "value": "68.20"} |
| 5 | HB_syn_000605 | Local Relations | Marker-Anchor Neighbor | Find the up neighbor of the highlighted in dark_yellow (hex #EFBC23) with value 76.83 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "11.89" |
| 6 | HB_syn_000606 | Markup Grounding | Marked Value Listing | Return all numbers highlighted in light_yellow (hex #FEF4B8) as a JSON array. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["68.20", "56.12", "61.45", "47.62", "69.15", "30.86", "49.99", "49.17", "27.50"] |
| 7 | HB_syn_000607 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 69.15. Return all values sharing its highlight color (light_yellow/#FEF4B8). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["68.20", "56.12", "61.45", "47.62", "69.15", "30.86", "49.99", "49.17", "27.50"] |
| 8 | HB_syn_000608 | Consistency & Missingness | Column Missingness | List all missing entries in column S-C↑ under Blk-D (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Baseline-Z [34]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000609 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Cost under Sect-C in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Method-F", "value": "82.81"}, {"item": "Baseline-F", "value": "68.25"}, {"item": "Baseline-X", "value": "65.64"}] |
| 2 | HB_syn_000610 | Local Relations | Neighbor Indexing | At (r=5,c=5) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "49.58", "up": "65.64", "down": "57.18", "left": "90.13", "right": "—"} |
| 3 | HB_syn_000611 | Consistency & Missingness | Cell Missingness | Is the cell (Model-E, Qual↑ under Slot-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_syn_000612 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are >= 41.20. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["62.07", "43.23", "68.26", "62.64", "70.33", "41.20"] |
| 5 | HB_syn_000613 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). Is the number at (r=2,c=8) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000614 | Constrained Retrieval | Cell Retrieval | What is the MetA↓ under Zone-C value for Baseline-Y [1]? Output only the number as a string. Example: "12.34". | "72.12" |
| 2 | HB_syn_000615 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column S-D under Grp-A? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-E", "value": "73.01"} |
| 3 | HB_syn_000616 | Local Relations | Neighbor Value Tracing | Locate value 12.00 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "12.00", "up": "58.69", "down": "74.99", "left": "81.18", "right": "15.49"} |
| 4 | HB_syn_000617 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "7" |
| 5 | HB_syn_000618 | Consistency & Missingness | Column Missingness | Return the items whose Eff↑ under Zone-C is missing (N/A or —). Output JSON array of strings. | ["Baseline-Y [1]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000619 | Aggregation & Comparison | Rank Selection | In MetAB, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-E [99]", "value": "69.53"}, {"item": "Baseline-D", "value": "60.56"}, {"item": "Variant-Z", "value": "49.13"}] |
| 2 | HB_syn_000620 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in MetAB and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Baseline-D", "value": "60.56"} |
| 3 | HB_syn_000621 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (left) to the highlighted in light_teal (hex #D3FAEE) with value 13.43 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "60.56" |
| 4 | HB_syn_000622 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "39" |
| 5 | HB_syn_000623 | Consistency & Missingness | Cell Missingness | Check whether Variant-X at MetE↓ is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000624 | Constrained Retrieval | Value Filtering | In column MetA↓, list all row labels in the leftmost row-label column with value <= 63.92. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Variant-Z", "value": "42.53"}, {"item": "Method-A", "value": "63.92"}, {"item": "System-E [99]", "value": "12.39"}, {"item": "Approach-Z [73]", "value": "30.95"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000625 | Aggregation & Comparison | Pairwise Comparison | Within row Method-X, choose the larger between MetB under Slot-D and MetE under Slot-D, output {item,value}. Exclude N/A and —. | {"item": "MetE under Slot-D", "value": "65.06"} |
| 2 | HB_syn_000626 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=7,c=10), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "47.29", "up": "26.57", "down": "54.13", "left": "34.35", "right": "28.28"} |
| 3 | HB_syn_000627 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["97.46", "2.94", "90.94", "94.41", "0.75", "99.17", "83.57", "63.13", "10.76", "1.70", "98.51", "13.24", "81.52", "6.24", "92.91"] |
| 4 | HB_syn_000628 | Consistency & Missingness | Column Missingness | In S-C under Set-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-F", "Method-X"] |
| 5 | HB_syn_000629 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those <= 65.06 as a JSON array of strings. | ["62.27", "4.00", "12.55", "27.28", "2.94", "59.95", "11.93", "0.75", "12.64", "63.13", "10.76", "19.19", "26.57", "6.49", "1.70", "20.96", "32.45", "13.24", "1.17", "62.14", "65.06", "6.24", "26.41"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000630 | Aggregation & Comparison | Argmax Reasoning | In column Stat under Blk-A, which row label in the leftmost row-label column has the maximum value? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-C", "value": "77.67"} |
| 2 | HB_syn_000631 | Aggregation & Comparison | Rank Selection | List the top-3 items by column MetA under Zone-C, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-C", "value": "90.94"}, {"item": "Variant-C", "value": "89.72"}, {"item": "Model-C [40]", "value": "87.14"}] |
| 3 | HB_syn_000632 | Local Relations | Neighbor Value Tracing | Locate value 31.58 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "31.58", "up": "21.70", "down": "91.86", "left": "79.14", "right": "N/A"} |
| 4 | HB_syn_000633 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=5,c=2). Return all values with the same highlight color (light_cyan_blue, #C8ECFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["75.65", "68.07", "30.15", "73.61", "87.14", "13.68", "65.67", "82.18", "84.59", "78.09"] |
| 5 | HB_syn_000634 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["95.03", "94.11", "89.05", "76.87", "86.24", "14.63", "10.74", "91.86", "76.14", "89.72"] |
| 6 | HB_syn_000635 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-B [63], Stat under Blk-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000636 | Aggregation & Comparison | Extremum Reasoning | Return the max of column Idx. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "92.63" |
| 2 | HB_syn_000637 | Local Relations | Neighbor Value Tracing | At the cell with value 60.19, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "60.19", "up": "46.36", "down": "83.54", "left": "92.63", "right": "90.87"} |
| 3 | HB_syn_000638 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 83.54, list all numbers highlighted with the same color (orange, hex #FF8C3A). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["70.71", "83.54", "40.32", "28.89", "62.33"] |
| 4 | HB_syn_000639 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "13" |
| 5 | HB_syn_000640 | Consistency & Missingness | Column Missingness | List all missing entries in column Idx under Cfg-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Variant-D"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000641 | Constrained Retrieval | Cell Retrieval | Locate row Method-Z and column S-C↓ under Grp-B. What is the cell value? Output only the number as a string. Example: "12.34". | "19.24" |
| 2 | HB_syn_000642 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in S-A. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "29.92" |
| 3 | HB_syn_000643 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in mid_cyan_blue (hex #95D9FF) with value 81.91 number and return the number immediately to its left. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "27.02" |
| 4 | HB_syn_000644 | Markup Grounding | Marked Value Listing | Which values are highlighted in dark_cyan_blue (#55BFFF)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["98.25", "12.56", "22.07", "91.64", "98.29", "5.70", "97.95", "8.10", "7.29", "0.37", "80.29", "97.73", "85.95", "93.31"] |
| 5 | HB_syn_000645 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-E [5], Qual under Cfg-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000646 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column S-D. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 13, "col": 5, "value": "99.16"} |
| 2 | HB_syn_000647 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=12,c=2), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "90.61", "up": "38.73", "down": "17.54", "left": "75.59", "right": "81.93"} |
| 3 | HB_syn_000648 | Markup Grounding | Marked Column Listing | Which entries in column Score↑ are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["81.64"] |
| 4 | HB_syn_000649 | Consistency & Missingness | Cell Missingness | Is the cell (System-A, Score↑) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000650 | Constrained Retrieval | Value Filtering | In column Score↑, list all row labels in the leftmost row-label column with value >= 38.73. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-B", "value": "59.73"}, {"item": "Method-C [24]", "value": "81.37"}, {"item": "Variant-A", "value": "41.68"}, {"item": "Approach-B [21]", "value": "81.64"}, {"item": "System-C", "value": "56.71"}, {"item": "Approach-D", "value": "38.73"}, {"item": "Baseline-X", "value": "90.61"}] |
| 6 | HB_syn_000651 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_purple(#E6D1FF), mid_purple(#B681F3), dark_purple(#8537EB). Is the number at (r=11,c=6) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000652 | Aggregation & Comparison | Pairwise Comparison | Between Method-A and Variant-Y for MetC under Sect-A, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Variant-Y", "value": "83.68"} |
| 2 | HB_syn_000653 | Local Relations | Neighbor Indexing | At (r=9,c=5) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "56.57", "up": "6.12", "down": "53.82", "left": "12.95", "right": "54.16"} |
| 3 | HB_syn_000654 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 15.90. Return all values sharing its highlight color (blue/#6DA1FD). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["57.19", "54.99", "15.90"] |
| 4 | HB_syn_000655 | Consistency & Missingness | Column Missingness | List all missing entries in column S-A↑ under Sect-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["System-Y"] |
