Breaks a 5-column business-results table by duplicating the 2025 header and merging the 2024 value and percent change into one cell; the Net sales row ends up with 254,811 and 2.6 smashed together.
What was measured
Table Preservation
Preserves complex table structures, including rows, columns, multi-row headers, and merged-cell relationships in markdown.
decisive for this rankingtransformation
Accurate Markdown conversion of complex PDFs depends on keeping table structure intact, not flattening it into plain text. (3 of 3 judges)
What was given, what came back
Test input: Financial Report - Table Heavy · pdf · group: financial-report-table-heavy
Input — what we sent
A table-heavy corporate financial report used to test extraction of dense, hierarchical financial statements with grouped columns, multi-row headers, segment-reporting tables, and narrative disclosures.
Why this input is hard
- · Multi-page financial statement extraction
- · Hierarchical table reconstruction
- · Grouped columns and multi-row headers
- · Reading order in a report with mixed narrative and tables
- · Document structure retention
- · Markdown usability
Output — unretouched
Also checked on this input — same tool, 7 other criteria
Advanced Features✓ WorkedSchema-driven extract.run fully recovers the broken Business Results table, separating the 2024 value and percent-change fields that parse.run had merged together.Advanced Features✗ FailedThe confidence field is nearly useless here: the only low-confidence flag is a false alarm on the correct page-18 section header, while the genuinely broken Business Results table stays high-confidence throughout.Complex Document Handling◐ MixedProcesses the 18-page report in 9.7 seconds with no truncation; the 9-column segment table at page 17 is flawless even though the first table on page 2 is broken.Markdown Quality✓ WorkedKeeps the syntax clean: no <signature>, <empty>, <b>, <i>, or <u> tags appear anywhere in the output, and the table-of-contents extract renders as a valid pipe table.Reading Order & Structure✗ FailedProvides zero markdown heading markup across all 18 pages; the Roman-numeral sections and numbered subsections are flattened into plain text despite the source’s visible section hierarchy.Text & OCR Completeness✓ WorkedConverts all 18 pages with no omitted text; a dense financial-condition paragraph preserves multiple JPY-billion values exactly, including JPY1,252.7 billion, JPY623.6 billion, JPY27.2 billion, JPY10.8 billion, JPY5.0 billion, and JPY9.7 billion.Visual Content Retention✓ WorkedRetains the recurring page-header logo consistently across all 17 pages that carry it; the returned crops on pages 2 and 18 are the same mark and wordmark, with only a 1px width difference (453x36 vs 452x36) consistent with rounding.
Provenance
- Observation
- f597cbbc-cbe9-4694-8b5e-078b1489986b
- Evidence run
- 6e3160de-fe46-4b45-b071-72560b5c5d0e
- Study
- Convert a Complex PDF into Clean Markdown with an API
- Research task
- 86b9h7t37
- Tested at
- not recorded
- Source
- first-party
- Evidence state
- verified
- Proof shown
- input + output shown
- Cost / latency
- not captured
- Repeat run
- not captured
- Tester
- not captured
The last three rows are honest blanks, not placeholders — our capture has no field for them yet.
Query this
get_evidence({
tool: "reducto",
scenario: "financial-report-table-heavy"
})MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Same input, same check — 8 other tools
measured on Table Preservation
Adobe API✗ FailedFlattens a dual-header financial table, losing the distinct row-header versus column-header roles in the extracted version.Extend AI◐ MixedKeeps the body rows aligned in a multilevel table, but fails to split compound header cells cleanly, so header roles are partially flattened.Landing AI✓ WorkedReconstructs a multi-column segment comparison table with previous-year, present-year, and year-over-year change columns aligned across the rows.LlamaParse✓ WorkedPreserves a multi-level segment-results table with grouped headers and year-over-year columns, keeping the hierarchy legible in the extracted output.Mistral AI✓ WorkedThe multilevel segment table is preserved with its hierarchical headers intact, maintaining the relationships between the grouped columns and their values.Nutrient.io✗ FailedLoses multi-level header organization in hierarchical tables, so parent-child column relationships are not reconstructed in the markdown output.Tensorlake✗ FailedBreaks on a more complex multi-header table by losing header hierarchy and omitting at least one header label, producing an incomplete and structurally incorrect reconstruction.Upstage AI✗ FailedMisaligns column headers with data regions in dense financial tables, producing a structurally inconsistent balance-sheet extraction rather than a clean header-to-row mapping.
This evidence is published in
Real inputs and real outputs, no retouching · every cell queryable via API & MCP · aidemos.com