Tables And Figures
Reducto · Reducto
Document Ingestion & Parsing for AI — Reducto
Reducto evals — Tables & Figures (relift v3 InfraRed)
About Reducto
Reducto is a document ingestion platform for AI pipelines that turns complex documents (PDFs, scans, spreadsheets) into clean, structured, layout-aware data. Its API parses documents into Markdown and typed content blocks, extracts structured fields against a user-defined schema with source citations, and splits bundled files into their constituent documents — feeding retrieval-augmented generation and document-automation workflows.
Employees
~50 (approx — verify)
Industry
Document AI / Data Ingestion
Headquarters
San Francisco, CA (verify)
Website
reducto.aiSample tests· showing 3 of 9
| # | Input | Expected behavior | Check |
|---|---|---|---|
| 01 | A financial statement has a dense table. The integrator gets back the table as a flat run of numbers with no row/column structure and tries to re-impose a grid by counting tokens. | Consume Reducto's structured table output (rows x columns, ideally as HTML/Markdown table or a cell matrix) rather than re-gridding a flattened number stream. Validate that row/column counts match the visible table on a sample. A mis-aligned grid silently maps values to wrong headers. | Pass / FailAi Platformhigh |
| 02 | A table spans pages 7-9 with the header row repeated on each page. The integrator treats each page's slice as an independent table and re-counts the header row as data three times. | Stitch continued tables across page breaks into one logical table and de-duplicate repeated header rows. Treating page slices independently double-counts headers and fragments the table for retrieval. Verify on a known multi-page table that the stitched row count is correct. Treat automatic table-s… | Pass / FailAi Platformmedium |
| 03 | A table has a header cell that spans three columns and a row label that spans two rows. The integrator assumes a rectangular grid and mis-aligns every value under the spanned header. | Handle merged/spanning cells per the documented table model (colspan/rowspan or repeated values) rather than assuming a flat rectangular grid. Verify spanned-header tables specifically, since they are the highest-risk for silent misalignment. Treat the exact spanning-cell representation as [REQUIRE… | Pass / FailAi Platformhigh |
How this eval is graded
Grade against expected.ideal_behavior and expected.rubric. Per-criterion pass requires mean >= 4.0 and no criterion below 3.
Rubric criteria
- Reducto
- Ai Platform
- Tables And Figures
Recommended for
Works with
Related evals
Run this eval in your workspace
Connect your data, configure thresholds, and review results with your team.