sample_idx int64 0 49 | model_a stringclasses 3 values | model_b stringclasses 3 values | winner stringclasses 3 values | reason stringlengths 119 777 | agreement stringclasses 1 value |
|---|---|---|---|---|---|
0 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Both outputs are faithful, complete, and accurate. Output B preserves the original spacing between 'Cab. 46' and '134 items' more precisely, which is a minor but correct formatting detail. Output A uses a single space, which is less accurate to the source. Therefore, B is slightly better in formatting fidelity. | 1/1 |
0 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A is better because it maintains the original line breaks and spacing, preserving the document's layout more accurately. Both outputs are faithful, complete, and accurate, but A's formatting is cleaner and more faithful to the source. Output B incorrectly combines 'Cab. 46' and '134 items' on the same line, which alters the document's structure. | 1/1 |
0 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | TIE | Both outputs are identical in content, accuracy, completeness, and formatting. They faithfully reproduce all visible text from the document without any errors or omissions. | 1/1 |
0 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | B | Output A is better because it maintains the original line breaks and spacing, preserving the document's layout more accurately. Both outputs are faithful, complete, and accurate, but A's formatting is cleaner and more faithful to the source image's structure. | 1/1 |
0 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Output A correctly preserves the spacing between 'Cab. 46' and '134 items', which appears as two distinct fields on the document. Output B merges them into a single line without the space, altering the document's layout. Both are faithful and complete, but A's formatting is more accurate to the source. | 1/1 |
0 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output A is better because it maintains the original line breaks and spacing, preserving the document's layout more accurately. Both outputs are faithful, complete, and accurate, but A's formatting is cleaner and more faithful to the source. Output B incorrectly combines 'Cab. 46' and '134 items' on the same line, which alters the document's structure. | 1/1 |
1 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is more faithful and complete as it includes the header text 'Schenck, H. F. 2' at the top of the page, which is clearly visible in the image. Output A omits this header, making it incomplete. Both outputs are accurate in the text they share, but B's completeness gives it the advantage. | 1/1 |
1 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A includes the header 'Schenck, H. F.' and the page number '2', which are visible in the document image. Output B omits these elements, making it incomplete. Both outputs are accurate in their text content and reading order, but A is more complete, satisfying the completeness criterion better than B. | 1/1 |
1 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output A is better because it correctly includes the author's name 'Schenck, H. F.' and the page number '2' at the top, which are visible in the document. Output B omits this information, making it incomplete. Both outputs are accurate in the rest of the text, but A's completeness gives it the advantage. Additionally, A's note about the document ending mid-sentence is a valid observation about the source, not an addition of text, and does not violate the faithfulness criterion as it is clearly marked as a note. | 1/1 |
1 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | B | Output A correctly preserves the underlined formatting of the book titles using markdown asterisks, which is a more accurate representation of the original document's emphasis. Output B omits this formatting, making it less faithful to the source. Both outputs are complete and accurate in terms of text content and reading order, but A's attention to formatting details gives it a slight edge in faithfulness. | 1/1 |
1 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Output B is more faithful to the original document's layout and text. It accurately preserves the line breaks and spacing as they appear in the image, including the two spaces after 'Schenck, H. F.' and the line breaks between sentences. Output A introduces formatting elements like italics and a note about the document ending mid-sentence, which are not part of the original text and violate the faithfulness criterion. While both outputs are complete and accurate, B's strict adherence to the visual structure makes it superior. | 1/1 |
1 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output A is better because it faithfully reproduces the document's text without adding interpretive commentary. Output B includes a note about the document ending mid-sentence and the black circle being a redaction, which is not part of the document's text and violates the faithfulness criterion. Both outputs are complete and accurate, but A adheres strictly to the rule of only including text from the document. | 1/1 |
2 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output A preserves the original spacing and layout more accurately, particularly the alignment of 'F- 3055' on the same line as the name, which is visually present in the document. Output B merges the name and code into a single line without the space, which is a minor formatting error. Both are faithful and complete, but A is more precise in its representation of the document's structure. | 1/1 |
2 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A is better because it preserves the original document's line breaks and spacing, which improves readability and accurately reflects the document's structure. Output B incorrectly merges multiple lines into a single line, losing the visual separation between the catalog number (F- 3055) and the author's name, which is a formatting error. Both outputs are complete and accurate in terms of text content, but A's formatting is more faithful to the source. | 1/1 |
