# Shared table typography · implementation brief

Baseline: `6c3186d2a4bac18f17acdbb63b3177759f32d6df`. This follows the two
explicit connection-quality integration findings. It is not permission to close
the complete typography, visual-benchmark or reader-comprehension standard.

Visual thesis: keep the cream/ink/terracotta comparison plane, with readable
authored-case headers and intact quantities; size and alignment carry hierarchy
without extra boxes, icons or decoration.

Content plan: preserve every table's exact labels, values, units, order, source
links and spoken explanation. Bring ordinary headers to the existing 16px
quantitative-header floor. Protect recognized number/unit expressions as inline
text groups, then give narrow row identities enough room to remain beside their
values. Do not turn independent rows into a process or change a quantity to fit.

Interaction thesis: retain keyboard/touch horizontal scrolling, sticky row
identity, accessible source disclosure and slide navigation. Existing
transitions remain reduced-motion aware. Static PDFs must keep complete content
without depending on a scroll or reveal; no new animation is needed.

## Intended contract

- All slide-table column headers use authored capitalization, 16px minimum,
  normal letter spacing and existing semantic `th` elements. Compact active
  narration must not restore the old 11px rule. Print at the existing
  13.333-inch page width gives approximately 12 effective pixels at 960px page
  fit, versus the previous 8.25. This extends an existing engineering floor; it
  is not a newly claimed reader-approved threshold.
- Native inline groups keep recognized number/unit expressions on one line.
  Preserve signs, decimal/grouping/range punctuation, spacing and source case;
  never normalize or convert the displayed quantity. Version labels, API paths,
  bare values and unknown words must not be guessed into units. Source escaping
  and optional identifier breaks remain intact outside protected quantities.
- Apply the shared treatment to affected tables across the library, not just
  `v2-quality-score`. Row identities must retain space for a readable value
  column. Resolve inadequate width through the table's shared sizing contract or
  meaningful authoring, not reduced type, clipped units or disabled checks.
  Mobile native cells use 14px right padding, retaining separation while making
  room for complete heading words. Ordinary row identities range from 120–146px;
  quantity rows and opted-in exact-source rows use 132–146px. The shared width
  calculation reserves a 200px value column when available, letting the row
  identity grow within its cap; at 320px it retains the smaller row floor. This
  accommodates enlarged “Dependency” and “Configuration” headings without
  reducing type or giving the row identity more than 55% of the reading width.
- Exact numerical/provenance data and narration stay unchanged. Source/table
  text remains selectable, searchable and available to assistive technology.

## Verification and review

Record the baseline and current affected table inventory. Add Python escaping,
quantity-recognition, exact-character, speech-preservation and PDF case/order
regressions. Extend the shared browser audit for header floors and intact
number/unit text ranges so full standalone and center suites exercise the rule.
Require deliberately small headers, split groups, hidden content and lost units
to fail independent checks.

Export coverage carries exact columns and rows for every native table, including
ordinary prose tables. The PDF checker validates their authored characters and
order, not only titles or typed quantitative rows. Poppler XML identifies the
actual terracotta header runs, authored case and integer point sizes, with a
12px effective floor at 960px page fit. Preserve the integer-size limitation;
these metadata checks do not measure precise glyph shape or optical balance.
Bind table-header font resources to the PDF's embedding and Unicode metadata. An
actual PDF exported with 11px headers must fail the size gate.

Optical inspection exposed a second export failure: the scrolling wrapper could
shrink vertically inside the print flex layout while the complete text remained
extractable. The Metis dimensions PDF clipped the final `g` descender exactly at
the wrapper boundary. A 1280px print-layout probe found nine overflowing tables,
four with final-row text crossing that boundary. Print now keeps each wrapper at
its natural table height with visible overflow. It retains all type sizes and
cell spacing. The complete-table guard requires no scrolling or clipped text,
page containment and at least 12 CSS px between the table and surrounding header
and takeaway. Full standalone and guide exports run it at the actual CSS page
width, not a wider screen viewport. The focused regression rejects the restored
shrink rule, a hidden/clipped last label, a table/takeaway collision and a wrong
print viewport. It also retains complete and deliberately clipped PDFs to show
why passing text extraction is insufficient. All 40 guides have now been
re-exported and the final PDF evidence distinguishes directly inspected current
pages, exact matches to inspected repair probes and exact unchanged-region
inheritance for restored outer-edge ink. The final 421-slide bank rerun passes
all twelve checks with observed exit zero; the
[evidence receipt](../verification/shared-table-typography-2026-09-07/README.md)
does not treat a passing text check as optical acceptance.

