pdf/corpus/README.md

Test PDF corpus

Reference PDFs used by rendering, text-extraction, and regression gates.

Do not commit large or confidential PDFs here. Binary PDFs are tracked by git but the corpus is expected to grow — once it does, move it to Git LFS or an external store and keep only a manifest here.

Structure (populated during Phase 1):

corpus/
  basic/        # clean, simple PDFs — smoke tests
  fonts/        # embedded fonts, CID/CJK, Arabic/RTL
  edge-cases/   # broken xrefs, scanned, PDF/A, encrypted
  golden/       # reference PNG renders for pixel-diff tests
  manifest.json # per-file metadata + expected results

Risk-register rule: every client PDF that surfaces a bug gets added to the corpus immediately.

Gates that consume this corpus:

• G1/G2/G3 (Phase 1): 30-PDF corpus, render + text extraction.
• G7 (Phase 3): 500-PDF regression corpus, SSIM > 0.95.