{"observation":{"id":"fcad1a88-bccf-4eea-8bf5-0d8e5dbc22ee","tool":"reducto","tool_name":"Reducto","criterion":"table-preservation","criterion_name":"Table Preservation","criterion_definition":"Preserves complex table structures, including rows, columns, multi-row headers, and merged-cell relationships in markdown.","criterion_evidence_type":"transformation","criterion_rank_role":"decisive","criterion_rank_role_reason":"Accurate Markdown conversion of complex PDFs depends on keeping table structure intact, not flattening it into plain text. (3 of 3 judges)","scenario":"scanned-research-paper","scenario_name":"Scanned Research Paper","group_tag":"scanned-research-paper","scenario_description":"An image-only scanned research paper used to stress OCR and layout recovery in a multi-column academic document with figures, charts, tables, captions, and references.","modality":"pdf","input_text":null,"input_artifact_refs":[{"alt":null,"url":"https://d3epheqghktydj.cloudfront.net/convert-a-complex-pdf-into-clean-markdow-scanned-research-pdf-7b86de49784d.pdf","role":"input","filename":"Scanned Research PDF.pdf"}],"stresses":["OCR on scanned pages","Multi-column reading order","Figure and chart handling","Table reconstruction from scans","Caption association","Reference extraction","Overall document structure retention"],"verdict":"mixed","score":null,"score_total":null,"note":"Partially reconstructs Table 1: most of the roughly 90 numeric values are exact, but literal 0 values in the 12-inch column become blanks, one mean cell picks up stray digits (33.0 830000), and a row-label-only section header is broadcast across all six columns in one instance.","evidence_state":"verified","source":null,"artifacts":[{"url":"https://cdn.futuresmart.ai/public/aidemos/463058512d444a0796cf3e61494de04e.png?v=1","role":"input","alt":null},{"url":"https://cdn.futuresmart.ai/public/aidemos/876729da74884fe3b81452a859362ad0.png?v=1","role":"output","alt":null},{"url":"https://cdn.futuresmart.ai/public/aidemos/44cf7295024f446e92d3d97336ecfeee.png?v=1","role":"input","alt":null}],"run_id":"6e3160de-fe46-4b45-b071-72560b5c5d0e","study_title":"Convert a Complex PDF into Clean Markdown with an API","study_kind":"generation","research_task":"86b9h7t37","tested_at":null,"completeness":"input-and-output","input":{"state":"files","text":null,"files":[{"url":"https://d3epheqghktydj.cloudfront.net/convert-a-complex-pdf-into-clean-markdow-scanned-research-pdf-7b86de49784d.pdf","filename":"Scanned Research PDF.pdf","alt":"Scanned Research Paper","role":"input"}],"modality":"pdf","stresses":["OCR on scanned pages","Multi-column reading order","Figure and chart handling","Table reconstruction from scans","Caption association","Reference extraction","Overall document structure retention"]},"tool_page_slug":"reducto","tool_url":"https://aidemos.com/tools/reducto","permalink":"https://aidemos.com/evidence/fcad1a88-bccf-4eea-8bf5-0d8e5dbc22ee","api_url":"https://ai.aidemos.com/v1/observations/fcad1a88-bccf-4eea-8bf5-0d8e5dbc22ee"},"peers":[{"id":"1dbde32e-95eb-4904-8433-272d8c5e1fdb","tool":"adobe-api","tool_name":"Adobe API","verdict":"worked","score":null,"score_total":null,"note":"Preserves a multi-column scientific table of tree treatments and before/after diameters, keeping the treatment rows and inch/cm subcolumns readable.","artifact_count":2,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/57a862fbcd21461ab376ee98f35f965b.png?v=1","evidence_url":"https://aidemos.com/evidence/1dbde32e-95eb-4904-8433-272d8c5e1fdb"},{"id":"1dcafead-b619-42e1-876d-e9a9b6902adb","tool":"extend-ai","tool_name":"Extend AI","verdict":"mixed","score":null,"score_total":null,"note":"Keeps the numeric table structure, but drops between-column annotations and leaves stray OCR characters in some numeric cells, so contextual information is partially lost.","artifact_count":3,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/6b3fa17009604505a41ee22a883265a2.png?v=1","evidence_url":"https://aidemos.com/evidence/1dcafead-b619-42e1-876d-e9a9b6902adb"},{"id":"0bfcd599-6c54-476e-94c8-d8773b3595ca","tool":"landing-ai","tool_name":"Landing AI","verdict":"mixed","score":null,"score_total":null,"note":"Partially reflects the text that sits between table columns, but the intervening content still disrupts the table structure in the parsed