2 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more accurate in its representation of the document's layout. The text 'F-3055' is correctly placed on a separate line, directly to the right of 'Munford, George Wythe', which matches the visual structure of the document. Output A incorrectly combines 'Munford, George Wythe' and 'F-3055' on the same line, which is a formatting error. Both outputs are complete and accurate in their text content, but B's formatting is superior. | 1/1 |
2 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output A correctly preserves the original layout and reading order, placing the identifier 'F- 3055' on the same line as the name 'Munford, George Wythe', which matches the document's visual structure. Output B separates these elements, disrupting the intended formatting. Both outputs are complete and accurate, but A's superior formatting fidelity makes it the better extraction. | 1/1 |
2 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Output B is more accurate in preserving the original spacing and layout of the text, particularly the alignment of 'F- 3055' and the spacing between '186 pp.', 'Boards & Calf', and '40 x 17 cm.' which reflects the document's structure more faithfully. Output A incorrectly separates 'F- 3055' onto a new line, which is a formatting error. Both outputs are complete and accurate in content, but B's formatting is superior. | 1/1 |
2 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output A is more faithful to the document's layout, preserving the line breaks and spacing as they appear, which improves readability and accurately reflects the document's structure. Output B combines multiple lines into a single line, losing the visual separation between '186 pp.', 'Boards & Calf', and '40 x 17 cm.', which is a formatting error. Both outputs are complete and accurate in terms of text content, but A's formatting is superior. | 1/1 |
3 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | A | Output A is more complete and accurate. It correctly captures the text 'Newbern North Carolina' and 'accessioned', while Output B contains errors: it incorrectly transcribes 'Newbern' as 'Northern' and 'accessioned' as 'accessed'. Additionally, Output B includes 'd5' which is a marginal note, but it omits the 'Treasure Room' header present in the image. Output A correctly includes all visible text in the correct reading order without adding extraneous information. | 1/1 |
3 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A is more complete as it includes all visible text from the document, including the header 'Treasure Room' and the marginalia 'S-329', 'TS-1', and 'dS'. Output B omits these elements, making it incomplete. Both outputs are accurate in the text they include, but A's completeness makes it superior. | 1/1 |
3 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | B captures all visible text, including marginalia and stamps (Treasure Room, S-329, TS-1, ds), which A omits. A also incorrectly omits the 'accessioned' label and the date is split incorrectly. B is more complete and faithful to the document. | 1/1 |
3 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | B | Output B is more accurate and complete. It correctly transcribes the handwritten 'S-329' and 'TS-1' in the top right corner, which Output A omits. It also correctly reads 'accessioned' and the lowercase 'dS' at the bottom right, whereas Output A has 'accessed' and 'd5'. Output B also correctly identifies 'Newbern' as the location, while Output A has 'Northern'. All text in B is accurate and complete, making it the better extraction. | 1/1 |
3 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output A is more accurate and complete. It correctly transcribes 'Newbern' and 'accessioned' from the document, while Output B contains errors: 'Northern' instead of 'Newbern' and 'accessed' instead of 'accessioned'. Output A also correctly captures the marginalia 'S-329', 'TS-1', and 'ds', which are present in the image. Output B incorrectly transcribes 'd5' instead of 'ds'. | 1/1 |
3 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output A is more complete and accurate. It correctly includes the handwritten 'S' in the top left corner, which is missing in Output B. Output A also preserves the correct reading order of the text, including the placement of the 'S-329' and 'TS-1' codes in the top right, which are correctly positioned relative to the main content. Output B incorrectly places 'S-329' and 'TS-1' as separate lines at the top, disrupting the document's layout and potentially misrepresenting the information. Additionally, Output A correctly captures the handwritten 'ds' in the bottom right, while Output B shows 'dS' with a capital S, which is a minor but incorrect alteration. The core content is identical, but A's superior handling of marginalia and layout makes it the better extraction. | 1/1 |
4 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is more accurate as it correctly preserves the line breaks and hyphenation of the original text, such as 'scar- city' and 'consump- tion', which are visually present in the document. Output A incorrectly merges these lines, losing the original formatting. Both outputs are faithful and complete, but B's adherence to the document's layout makes it superior in accuracy and readability. | 1/1 |
4 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A is better because it accurately preserves the line breaks and hyphenation of the original text, such as 'scar- city' and 'consump- tion', which are correctly represented. Output B incorrectly merges these lines, losing the original formatting and potentially altering the reading. Both outputs are faithful and complete, but A's handling of the text layout is more accurate. | 1/1 |
