Skip to main content
Ctrl+K
Tabulus - Home Tabulus - Home
  • Installation And Setup
    • Windows CPU Setup
    • GPU Server Installation
    • Python Library Setup
  • Tutorial
    • Core Pipeline Overview
    • Step 1: PDF Profiling
    • Step 2: Table Reconstruction
    • Step 3: Reference-Table Classification
    • Step 4: Bibliography Extraction
    • Step 5: Reference Matching
    • Step 6: Scholarly Reference Resolution
    • Step 7: Resolved CSV Export
    • Planned Step: Run Report And QA Bundle
  • Components
    • Component Overview
    • PDF Profilers
    • PDF Readers
    • Page Renderers
    • Layout Detectors
    • Table Detectors
    • Table Reconstruction Adapters
    • Bibliography Extractors
    • Reference Matchers
    • DOI Resolvers
  • Workflows
    • Single PDF Local Workflow
    • GPU Server Run
    • MinerU GPU Execution
    • End-To-End Run
    • Debugging A Failed Step
  • External Tools
    • Step 1 PDF Profiling Tool
      • MinerU
    • Step 2 Table-Reconstruction Adapters
      • PaddleOCR-VL
      • Chandra OCR 2
      • NuExtract3
      • Tesseract + Table Transformer
      • Docling TableFormer
      • Granite Vision 4.1 4B
      • TRivia-3B
      • GLM-OCR
      • Dolphin-v2
      • DeepSeek-OCR-2
      • Nanonets-OCR-s
      • MonkeyOCRv2-B-Parsing
      • NVIDIA Nemotron Parse v1.2
      • HunyuanOCR-1.5
      • dots.mocr
      • InternVL3.5-8B
    • Step 4 Bibliography Extraction Tool
      • GROBID
  • Data Contracts
    • Run Directory
    • PDF Profiling / Step 1
      • MinerU Output Files
      • pdf_profile.json
      • tables_index.json
    • Table Reconstruction / Step 2
      • ocr_tables.json
      • Table Prediction CSV
    • Bibliography & Reference Processing
      • bibliography.json
      • reference_matches.json
      • reference_resolution.json
      • Resolved CSV
  • Evaluation
    • Evaluation Overview
    • Reference-Table Classification Quality
    • Table Reconstruction Quality
    • Bibliography / Reference Extraction Quality
    • Reference Matching Quality
  • Project Notes
    • Current State
    • Branding
    • Containerization Later
  • .md

Data Contracts

Data Contracts#

Data contract pages describe the stable files exchanged between Tabulus steps. Tutorial pages explain when and why a file is produced; these pages describe what the file contains and how downstream consumers should interpret it.

  • Run Directory
    • Current Output Hierarchy
    • Step Dependencies
    • Current Profiling Output Convention
    • Canonical Table-Crop Directory
    • Reconstruction Directory
    • Continued-Table Semantics
    • Reconstruction Reruns
    • Reference-Table Classification Boundary
  • PDF Profiling / Step 1
    • MinerU Output Files
    • pdf_profile.json
    • tables_index.json
  • Table Reconstruction / Step 2
    • ocr_tables.json
    • Table Prediction CSV
  • Bibliography & Reference Processing
    • bibliography.json
    • reference_matches.json
    • reference_resolution.json
    • Resolved CSV

previous

GROBID

next

Run Directory

By Tabulus contributors