PDF to PDF/A: archival-compliant PDF

[mithun-srinivas]

Short Description

Convert a normal PDF into a PDF/A file for long-term archiving (self-contained, embedded fonts, no external dependencies). Sejda offers this. Needs research to confirm what level of compliance is achievable in-browser.

Scope

• Follow the plugin pattern: a self-contained src/operations/<id>/{meta.js,index.jsx,helpers.js} folder (auto-discovered, no central wiring), reusing shared pieces (Dropzone, useJob, Note, Progress, ResultGallery/ImageResult).
• New operation src/operations/pdf-to-pdfa/.
• Research client-side PDF/A conformance with pdf-lib: embed all fonts, set the OutputIntent / ICC color profile, add the PDF/A XMP metadata, and remove disallowed features (encryption, external references, transparency where required).
• Target a realistic level first (for example PDF/A-2b) and state it clearly.
• Be honest in meta.notes that this is a best-effort conversion, not a certified validator.
• Must stay 100% client-side: no network calls, no CDNs, no external assets (respect the strict CSP connect-src 'self' in index.html).

Acceptance criteria

• Spike documents which PDF/A level is achievable client-side and how it is verified.
• Exported file carries embedded fonts, an OutputIntent, and PDF/A XMP metadata.
• The targeted PDF/A level is stated in the tool's notes.
• Runs fully offline.

[slegarraga]

Taking this one — spike complete first, and it is viable:

Approach: rasterize every page with pdf.js (the same path grayscale-pdf uses), rebuild the PDF as page images with pdf-lib, then inject the PDF/A XMP metadata package by byte-level post-processing (pdf-lib's save() drops custom indirect objects registered after load — verified empirically; the metadata object is appended before the xref table and the catalog gains /Metadata N 0 R).

Target level: PDF/A-2b, stated plainly.

Honest limits, stated in meta.notes: pages become images (text no longer selectable/searchable), fonts are inherently self-contained because they are rasterized away, and this is best-effort structure conformance — not a certified validator run.

PoC already proves: catalog injection works (/Type /Catalog /Metadata N 0 R), XMP with pdfaid:part=2, conformance=B survives round-trip. Full implementation next.

[slegarraga]

Spike findings from the first attempt (#181, now a draft after maintainer review):

• Rasterizing pages through pdf.js + canvas sidesteps the hardest PDF/A problems entirely (fonts, transparency, encryption), so a real PDF/A-2b for the raster path looks achievable client-side.
• The pieces that actually matter: (1) a genuine OutputIntent with an embedded sRGB ICC profile bundled in-repo so it stays offline, registered via /OutputIntents on the catalog and page resources; (2) XMP attached through pdf-lib's object model as a proper stream object with correct xref entries; (3) page bytes via canvas.toBlob to respect connect-src 'self' (no fetch on data: URLs).
• Verification plan: run veraPDF against generated files during development and state the verdict in meta.notes, per the acceptance criteria.

Working through the OutputIntent piece now; notes to follow here.