| 5 | HB_syn_000656 | Markup Grounding | Highlight Presence | Check whether value 17.20 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000657 | Aggregation & Comparison | Extremum Reasoning | In row Method-Y, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "35.20" |
| 2 | HB_syn_000658 | Local Relations | Neighbor Value Tracing | At the cell with value 3.36, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "3.36", "up": "66.77", "down": "93.35", "left": "44.80", "right": "73.82"} |
| 3 | HB_syn_000659 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "18" |
| 4 | HB_syn_000660 | Consistency & Missingness | Column Missingness | List all missing entries in column Score under Blk-C (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Method-D"] |
| 5 | HB_syn_000661 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are <= 85.54. Output JSON array of strings. | ["10.91", "4.74", "15.96", "2.42", "7.58", "6.76", "3.36", "8.63", "5.08", "85.54"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000662 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in Err. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "2.75" |
| 2 | HB_syn_000663 | Aggregation & Comparison | Extremum Reasoning | In column MetA, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "78.28" |
| 3 | HB_syn_000664 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column MetA under Set-A. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-E", "value": "78.28"} |
| 4 | HB_syn_000665 | Aggregation & Comparison | Extremum Reasoning | In row Model-A, return the greatest numeric entry. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "90.73" |
| 5 | HB_syn_000666 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the boldfaced with value 97.24 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "8.80" |
| 6 | HB_syn_000667 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 7, column 4, list all numbers highlighted with the same color (purple, hex #B07CFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["93.66", "43.60", "71.85", "78.19", "53.53", "20.74"] |
| 7 | HB_syn_000668 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["17.48", "13.22", "+2.18 ±0.64", "61.58", "90.73", "87.30", "20.19", "67.85", "20.52"] |
| 8 | HB_syn_000669 | Consistency & Missingness | Cell Missingness | Is the cell (Model-D, Stat↓ under Zone-B) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000670 | Aggregation & Comparison | Argmax Reasoning | In Qual, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 12, "col": 4, "value": "88.49"} |
| 2 | HB_syn_000671 | Aggregation & Comparison | Pairwise Comparison | On metric Qual, which is higher: Method-X or Model-E [48]? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-X", "value": "79.30"} |
| 3 | HB_syn_000672 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (down) to the highlighted in mid_teal (hex #83EDC4) with value 98.09 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "34.01" |
| 4 | HB_syn_000673 | Markup Grounding | Marked Column Listing | Which entries in column Rank↓ are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["5.04"] |
| 5 | HB_syn_000674 | Consistency & Missingness | Cell Missingness | Is the cell (Method-E, Qual) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_syn_000675 | Constrained Retrieval | Value Filtering | Filter Val↓ by < 45.70. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-F", "value": "22.13"}, {"item": "Approach-A [67]", "value": "12.69"}, {"item": "System-F [48]", "value": "36.95"}, {"item": "Model-A", "value": "27.28"}, {"item": "Variant-C", "value": "39.84"}, {"item": "System-E [2]", "value": "43.71"}, {"item": "Baseline-D", "value": "0.70"}, {"item": "Model-E [48]", "value": "22.05"}, {"item": "Baseline-E", "value": "18.92"}, {"item": "System-A", "value": "20.93"}] |
| 7 | HB_syn_000676 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_teal(#D3FAEE), mid_teal(#83EDC4), dark_teal(#22D28E). Is the number with value 54.75 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000677 | Constrained Retrieval | Cell Retrieval | For item System-B, what is the value in column MetD? Output only the number as a string. Example: "12.34". | "44.28" |
| 2 | HB_syn_000678 | Aggregation & Comparison | Pairwise Comparison | Within row Method-Y, choose the larger between Val and S-C, output {item,value}. Exclude N/A and —. | {"item": "S-C", "value": "48.90"} |
| 3 | HB_syn_000679 | Local Relations | Neighbor Indexing | At (r=6,c=10) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "36.55", "up": "16.13", "down": "56.24", "left": "44.92", "right": "70.06"} |
| 4 | HB_syn_000680 | Consistency & Missingness | Column Missingness | Return the items whose Qual↑ is missing (N/A or —). Output JSON array of strings. | ["Baseline-B"] |
| 5 | HB_syn_000681 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_cyan_blue(#C8ECFF), mid_cyan_blue(#95D9FF), dark_cyan_blue(#55BFFF). Is the number at (r=4,c=2) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000682 | Aggregation & Comparison | Rank Selection | In Val, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-X", "value": "93.94"}, {"item": "Method-Y", "value": "93.79"}, {"item": "Approach-Y", "value": "90.07"}] |
| 2 | HB_syn_000683 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Val and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Method-Y", "value": "93.79"} |
| 3 | HB_syn_000684 | Local Relations | Neighbor Value Tracing | At the cell with value 90.05, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "90.05", "up": "45.44", "down": "93.96", "left": "56.27", "right": "61.14"} |
| 4 | HB_syn_000685 | Markup Grounding | Marked Value Listing | Which values are highlighted in mid_cyan_blue (#95D9FF)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["93.79", "87.98", "90.05"] |
| 5 | HB_syn_000686 | Consistency & Missingness | Column Missingness | Return the items whose Val is missing (N/A or —). Output JSON array of strings. | ["Baseline-E"] |
| 6 | HB_syn_000687 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are >= 90.07. Output JSON array of strings. | ["93.94", "93.79", "93.96", "90.81", "90.07"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000688 | Aggregation & Comparison | Pairwise Comparison | On metric Stat under Part-A, which is higher: Method-E or Method-D? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-E", "value": "88.91"} |
| 2 | HB_syn_000689 | Aggregation & Comparison | Pairwise Comparison | Within row Method-D, choose the larger between Eff under Slot-B and MetC under Cfg-C, output {item,value}. Exclude N/A and —. | {"item": "MetC under Cfg-C", "value": "83.31"} |
| 3 | HB_syn_000690 | Local Relations | Neighbor Indexing | At (r=6,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "90.05", "up": "11.25", "down": "88.91", "left": "33.07", "right": "—"} |
| 4 | HB_syn_000691 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 5, column 7, list all numbers highlighted with the same color (light purple, hex #E593FC). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["2.77", "3.79", "85.76"] |
| 5 | HB_syn_000692 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "8" |
| 6 | HB_syn_000693 | Consistency & Missingness | Cell Missingness | Check whether Model-B at MetC under Cfg-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000694 | Aggregation & Comparison | Extremum Reasoning | In row Model-Z, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "6.08" |
| 2 | HB_syn_000695 | Local Relations | Neighbor Value Tracing | At the cell with value 68.11, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "68.11", "up": "N/A", "down": "52.54", "left": "78.23", "right": "70.69"} |
| 3 | HB_syn_000696 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 90.52, list all numbers highlighted with the same color (light_teal, hex #D3FAEE). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["28.42", "90.52", "45.01", "93.67", "78.87", "59.85", "53.83", "86.46", "83.24", "71.00"] |
| 4 | HB_syn_000697 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "10" |
| 5 | HB_syn_000698 | Consistency & Missingness | Column Missingness | In MetA↑ under Set-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-Z", "Model-X"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000699 | Constrained Retrieval | Cell Retrieval | Locate row Method-X [77] and column Val↓ under Sect-B. What is the cell value? Output only the number as a string. Example: "12.34". | "55.67" |
| 2 | HB_syn_000700 | Aggregation & Comparison | Pairwise Comparison | For Method-B, compare Stat under Slot-A and Err under Set-C. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "Err under Set-C", "value": "91.35"} |
| 3 | HB_syn_000701 | Local Relations | Marker-Anchor Neighbor | Find the left neighbor of the highlighted in light_cyan_blue (hex #C8ECFF) with value 55.43 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "47.68" |
| 4 | HB_syn_000702 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Val↓ under Sect-B. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["10.07"] |
| 5 | HB_syn_000703 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-D, MetC↑ under Set-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000704 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column Score under Slot-C? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-X [47]", "value": "88.33"} |
| 2 | HB_syn_000705 | Aggregation & Comparison | Pairwise Comparison | Within row Model-F, choose the larger between Stat under Slot-C and MetC under Set-B, output {item,value}. Exclude N/A and —. | {"item": "MetC under Set-B", "value": "48.45"} |
| 3 | HB_syn_000706 | Local Relations | Marker-Anchor Neighbor | Find the left neighbor of the highlighted in dark_green (hex #72D46A) with value 96.71 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "60.99" |
| 4 | HB_syn_000707 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-Y [13], MetC under Set-B) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000708 | Constrained Retrieval | Value Filtering | In column S-A↑ under Cfg-A, list all row labels in the leftmost row-label column with value >= 52.83. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-Z", "value": "99.53"}, {"item": "System-X [36]", "value": "89.76"}, {"item": "System-C", "value": "52.83"}, {"item": "Baseline-X [47]", "value": "64.63"}, {"item": "Variant-F", "value": "61.28"}, {"item": "Approach-Z", "value": "71.77"}] |
| 6 | HB_syn_000709 | Markup Grounding | Highlight Presence | Check whether (r=3,c=8) is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_green(#D7F6D1), mid_green(#AEEA9F), dark_green(#72D46A). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000710 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Idx. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 14, "col": 5, "value": "79.52"} |