4 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | A | Output B is more accurate as it correctly preserves the original line breaks and word spacing, particularly in the phrase 'scarcity of textile goods for private consumption' which is split incorrectly in Output A. Output A also has a minor formatting issue with the header line spacing. Both capture all text, but B's representation is more faithful to the document's layout. | 1/1 |
4 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | TIE | Both outputs are identical in content, accuracy, completeness, and formatting. They faithfully reproduce all text from the document, including the header, body, and the date stamp, with no omissions or errors. | 1/1 |
4 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more accurate in preserving the original formatting and punctuation of the document. It correctly includes the period after '7 pieces' and maintains the line breaks and hyphenation as they appear in the image, which is more faithful to the source. Output A omits the period and has slightly different line breaks, making B the better transcription. | 1/1 |
4 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output B is more accurate as it correctly includes the period after '7 pieces' which is visible in the original document. Output A omits this punctuation mark, making it less faithful to the source. Both outputs are otherwise identical in content, accuracy, and formatting. | 1/1 |
5 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is more complete as it includes the header 'Field-Musgrave Family' and the page number '56', which are visible in the document. Output A omits this information, making it less faithful to the original document. Both outputs are accurate in the text they share, but B's completeness makes it superior. | 1/1 |
5 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output B is better because it accurately preserves the line breaks and formatting of the original document, which is a key aspect of faithful representation. Both outputs are complete and accurate, but B's formatting more closely reflects the visual structure of the source. Output A incorrectly combines lines, losing the original paragraph structure. | 1/1 |
5 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more complete as it includes the header 'Field-Musgrave Family 56' which is visible at the top of the document. Output A omits this header, making it less complete. Both outputs are accurate in their text content and follow the correct reading order. Therefore, B is better due to completeness. | 1/1 |
5 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output A correctly includes the header 'Field-Musgrave Family' and the page number '56', which are visible at the top of the document. Output B omits this header and page number, making it incomplete. Both outputs are accurate in the text they share, but A is more complete and faithful to the original document. | 1/1 |
5 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output A correctly includes the header 'Field-Musgrave Family 56' as it appears in the document, while Output B omits the page number '56'. Both outputs are otherwise identical in content, accuracy, and reading order. However, A is more complete as it captures all visible text, including the marginal number, which is a key part of the document's structure. | 1/1 |
5 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output B is more complete as it includes the header 'Field-Musgrave Family 56' which is visible at the top of the document. Output A omits this header, making it incomplete. Both outputs are accurate in their text content and reading order, but B's completeness gives it the advantage. | 1/1 |
6 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is more faithful and complete as it includes the header text 'Family' and the line 'Hemphill, A WXXX XXXXXXXX and XXX RXXX' which are visible in the document. Output A omits this header information, making it incomplete. Both outputs are accurate in the text they share, but B's inclusion of all visible text makes it the better extraction. | 1/1 |
6 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output B is more faithful and complete as it includes the header text 'Family', the name 'Hemphill' with the obscured characters, and the number '35' which are all visible in the document. Output A omits these elements, making it incomplete. While Output B's OCR of the obscured text is not perfect, it is a more accurate representation of the visual content than Output A's omission. Therefore, B is the better extraction overall. | 1/1 |
6 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B correctly includes the first line of text 'Hemphill, W. R. Cathcart, and S. J. Barrows.' which is present in the image, while Output A omits it. Output A also contains a typographical error, missing the 'H' in 'Hemphill' at the beginning. Output B is more complete and accurate, capturing all visible text correctly. | 1/1 |
6 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output A is more accurate and complete. It correctly transcribes the name 'Hemphill, A WXXX XXXXXXXX and XXX RXXX' from the top line, which is partially and incorrectly rendered as 'Hemphill, WXXXIXXXIXXXIXXXIXXX' in Output B. Output A also correctly includes the number '35' on the right side of the header, which is missing in Output B. Both outputs are otherwise similar in structure and content, but A's superior accuracy and completeness make it the better extraction. | 1/1 |
6 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Output A includes the header text 'Family' and the line 'Hemphill, A WXXX XXXXXXXX and XXX RXXX' which are visible at the top of the document, making it more complete. Output B omits these lines, resulting in incomplete text capture. Both outputs are accurate in the text they include, but A is more faithful to the original document's content. | 1/1 |
