viz: link the localized gallery examples, and fix the two dashboard bugs they exposed

[jqnatividad]

Both localized smart dashboards were already deployed to GitHub Pages — the viz-gallery-pages workflow stages examples/viz wholesale — but gallery.html never linked them, so they were reachable only by direct URL. The commits that added them (5a2ba093e, d498b8873) didn't touch gen_gallery.py, so the generator had no entry to render.

Fixing that surfaced two dashboard bugs, which are the second commit.

1. Gallery entries (b2d74cda0)

Added as SCREENSHOTS link-outs rather than embedded iframes: at 17.0 MB and 5.0 MB they're far past the 0.1–1.5 MB range of the embedded figures, so embedding would have added ~22 MB to every gallery page load. Matches the existing Pittsburgh/Boston treatment. Neither dataset is committed, so both carry a verbatim cmd.

README.md was stale on two counts gen_gallery.py's own consistency check flags — the link-out count (three → five) and the page size range (3.6–6.9 MB → 3.9–20.6 MB) — plus the dictionary table was presented as a complete list while omitting both new dictionaries.

2. Notable Characteristics rendered in English (637001ba5)

Inside otherwise fully Spanish/Portuguese dictionary drawers.

The obvious fix would have broken OKF export. okf_short_description (describegpt.rs:5513) finds that exact English literal to cut the lead prose it needs for OKF's frontmatter description: key. The string is load-bearing and has to survive in the stored schema — prompting the model to translate it would silently break OKF short-description extraction on every non-English run.

So it's localized at render time, after all marker parsing: localize_dict_heading swaps the one structural literal for a curated catalog string at the single ATX heading render site. Verified on the Colombia dashboard — 0 occurrences of the English heading in its HTML, 1 still in its schema. Uncurated locales and reworded headings fall through to English, i.e. to today's behavior.

3. KPI tile labels collided with their neighbours

wrap_kpi_label wraps against a character budget fixed at generation time, but the KPI row spans ~550px to ~1800px on the same page — narrow windows, and the --dict-info drawer opening, which shrinks the plot with no window resize event. No character count is right at both ends.

KPI_REWRAP_SCRIPT re-wraps against the tile's measured pixel width on every resize, via a ResizeObserver on the plot div (a window listener would miss the drawer case). It finds its labels by a name marker rather than annotation index, since the two emission sites produce different index layouts, and derives tile width from the pitch between adjacent label centers so nothing extra is serialized. The Rust wrap stays as the pre-JS baseline, ceiling lowered 36 → 28 (measured: ~6.7 px/char, tiles ~183px at the narrow case).

Verified in-browser, not just compiled

condition	plot	tile	result
1070px, drawer open	549px	183px	no overlap; widest label 171px ≤ 183px
2360px, drawer open	1808px	~600px	labels re-join into single lines

The wide case is what proves the JS is doing the work — the static wrap alone would leave them wrapped forever.

Regenerated artifacts

The Colombia dashboard, its preview image, and the embedded gallery figures, so the fixes are live in the published examples. Per page the diff is the script block, a Compiled: timestamp, and name attributes only where a KPI row exists — confirmed byte-exact on three pages. Colombia's extra 26 characters are two new <br>s from the tightened baseline wrap.

Not included

Six committed artifacts still render the English Notable Characteristics, because gen_gallery.py cannot refresh them without their uncommitted source data: pitt311data, smart_boston_311_2025, smart_dict_sunburst, smart_dict_treemap, smart_geospatial, smart_world_choropleth. Two of those (pitt311data, smart_boston_311_2025) are link-out dashboards reachable from the gallery, so this is a user-visible residual gap, not just an internal one. Refreshing them needs the datasets and each entry's cmd.

Testing

295 unit + 395 integration viz tests pass; cargo clippy --bin qsv -F all_features clean.

One test needed a fix worth noting: render_dict_markdown_atx_headings began failing by returning Italian — localizing that render path made the assertion locale-sensitive, and it wasn't holding the english_locale() mutex against concurrently-running tests. Added the guard plus two new tests covering the localization (including that non-marker headings pass through untouched) and the marker/script agreement.

🤖 Generated with Claude Code

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}