output.","artifact_count":2,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/fc95e14e58844709b2619a3f3621bc7a.png?v=1","evidence_url":"https://aidemos.com/evidence/0bfcd599-6c54-476e-94c8-d8773b3595ca"},{"id":"6893a7fc-87db-41b5-a8fc-06a4f67d61d5","tool":"llamaparse","tool_name":"LlamaParse","verdict":"worked","score":null,"score_total":null,"note":"Preserves a nested treatment table with the 7-inch, 10-inch, 12-inch, 100-leave-tree, and clearcut columns and the acres/live-lodgepole rows.","artifact_count":2,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/d1659c50c6c449afa114d98746024dbb.png?v=1","evidence_url":"https://aidemos.com/evidence/6893a7fc-87db-41b5-a8fc-06a4f67d61d5"},{"id":"de43911e-dc72-4aee-a02a-2145d2d7acfb","tool":"mistral-ai","tool_name":"Mistral AI","verdict":"failed","score":null,"score_total":null,"note":"Broken column boundaries and disrupted value alignment make the reconstructed table significantly less faithful to the source.","artifact_count":4,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/56ad045174884d278b8ec196b932b07f.png?v=1","evidence_url":"https://aidemos.com/evidence/de43911e-dc72-4aee-a02a-2145d2d7acfb"},{"id":"d9350bf4-9574-4634-8293-5ced2ab60d9c","tool":"nutrient-io","tool_name":"Nutrient.io","verdict":"failed","score":null,"score_total":null,"note":"Breaks more complex multi-level tables, with rows and merged cells misaligned or lost once the hierarchy becomes dense.","artifact_count":2,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/aca6328506da49aeb716e311b50a278c.png?v=1","evidence_url":"https://aidemos.com/evidence/d9350bf4-9574-4634-8293-5ced2ab60d9c"},{"id":"c19621dc-83eb-467f-9210-161a2adcac66","tool":"tensorlake","tool_name":"Tensorlake","verdict":"failed","score":null,"score_total":null,"note":"Struggles with hierarchical scanned tables, misplacing column headers and producing unreliable reconstructions on both the multicolumn table and the denser complex table.","artifact_count":6,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/57a862fbcd21461ab376ee98f35f965b.png?v=1","evidence_url":"https://aidemos.com/evidence/c19621dc-83eb-467f-9210-161a2adcac66"},{"id":"6db07c2d-fbc7-464c-ba3c-822a7495c3d5","tool":"upstage-ai","tool_name":"Upstage AI","verdict":"mixed","score":null,"score_total":null,"note":"Keeps the table values intact but reconstructs the headers incorrectly, leaving a grouped-column table with inconsistent structure.","artifact_count":2,"thumbnail":"https://cdn.futuresmart.ai/public/aidemos/57a862fbcd21461ab376ee98f35f965b.png?v=1","evidence_url":"https://aidemos.com/evidence/6db07c2d-fbc7-464c-ba3c-822a7495c3d5"}],"other_criteria":[{"id":"73354714-3c90-4a35-a5c6-c27f03292e80","criterion":"advanced-features","criterion_name":"Advanced Features","rank_role":"context","verdict":"worked","score":null,"score_total":null,"note":"Schema-driven extract.run fully recovers the corrupted Table 4 rows, returning all 12 checked rows exactly, including the 10-inch-cut 1979 MPB row whose parse.run output was badly garbled.","artifact_count":3,"evidence_url":"https://aidemos.com/evidence/73354714-3c90-4a35-a5c6-c27f03292e80"},{"id":"6e5c93ad-2b5d-4c86-89f8-0503d11424d6","criterion":"advanced-features","criterion_name":"Advanced Features","rank_role":"context","verdict":"mixed","score":null,"score_total":null,"note":"The confidence signal flags the faint handwritten margin note and the garbled Table 4 as low confidence, but it misses the hallucinated 'USA' word in the title, which is still rated 0.9354 confident.","artifact_count":4,"evidence_url":"https://aidemos.com/evidence/6e5c93ad-2b5d-4c86-89f8-0503d11424d6"},{"id":"3bec7ed4-9b8e-44f3-b548-b76f0215b34c","criterion":"complex-document-handling","criterion_name":"Complex Document Handling","rank_role":"decisive","verdict":"struggled","score":null,"score_total":null,"note":"Handles the 12-page scan in 10.6 seconds with no truncation, but the densest 17-column table has unreadable stretches with injected glyphs and run-on numbers.","artifact_count":2,"evidence_url":"https://aidemos.com/evidence/3bec7ed4-9b8e-44f3-b548-b76f0215b34c"},{"id":"39ecbd6f-1599-4ba1-ae17-16b29b9c07cb","criterion":"markdown-quality","criterion_name":"Markdown Quality","rank_role":"decisive","verdict":"worked","score":null,"score_total":null,"note":"Uses no invented tags or malformed markdown; the only HTML seen is legitimate <br /> inside table cells, so the syntax stays clean even though the document does not surface real heading markup.","artifact_count":2,"evidence_url":"https://aidemos.com/evidence/39ecbd6f-1599-4ba1-ae17-16b29b9c07cb"},{"id":"7840c24e-489b-4551-b576-148911ebe640","criterion":"reading-order-structure","criterion_name":"Reading Order & Structure","rank_role":"decisive","verdict":"struggled","score":null,"score_total":null,"note":"Displaces the byline by a full column: the author line that sits above the two-column split in the source is emitted only after the entire left column, and its footnote markers are rendered inconsistently.","artifact_count":2,"evidence_url":"https://aidemos.com/evidence/7840c24e-489b-4551-b576-148911ebe640"},{"id":"b0876a41-66cd-455c-b38c-421968bf4e16","criterion":"text-ocr-completeness","criterion_name":"Text & OCR Completeness","rank_role":"decisive","verdict":"worked","score":null,"score_total":null,"note":"Converts all 12 scanned pages with no gaps; the page-11-to-page-12 handoff is preserved verbatim, and a separate page-marker output shows pages 1 through 12 present with no missing markers.","artifact_count":4,"evidence_url":"https://aidemos.com/evidence/b0876a41-66cd-455c-b38c-421968bf4e16"},{"id":"ab812bbc-7ab3-4b0c-82e1-e142cb5375fe","criterion":"visual-content-retention","criterion_name":"Visual Content Retention","rank_role":"decisive","verdict":"worked","score":null,"score_total":null,"note":"Segments a shield logo out of a single full-page raster scan and also retains Figure 1 as an image with an accurate synthesized caption; the returned logo crop is a tight 77x81px cut with legible shield text.","artifact_count":5,"evidence_url":"https://aidemos.com/evidence/ab812bbc-7ab3-4b0c-82e1-e142cb5375fe"}],"appears_in":[{"page_type":"ranking","slug":"pdf-to-markdown-apis","title":"Best AI Tools to Convert Complex PDFs into Clean Markdown with an API","url":"https://aidemos.com/best/pdf-to-markdown-apis","binding":"run"}],"same_scenario":[{"id":"1dbde32e-95eb-4904-8433-272d8c5e1fdb","tool":"adobe-api","tool_name":"Adobe API","verdict":"worked","score":null,"score_total":null,"note":"Preserves a multi-column scientific table of tree treatments and before/after diameters, keeping the treatment rows and inch/cm subcolumns readable."},{"id":"1dcafead-b619-42e1-876d-e9a9b6902adb","tool":"extend-ai","tool_name":"Extend AI","verdict":"mixed","score":null,"score_total":null,"note":"Keeps the numeric table structure, but drops between-column annotations and leaves stray OCR characters in some numeric cells, so contextual information is partially lost."},{"id":"0bfcd599-6c54-476e-94c8-d8773b3595ca","tool":"landing-ai","tool_name":"Landing AI","verdict":"mixed","score":null,"score_total":null,"note":"Partially reflects the text that sits between table columns, but the intervening content still disrupts the table structure in the parsed output."},{"id":"6893a7fc-87db-41b5-a8fc-06a4f67d61d5","tool":"llamaparse","tool_name":"LlamaParse","verdict":"worked","score":null,"score_total":null,"note":"Preserves a nested treatment table with the 7-inch, 10-inch, 12-inch, 100-leave-tree, and clearcut columns and the acres/live-lodgepole rows."},{"id":"de43911e-dc72-4aee-a02a-2145d2d7acfb","tool":"mistral-ai","tool_name":"Mistral AI","verdict":"failed","score":null,"score_total":null,"note":"Broken column boundaries and disrupted value alignment make the reconstructed table significantly less faithful to the source."},{"id":"d9350bf4-9574-4634-8293-5ced2ab60d9c","tool":"nutrient-io","tool_name":"Nutrient.io","verdict":"failed","score":null,"score_total":null,"note":"Breaks more complex multi-level tables, with rows and merged cells misaligned or lost once the hierarchy becomes dense."},{"id":"c19621dc-83eb-467f-9210-161a2adcac66","tool":"tensorlake","tool_name":"Tensorlake","verdict":"failed","score":null,"score_total":null,"note":"Struggles with hierarchical scanned tables, misplacing column headers and producing unreliable reconstructions on both the multicolumn table and the denser complex table."},{"id":"6db07c2d-fbc7-464c-ba3c-822a7495c3d5","tool":"upstage-ai","tool_name":"Upstage AI","verdict":"mixed","score":null,"score_total":null,"note":"Keeps the table values intact but reconstructs the headers incorrectly, leaving a grouped-column table with inconsistent structure."}]}