6 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output A is more accurate and complete. It correctly transcribes the text 'Hemphill, W. R. Cathcart, and S. J. Barrows' and the full names of the other writers, whereas Output B contains garbled text 'Hemphill, WXXXIXXXIXXXIXXX' and omits the 'R.' in Cathcart. Output A also correctly includes the 'Family' header and the number '35' which are visible in the image, while Output B separates them incorrectly. Output A maintains the correct reading order and is more faithful to the original document. | 1/1 |
7 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output A is better because it includes the header text at the top of the document, which is visible in the image. Output B omits this header, making it incomplete. Both outputs are accurate in their text and reading order, but completeness is a higher priority criterion. | 1/1 |
7 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A includes the header text at the top of the document, which is visible in the image and is part of the complete record. Output B omits this header, making it incomplete. While both outputs are accurate in their shared content, A is more complete and faithful to the original document. | 1/1 |
7 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output A is better because it includes the header text at the top of the document, which is visible in the image. Output B omits this header, making it incomplete. Both outputs are accurate in their text and reading order, but completeness is a higher priority criterion. | 1/1 |
7 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output B correctly transcribes 'Lisetta' while Output A incorrectly transcribes it as 'Lissetta'. This is a factual error in the name, making B more accurate. Both outputs are complete, faithful, and correctly ordered, but B is superior in accuracy. | 1/1 |
7 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Both outputs are nearly identical, but Output A includes the header text 'FRIES, Henry Elias. Letters. 1869-1884. Salem, North Carolina. 29 pieces. Sketch.' which is present in the image and is part of the complete document. Output B omits this header, making it incomplete. Therefore, A is better due to completeness. | 1/1 |
7 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output B correctly transcribes 'Lissetta' while Output A has a typo 'Lisetta'. Both outputs are faithful, complete, and accurate in all other aspects, but B is more accurate in this specific detail. | 1/1 |
8 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | A | Output A is more accurate and faithful to the document. It correctly transcribes 'copy' instead of the misspelled 'opp' in Output B. It also correctly captures the date as 'APR 15 1936' with the number 15, whereas Output B incorrectly separates the digits as 'APR 1 5 1936'. Output A also includes the header 'Green, Duff' which is present in the image, making it more complete. While both outputs have the same reading order and formatting, A's superior accuracy and completeness make it the better extraction. | 1/1 |
8 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | A | Both outputs are nearly identical, but B is slightly better because it correctly preserves the line break after 'Falmouth, Va.' and before '304', which matches the document's layout more accurately. A adds an unnecessary line break after 'Falmouth Mill' that is not present in the original. Both are faithful, complete, and accurate, but B has a minor formatting advantage in reading order and structure. | 1/1 |
8 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output A is more complete as it includes the top-left text 'Green, Duff' and the top-right text 'L-1216', which are visible in the document. Output B omits these elements, making it less faithful to the original image. Both outputs are accurate in the text they include and maintain correct reading order, but A's completeness gives it a clear advantage. | 1/1 |
8 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | B | Output A is more accurate and complete. It correctly transcribes 'copy' instead of the misspelled 'opp' in Output B, and it correctly formats the date as 'APR 15 1936' with a space between the day and year, matching the document. Output B incorrectly includes 'Green, Duff' at the top, which is a marginal note, and fails to accurately capture the text 'copy' and the date format. | 1/1 |
8 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output A is more accurate as it correctly transcribes 'cpy' instead of 'opp' and '15' instead of '1 5', which are visible in the image. It also includes the full text from the document, including the number 'L-1216' in the top right corner, which is missing in Output B. Output A maintains the correct reading order and formatting. | 1/1 |
8 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output B is more complete as it includes all text from the document, including the header 'Green, Duff' and the marginalia 'L-1216', which are missing in Output A. Both outputs are accurate and maintain correct reading order, but B's completeness makes it superior. | 1/1 |
9 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is better because it correctly includes the header 'Tyson, Bryan' which is visible at the top of the document, making it more complete. It also accurately transcribes the date as '5:21-52' with a colon, which matches the image, whereas Output A uses a hyphen. Output B correctly captures the garbled text 'xxaxxxx' as it appears, while Output A incorrectly replaces it with 'axxxxxxx'. Output A omits the header and contains a minor error in the date format. Both outputs are incomplete at the end, but B is more faithful to the visible text. | 1/1 |
