# Complete navigation in short viewports

The center now keeps its complete guide title, Topics and Guide details in one
row at viewport heights of at most 480 CSS pixels. Secondary actions and guide
metadata remain in the existing dialog, which opens at its actions. The short
drawer scrolls as one document. Focus survives controls being hidden by a
breakpoint, and guide selection moves focus to the presentation.

Baseline: `a0a201457bcefa5e05cfd47afccb859a6d26330f`. Source pin:
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`. The library remains 40 guides, 421
unique slides, 447 guide instances and 287 cited sources. The
[implementation brief](../../authoring/center-short-viewports.md) applies the
mandatory frontend skill to the existing cream/ink/terracotta workspace.

At actual 400% tab zoom in a 1280×900 Chromium window, the content viewport is
320×203 CSS pixels. All 40 guide titles fit at 14px in a 57px toolbar; the
iframe retains about 146 CSS pixels of height. The baseline left only 100 and
truncated the selected title. These measurements describe this fixture, not
whole-slide first-screen fit or reader-approved legibility. Ordinary-height
layouts retain the existing two toolbar rows.

## Verification

- The new Playwright suite checks all 40 titles and details at actual 200%/400%
  tab zoom, plus narrow, landscape, short desktop, ordinary desktop/mobile and
  breakpoint transitions. It verifies zoom settings, DPR, physical screenshot
  dimensions and ink; exact metadata and URLs; keyboard traps and focus; search,
  history and prerequisites; a real hosted PDF download and separate window; and
  audio interruption. Its 84 title records, 80 details records, twelve named
  interaction/accessibility checks and 102 captures pass. Six deliberate defects
  are rejected. Viewport emulation is labeled separately.
- The complete existing center suite passes all fourteen checks: all 40 guides,
  894 embedded slide captures, 80 real audio end/replay boundaries,
  accessibility, navigation timing and all 40 fresh PDFs. No subset flags are
  used.
- The complete table suite passes 1,180 checks across 109 tables, with 310
  captures and ten rejected defects. It includes real browser zoom, embedded
  playback, enlarged text and the existing font probes.
- All 226 Python and 274 Node tests pass, together with scoped Ruff and ESLint,
  builder/preview freshness, documentation coverage, repository discovery,
  workflow inventory and coverage, and all 421 audio
  integrity/decoding/duration/ encoded-pause checks. There are no TypeScript or
  app/native API changes.

Checks run serially through the repository resource supervisor. Logs preserve
memory/disk observations; published copies omit PID-prefixed host-process rows
and normalize ANSI escapes, trailing whitespace and blank EOF lines. Raw and
retained hashes and transformation counts are recorded. Terminal exit records
establish completion independently of an early report's `completed` field.

Development attempts remain historical in `history/` and the logs. They include
harness corrections for iframe loading, per-tab zoom reset, actual Narrate
controls, HTTP download semantics and isolated negative controls; real resize
focus failures; and the stale dialog scroll found visually. The first full
center run was deliberately stopped before that dialog repair and superseded by
the complete passing final run. Its partial exports were not silently restored.
An apparent alignment discrepancy did not reproduce in computed styles or exact
PNG comparisons; the harness additionally waits for iframe fonts/layout before
capturing. Archived diagnostic scripts retain their original local paths.

## Review and preservation

All 80 selected-title headers were inspected as unscaled physical pixel crops.
The final crops exactly match the sixteen inspected sheets;
`verify-header-crops.py` rechecks every pixel. All 22 additional
UI/dialog/drawer/reading captures and all 72 changed table captures were
inspected directly at their physical sizes. Initial, reopened, top, ending,
table and last-column states stay separately labeled; the short suite also
records focus and scroll positions. The 238 other table captures have exact PNG
equality to the prior reviewed receipt.

All 894 ordinary center captures are reconciled. Of these, 730 have exact PNG
equality to the prior full center. The other 164 differ by only 297 pale edge
pixels, 1–10 per image, at the guide-position or left table-rule edge. Complete
before/after values and current PNGs are retained; `verify-center-pixels.py`
reconstructs the hash-verified prior bitmap and requires equality everywhere
else. This is bounded unchanged-output evidence, not a fresh every-slide optical
review or reader comprehension result.

All 835 preserved authored/delivery files, including recordings and standalone
HTML, match the baseline. Only `render_center` changes in the builder's
top-level AST. The complete standalone bank's prior twelve-check, exit-zero
result is inherited only through its exact fixture bytes and 67 unchanged
browser/renderer inputs. No fresh bank, native implementation or model pipeline
execution is claimed.

The workflow freshness check also exposed a pre-existing stale repository-index
hash. Regeneration changes only that hash and the dependent teaching-report
index hash; the preservation checker proves these two exact substitutions.
Workflow inventory and coverage checks then pass. No declaration units or
teaching assignments change, and the 3,718 unassigned units remain open.

The forty PDF exports preserve every byte except CreationDate/ModDate and, for
one guide, a bijective renumbering of structure-node identifiers. Streams,
content and tag-reference relationships remain identical. The PDF preservation
checker rejects changed payloads and changed reference relationships. Existing
page review transfers through this exact allowed-difference proof; no new PDF
optical or reader review is claimed.

Whole-library typography and composition, isolated filename fragments, listening
at normal speed, willing-reader tasks, source expansion, Eve integration and the
remaining exhaustive checklist are still open. This receipt completes the
specific short-viewport navigation repair.

## Recheck retained evidence

From the repository root, with the ignored bank fixture regenerated if needed:

```sh
node docs/presentations/presentation-center/verification/center-short-viewports-2026-09-07/scripts/verify-retained.mjs
python3 docs/presentations/presentation-center/verification/center-short-viewports-2026-09-07/scripts/verify-header-crops.py
python3 docs/presentations/presentation-center/verification/center-short-viewports-2026-09-07/scripts/verify-center-pixels.py
```

The pixel checks use Pillow. The preservation scripts can additionally recompare
against the baseline Git revision. Fresh browser reruns require the repository's
Playwright/Chromium and axe dependencies and a fresh output directory. The full
center test re-exports the delivered PDFs, whose new bytes require a new
receipt. Historical receipts are immutable.
