07 — Brainstorm: extending the hold on the PDF world
Ideas beyond the feature list, ranked by leverage per effort. Each names the proof that would make it real.
- E-invoicing as the wedge. The French reform is live (reception for all since
2026-09-01), Germany/Belgium/Poland/Saudi/Malaysia follow in 2026, SMEs in 2027. Every
accounting package needs a validated PDF/A-3 hybrid. We already have PDF/A-3,
/AF, XMP extension schemas; missing: EN 16931 XML and the national validators. First customers, first revenue, first "reference" claim (Mustang/KoSIT/ZATCA 100 %). - Post-quantum PAdES, first. RFC 9882 exists, ETSI TR 119 619 is a draft, Adobe has nothing shipped. A validated PQ/hybrid-signed PDF (two independent verifiers) positions ASKLERA ahead of Adobe and in line with the PQC-hybrid-by-default PRD.
- Generic smartcard/HSM layer as a product. PC/SC + PKCS#15/IAS-ECC + PKCS#11 + CNG + CTK + SKF + cloud KMS + CSC behind one trait; OpenSC as oracle. Enterprises pay three vendors for this today; libcps-rs becomes one module in a universal stack.
- Hardening report. Replay PDFex, shadow attacks, signature wrapping, JBIG2/JPX decoder attacks (FORCEDENTRY class) and publish the method. "More secure than Adobe" becomes a measurement.
- Contribute the deviations we found. Arlington, veraPDF (59 TS 32005 pairs, 8.10.3.5-1), jbig2dec, poppler (JBIG2, CCITT), ITU T.88 vectors (TT5–TT8 violate Table 31), PRC WD inconsistencies, pdf-issues errata — each ticket with file and clause is public proof of expertise.
- Extension prefix
ASKL. Register with the ISO 32000 registration authority and publish via the PDF Association registry: PQ signature metadata, Zstandard filter proposal, typed/AFMIME registry, signed-2D-barcode binding. Whoever writes the extension writes the standard. - MRC compression. We own JBIG2 and JPX encoders (rare in the open); a scan compressor beating LuraTech-class ratios is measurable and sells to every scanner and ECM vendor.
- SafeDocs at scale. Read 8 million PDFs without a panic and publish feature-usage statistics; nobody else does it with an engine they fully control.
- CUPS filter and OS hooks. Ghostscript and MuPDF are AGPL; printer vendors pay
Artifex to avoid it. An
apdftorasterunder an acceptable licence (D-1) is an industrial adoption vector; IFilter/QuickLook/thumbnailers put us on every desktop. - The licence gap. The open market is split between permissive-but-incomplete (pdfium, pdf.js, qpdf, pdfcpu) and complete-but-AGPL/commercial (MuPDF, Ghostscript, iText, PyMuPDF). A complete library under an enterprise-readable licence takes that seat.
- OFD ↔ PDF. China's fixed-layout format (GB/T 33190) with its e-seal (GB/T 38540) is served by no Western library; a converter/validator (oracle: ofdrw) opens a market the others do not even see.
- Signed 2D barcodes as one engine. 2D-Doc, ICAO VDS/VDS-NC, EU DCC, SMART Health Cards, Aadhaar QR, ZATCA TLV, IRN JWT, Swiss QR-bill, EPC QR, BCBP, AAMVA — one encode/decode/verify module with trust lists; every government and airline document carries one.
- AFP/XPS/CGM importers. Mainframe output (AFP), Windows print (XPS) and aerospace/defence illustrations (CGM in S1000D/ATA) are converted by expensive niche vendors; importers into our writer capture that spend.
- CDR by construction. Rebuilding a PDF from the validated object model is Content
Disarm & Reconstruction; the CDS market (OPSWAT, Votiro, Glasswall) is priced per
gateway.
apdf scrub --rebuildwith the NSA/NCSC guidance as the checklist. - eCTD PDF profile. Every pharma submission worldwide must satisfy the ICH/FDA PDF specification; a linter profile with Lorenz-class accuracy is a product on its own.