9 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | A | Output A is more accurate and faithful to the document. It correctly captures the text without adding garbled characters like 'رعانع واعده واعده واعده' in Output B. Output A also correctly includes the name 'Tyson, Bryan' at the top, which is visible in the image, while Output B omits it and instead includes the number '6' on a separate line, which is not part of the main text. Output A maintains the correct reading order and is more complete and accurate. | 1/1 |
9 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output A is better because it includes the header 'Tyson, Bryan' and the page number '6', which are visible in the document. Output B omits this information, making it incomplete. Both outputs are otherwise similar in accuracy and reading order, but completeness is a higher priority criterion. | 1/1 |
9 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output A is more accurate and faithful to the document. It correctly captures the name 'Tyson, Bryan' at the top, which is present in the image. Output B incorrectly replaces the name with '6' and contains garbled text 'رعانع واعده واعده واعده' which is not in the original document. Output A also correctly transcribes the date as '5:21-52' (though it should be '5-21-52'), while Output B has the correct date format but fails in other critical areas. The text in Output A is more complete and accurate, making it the better extraction. | 1/1 |
9 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output A is more accurate and complete. It correctly transcribes the first word as 'lately' (Output B has 'Iately'), and it correctly represents the date format as '5-21-52' (Output B uses '5:21-52'). Output A also correctly includes the name 'Bryan Tyson' in the list of letter recipients, while Output B has 'xxaxxxx' which is a garbled transcription of the original text. Output A captures the full text of the document without adding commentary, adhering to the faithfulness criterion. | 1/1 |
9 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output A is more accurate and faithful to the document. It correctly captures the name 'Tyson, Bryan' and the number '6' at the top. Output B omits the name and number, and contains a significant error: it replaces the garbled text 'BryanxTysonxgxxAxRaxxoxkxkxgxtxx' with 'رعانع واعده واعده واعده', which is not present in the original image and is a hallucination. Output A correctly represents the garbled text as 'Bryan Tyson' and 'BryanxTysonxgxxAxRaxxoxkxkxgxtxx', which is a more faithful representation of the OCR's attempt to read the original. Output A is also more complete as it includes the header text. | 1/1 |
10 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is more complete as it includes the header text 'C.S.A. ARCHIVES. ARMY UNITS. ARMY OF N. VIRGINIA' and the page number '3' which are visible in the document image. Output A omits this information, making it less faithful to the original document. Both outputs are accurate in the text they share, but B's completeness makes it superior. | 1/1 |
10 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output B is more complete as it includes the header text 'C.S.A. ARCHIVES. ARMY UNITS. ARMY OF N. VIRGINIA' and the page number '3' which are visible in the document image. Output A omits this information, making it less faithful to the original document. Both outputs are accurate in the text they share, but B's completeness makes it superior. | 1/1 |
10 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more complete as it includes the header text 'C.S.A. ARCHIVES. ARMY UNITS. ARMY OF N. VIRGINIA' and the page number '3' which are visible in the document image. Output A omits this information, making it less faithful to the original document. Both outputs are accurate in the text they share, but B's completeness makes it superior. | 1/1 |
10 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Both outputs are identical and accurate. They correctly capture all text from the document, including the header and the paragraph, with no omissions or errors. The text is presented in the correct reading order and is faithful to the source. Therefore, it is a tie. | 1/1 |
10 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Both outputs are identical and accurate. They correctly capture all text from the document, including the header and the paragraph, with no omissions or errors. The text is presented in the correct reading order and is faithful to the source. Therefore, it is a tie. | 1/1 |
10 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Both outputs are identical and accurate. They correctly capture all text from the document, including the header and the paragraph, with no omissions or errors. The text is presented in the correct reading order and is faithful to the source. Therefore, it is a tie. | 1/1 |
11 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output A is more faithful to the document's layout by preserving the line breaks and spacing between distinct lines of text, which improves readability and accurately reflects the document's structure. Both outputs are complete and accurate, but A's formatting better represents the original visual hierarchy. | 1/1 |
11 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A is more faithful to the document's layout by preserving the line breaks and spacing between the '18-E' and '10-23-73' lines, as well as the '1 item' label on a separate line. Output B incorrectly combines '18-E' and '1 item' on the same line, which alters the document's structure and could be misleading. Both outputs are complete and accurate, but A's formatting better reflects the original. | 1/1 |
11 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | A | Output B correctly transcribes 'Macneile' as it appears in the image, while Output A contains a typo, writing 'Macneille' instead. This makes B more faithful to the source document. | 1/1 |
