C2PA/JUMBF Support Plan

Architecture (from ExifTool research)

Container Format → JUMBF Detection → Jpeg2000::Main (JUMBF boxes) → CBOR.pm (manifest/merkle)
       │                                            │
       │                      ┌────────────────────┴────────────────────┐
       │                      ▼                                         ▼
       │               JUMD boxes (description)                  CBOR data (manifest)
       │                      │                                         │
       │                      ▼                                         ▼
       └────────────────── JUMDType, JUMDLabel,                claim_generator, actions,
                          JUMDToggles, JUMDID,                    assertions (actions, etc.),
                          JUMDSignature                           exclusions, items, etc.

Key modules in ExifTool:

• Jpeg2000.pm — core JUMBF box parser (ProcessJUMB, ProcessJUMD), JUMD tag table
• CBOR.pm — CBOR decoder with C2PA-specific tag table (dc:title, authorName, documentID, etc.)
• JPEG.pm / PNG.pm / PDF.pm / QuickTime.pm — format-specific JUMBF detection, hand off to Jpeg2000

Our test assets currently lack C2PA. Library sweep found 1 file with C2PA tags (the MISSING tags above).

---

Implementation Scope

Phase 1: JUMBF Box Infrastructure (core)

• src/format/jumbf.ts — JUMBF box parser (mirrors ProcessJUMB/ProcessJUMD)
  • JUMBF box header parsing (size, type, UUID)
  • JUMD description box parsing (type UUID, flags, label, ID, signature)
  • Sub-box dispatch (C2PA manifest = CBOR, merkle = CBOR, etc.)
  • Multi-level sub-document tracking (jumd_level / DOC_NUM)

Phase 2: Format Integration

• src/format/jpeg.ts — detect APP11 "JP" prefix, delegate to JUMBF parser
• src/format/png.ts — detect caBX chunk, delegate
• src/format/quicktime.ts — detect C2PA UUID (d8fec3d6-1b0e-483c-9297-5828877ec481), delegate
• src/format/pdf.ts — (later; PDF embedded files)

Phase 3: CBOR Parser (for manifest/merkle)

• src/exif/cbor.ts — CBOR decoder (RFC 7049 subset for C2PA)
  • Major types 0-7, indefinite length not needed for C2PA
  • Tag handling (date/time string, epoch time, bignum, decimal fraction, COSE, URI, base64, etc.)
  • Output into tag table (dc:title → Title, authorName → AuthorName, etc.)

Phase 4: C2PA Tag Tables

• JUMD tags: JUMDType, JUMDLabel, JUMDToggles, JUMDID, JUMDSignature
• CBOR/manifest tags: ClaimGenerator, ClaimGeneratorInfo*, Actions*, Exclusions*, Assertions*, Signature, Items*
• Map ExifTool tag names → our camelCase convention

Phase 5: Test & Verification

• Add C2PA test asset (or use library file)
• Parity sweep: reduce MISSING C2PA tags from ~25 → 0
• Integration tests for JPEG/PNG/QuickTime

---

Scope Notes

Area	Effort	Notes
JUMBF core	Medium	~300-400 lines; box parsing is well-defined
Format hooks	Low	3-4 format files, ~20 lines each
CBOR decoder	Medium	~200 lines; RFC 7049 subset; already have JSON helpers
Tag tables	Low	Straight mapping from ExifTool CBOR::Main + C2PA spec
Total	~1-2 days	Can be incremental; each format works independently

---

Decision Points

1. CBOR implementation: Full RFC 7049 decoder vs C2PA subset?

   • Recommendation: Full subset (major types 0-5, 7, tags 0-36, 55799). ~200 lines. ExifTool's CBOR.pm is ~500 lines with COSE/encoding. Our subset can be smaller.

2. JUMBF as block save: ExifTool has -jumbf:all to extract raw JUMBF blob. We should support -b/--binary for JUMBF tag.

3. Multi-format test assets: Need at least one JPEG with C2PA, one PNG with C2PA. Can extract from ExifTool test suite or create minimal synthetic.

4. Parser integration point: Current parseXMP / parseICC pattern → new parseJUMBF(data, format) called from format parsers.

---

Acceptance Criteria

• exiftool -jumbf:all -G3 file.jpg equivalent output for tags:
  • JUMBF:JUMDType, JUMBF:JUMDLabel, JUMBF:ClaimGenerator, JUMBF:ActionsAction, etc.
• Library parity: C2PA MISSING tags → 0
• Coverage: 100% for new modules
• No regression on existing 478 tests