| 2 | HB_syn_000711 | Local Relations | Neighbor Indexing | At (r=11,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "14.53", "up": "17.54", "down": "56.16", "left": "68.05", "right": "56.74"} |
| 3 | HB_syn_000712 | Markup Grounding | Marked Value Listing | List numbers highlighted in dark_red (hex #FC5255). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["91.13", "97.59", "96.52", "79.52", "88.80"] |
| 4 | HB_syn_000713 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 82.30. Return all values sharing its highlight color (light_red/#FFD6D6). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["70.95", "71.21", "85.30", "82.30", "86.48"] |
| 5 | HB_syn_000714 | Consistency & Missingness | Column Missingness | In MetC↑, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["System-B", "Method-Y"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000715 | Aggregation & Comparison | Extremum Reasoning | Return the max of column Score. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "95.75" |
| 2 | HB_syn_000716 | Aggregation & Comparison | Rank Selection | List the top-3 items by column Score under Slot-A, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-X", "value": "95.75"}, {"item": "System-E", "value": "93.51"}, {"item": "Approach-Y", "value": "86.48"}] |
| 3 | HB_syn_000717 | Aggregation & Comparison | Extremum Reasoning | For item Method-A [18], what is the maximum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "85.33" |
| 4 | HB_syn_000718 | Local Relations | Neighbor Value Tracing | At the cell with value 88.77, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "88.77", "up": "23.63", "down": "38.58", "left": "63.05", "right": "42.81"} |
| 5 | HB_syn_000719 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 8, column 4, list all numbers highlighted with the same color (orange, hex #FF8C3A). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["58.92", "60.51", "67.50"] |
| 6 | HB_syn_000720 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "9" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000721 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Eff under Zone-B. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-Z", "value": "95.50"} |
| 2 | HB_syn_000722 | Aggregation & Comparison | Pairwise Comparison | Within row Baseline-E, choose the larger between Eff under Zone-B and Qual under Slot-D, output {item,value}. Exclude N/A and —. | {"item": "Qual under Slot-D", "value": "57.38"} |
| 3 | HB_syn_000723 | Local Relations | Neighbor Value Tracing | At the cell with value 90.36, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "90.36", "up": "56.85", "down": "16.62", "left": "3.77", "right": "79.82"} |
| 4 | HB_syn_000724 | Consistency & Missingness | Column Missingness | Return the items whose Rate under Sect-E is missing (N/A or —). Output JSON array of strings. | ["Model-A", "Method-E [37]"] |
| 5 | HB_syn_000725 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are <= 57.84. Output JSON array of strings. | ["37.78", "52.10", "15.31", "22.78", "57.84", "3.02", "46.63", "48.37", "46.04", "52.37"] |
| 6 | HB_syn_000726 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number with value 48.37 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "orange", "hex": "#FF8C3A"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000727 | Constrained Retrieval | Cell Retrieval | What is the MetD↑ under Part-A value for Variant-D [62]? Output only the number as a string. Example: "12.34". | "43.32" |
| 2 | HB_syn_000728 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Baseline-B [91]. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "4.15" |
| 3 | HB_syn_000729 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in light purple (hex #E593FC) with value 17.38 number and return the number immediately below it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "83.29" |
| 4 | HB_syn_000730 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "13" |
| 5 | HB_syn_000731 | Consistency & Missingness | Cell Missingness | Check whether Variant-Z [23] at MetD↑ under Part-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000732 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column MetD? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "13.66" |
| 2 | HB_syn_000733 | Local Relations | Neighbor Indexing | At (r=8,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "91.01", "up": "93.07", "down": "47.67", "left": "42.57", "right": "6.89"} |
| 3 | HB_syn_000734 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["27.12", "13.41", "73.81", "91.13", "64.96", "18.27", "47.67", "12.20", "5.81"] |
| 4 | HB_syn_000735 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "27" |
| 5 | HB_syn_000736 | Consistency & Missingness | Cell Missingness | Check whether Method-A at MetD under Set-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_syn_000737 | Constrained Retrieval | Value Filtering | In column MetD↑ under Sect-D, list all row labels in the leftmost row-label column with value <= 51.02. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-X [69]", "value": "51.02"}, {"item": "Variant-D", "value": "17.00"}, {"item": "Model-Z [92]", "value": "25.54"}, {"item": "Method-Z [93]", "value": "15.07"}, {"item": "Model-D [62]", "value": "8.23"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000738 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Idx? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "1.94" |
| 2 | HB_syn_000739 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Idx under Sect-A. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-A", "value": "91.32"} |
| 3 | HB_syn_000740 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Idx under Sect-A in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Method-A", "value": "91.32"}, {"item": "Baseline-E", "value": "89.03"}, {"item": "Variant-A", "value": "86.65"}] |
| 4 | HB_syn_000741 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Baseline-X [62], what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "73.35" |
| 5 | HB_syn_000742 | Local Relations | Marker-Anchor Neighbor | Locate the text-colored in red (hex #C62828) with value −0.24 ±0.65 number and return the number immediately to its left. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "89.55" |
| 6 | HB_syn_000743 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 1.62, list all numbers highlighted with the same color (orange, hex #FF8C3A). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["31.85", "63.41", "60.45", "1.62"] |
| 7 | HB_syn_000744 | Markup Grounding | Marked Column Listing | Which entries in column Cost under Cfg-C are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["89.55"] |
| 8 | HB_syn_000745 | Consistency & Missingness | Column Missingness | In MetC↑ under Cfg-C, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-Y [77]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000746 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in MetB. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.36" |
| 2 | HB_syn_000747 | Local Relations | Neighbor Value Tracing | At the cell with value 1.34, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "1.34", "up": "38.97", "down": "25.89", "left": "12.88", "right": "16.02"} |
| 3 | HB_syn_000748 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=8,c=3). Return all values with the same highlight color (dark_cyan_blue, #55BFFF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["99.54", "79.26", "87.15", "1.90", "97.35", "1.34", "95.68", "99.36", "5.64", "8.82", "99.64"] |
| 4 | HB_syn_000749 | Consistency & Missingness | Column Missingness | In Eff↑ under Sect-D, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Baseline-E"] |
| 5 | HB_syn_000750 | Markup Grounding | Highlight Presence | Check whether value 23.08 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_cyan_blue(#C8ECFF), mid_cyan_blue(#95D9FF), dark_cyan_blue(#55BFFF). The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000751 | Aggregation & Comparison | Argmax Reasoning | In Err under Slot-A, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 5, "col": 4, "value": "98.05"} |
| 2 | HB_syn_000752 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Err under Slot-A and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Approach-D", "value": "85.24"} |
| 3 | HB_syn_000753 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Approach-B, what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "1.28" |
| 4 | HB_syn_000754 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=6,c=9), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "85.94", "up": "31.95", "down": "58.79", "left": "30.53", "right": "42.00"} |
| 5 | HB_syn_000755 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["91.15", "97.59", "87.87", "87.13", "50.42", "85.94", "85.64", "18.86", "85.24", "21.55"] |
| 6 | HB_syn_000756 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-F, Qual↑ under Zone-B) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 7 | HB_syn_000757 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are >= 85.64. Output JSON array of strings. | ["91.15", "97.59", "88.15", "87.87", "87.13", "98.05", "93.07", "91.34", "88.27", "85.94", "85.64", "92.62", "98.24", "91.69", "88.14"] |
| 8 | HB_syn_000758 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_green(#D7F6D1), mid_green(#AEEA9F), dark_green(#72D46A). Is the number at (r=8,c=10) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "mid_green", "hex": "#AEEA9F"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000759 | Constrained Retrieval | Cell Retrieval | In the Err↓ under Blk-C column, read the entry for Approach-D [31]. Output only the number as a string. Example: "12.34". | "82.34" |
| 2 | HB_syn_000760 | Aggregation & Comparison | Extremum Reasoning | In column Score, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "10.31" |
| 3 | HB_syn_000761 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the underlined with value 77.75 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "86.77" |
| 4 | HB_syn_000762 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "11" |
| 5 | HB_syn_000763 | Consistency & Missingness | Column Missingness | Return the items whose Eff↑ under Blk-C is missing (N/A or —). Output JSON array of strings. | ["Method-Z"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000764 | Aggregation & Comparison | Pairwise Comparison | In row System-Z, is S-D under Slot-C higher or Cost under Slot-D higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "Cost under Slot-D", "value": "50.24"} |
| 2 | HB_syn_000765 | Markup Grounding | Marked Value Listing | Which values are highlighted in mid_teal (#83EDC4)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["96.21", "96.46", "36.03", "94.60", "66.84", "87.35", "96.54", "6.78", "80.03", "7.08", "17.15", "24.66", "86.95"] |
