jedarden/pdftract
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
pdftract/notes/pdftract-206o6.md
jedarden 21fcd902d1 feat(pdftract-2vajs): implement slide_deck profile with fixtures and tests 
Implements the slide_deck document profile for PowerPoint/Keynote/Google
Slides exports as PDF. Includes 5 fixtures, expected outputs, and regression
tests.

Components:
- profiles/builtin/slide_deck/profile.yaml - Profile configuration
- tests/fixtures/profiles/slide_deck/ - 5 PDF fixtures with expected outputs
- crates/pdftract-cli/tests/test_slide_deck.rs - Regression tests (12 PASS)

Fixtures cover:
1. pitch_deck - Sales pitch (10 slides)
2. academic_lecture - Academic lecture (40 slides)
3. corporate_kickoff - Corporate kickoff (15 slides)
4. bilingual_deck - Bilingual EN/ES (12 slides)
5. googleslides_handout - Google Slides handout mode (4 pages, 3 slides/page)

Extracted fields: title, presenter, date, slide_titles

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-27 21:12:24 -04:00

3.8 KiB
Raw Blame History

Bead pdftract-206o6: Scientific Paper Profile Implementation

Summary

Implemented the scientific_paper document profile per Phase 7.10 YAML schema with 5 fixtures and regression tests.

Files Created/Modified

Profile Configuration

  • profiles/builtin/scientific_paper/profile.yaml - Updated to Phase 7.10 schema with:
    • name: scientific_paper
    • description: Academic papers from arXiv, journals, conference proceedings
    • priority: 30
    • match predicates: text_contains (Abstract, References, doi:, arXiv:, Bibliography), heading_matches, structural (has_math, heading_depth, page_count)
    • extraction tuning: xy_cut reading order for 2-column layout, readability_threshold 0.5
    • fields: title, authors, abstract, doi, journal, publication_date, references

Fixtures (5 expected outputs)

  • tests/fixtures/profiles/scientific_paper/arxiv_paper-expected.json
  • tests/fixtures/profiles/scientific_paper/plos_one_paper-expected.json
  • tests/fixtures/profiles/scientific_paper/ieee_paper-expected.json
  • tests/fixtures/profiles/scientific_paper/nature_paper-expected.json
  • tests/fixtures/profiles/scientific_paper/conference_paper-expected.json
  • tests/fixtures/profiles/scientific_paper/README.md
  • tests/fixtures/profiles/scientific_paper/PROVENANCE.md

Tests

  • crates/pdftract-cli/tests/test_scientific_paper.rs - Comprehensive regression tests including:
    • Profile schema validation
    • Fixture structure verification
    • Expected output consistency checks
    • Match predicate validation
    • Fixture diversity verification
    • xy_cut reading order verification
    • DOI regex format validation

Acceptance Criteria Status

PASS

  • profiles/builtin/scientific_paper.yaml validates (follows Phase 7.10 schema)
  • 5+ fixtures with expected outputs (5 fixtures covering arXiv, PLOS ONE, IEEE, Nature, conference proceedings)
  • tests/profiles/test_scientific_paper.rs passes (10/10 tests passing, 2 skipped integration tests awaiting Phase 7.10 implementation)

Test Results (2026-05-27)

cargo nextest run -p pdftract-cli --test test_scientific_paper
────────────
     Summary [   0.011s] 10 tests run: 10 passed, 2 skipped

All tests pass:

  • test_doi_regex_format
  • test_provenance_completeness
  • test_scientific_paper_profile_exists
  • test_scientific_paper_profile_schema
  • test_xy_cut_reading_order
  • test_scientific_paper_match_predicates
  • test_fixture_diversity
  • test_fixture_count
  • test_scientific_paper_fixture_structure
  • test_expected_output_consistency

Skipped (awaiting Phase 7.10 profile loader):

  • integration_tests::test_load_scientific_paper_profile
  • integration_tests::test_scientific_paper_extraction_accuracy

Profile Fields

Field Extraction Strategy
title region: top_quarter, pick: largest_font
authors region: top_quarter, pick: nearest_below, after: title
abstract near: ["Abstract"], region: top_half
doi regex: 'doi[:.]\s*(10.\d{4,9}/[\w-._;()/:]+)', parse: string
journal region: top_eighth, pick: first
publication_date near: ["Published", "Received", "Accepted"], parse: date
references region: bottom_half, after_heading: References

Notes

  • 2-column layout handling via xy_cut reading order is critical for IEEE-style papers
  • DOI regex matches canonical doi.org format (10.NNNN/...)
  • Authors extraction captures verbatim author block; downstream parsing handles name decomposition
  • References extraction is best-effort at v1.0 (single text block from References heading to end)
  • Math equations handled by Phase 7 OpenType Math (structural.has_math signal)

TODO for Future

  • Add arxiv_id field for arXiv-specific paper IDs
  • Per-field accuracy testing once extraction implementation is complete
  • Classifier corpus evaluation (50-paper subset) for precision/recall metrics