The Experimental row showed the titleized id ("Gospel Internet") in every
language: tabLabel() translates a discovered tab only when strings.js has
'tab.<id>', and the manifest carries no title. Add 'tab.gospel-internet'
after 'tab.bless-india-posters' in each locale, with the plugin's own
strings. Rebuilt the gallery bundle; it differs from the pilot's only by
these 11 strings.
A host that serves the tree with Last-Modified and no Cache-Control lets a browser run its cached files for
hours. Measured: a typed-URL revisit after a deploy requested none of the gallery, the manifest or the bundles,
so a returning reader ran the old files, or an old manifest against a new bundle.
- The manifest is fetched with cache 'no-cache', so it is always revalidated (a 304 when unchanged).
- A fingerprint of the manifest (every bundle's integrity) is kept per tree. When it differs from the reader's
last visit, every bundle the browser already holds gets one conditional request before any tab mounts:
a 304 refreshes it, and a 200 is cancelled at its headers, which drops the stale copy. The plain <script>
that follows, the gallery's own or a composite's for its children, then runs the new file at the same URL.
- A version query is not used: loadScriptOnce keys its registry by the full href, so X.js?v= next to a
composite's plain X.js would run the bundle twice.
- A first visit with an empty cache costs nothing extra (only-if-cached finds nothing held), and so does an
ordinary visit with an unchanged manifest. The mount waits at most 8 s for the check.
The gallery fills the tab layout's two new slots (the end of the phone bar, and the bottom of the desktop
rail) with the shared ShellLanguageMenu, showing the language in effect. The menu first asks the host with a
cancelable doxa-language-request. A host that takes it changes the language itself and sets `language`.
Otherwise the gallery switches on its own, and that pick holds over a `language` the host never updates, until
the host changes it. A saved choice never outranks the host page's own language. EMBED.md shows the listener.
- library/components/WARNING.md, library/dashboards/README.md and
documentation/contributing/README.md taught "graduate into the library at the second
consumer". They now teach the plugin-designer skill's rule (section 3): a component that
could serve more than one map is built once in library/, driven by props, and landed through
the build owner; one-off or experimental pieces stay in a map's folder; a new pattern is a
pilot, then approved, then rolled out. The contributing guide's "full reasoning" link pointed
to an index without it; it now points to the skill.
- The app shell doxa-maps-spa is labelled Deep Research. Ten docs still called it the Research
Gallery (15 mentions): DEPLOY.md, the desktop how-to, embeddability, develop-a-plugin,
reference-universal-props, reference.md, the shell's EMBED.md, and the plugin-designer skill
with two of its references. Identifiers (doxa-maps-spa, branch and route names) are unchanged;
README.md is not touched.
The case-insensitive census with every plugin agent's name found 30 more lines: library (ProfileLoader,
useMapProvider, ApproxPinNote, listSelection, colocation), the gallery's embed-contract card id, the gospel
plugins' source notes and build scripts, a regions script note and one gate description. Comments and notes
only, except the Gospel x Internet CSV export's provenance line, which no longer names who screened the links.
Quoted remarks, attributions and person-tied dates in comments become plain statements of the same
technical fact, and the names of internal tools and roles become neutral descriptions of the check that
was run. No code, markup or style changes: the rebuilt bundles are byte-identical apart from Vue's
scoped-style ids, which hash each component's source, comments included.
Bible Posters lives on Posterverse's Experimental shelf, so the gallery excludes it, as it already excludes
Posterverse itself. One poster plugin, shown once.
Prayer Mobilization now reads Prayer Progress, UPG 100 List Posters, Country Posters, then Poster Design.
The row mounts the Styles + Sizes composite (bundle poster-design, profile poster-design-mfe), uses the same
tab id as the Posterverse shell, and carries the same label and subtitle in all 11 languages. As a curated
row it never appears on the Experimental shelf.