| 3 | HB_syn_000766 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-Y [56], Err↓ under Slot-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_syn_000767 | Constrained Retrieval | Value Filtering | Filter MetD↑ under Zone-A by >= 41.09. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Baseline-Y [56]", "value": "55.82"}, {"item": "Method-E [1]", "value": "66.84"}, {"item": "Model-Z", "value": "41.09"}, {"item": "Model-A [80]", "value": "41.90"}, {"item": "Model-F [42]", "value": "88.91"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000768 | Aggregation & Comparison | Pairwise Comparison | Compare Approach-F vs Model-A on MetB under Slot-B. Which one is larger? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-F", "value": "98.21"} |
| 2 | HB_syn_000769 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=11,c=3), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "50.52", "up": "55.76", "down": "27.58", "left": "95.74", "right": "57.16"} |
| 3 | HB_syn_000770 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 75.71, list all numbers highlighted with the same color (light_yellow, hex #FEF4B8). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["84.78", "84.74", "7.72", "88.44", "9.82", "95.74", "83.57", "75.71"] |
| 4 | HB_syn_000771 | Markup Grounding | Marked Column Listing | Which entries in column Stat↑ under Slot-B are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["90.05"] |
| 5 | HB_syn_000772 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-Y, Idx under Cfg-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000773 | Aggregation & Comparison | Extremum Reasoning | Find the highest metric value for Approach-A. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "81.34" |
| 2 | HB_syn_000774 | Local Relations | Neighbor Value Tracing | At the cell with value 14.50, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "14.50", "up": "88.01", "down": "87.27", "left": "53.92", "right": "66.34"} |
| 3 | HB_syn_000775 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=4,c=5) as an anchor, list all cells sharing its highlight color (mid_pink/#FC88BC). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["20.40", "80.64", "96.26", "2.06", "92.56", "84.86", "17.96", "85.32"] |
| 4 | HB_syn_000776 | Consistency & Missingness | Column Missingness | In Idx↑ under Part-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Variant-D"] |
| 5 | HB_syn_000777 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_pink(#FFD1E7), mid_pink(#FC88BC), dark_pink(#F74B98). Is the number at (r=9,c=1) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000778 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in Qual. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "95.92" |
| 2 | HB_syn_000779 | Aggregation & Comparison | Extremum Reasoning | For item System-X, what is the maximum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "97.85" |
| 3 | HB_syn_000780 | Aggregation & Comparison | Extremum Reasoning | In row System-X, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "16.74" |
| 4 | HB_syn_000781 | Local Relations | Marker-Anchor Neighbor | Find the up neighbor of the underlined with value 3.76 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "53.80" |
| 5 | HB_syn_000782 | Markup Grounding | Marked Value Listing | List numbers highlighted in light_orange (hex #FFE6C7). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["92.80", "83.65", "37.18", "56.96", "31.31", "92.32", "1.89", "51.84", "43.66"] |
| 6 | HB_syn_000783 | Consistency & Missingness | Column Missingness | List all missing entries in column Qual under Zone-B (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["System-E"] |
| 7 | HB_syn_000784 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those > 43.66 as a JSON array of strings. | ["99.66", "92.80", "83.65", "82.20", "56.96", "92.32", "51.84"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000785 | Constrained Retrieval | Cell Retrieval | Give the number at (System-F, Rank under Slot-C). Output only the number as a string. Example: "12.34". | "31.33" |
| 2 | HB_syn_000786 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column S-A under Part-A. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 2, "col": 1, "value": "60.98"} |
| 3 | HB_syn_000787 | Local Relations | Neighbor Value Tracing | At the cell with value 39.38, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "39.38", "up": "60.01", "down": "66.89", "left": "48.42", "right": "37.37"} |
| 4 | HB_syn_000788 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["51.91", "16.41", "38.98", "67.26", "39.38", "17.79", "77.91", "42.93", "26.76", "83.53", "74.60", "86.51", "52.35", "66.24"] |
| 5 | HB_syn_000789 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-A, Score↑ under Cfg-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000790 | Markup Grounding | Highlight Presence | Check whether value 28.48 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_teal(#D3FAEE), mid_teal(#83EDC4), dark_teal(#22D28E). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "light_teal", "hex": "#D3FAEE"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000791 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest MetC? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Model-C", "value": "96.36"} |
| 2 | HB_syn_000792 | Aggregation & Comparison | Pairwise Comparison | On metric MetC, which is higher: Method-D [83] or Method-X? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-D [83]", "value": "95.91"} |
| 3 | HB_syn_000793 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=7,c=4), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "94.00", "up": "10.22", "down": "42.75", "left": "54.46", "right": "78.91"} |
| 4 | HB_syn_000794 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "21" |
| 5 | HB_syn_000795 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "7" |
| 6 | HB_syn_000796 | Consistency & Missingness | Cell Missingness | Is the cell (System-X, S-C↑) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 7 | HB_syn_000797 | Constrained Retrieval | Value Filtering | Filter S-A↑ by > 52.00. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-B", "value": "97.02"}, {"item": "Model-F [74]", "value": "74.36"}, {"item": "Method-Z", "value": "98.90"}, {"item": "Model-E", "value": "86.05"}, {"item": "Model-C", "value": "93.47"}, {"item": "Approach-E [80]", "value": "76.40"}, {"item": "Method-Y", "value": "70.19"}, {"item": "Baseline-C", "value": "57.17"}, {"item": "System-X", "value": "63.01"}] |
| 8 | HB_syn_000798 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_red(#FFD6D6), mid_red(#F86D7B), dark_red(#FC5255). Is the number at (r=11,c=2) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "dark_red", "hex": "#FC5255"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000799 | Aggregation & Comparison | Rank Selection | List the top-3 items by column MetA, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-F", "value": "87.34"}, {"item": "Baseline-A [49]", "value": "87.01"}, {"item": "Variant-B", "value": "52.73"}] |
| 2 | HB_syn_000800 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column MetA? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-A [49]", "value": "87.01"} |
| 3 | HB_syn_000801 | Local Relations | Marker-Anchor Neighbor | Find the right neighbor of the boldfaced with value 4.13 cell. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "48.39" |
| 4 | HB_syn_000802 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=1,c=1). Return all values with the same highlight color (light_purple, #E6D1FF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["15.25", "52.73"] |
| 5 | HB_syn_000803 | Markup Grounding | Highlight Presence | Check whether value 87.01 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_purple(#E6D1FF), mid_purple(#B681F3), dark_purple(#8537EB). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "mid_purple", "hex": "#B681F3"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000804 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Val? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "96.89" |
| 2 | HB_syn_000805 | Local Relations | Neighbor Indexing | At (r=4,c=9) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "11.87", "up": "90.81", "down": "76.12", "left": "44.43", "right": "91.66"} |
| 3 | HB_syn_000806 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 46.68. Return all values sharing its highlight color (pink/#FFB7D6). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["54.25", "44.43", "58.41", "95.14", "−0.04 ±0.57", "46.68", "17.85", "86.98", "36.41", "33.74", "41.97", "16.32", "0.97", "60.73", "+2.36 ±0.67", "19.84", "54.42", "74.06"] |
| 4 | HB_syn_000807 | Markup Grounding | Marked Column Listing | In Score↑ under Blk-D, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["89.29"] |
| 5 | HB_syn_000808 | Consistency & Missingness | Column Missingness | List all missing entries in column Score under Grp-B (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Approach-E [65]", "Baseline-X"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000809 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column MetB? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "43.54" |
| 2 | HB_syn_000810 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "10" |
| 3 | HB_syn_000811 | Consistency & Missingness | Column Missingness | In Err↑ under Part-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Baseline-F", "Method-C"] |
| 4 | HB_syn_000812 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are < 81.78. Output JSON array of strings. | ["79.95", "18.14", "81.66", "75.66", "65.12", "71.00", "17.62", "69.65", "77.19", "42.18", "67.92", "29.73", "32.57", "81.29", "22.47"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000813 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column Err under Blk-D. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 1, "col": 11, "value": "79.56"} |
| 2 | HB_syn_000814 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "36" |
| 3 | HB_syn_000815 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-A, Err under Blk-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 4 | HB_syn_000816 | Constrained Retrieval | Value Filtering | In column S-C↓ under Set-B, list all row labels in the leftmost row-label column with value < 44.65. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-Y", "value": "20.19"}, {"item": "Method-D", "value": "28.23"}, {"item": "Model-C", "value": "42.14"}, {"item": "Method-Z", "value": "19.50"}, {"item": "Variant-A", "value": "8.18"}, {"item": "Method-F", "value": "22.90"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000817 | Constrained Retrieval | Cell Retrieval | For item Baseline-Y, what is the value in column Score↑ under Cfg-A? Output only the number as a string. Example: "12.34". | "15.33" |
