PDF QA — the Pantheon canonical record

What was reviewed

Checklist

• Leaked placeholders & ugly markers

  Pass

  No [DATE], [TBD], [X], undefined, null, NaN or raw underscore fill-lines anywhere in the nine pages.

• Signature / date block

  N/A

  A reading document, not a letter or filing — there is no signature block to vet.

• Page layout & overflow (widow / orphan / split)

  Advisory

  No row or cell splits across a page break and the closing section is not orphaned; two cosmetic balance notes only (page-1 whitespace, no page numbers).

• General professionalism (headings, diacritics, footer)

  Pass

  Consistent heading hierarchy; Turkish diacritics correct; no debug or internal-note text in the footer.

• Rendering fidelity (icons, copy layer)

  Advisory

  Uneven emoji rendering and a copy-layer that drops spaces at in-cell line wraps — both cosmetic / machine-only, neither visible on the page.

Findings & corrections (4)

1. 1. Copy / text layer

   Advisory

   Inside the table, line wraps drop the space between wrapped tokens in the PDF's copy layer, so an agent id that reads correctly on the page (e.g. "gu-holding-ceo") extracts as a run-together string ("gu-holdingceo"). The visible page is clean; only copy-paste / parsing is affected.

   Fix No source change needed for human delivery. For a machine-readable roster, point parsers at uzun.uk/roster.json (already published) rather than the PDF copy layer — the PDF is the reading surface, the JSON is the data surface.

2. 2. Emoji rendering

   Advisory

   Icon glyphs render unevenly — some in full colour, some as a monochrome / outline fallback — a known font-substitution limitation of the PDF pipeline's emoji handling. Cosmetic only; every row is still fully legible from its code-name and function.

   Fix Acceptable as-is. For uniformity, standardise the pipeline on one emoji font, or drop the decorative icon column from the PDF entirely (the code-name and function carry the meaning).

3. 3. Pagination furniture

   Advisory

   A nine-page reference record carries no page numbers; the generation timestamp and source-file footer appear on the final page only.

   Fix Optional: add a running "Page X / 9" footer so a canonical record can be cited page by page. Not required for delivery.

4. 4. Page-1 balance

   Advisory

   After the title, theme and Human-Gate prose, only the first table row fits on page 1, leaving noticeable trailing whitespace before the table continues on page 2.

   Fix Optional: place `[[pagebreak]]` on its own line before the "Panteon (39 ajan)" heading in the source so the full table starts clean on its own page. Leaving it as-is is also acceptable.

Verified clean

• No leaked placeholders — no [DATE], [TBD], undefined, null, NaN or raw underscore fill-lines across the nine pages.
• Turkish diacritics render correctly — İ, ş, ğ, ı, ö, ü and the circumflex â (yapay-zekâ) are all intact.
• The six-column table header repeats at the top of every page (pages 1 to 9).
• No row and no cell is split across a page break; the closing "Uyum notu" section sits with the table on the final page rather than being orphaned.
• Heading hierarchy is consistent (title, then section headings, then table); no debug or internal-note text leaked into the footer.
• The title date (2026-07-12) versus the build timestamp (2026-07-11 17:16) is a deliberate forward-date for a canonical record — verified consistent by design, not an error.

Provenance