The tab layout takes an optional homeHref and homeLabel from its host. When a URL is given it draws a
link directly above the rail's Collapse button, in the same box, and a phone shows it at the top of the
drawer. It mirrors in Arabic and slims with the rail. With no URL nothing changes, so the Plugin Studio
and every other host look exactly as before. A javascript:, data: or vbscript: URL is ignored.
The gallery accepts home-href and home-label (or profile-config homeHref and homeLabel) and passes them
on; its own "Back to homepage" is the fallback label in all 11 languages.
Prayer Mobilization now reads Prayer Progress, UPG 100 List Posters (renamed in all 11 languages), then
Country Posters, which mounts its composite view and leaves the Experimental shelf. Posterverse, the
second shell, is kept off this shelf so a shell never opens inside a shell.
The Regions header takes Big Picture's header recipe: its own MapInfoButton and ShareButton in the header form
(shape rounded-square, the vertical tabs' radius), side by side at the header's end, 28 px with a 44 px tap
target; on a phone the ⓘ takes the header's start edge, since the centred view switcher leaves no room for
two at one end. The gallery hands the tab's MapInfo to that button (trigger 'control'), with the floating ⓘ as
the fallback. The map-print progress card no longer shows the renderer's English phase messages.
The header rollout of Big Picture's header-button form. UPG-100's header
now carries a rounded-square info button to the left of a rounded-square
Share (MapInfoButton + ShareButton with shape="rounded-square", the
vertical tabs' radius, Big Picture's .rd-header-btns metrics), replacing
the round Share and the floating round info button. The tab's MapInfo
takes the header button as its trigger (tabInfo 'upg-100'
trigger:'control'), so the floating one stays off while the header one
shows, and a tap on it opens the panel.
On a phone the centred view switcher and two buttons do not share the
row, so below 481px the switcher moves to the start and the buttons keep
the end. Desktop keeps the switcher centred. Rebuilt: my-upg-100-list,
doxa-maps-spa; the tree and SRI regenerated.
The strategy (bottom line + points, 41 keys per locale) moves out of the
mapInfo area into a new mapStrategy area (library/i18n/locales/<lc>/
mapStrategy.json, keyed mapStrategy.tab.<id>), and tabInfo's strategyKey
rows name it. The two tab labels stay in mapInfo.
The build ships an area only to bundles whose code names it, and only
the gallery shell names mapStrategy. mapInfo ships to every map bundle
for its standalone info button, so before this change each map bundle
carried every map's strategy in eleven languages without rendering it.
Measured on the rebuilt bundles against development 63a64d1bf, gzip:
doxa-research-map +3.7 KB (was +47.9), cities-internet +4.0 (was +47.9),
gospel-resources +6.7, research-dashboard +3.8, dashboard-plus-grid
+3.7, doxa-simple-map +3.5; the gallery shell +33.9 KB, where the
strategy renders. check:i18n-treeshake PASS (26 bundles); the strategy
text is in the shell and in none of the six map bundles.
The moved values are identical to the ones they replace (a parsed-JSON
comparison per locale). Rebuilt the seven mapInfo carriers, then the
tree and SRI; pages and manifests change only their integrity.
Each tabInfo row names its strategy literally (strategyKey), for the
area scan, and MapInfo reads the bottom line and points from it. The
rebuild of the consumers follows at landing, on the landed train.
- Simple Research drops the example packet (a UPG-100 zone's, not
something its own view makes) and its takeaway line, which promised a
poster and a prayer packet a reader cannot reach from this map.
- Every tab whose map draws people-group pins (UPG-100, Simple Research,
Prayer Progress, Cities with Internet, Gospel Resources) carries the
approximate-location caveat as its info note (pins.approximate), so
it stays findable where a phone legend no longer states it.
- The Regions comment no longer says China is withheld: China prints.
The gallery bundle grows by about 3.6 KB raw (1.1 KB gzipped) for the pins
catalogue area; the rebuild is in the commit that closes this series.