| 2 | HB_syn_000818 | Aggregation & Comparison | Extremum Reasoning | Across all columns for System-E [65], what is the minimum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "42.19" |
| 3 | HB_syn_000819 | Local Relations | Neighbor Value Tracing | Locate value 94.37 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "94.37", "up": "64.27", "down": "59.83", "left": "11.83", "right": "46.62"} |
| 4 | HB_syn_000820 | Markup Grounding | Marked Value Listing | Which values are highlighted in mid_orange (#F1B17D)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["82.76", "84.85", "88.80", "10.52", "82.53"] |
| 5 | HB_syn_000821 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["82.76", "84.85", "88.80", "10.52", "82.53"] |
| 6 | HB_syn_000822 | Consistency & Missingness | Column Missingness | List all missing entries in column S-D↑ under Cfg-B (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Baseline-Y", "System-E [65]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000823 | Aggregation & Comparison | Pairwise Comparison | Between System-Z and Variant-Y for S-D, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Variant-Y", "value": "34.82"} |
| 2 | HB_syn_000824 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 91.76. Return all values sharing its highlight color (purple/#B07CFF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["40.20", "91.76", "82.47"] |
| 3 | HB_syn_000825 | Consistency & Missingness | Cell Missingness | Check whether System-E [34] at S-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 4 | HB_syn_000826 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number with value 77.20 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000827 | Aggregation & Comparison | Rank Selection | List the top-3 items by column MetA under Sect-C, sorted from highest to lowest. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-E", "value": "99.45"}, {"item": "Baseline-Y", "value": "89.59"}, {"item": "System-F", "value": "77.66"}] |
| 2 | HB_syn_000828 | Local Relations | Neighbor Value Tracing | Locate value 92.60 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "92.60", "up": "73.85", "down": "34.23", "left": "55.43", "right": "60.58"} |
| 3 | HB_syn_000829 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=12,c=7) as an anchor, list all cells sharing its highlight color (dark_teal/#22D28E). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["81.85", "1.56", "89.31", "99.45", "98.46", "2.21", "96.21"] |
| 4 | HB_syn_000830 | Markup Grounding | Marked Column Listing | Which entries in column Score↓ under Sect-C are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["22.46"] |
| 5 | HB_syn_000831 | Consistency & Missingness | Cell Missingness | Is the cell (Model-Y [12], Idx↑ under Grp-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000832 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Qual under Slot-C. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Baseline-Z", "value": "92.84"} |
| 2 | HB_syn_000833 | Local Relations | Neighbor Indexing | At (r=3,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "89.43", "up": "80.11", "down": "56.95", "left": "75.10", "right": "45.67"} |
| 3 | HB_syn_000834 | Markup Grounding | Marked Value Listing | Find all orange-highlighted values (hex #F9B163) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["34.30"] |
| 4 | HB_syn_000835 | Consistency & Missingness | Column Missingness | List all missing entries in column Qual under Slot-C (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["System-Y"] |
| 5 | HB_syn_000836 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are >= 55.20. Output JSON array of strings. | ["97.88", "55.20", "78.71", "78.91", "69.35"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000837 | Constrained Retrieval | Cell Retrieval | In the Score↓ column, read the entry for System-E. Output only the number as a string. Example: "12.34". | "69.96" |
| 2 | HB_syn_000838 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in Qual. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "99.66" |
| 3 | HB_syn_000839 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column Qual? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Variant-Z", "value": "98.47"} |
| 4 | HB_syn_000840 | Local Relations | Marker-Anchor Neighbor | Locate the boldfaced with value 0.65 number and return the number immediately above it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "13.83" |
| 5 | HB_syn_000841 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["0.69", "79.50", "98.47", "93.38"] |
| 6 | HB_syn_000842 | Consistency & Missingness | Column Missingness | In Val↑, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-F"] |
| 7 | HB_syn_000843 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_green(#D7F6D1), mid_green(#AEEA9F), dark_green(#72D46A). Is the number at (r=7,c=4) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "mid_green", "hex": "#AEEA9F"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000844 | Aggregation & Comparison | Pairwise Comparison | For System-A, compare MetC under Slot-B and MetD under Slot-B. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "MetD under Slot-B", "value": "50.15"} |
| 2 | HB_syn_000845 | Local Relations | Marker-Anchor Neighbor | Locate the text-colored in blue (hex #1565C0) with value 79.17 number and return the number immediately to its right. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "63.14" |
| 3 | HB_syn_000846 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "12" |
| 4 | HB_syn_000847 | Consistency & Missingness | Cell Missingness | Check whether Method-X [27] at MetC under Slot-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000848 | Constrained Retrieval | Value Filtering | Filter Err under Grp-C by <= 50.27. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-Y", "value": "8.18"}, {"item": "Method-B [64]", "value": "50.27"}, {"item": "Baseline-Z [20]", "value": "0.60"}, {"item": "System-F [80]", "value": "1.44"}, {"item": "Method-X [27]", "value": "33.86"}, {"item": "System-A", "value": "7.25"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000849 | Aggregation & Comparison | Pairwise Comparison | For Approach-D, compare MetB under Blk-D and Score under Blk-D. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "MetB under Blk-D", "value": "34.80"} |
| 2 | HB_syn_000850 | Local Relations | Neighbor Indexing | At (r=2,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "1.19", "up": "62.89", "down": "14.55", "left": "56.94", "right": "25.75"} |
| 3 | HB_syn_000851 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=4,c=8) as an anchor, list all cells sharing its highlight color (light purple/#E593FC). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["72.49", "74.84", "36.06", "50.51"] |
| 4 | HB_syn_000852 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "11" |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000853 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Rate under Cfg-B. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Method-Y", "value": "91.99"} |
| 2 | HB_syn_000854 | Aggregation & Comparison | Extremum Reasoning | For item Baseline-X, what is the maximum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "90.87" |
| 3 | HB_syn_000855 | Local Relations | Neighbor Value Tracing | At the cell with value 72.36, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "72.36", "up": "30.42", "down": "2.21", "left": "87.51", "right": "69.15"} |
| 4 | HB_syn_000856 | Markup Grounding | Marked Value Listing | Return all numbers highlighted in mid_blue (hex #8DB3FF) as a JSON array. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["2.66", "27.81", "15.77", "31.76", "86.48", "87.51", "69.15"] |
| 5 | HB_syn_000857 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 33.18, list all numbers highlighted with the same color (light_blue, hex #CFE0FF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["63.66", "86.47", "81.25", "23.97", "5.12", "33.92", "33.18"] |
| 6 | HB_syn_000858 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-Z, Score↓ under Slot-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000859 | Aggregation & Comparison | Extremum Reasoning | Return the min of column Score. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "19.32" |
| 2 | HB_syn_000860 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Score in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Variant-C", "value": "85.37"}, {"item": "Method-X", "value": "62.77"}, {"item": "Method-Y [86]", "value": "56.36"}] |
| 3 | HB_syn_000861 | Local Relations | Neighbor Value Tracing | At the cell with value 62.77, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "62.77", "up": "56.36", "down": "35.77", "left": "0.32", "right": "49.33"} |
| 4 | HB_syn_000862 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["33.68", "38.37", "62.77", "+1.71 ±0.37", "48.96"] |
| 5 | HB_syn_000863 | Consistency & Missingness | Column Missingness | Return the items whose Err is missing (N/A or —). Output JSON array of strings. | ["Method-X"] |
| 6 | HB_syn_000864 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are >= 34.78. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["34.78"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000865 | Constrained Retrieval | Cell Retrieval | Locate row Method-Y and column Qual↑ under Cfg-A. What is the cell value? Output only the number as a string. Example: "12.34". | "32.66" |
| 2 | HB_syn_000866 | Aggregation & Comparison | Pairwise Comparison | Between System-F [70] and Variant-E for Err under Part-C, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Variant-E", "value": "85.15"} |
| 3 | HB_syn_000867 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in dark_orange (hex #E47B30) with value 5.25 number and return the number immediately to its right. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "18.91" |
| 4 | HB_syn_000868 | Consistency & Missingness | Column Missingness | List all missing entries in column Qual↑ under Cfg-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Variant-E"] |
| 5 | HB_syn_000869 | Markup Grounding | Highlight Presence | Check whether value 79.43 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_orange(#FCE0C7), mid_orange(#F1B17D), dark_orange(#E47B30). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "mid_orange", "hex": "#F1B17D"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000870 | Aggregation & Comparison | Argmax Reasoning | In Rate under Set-D, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 6, "col": 10, "value": "94.07"} |
