Multi-Column PDF and Table Extraction Accuracy Across Leading APIs
Geometry, not model quality, determines whether PDF table extraction actually works in production.
Marcus Lien
Section
Extraction Evals
6 stories in Extraction Evals.
Ground Truth Dataset Construction for Document Extraction Tasks
How hidden choices in building ground truth datasets mask the real causes of extraction failures.
Priya Subramaniam
Error Analysis for Low-Accuracy Document Extraction Fields
Four structural causes drive extraction failures on fields that business workflows depend on.
Marcus Lien
Field-Level Accuracy Metrics for Structured Document Extraction
Vendors mix four different metrics into one accuracy claim, causing pipelines to fail in production.
Marcus Lien
Confidence Scoring Architecture for Document Extraction Models
Understanding three layers reveals why confidence scores don't mean what they appear to.
Celeste Okafor
Benchmarking Document Extraction Accuracy on Production Document Sets
Real-world documents reveal why benchmark accuracy scores collapse once models enter production.
Tobias Krenn