- VPAT/ACR generator. ADA Title II (2026/2027) and the EAA make accessibility reports mandatory purchases; we already have the findings, we only need the report.
- "MDN of PDF" documentation. Clause-by-clause public documentation with our tests as examples: search traffic, training, authority.
- Structure-first extraction for AI. Sell proven reading order (UA, bidi) to RAG pipelines that guess with models; benchmark publicly on OmniDocBench.
- PDF time machine. Revision forensics as a product (who signed what, when, what changed): legal, insurance, banking, health.
- Reflow and braille. PDF → EPUB/HTML/BRF via tagged structure: mobile reading and disability markets where Acrobat is weak.
- Government form catalogues. cerfa, IRS, HMRC, Elster, ATO pre-fill with a JS-subset sandbox: 100 % of real forms usable without Acrobat.
- Moodle/LMS and Office add-ins. Replace Ghostscript in Moodle's annotation pipeline, add "export verified PDF/A-UA" to Word/LibreOffice: adoption through the tools people already use.
- GPU rasteriser and portable SIMD once F-004 says where it pays; no
unsafe, ADR for wgpu. no_stdcore (apdf-object,apdf-parse) for printer and scanner firmware.- Conformance badge. Publish the open suite (F-061), let engines run it, issue "apdf-verified" for producers whose output passes — a de-facto certification nobody currently offers below veraPDF's PDF/A scope.
- Interop workshops. Attend PDF Association PDF 2.0 interop events with our engine; file the results in the Observatory.
- Bug bounty once distribution allows: "break apdf" on the hardened targets.
Added at the pause (2026-09-14 evening) — what the vendoring authorisation (D-13) unlocks, and last ideas
- A complete font safety net. With Noto CJK, Noto Sans/Serif, DejaVu and Liberation shipped (all OFL/permissive), no document renders blank for lack of a font — a claim poppler and mupdf cannot make (they depend on the host). Measured target: 0 pages in the corpora "without substitute".
- Emoji and colour fonts flattened. Noto Color Emoji (OFL) / Twemoji (CC-BY) vectors embedded as Type 3 or image glyphs by the writer; COLR/SVG fonts flattened when embedding.
- Offline trust stores with provenance. Dated snapshots of the EU LOTL/TSLs, AATL, ICP-Brasil, CCA India, JPKI roots and the Mozilla store shipped as data with SHA-256 and a refresh command — signature validation that works air-gapped (defence, archives).
- Format identification built in. PRONOM signatures (open) vendored:
apdf info --puididentifies the exact PDF flavour the way archivists' tools (DROID/Siegfried) do — and we can extend the PRONOM registry with the flavours we detect (PDF/UA-2, PDF/A-4e…). - Code lists for e-invoicing (EN 16931, UN/CEFACT, Peppol, ISO 4217/3166) vendored: the XML half of a hybrid invoice is validated offline, no network at signing time.
- Hyphenation for 70+ languages from the permissive subset of TeX hyph-utf8 patterns, with the licence of each language recorded; the writer breaks lines like TeX.
- A "reference viewer" scoreboard. With Acrobat Pro as a manual tie-breaker (D-24), the Observatory can publish, per feature, the viewer that agrees with the ISO text — the first public, reproducible "who is right" table for PDF.
- Corpus of our own. Every synthetic fixture and negative control written by the lanes (thousands by now) becomes the open conformance suite (F-061); under D-13 the permissive test vectors (UCD, ICAO, ANTS, NIST PIV, ITU/ISO conformance streams where licensed) join it, the third-party corpora never do.
- Reference data as a product. The atlas of national profiles (03a/03b) maintained as machine-readable data (JSON: jurisdiction → signature law, formats, validators, dates) is itself sellable and citable — the "Nutshell" the industry lacks.