Summary

• Sync evaluation pipeline with the latest version (NID filter, Hungarian matching, merged tables)
• Sync inference scripts (aws, google, llamaparse, microsoft, unstructured, upstage)
• Refresh dataset/reference.json and dataset/sample_results/ with 2026-02-09 inference runs (7 providers)
• Update main README usage and leaderboard; update scripts/README.md

Changes

Evaluation pipeline

• src/: sync layout_evaluation.py, table_evaluation.py, utils.py; add geometry.py (isolated is_included); add _normalize_coords shim to handle both dict/list coordinate formats
• evaluate.py: new CLI flags (--ref-path, --pred-path, --mode {layout,table,all,speed}, --evaluate-merged-table, --filter-by-gt-area, --min-match-score, --max-workers); writes <pred>.eval.json

Inference scripts

• scripts/: add base.py, doc_grouping.py; sync infer_{aws,google,llamaparse,microsoft,unstructured,upstage}.py and utils.py

Dataset

• dataset/reference.json: updated to the latest version
• dataset/sample_results/
  • Add 7 new samples (2026-02-09): aws_260209.json, google_260209.json, llamaparse_260209.json, microsoft_260209.json, unstructured_260209.json, upstage_standard_260209.json, upstage_enhanced_260209.json

Docs & misc

• README.md: update Evaluation section to match new CLI (--ref-path, --pred-path, --mode); refresh leaderboard (7 rows, 2026-02-09)
• scripts/README.md: updated to the latest version
• .gitignore: ignore predictions/, __pycache__/, *.eval.json

##Leaderboard (2026-02-09)

LGTM!