Whole-inventory browser checks cover ordinary, quantitative and exact-source
tables at 1440×900, 1280×720 with actual playback, 390×844 and 320px narrow
reading, plus focused actual expanded/collapsed embedded frames. Add
whole-inventory 20px text enlargement and explicit missing/preferred/fallback
font probes. Record actual platform fonts and PDF font embedding/substitution
where available; font-family declarations alone are not evidence of the face
used. Label CSS enlargement separately from browser zoom and do not claim
unsupported accommodation states.

Use a disposable Chromium profile and a permission-free local test extension for
actual 200% and 400% tab zoom in a 1280×900 browser window. Assert the
browser-reported zoom, changed CSS viewport and device pixel ratio, unchanged
window dimensions, CSS zoom of 1 and visual-viewport scale of 1. Check all
native tables, plus focused embedded and playing contexts. Retain separately
labeled initial, table-start and table-end viewport captures; a scrolled zoomed
view is not a whole-slide fit claim. This follows the
[Playwright extension harness](https://playwright.dev/docs/chrome-extensions)
and
[Chromium tab-zoom API](https://developer.chrome.com/docs/extensions/reference/api/tabs#method-setZoom).
The fixture is never installed into a user's profile or shipped with the center.
Zoom images use direct Chromium viewport capture. Verify their physical pixel
dimensions and visible ink, and deliberately reject cropped and empty captures;
the ordinary screenshot helper produced a blank 320×203 crop at true 400% zoom
on this host. Record inner/outer dimensions, pixel ratio and scroll positions
alongside the full physical viewport image.

When a focused narrow table overflows horizontally, retain explicitly labeled
last-column captures in addition to the initial left edge. Embedded and zoomed
contexts include both the header and final-row vertical positions at that last
column. Recorded scroll offsets distinguish horizontal completion from a
vertical table-end capture; neither is a whole-slide first-screen fit claim.

The shared authored-case header rule also reaches source excerpts. Their
existing 15px source-text rule and independent keyboard-focusable scroll
container remain separate from slide-table sizing and sticky row identity. The
focused source dialog regression exercises the pinned Tara surface table at
desktop, mobile and 320px widths, checks exact disclosed cells and
accessibility/focus return, and rejects small/uppercased headers or leaked
sticky positioning. At 320px it also exposed a baseline provenance-note
overflow: the unbroken forty-character revision extended beyond the dialog. The
provenance note now permits wrapping without changing, hiding or shortening that
exact revision. A restored unwrapped note must fail actual text-range
containment. Narrow optical inspection also found ordinary source headers and
row labels broken into short fragments. Source cells now retain a 160px minimum
reading width within their own horizontal scroll container. Actual arrow-key
scrolling, left/right endpoints and intact header words are checked;
deliberately removed minimum widths must fail. The source table does not inherit
slide-table sticky row positioning. All 482 embedded source-table specimens from
124 source templates are also checked in the dialog's desktop, mobile and narrow
reading widths, with every column visited. This isolates table geometry; it is
not a whole-source flow or optical acceptance claim. Captured headers and final
rows must be placed below the sticky dialog header, not merely scrolled into the
browser's layout bounds. Their endpoints need their own review; this is not a
new 15px slide-header exception or source-reading acceptance threshold.

Review current full-size and 960px page-fit PDFs, normal-size browser captures,
quantity-row identity after scrolling, and complete-guide context. Run all
applicable existing focused regressions, complete bank/center suites, audio
integrity and freshness gates serially under the resource supervisor. Keep
invalidated historical receipts immutable and retain current evidence
separately.

Willing-reader interpretation, normal-speed listening and the rest of the
complete typography specification remain required. No automated size or geometry
score establishes optical balance or comprehension by itself.