| 2 | HB_syn_000871 | Local Relations | Neighbor Indexing | At (r=5,c=7) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "27.65", "up": "14.44", "down": "89.05", "left": "61.37", "right": "74.40"} |
| 3 | HB_syn_000872 | Markup Grounding | Marked Column Listing | List all underlined numbers in column S-C↓ under Slot-A. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["14.70"] |
| 4 | HB_syn_000873 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "11" |
| 5 | HB_syn_000874 | Consistency & Missingness | Cell Missingness | Check whether Baseline-Y at Val↑ under Sect-B is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": true} |
| 6 | HB_syn_000875 | Constrained Retrieval | Value Filtering | In column Cost under Blk-C, list all row labels in the leftmost row-label column with value >= 24.72. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Model-E", "value": "65.53"}, {"item": "Baseline-Z", "value": "24.72"}, {"item": "System-X", "value": "57.39"}, {"item": "Model-F", "value": "74.40"}] |
| 7 | HB_syn_000876 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number at (r=3,c=6) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "purple", "hex": "#B07CFF"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000877 | Aggregation & Comparison | Argmax Reasoning | Find the location (row,col) and value of the maximum cell in column MetD. Output JSON {"row":<int>,"col":<int>,"value":<string>}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 12, "col": 6, "value": "99.25"} |
| 2 | HB_syn_000878 | Aggregation & Comparison | Rank Selection | Who ranks 2nd in column S-D? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Variant-X", "value": "96.14"} |
| 3 | HB_syn_000879 | Aggregation & Comparison | Extremum Reasoning | In row System-E, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "0.56" |
| 4 | HB_syn_000880 | Local Relations | Marker-Anchor Neighbor | Locate the boldfaced with value 99.25 number and return the number immediately to its left. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "29.54" |
| 5 | HB_syn_000881 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 10, column 1, list all numbers highlighted with the same color (mid_yellow, hex #F7DA67). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["8.08", "98.59", "2.35", "89.17", "6.54", "96.14"] |
| 6 | HB_syn_000882 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "18" |
| 7 | HB_syn_000883 | Consistency & Missingness | Cell Missingness | Is the cell (Method-F, MetD) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000884 | Aggregation & Comparison | Extremum Reasoning | In row System-B, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "93.83" |
| 2 | HB_syn_000885 | Aggregation & Comparison | Pairwise Comparison | Between Model-Y and Approach-D [91] for RateB, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Approach-D [91]", "value": "80.40"} |
| 3 | HB_syn_000886 | Local Relations | Neighbor Value Tracing | Locate value 8.36 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "8.36", "up": "23.07", "down": "49.62", "left": "59.67", "right": "34.72"} |
| 4 | HB_syn_000887 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 34.44. Return all values sharing its highlight color (purple/#B07CFF). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["48.47", "27.91", "73.99", "10.27", "55.06", "34.44", "23.39", "68.33"] |
| 5 | HB_syn_000888 | Markup Grounding | Marked Column Listing | List all underlined numbers in column Rate↓. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["30.44"] |
| 6 | HB_syn_000889 | Consistency & Missingness | Column Missingness | List all missing entries in column RateB (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Model-D", "Variant-A", "Variant-C [83]"] |
| 7 | HB_syn_000890 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: pink(#FFB7D6), light purple(#E593FC), purple(#B07CFF), blue(#6FA0FF). Is the number with value 67.50 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "light purple", "hex": "#E593FC"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000891 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in Qual. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "92.92" |
| 2 | HB_syn_000892 | Aggregation & Comparison | Pairwise Comparison | Within row Method-A, choose the larger between MetD under Sect-E and Idx under Slot-A, output {item,value}. Exclude N/A and —. | {"item": "MetD under Sect-E", "value": "51.61"} |
| 3 | HB_syn_000893 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=8,c=9), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "25.59", "up": "43.43", "down": "N/A", "left": "80.32", "right": "55.53"} |
| 4 | HB_syn_000894 | Consistency & Missingness | Cell Missingness | Is the cell (Variant-D, Rank under Set-D) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 5 | HB_syn_000895 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are < 76.04. Output JSON array of strings. | ["5.90", "55.85", "51.61", "27.80", "34.34", "72.88", "67.12", "63.27", "49.31", "36.43", "66.17", "32.15", "67.81", "62.46", "2.45", "68.07", "11.41", "43.43", "2.26"] |
| 6 | HB_syn_000896 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_red(#FFD6D6), mid_red(#F86D7B), dark_red(#FC5255). Is the number at (r=1,c=7) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "dark_red", "hex": "#FC5255"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000897 | Constrained Retrieval | Cell Retrieval | What is the Idx↑ under Slot-D value for Model-B? Output only the number as a string. Example: "12.34". | "45.39" |
| 2 | HB_syn_000898 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in Err. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "4.34" |
| 3 | HB_syn_000899 | Aggregation & Comparison | Pairwise Comparison | For Model-D, compare Qual under Cfg-B and Err under Cfg-B. Which metric value is larger? Return JSON {"item":<metric_name>,"value":<string>}. Exclude N/A and —. | {"item": "Err under Cfg-B", "value": "86.71"} |
| 4 | HB_syn_000900 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in light_pink (hex #FFD1E7) with value 50.03 number and return the number immediately above it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "67.83" |
| 5 | HB_syn_000901 | Markup Grounding | Marked Value Listing | Return all numbers highlighted in light_pink (hex #FFD1E7) as a JSON array. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["36.50", "60.73", "71.63", "31.15", "67.59", "20.29", "57.65", "8.92", "50.03"] |
| 6 | HB_syn_000902 | Consistency & Missingness | Column Missingness | In MetD↑ under Slot-D, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-Z [57]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000903 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Stat under Slot-A. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "System-Y", "value": "89.63"} |
| 2 | HB_syn_000904 | Local Relations | Neighbor Indexing | At (r=3,c=2) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "34.89", "up": "36.22", "down": "89.63", "left": "61.83", "right": "23.12"} |
| 3 | HB_syn_000905 | Markup Grounding | Marked Value Listing | Give a JSON array of all underline-marked numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["7.52", "79.38", "6.49", "97.03", "87.43", "+2.30 ±0.17"] |
| 4 | HB_syn_000906 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "6" |
| 5 | HB_syn_000907 | Consistency & Missingness | Column Missingness | Return the items whose Stat↑ under Part-B is missing (N/A or —). Output JSON array of strings. | ["Model-Z", "Method-Z", "Model-E"] |
| 6 | HB_syn_000908 | Constrained Retrieval | Value Filtering | In column Stat↑ under Part-B, list all row labels in the leftmost row-label column with value >= 48.19. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-D", "value": "63.66"}, {"item": "System-Y", "value": "57.71"}, {"item": "Baseline-C", "value": "59.70"}, {"item": "Variant-E", "value": "66.99"}, {"item": "Variant-D", "value": "76.04"}, {"item": "Baseline-A", "value": "79.38"}, {"item": "Approach-Z", "value": "99.33"}, {"item": "Approach-Y [10]", "value": "74.10"}, {"item": "Method-A", "value": "48.19"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000909 | Aggregation & Comparison | Extremum Reasoning | Find the lowest metric value for Model-Z. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "41.63" |
| 2 | HB_syn_000910 | Local Relations | Neighbor Value Tracing | At the cell with value 89.82, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "89.82", "up": "64.28", "down": "25.41", "left": "92.56", "right": "91.42"} |
| 3 | HB_syn_000911 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=4,c=6). Return all values with the same highlight color (mid_cyan_blue, #95D9FF). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["94.17", "7.69", "30.67", "3.47", "77.15", "94.95", "92.56"] |
| 4 | HB_syn_000912 | Markup Grounding | Marker Counting | How many highlighted cells are in the table? Output only the count as a string. Example: "7". | "21" |
| 5 | HB_syn_000913 | Consistency & Missingness | Cell Missingness | Is the cell (Baseline-D, Stat↓) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000914 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column S-D? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "0.01" |
| 2 | HB_syn_000915 | Aggregation & Comparison | Extremum Reasoning | Across all columns for Approach-E, what is the maximum? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "93.47" |
| 3 | HB_syn_000916 | Local Relations | Neighbor Value Tracing | Locate value 43.03 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "43.03", "up": "30.05", "down": "59.72", "left": "44.01", "right": "99.74"} |
| 4 | HB_syn_000917 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 60.24. Return all values sharing its highlight color (orange/#FF8C3A). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["60.24", "11.11", "3.88", "11.94"] |
| 5 | HB_syn_000918 | Consistency & Missingness | Cell Missingness | Check whether Approach-E at Qual↓ under Slot-A is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| 6 | HB_syn_000919 | Markup Grounding | Highlight Presence | Check whether value 53.31 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). The value appears exactly once in the numeric data grid. | {"highlighted": true, "color": "red", "hex": "#FF5A5C"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000920 | Aggregation & Comparison | Pairwise Comparison | In row System-D, is MetC under Sect-D higher or Val under Sect-B higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "MetC under Sect-D", "value": "80.98"} |