11 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | B | Output A maintains the correct reading order of the text as it appears in the document, with the date '10-23-73' positioned below '18-E' and above '1 item', which matches the visual layout. Output B incorrectly places '10-23-73' at the end, disrupting the logical flow and reading order. Both outputs are faithful, complete, and accurate in their content, but A's superior formatting and reading order make it the better extraction. | 1/1 |
11 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Output B is more accurate in preserving the original text. The name in the document is 'Macneile', not 'Macneille' as in Output A. Output B correctly separates '18-E' and '1 item' onto two lines, matching the document's layout, while Output A combines them on one line. Both outputs are complete and faithful, but B's accuracy in the name and formatting makes it superior. | 1/1 |
11 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output A is more faithful to the document's layout, preserving the separation between the '18-E' and '1 item' fields as they appear on the card. Output B incorrectly combines '18-E' and '1 item' on the same line, which alters the original formatting and could be misleading. Both outputs are complete and accurate, but A's formatting is superior. | 1/1 |
12 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | TIE | Both outputs are identical and accurately capture all visible text from the document with perfect faithfulness, completeness, accuracy, and reading order. There is no difference in quality between them. | 1/1 |
12 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | A | Both outputs are faithful, complete, and accurate. Output B is superior because it correctly preserves the single sentence structure of the paragraph without unnecessary line breaks, maintaining the original formatting more precisely. Output A introduces artificial line breaks that do not exist in the source document. | 1/1 |
12 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | TIE | Both outputs are identical and accurately capture all visible text from the document with perfect faithfulness, completeness, accuracy, and reading order. There is no difference in quality between them. | 1/1 |
12 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Both outputs are faithful, complete, and accurate. Output B is superior because it correctly preserves the single sentence structure of the paragraph without unnecessary line breaks, maintaining the original formatting more precisely. Output A introduces artificial line breaks that do not exist in the source document. | 1/1 |
12 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | TIE | Both outputs are identical and accurately capture all visible text from the document with perfect faithfulness, completeness, accuracy, and reading order. There is no difference in quality between them. | 1/1 |
12 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output B is more accurate in preserving the line breaks and formatting of the original text, which is a key aspect of faithful representation. Both outputs are complete and accurate, but B's structure better reflects the document's layout. Output A combines the text into a single paragraph, which is less faithful to the original formatting. | 1/1 |
13 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output A includes the page number '4' in the top right corner, which is visible in the image, while Output B omits it. This makes Output A more complete and faithful to the original document. | 1/1 |
13 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A includes the page number '4' which is visible in the top right corner of the document, making it more complete than Output B. Both outputs are accurate and faithful to the text, but A captures all visible elements, including the marginal number, thus fulfilling the completeness criterion better. | 1/1 |
13 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more faithful to the original document as it includes the number '4' in the header, which is visible in the image. Output A omits this detail, making it incomplete. Both outputs are accurate and complete in their text content, but B's inclusion of the number '4' makes it a more faithful transcription. | 1/1 |
13 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | B | Output A includes the page number '4' which is visible in the top right corner of the document, making it more complete than Output B. Both outputs are accurate and faithful to the text, but A captures all visible elements, including the marginal number, thus fulfilling the completeness criterion better. | 1/1 |
13 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more faithful to the original document as it includes the number '4' in the header, which is visible in the image. Output A omits this detail, making it incomplete. Both outputs are accurate and complete in their text content, but B's inclusion of the number '4' makes it a more faithful transcription. | 1/1 |
13 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output A correctly separates the header 'Mills, Robert S.' and the page number '4' onto distinct lines, preserving the document's layout more accurately than Output B, which merges them. Both outputs are otherwise faithful, complete, and accurate in their text content. However, A's superior formatting fidelity makes it the better extraction. | 1/1 |
14 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output A is better because it includes the text '1 item' which is visible in the document and is missing from Output B. Both outputs are faithful, accurate, and maintain correct reading order, but A is more complete. | 1/1 |