The research map draws the ⓘ as the first button of its own control column (MapInfoButton), and the
gallery's Simple Research row switches to trigger: 'control' in the same commit, so the tab never
shows two ⓘ, and never shows none. At mount the map tells the gallery that its control is coming, so
the floating ⓘ does not flash while a slow phone loads. A failure or an unmount withdraws that
promise, and the floating ⓘ comes back at once. On a page with no gallery nothing answers, and the
map is unchanged.
The ⓘ becomes the first round control in each map's own column (MapInfoButton, as Prayer Progress
has it), and each tab's gallery row says trigger: 'control', so there is exactly one ⓘ: the control.
MapInfo keeps its fail-open, so a page where the control never arrives still gets the floating one.
Rebuilds cities-internet, gospel-resources and doxa-maps-spa, with their pages and the two manifests.
Every map whose ⓘ is its own control (MapInfo trigger 'control') now gets the same pointer:
it shows the first time that control is on screen in a page load, and goes for the rest of the
load once the reader opens the popup or leaves the tab. A full reload shows it again. This is
page-load memory, never storage, so the popup offers no checkbox in control mode: the "Keep
tooltip turned on" box and its string (all 11 locales) are removed, along with the pointerKeep
prop that selected this model per tab. Floating-mode maps keep their stored model and their
"don't point again" box, unchanged.
Leaving the tab is read from the pane (hidden or inert), on the control's intersection change
and the existing 1.5 s measurement tick.
Rebuilt doxa-simple-map (it ships the mapInfo catalogue) and doxa-maps-spa, then tree + SRI.
The dashboard header gains an info button ("What can this map do for you?") to
the left of Share. Both are rounded squares at the vertical tabs' radius
(--vtl-radius), matching the app's own roundedness; map controls stay round.
Built as a reusable, parameterized header-button form so other dashboards adopt
it by prop, not by copy:
- MapControlButton: new shape="rounded-square" (border-radius: var(--vtl-radius,8px)).
- ShareButton / MapInfoButton: a shape prop forwarded to MapControlButton;
ShareButton also squares its rotating ring so a round ring never wraps a square.
Defaults stay 'round' — every existing consumer is unchanged in behaviour.
- research-dashboard DashboardTopBar mounts the ⓘ and Share as one header-button
group, sized to the bar with the 44px tap target kept for both. DOM order
[ⓘ, Share] puts ⓘ at the logical start: left in English, right of Share in
Arabic (the flex row reverses under rtl).
- doxa-maps-spa dashboard-plus-grid tab -> trigger:'control', so the tab's
MapInfo is driven by the header ⓘ and the gallery's floating ⓘ is suppressed:
exactly one ⓘ, fail-open. No new strings (reuses mapInfo.open, 11 locales).
- Prayer Progress's column becomes the engine's square grouped card (controls="square"). Its first
control is MapInfoButton, which opens the gallery's MapInfo for the tab. The gallery's tabInfo
'prayer-map' entry now uses trigger 'control', so the floating button hides while that control is
on screen, and shows again (fail-open) whenever it is not.
- MapInfo, control mode: the first-visit pointer now points at the control. Its place is measured
against the pane (which scrolls) and goes on the control's physical inner side, so an Arabic page,
whose column mirrors to the left, gets the pointer on the right with the arrow reversed. It shows at
every width: the under-1200px rule was measured for the floating pointer crossing each tab's top
strip, and does not apply here.
- MapInfo, pointerKeep (prayer-map only): the popup offers "Keep tooltip activated" instead of the
every-map "don't point again" box. It is off by default, and opening the popup retires the pointer.
Checked, the pointer keeps showing on every load. It is stored per map.
- mapInfo.keepPointer in all 11 locales.
- Rebuilt doxa-simple-map and doxa-maps-spa, then tree and fix:sri.