| 2 | HB_syn_000921 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=5,c=3), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "78.84", "up": "70.15", "down": "14.91", "left": "25.69", "right": "6.44"} |
| 3 | HB_syn_000922 | Markup Grounding | Marked Column Listing | In MetB under Grp-A, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["83.98"] |
| 4 | HB_syn_000923 | Consistency & Missingness | Column Missingness | Return the items whose Stat under Blk-C is missing (N/A or —). Output JSON array of strings. | ["Baseline-D [98]"] |
| 5 | HB_syn_000924 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those < 14.91 as a JSON array of strings. | ["7.92", "3.46", "8.79", "9.03", "4.94", "11.98", "8.84"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000925 | Constrained Retrieval | Cell Retrieval | In the MetB↑ under Set-B column, read the entry for Model-C. Output only the number as a string. Example: "12.34". | "75.36" |
| 2 | HB_syn_000926 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Score under Set-B in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "System-A", "value": "36.26"}, {"item": "Baseline-E", "value": "22.20"}, {"item": "Baseline-F [94]", "value": "16.50"}] |
| 3 | HB_syn_000927 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (down) to the highlighted in dark_yellow (hex #EFBC23) with value 36.26 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "0.26" |
| 4 | HB_syn_000928 | Markup Grounding | Marked Value Listing | Which values are highlighted in mid_yellow (#F7DA67)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["62.21", "86.96", "75.36", "20.67", "75.93", "18.09", "12.30", "68.44", "22.20", "82.52", "12.13", "86.99", "53.71"] |
| 5 | HB_syn_000929 | Consistency & Missingness | Cell Missingness | Is the cell (Approach-A, Rate↓ under Part-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000930 | Aggregation & Comparison | Pairwise Comparison | Within row Method-Y, choose the larger between MetA under Set-B and Idx under Zone-D, output {item,value}. Exclude N/A and —. | {"item": "MetA under Set-B", "value": "45.76"} |
| 2 | HB_syn_000931 | Local Relations | Marker-Anchor Neighbor | Locate the highlighted in pink (hex #FFB7D6) with value 39.26 number and return the number immediately above it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "85.83" |
| 3 | HB_syn_000932 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "12" |
| 4 | HB_syn_000933 | Consistency & Missingness | Column Missingness | In MetA under Set-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Variant-E [67]"] |
| 5 | HB_syn_000934 | Constrained Retrieval | Value Filtering | Filter F1↑ under Zone-D by > 91.39. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-Y", "value": "92.12"}, {"item": "Variant-E [67]", "value": "96.31"}, {"item": "Baseline-B [33]", "value": "96.89"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000935 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in S-D under Cfg-A and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "System-Y", "value": "70.66"} |
| 2 | HB_syn_000936 | Local Relations | Marker-Anchor Neighbor | Locate the boldfaced with value 8.45 number and return the number immediately below it. The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "47.52" |
| 3 | HB_syn_000937 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 27.34. Return all values sharing its highlight color (red/#FF5A5C). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["27.34"] |
| 4 | HB_syn_000938 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number at (r=3,c=1) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": true, "color": "red", "hex": "#FF5A5C"} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000939 | Aggregation & Comparison | Pairwise Comparison | Between Baseline-F and Approach-F for S-C under Blk-B, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Baseline-F", "value": "80.94"} |
| 2 | HB_syn_000940 | Aggregation & Comparison | Pairwise Comparison | In row Approach-F, is F1 under Slot-C higher or S-C under Blk-B higher? Return {item,value} where item is the winning metric. Exclude N/A and —. | {"item": "F1 under Slot-C", "value": "44.42"} |
| 3 | HB_syn_000941 | Local Relations | Neighbor Value Tracing | Locate value 4.37 in the numeric part of the table and return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "4.37", "up": "43.32", "down": "25.59", "left": "55.12", "right": "44.42"} |
| 4 | HB_syn_000942 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "10" |
| 5 | HB_syn_000943 | Consistency & Missingness | Column Missingness | In Val under Slot-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Baseline-Y"] |
| 6 | HB_syn_000944 | Constrained Retrieval | Value Filtering | Take all highlighted numbers (any color) and return those < 59.26 as a JSON array of strings. | ["23.03", "19.28", "25.98", "0.37", "14.39"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000945 | Aggregation & Comparison | Argmax Reasoning | In MetD under Zone-A, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 9, "col": 2, "value": "98.48"} |
| 2 | HB_syn_000946 | Aggregation & Comparison | Rank Selection | In Cost under Zone-A, who are the top 3 (highest to lowest)? Output JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-B", "value": "88.04"}, {"item": "System-A [89]", "value": "74.51"}, {"item": "Variant-C", "value": "69.28"}] |
| 3 | HB_syn_000947 | Local Relations | Neighbor Indexing | At (r=10,c=6) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "53.01", "up": "98.97", "down": "15.91", "left": "18.36", "right": "45.38"} |
| 4 | HB_syn_000948 | Constrained Retrieval | Color-Anchor Retrieval | Take the cell at (r=1,c=2). Return all values with the same highlight color (light purple, #E593FC). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["97.45", "+2.07 ±0.61"] |
| 5 | HB_syn_000949 | Markup Grounding | Marked Value Listing | List all underlined numbers. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["95.08", "77.52", "2.99", "92.82", "99.81", "88.04", "+2.45 ±0.80", "98.48", "96.39", "96.68"] |
| 6 | HB_syn_000950 | Consistency & Missingness | Cell Missingness | Is the cell (System-Y, Cost↑ under Cfg-C) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000951 | Aggregation & Comparison | Extremum Reasoning | In column MetA, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "90.96" |
| 2 | HB_syn_000952 | Consistency & Missingness | Cell Missingness | Is the cell (System-A, S-C↑ under Set-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 3 | HB_syn_000953 | Constrained Retrieval | Value Filtering | In column Val↑ under Slot-B, list all row labels in the leftmost row-label column with value < 53.98. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "System-A", "value": "22.18"}, {"item": "Baseline-Y", "value": "27.58"}, {"item": "Model-E", "value": "29.22"}, {"item": "Variant-X [42]", "value": "12.41"}, {"item": "Model-Z [88]", "value": "31.87"}] |
| 4 | HB_syn_000954 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number at (r=9,c=7) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000955 | Constrained Retrieval | Cell Retrieval | For item Method-A, what is the value in column MetC↑ under Zone-C? Output only the number as a string. Example: "12.34". | "66.99" |
| 2 | HB_syn_000956 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest Acc under Grp-B? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Variant-F [15]", "value": "98.63"} |
| 3 | HB_syn_000957 | Local Relations | Neighbor Value Tracing | At the cell with value 5.35, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "5.35", "up": "0.60", "down": "N/A", "left": "99.32", "right": "37.10"} |
| 4 | HB_syn_000958 | Markup Grounding | Marked Value Listing | Find all light_teal-highlighted values (hex #D3FAEE) and list them. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["30.70", "61.30", "80.67", "27.86", "51.63", "92.18", "41.36"] |
| 5 | HB_syn_000959 | Markup Grounding | Marked Column Listing | In Acc under Grp-B, return the values that are underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["86.09"] |
| 6 | HB_syn_000960 | Consistency & Missingness | Column Missingness | List all missing entries in column Idx under Slot-A (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Approach-Y", "Variant-B"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000961 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Qual under Part-B and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Baseline-A", "value": "76.05"} |
| 2 | HB_syn_000962 | Constrained Retrieval | Color-Anchor Retrieval | For the number at row 7, column 6, list all numbers highlighted with the same color (light_yellow, hex #FEF4B8). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["41.11", "74.96", "55.26", "20.27", "38.42", "29.17", "77.34"] |
| 3 | HB_syn_000963 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "21" |
| 4 | HB_syn_000964 | Consistency & Missingness | Cell Missingness | Check whether Method-X at Qual↓ under Sect-C is missing (N/A or —). Output JSON {"missing":true/false}. | {"missing": false} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000965 | Aggregation & Comparison | Extremum Reasoning | In row Approach-X, what is the smallest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "11.18" |
| 2 | HB_syn_000966 | Aggregation & Comparison | Extremum Reasoning | In row Variant-B [51], return the greatest numeric entry. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "91.37" |
| 3 | HB_syn_000967 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=15,c=5), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "85.09", "up": "64.33", "down": "46.39", "left": "10.57", "right": "12.37"} |
| 4 | HB_syn_000968 | Constrained Retrieval | Color-Anchor Retrieval | For the number with value 99.39, list all numbers highlighted with the same color (dark_purple, hex #8537EB). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["96.84", "0.06", "93.23", "2.91", "98.86", "99.39"] |
| 5 | HB_syn_000969 | Markup Grounding | Marker Counting | Count all underline-marked cells and output the count as a string. Example: "5". | "6" |
| 6 | HB_syn_000970 | Consistency & Missingness | Column Missingness | Return the items whose Eff↑ under Cfg-C is missing (N/A or —). Output JSON array of strings. | ["Method-A"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000971 | Aggregation & Comparison | Argmax Reasoning | Find the row label with the maximum value in column Qual under Part-D. Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Model-X", "value": "58.91"} |
| 2 | HB_syn_000972 | Local Relations | Neighbor Value Tracing | At the cell with value 95.67, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "95.67", "up": "72.80", "down": "72.34", "left": "16.81", "right": "71.49"} |