14 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output A includes the line '1 item' which is visible in the document and is missing from Output B. This makes Output A more complete. Both outputs are accurate and faithful to the text, but A captures all visible text, including the item count, which is a key piece of information in a catalog record. | 1/1 |
14 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is more accurate as it correctly includes the text '1 item' which is visible in the document, whereas Output A omits it. Both outputs are faithful, complete, and accurate in their other aspects, but B captures all visible text. | 1/1 |
14 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output A maintains the correct reading order of the text, placing the date '4-8-60' at the bottom as it appears in the document, while Output B incorrectly places it at the end of the list of lines. Both outputs are faithful and complete, but A's reading order is more accurate to the source layout. | 1/1 |
14 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | B | Output A is more accurate and faithful to the document. It correctly captures the text '1 item' on the same line as 'MSS. Div.' as it appears in the image, while Output B separates '1 item' onto a new line, altering the document's layout. Output A also maintains the correct reading order and includes all visible text without omissions or additions. | 1/1 |
14 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | A | Output B is more accurate in its representation of the document's layout. The text 'History of Medicine--MSS. Div.' is correctly split across two lines, matching the visual structure of the original image. Output A incorrectly combines these lines into a single line, which is a formatting error. Both outputs are faithful, complete, and accurate in their content, but B better preserves the original document's line breaks. | 1/1 |
15 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output A is better because it includes the header text at the top of the document, which is visible in the image. Output B omits this header, making it incomplete. Both outputs are accurate in their shared content, but completeness is a higher priority criterion. | 1/1 |
15 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | TIE | Both outputs are identical in content, accuracy, completeness, and formatting. They capture all visible text from the document without any errors or omissions. | 1/1 |
15 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is better because it includes the header text at the top of the document, which is visible in the image. Output A omits this header, making it incomplete. Both outputs are accurate in their text and formatting, but B is more complete, which is a higher priority in the evaluation criteria. | 1/1 |
15 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output B is more complete as it includes the header text at the top of the document, which is visible in the image. Output A omits this header, making it incomplete. Both outputs are accurate in their captured text and maintain correct reading order, but completeness is a higher priority criterion. Therefore, B is the better extraction. | 1/1 |
15 | deepseek-ai/DeepSeek-OCR | lightonai/LightOnOCR-2-1B | A | Output B is more accurate in preserving the line breaks and hyphenation as they appear in the original document, particularly the word 'Wilmington' split across two lines and the hyphenated word 'Wil-mington'. Output A incorrectly combines these lines and omits the hyphenation, which is a form of inaccurate transcription. Both outputs are complete and faithful to the text, but B's handling of the document's layout is superior. | 1/1 |
15 | rednote-hilab/dots.ocr | lightonai/LightOnOCR-2-1B | B | Output A includes the header text 'Jones, Oliver H. Papers, 1832-1861. Wilmington N. C. 253 items. Sketch' which is present in the image, while Output B omits it. Output A is therefore more complete. Both outputs are accurate in their text and follow the correct reading order. Since completeness is a higher priority than formatting, A is the better extraction. | 1/1 |
16 | zai-org/GLM-OCR | deepseek-ai/DeepSeek-OCR | B | Output B is better because it is more complete, capturing the header text 'Pettet, William. Papers. Bellefonte, Centre Co., Pennsylvania' which is clearly visible at the top of the document. Output A omits this header, making it incomplete. Both outputs are accurate and faithful to the text they contain, but completeness is a higher priority. Output B also maintains the correct reading order and formatting. | 1/1 |
16 | zai-org/GLM-OCR | rednote-hilab/dots.ocr | B | Output B is more complete as it includes the header text 'Pettet, William. Papers. Bellefonte, Centre Co., Pennsylvania' which is visible in the image. Output A omits this header, making it incomplete. Both outputs are accurate in the body text, but B's completeness makes it superior. | 1/1 |
16 | zai-org/GLM-OCR | lightonai/LightOnOCR-2-1B | B | Output B is better because it is more complete, capturing the header text 'Pettet, William. Papers. Bellefonte, Centre Co., Pennsylvania' which is clearly visible at the top of the document. Output A omits this header, making it incomplete. Both outputs are accurate and faithful to the text they contain, but completeness is a higher priority. Output B also maintains the correct reading order and formatting. | 1/1 |
16 | deepseek-ai/DeepSeek-OCR | rednote-hilab/dots.ocr | A | Output A is more accurate and complete. It correctly transcribes the header text 'Pettet, William. Papers. Bellefonte, Centre Co., Pennsylvania' as a single line, matching the document's layout. Output B incorrectly splits this line into two, which is a formatting error. Additionally, Output A correctly spells 'campaign' while Output B contains a typo ('compaign'). | 1/1 |
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 20