| 3 | HB_syn_000973 | Consistency & Missingness | Column Missingness | In MetD↑ under Grp-A, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Model-B"] |
| 4 | HB_syn_000974 | Constrained Retrieval | Value Filtering | Among highlighted cells (any color), list the values that are > 60.29. Output JSON array of strings. | ["62.37", "70.54", "69.99", "69.67", "68.27"] |
| 5 | HB_syn_000975 | Markup Grounding | Highlight Presence | Check whether value 39.12 is highlighted. Return JSON {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Palette: light_blue(#E6F2FF), blue(#6DA1FD), light_orange(#FFE6C7), orange(#F9B163). The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000976 | Aggregation & Comparison | Argmax Reasoning | In Cost, where is the maximum located? Return JSON with row, col, and value. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Exclude N/A and —. | {"row": 12, "col": 5, "value": "95.01"} |
| 2 | HB_syn_000977 | Aggregation & Comparison | Rank Selection | Find the 2nd highest entry in Cost and return JSON {"item":"...","value":"..."}. Exclude N/A and —. | {"item": "Variant-B", "value": "90.43"} |
| 3 | HB_syn_000978 | Aggregation & Comparison | Pairwise Comparison | Between Approach-Z and System-D for S-C, pick the larger value and return {item,value}. Exclude N/A and —. | {"item": "Approach-Z", "value": "72.77"} |
| 4 | HB_syn_000979 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (right) to the underlined with value 92.58 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "85.99" |
| 5 | HB_syn_000980 | Markup Grounding | Marked Value Listing | Which values are highlighted in orange (#FF8C3A)? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["96.07", "60.36", "+2.30 ±0.28", "52.52"] |
| 6 | HB_syn_000981 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["90.43", "92.58", "12.23", "+2.30 ±0.28", "98.88", "91.62"] |
| 7 | HB_syn_000982 | Consistency & Missingness | Cell Missingness | Is the cell (Method-Y, S-A↓) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": false} |
| 8 | HB_syn_000983 | Constrained Retrieval | Value Filtering | In column Rank, list all row labels in the leftmost row-label column with value < 75.54. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Method-Y", "value": "37.10"}, {"item": "Model-Y [52]", "value": "6.75"}, {"item": "Baseline-B", "value": "73.24"}, {"item": "Approach-Z", "value": "60.00"}, {"item": "Method-D [32]", "value": "59.71"}, {"item": "System-A [80]", "value": "62.93"}, {"item": "Variant-D [48]", "value": "10.08"}, {"item": "Approach-X", "value": "66.29"}, {"item": "System-Y", "value": "70.99"}] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000984 | Constrained Retrieval | Cell Retrieval | Return the value where Variant-E intersects Eff. Output only the number as a string. Example: "12.34". | "95.15" |
| 2 | HB_syn_000985 | Aggregation & Comparison | Extremum Reasoning | Find the highest entry in Val. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "93.98" |
| 3 | HB_syn_000986 | Local Relations | Neighbor Indexing | At (r=3,c=4) in the numeric part of the table, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "50.59", "up": "59.91", "down": "31.54", "left": "76.32", "right": "47.30"} |
| 4 | HB_syn_000987 | Markup Grounding | Marker Counting | Count the number of colored/highlighted cells. Output only the count as a string. Example: "7". | "15" |
| 5 | HB_syn_000988 | Consistency & Missingness | Cell Missingness | Is the cell (Model-D, Val) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000989 | Aggregation & Comparison | Extremum Reasoning | In row Variant-Y, what is the largest number? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "90.34" |
| 2 | HB_syn_000990 | Aggregation & Comparison | Extremum Reasoning | For item Model-A, what is the minimum value across all metrics? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "24.89" |
| 3 | HB_syn_000991 | Local Relations | Neighbor Indexing | For the the numeric part of the table cell at (r=8,c=4), return the target value and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. | {"target": "31.24", "up": "22.94", "down": "95.97", "left": "90.80", "right": "28.79"} |
| 4 | HB_syn_000992 | Constrained Retrieval | Color-Anchor Retrieval | Find the cell containing 11.28. Return all values sharing its highlight color (orange/#FF8C3A). Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["11.28"] |
| 5 | HB_syn_000993 | Consistency & Missingness | Column Missingness | In Err↑, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Baseline-B"] |
| 6 | HB_syn_000994 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: red(#FF5A5C), orange(#FF8C3A), yellow(#FFE9A3). Is the number with value 54.71 highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. The value appears exactly once in the numeric data grid. | {"highlighted": false, "color": null, "hex": null} |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_000995 | Aggregation & Comparison | Argmax Reasoning | Which item (row label in the leftmost row-label column) achieves the highest Val under Set-D? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Approach-B [14]", "value": "97.85"} |
| 2 | HB_syn_000996 | Local Relations | Neighbor Value Tracing | At the cell with value 37.47, return the target and its four neighbors (up, down, left, right) as JSON {"target":"12.34","up":"11.11","down":"56.78","left":"N/A","right":"90.12"}. The numeric part of the table includes numeric/data cells only (no headers; no left header columns). If a neighbor cell contains N/A or —, return that token string. The target value appears exactly once in the numeric part of the table. | {"target": "37.47", "up": "11.66", "down": "17.95", "left": "13.66", "right": "69.52"} |
| 3 | HB_syn_000997 | Constrained Retrieval | Color-Anchor Retrieval | Using (r=5,c=6) as an anchor, list all cells sharing its highlight color (dark_purple/#8537EB). Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["93.12", "1.17", "2.80", "2.19", "8.40", "97.00", "97.85", "11.66", "83.49", "1.14"] |
| 4 | HB_syn_000998 | Markup Grounding | Marker Counting | How many underlined numbers are in the table? Output only the count as a string. Example: "5". | "10" |
| 5 | HB_syn_000999 | Consistency & Missingness | Column Missingness | List all missing entries in column Val under Set-D (N/A or —). Output only a JSON array of strings. Example: ["System-A","System-B"]. | ["Model-X [27]"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_001000 | Aggregation & Comparison | Extremum Reasoning | What is the maximum value in column Score? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "81.15" |
| 2 | HB_syn_001001 | Aggregation & Comparison | Extremum Reasoning | What is the minimum value in column Score? Exclude N/A and —. Output only the number as a string. Example: "12.34". | "14.47" |
| 3 | HB_syn_001002 | Aggregation & Comparison | Pairwise Comparison | On metric Score under Sect-A, which is higher: Model-B or Method-F? Return JSON {"item":"<row label in the leftmost row-label column>","value":"<cell value>"}. Exclude N/A and —. | {"item": "Model-B", "value": "81.15"} |
| 4 | HB_syn_001003 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (up) to the underlined with value 77.95 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "1.02" |
| 5 | HB_syn_001004 | Markup Grounding | Marked Value Listing | Return every value that is underlined. Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["61.13", "63.50", "77.95", "59.16", "97.52", "+1.41 ±0.24", "15.60"] |
| 6 | HB_syn_001005 | Consistency & Missingness | Cell Missingness | Is the cell (Method-F, Val↑ under Sect-A) missing (N/A or —)? Answer JSON {"missing":true/false}. | {"missing": true} |
| 7 | HB_syn_001006 | Constrained Retrieval | Value Filtering | Among highlighted standalone numeric values (any color), list the values that are >= 52.32. Within the highlighted cells, exclude entries written as signed change/uncertainty values, such as "+1.15 ± 0.63". Output only a JSON array of strings. | ["52.32", "82.65", "81.15"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_001007 | Constrained Retrieval | Cell Retrieval | Give the number at (Variant-A, Eff under Zone-B). Output only the number as a string. Example: "12.34". | "65.89" |
| 2 | HB_syn_001008 | Aggregation & Comparison | Pairwise Comparison | Within row Approach-D, choose the larger between Rate under Grp-A and Eff under Zone-B, output {item,value}. Exclude N/A and —. | {"item": "Rate under Grp-A", "value": "66.23"} |
| 3 | HB_syn_001009 | Markup Grounding | Marked Column Listing | Which entries in column ΔDelta↑ under Cfg-C are underlined? Output only a JSON array of strings. Example: ["12.34","56.78"]. | ["+2.01 ±0.38"] |
| 4 | HB_syn_001010 | Consistency & Missingness | Column Missingness | In Eff under Zone-B, which rows have missing values (N/A or —)? Return a JSON array of row names. Example: ["System-A","System-B"]. | ["Method-B"] |
| # | qid | task | subtask | question | answer |
|---|---|---|---|---|---|
| 1 | HB_syn_001011 | Aggregation & Comparison | Extremum Reasoning | Find the lowest entry in Score. Exclude N/A and —. Output only the number as a string. Example: "12.34". | "15.00" |
| 2 | HB_syn_001012 | Aggregation & Comparison | Rank Selection | Return the 3 highest rows in Score in descending order as a JSON array of {"item":"...","value":"..."} objects. Exclude N/A and —. | [{"item": "Model-Z", "value": "78.69"}, {"item": "Approach-Y", "value": "76.59"}, {"item": "Method-C", "value": "73.88"}] |
| 3 | HB_syn_001013 | Local Relations | Marker-Anchor Neighbor | What is the value adjacent (right) to the highlighted in mid_teal (hex #83EDC4) with value 76.59 cell? The requested neighbor is guaranteed to be a numeric value (not N/A or —). Output only the number as a string. Example: "12.34". | "91.37" |
| 4 | HB_syn_001014 | Constrained Retrieval | Value Filtering | Filter Score by > 48.19. Return a JSON array of {"item":"<row label in the leftmost row-label column>","value":"<cell value>"} objects. Exclude N/A and —. | [{"item": "Approach-Y", "value": "76.59"}, {"item": "Method-C", "value": "73.88"}, {"item": "Model-Z", "value": "78.69"}, {"item": "Approach-A", "value": "61.94"}] |
| 5 | HB_syn_001015 | Markup Grounding | Highlight Presence | In this table, some numbers are highlighted using the following colors: light_teal(#D3FAEE), mid_teal(#83EDC4), dark_teal(#22D28E). Is the number at (r=3,c=2) highlighted? Answer in JSON: {"highlighted":true/false,"color":<color_name or null>,"hex":<hex or null>}. Example: {"highlighted":true,"color":"light_blue","hex":"#CFE0FF"}. Row/col indices refer to the numeric data grid only; exclude all headers/section headers and left header columns. | {"highlighted": false, "color": null, "hex": null} |