At top 46px the button lay in the empty end of Big Picture's filter area, but that bar is a
positioned layer (z-index 55) painted over the button (z-index 20): nothing opened on click (rega,
1440, en and ar). The filter area wraps with width, so no spot below it is free at every width (at
1024 the stats row moves under a fixed offset). pos.raise lifts the BUTTON to z-index 60; the one-time
pointer stays under the bar, so no filter is covered even visually. Measured on the built tree: no
interactive element or text under the button, the button topmost at its centre, and a click opens the
popup, at 1440, 1024 and 800, en and ar.
At 390 Cities' legend spans the same width as Gospel's (344px), and side 56px put the button on
the end of the legend title's first line (gospel's measurement; a probe that also counts visible
text under the button reports the same, 236px2). narrow side 4px, as Gospel.
UPG-100's info popup lists its Middle East Core poster, then the full packet's 12 pages and the 70-card
deck (upg100's pre-generated files). Simple Research shows the same 12 pages, captioned "Example
packet: Middle East Core": same renderer, method and 'full' variant, but a UPG-100 zone's data, so it
is labelled as an example rather than passed off as a Simple Research selection.
At 390 the desktop offsets landed on each composite's own top chrome: Regions' and UPG-100's view
selector and Gospel's search box (probe, en and ar). A tab's pos can now carry `narrow`, laid over
it while the shell is narrow (max-width:760px): Regions and UPG-100 put the button at the START of
their strip, which is empty on a phone; Gospel takes the corner above its zoom column. The same
probe then reports no interactive element under the button on any tab, 1440 and 390, en and ar.
The one-time pointer still lies over the selector or search box until first opened; it takes no
clicks (pointer-events:none).
A geometry probe (every visible interactive element in every shadow root, intersected with the
button's box, 7 tabs in en and ar at 1440) found the default position covering each composite's Share
button exactly, which no screenshot showed. They take side 56px like the map tabs. With that, the
probe reports no element under the button on any tab in either direction; before this branch it
reported Zoom in/out under it on Prayer Progress, Cities and Gospel and Share on Big Picture.
At the default 12px from the top and the end of the line, the button covered the zoom-in button on
Prayer Progress, Cities with Internet and Gospel Resources (a control nobody could click), Big
Picture's header buttons, and Simple Research's tab bar, in English and, mirrored, in Arabic
(rega's 1440 captures, exp's second read). Each tab now carries its placement in src/tabInfo.js,
passed as MapInfo's own --map-info-top/--map-info-side: 56px from the end beside a zoom column,
46px down below Big Picture's header bar, 56px down below Simple Research's tab bar. Regions and
UPG-100 keep the default. MapInfo itself now places by the logical end of the line (upg100), so the
gallery passes no direction.
The four headers are copied verbatim, in all eleven locales, from dashboard.json into mapInfo.col
(peopleGroup, country, region, population); no string is retranslated. Measured with areasForBundle:
doxa-maps-spa carried [buttons, dashboard, grid, mapInfo, messages] and now carries [buttons, grid,
mapInfo, messages]. The bundle drops from 420,172 to 371,064 bytes (136,998 to 124,005 gzipped).
Big Picture's export is a spreadsheet, so its example is a kind:'table' excerpt (sr): the six largest
groups by population, of 2,107, with the live table's own column keys (dashboard.people_group,
.country, .region, .population), so the popup header and the table's header cannot drift.
Cost, measured: those keys bring the 'dashboard' area into the gallery bundle, which grows from
369,635 to 420,172 bytes (123,536 to 136,998 gzipped, +13.5 KB). The alternative, header keys
duplicated into 'mapInfo' in eleven locales, is a translation job for after the demo.
The Deep Research page publishes the rail as Dashboards (Big Picture, Simple Research, Regions),
Prayer Mobilization (Prayer Progress, UPG-100 List), Gospel Research (Cities with Internet, Gospel
Resources), then the Experimental folder. Only the Prayer Mobilization pair moves. Big Picture now
says "desktop only" on every width, not only once a phone has disabled it.
check:spa-tab-order asserted the old four-name order with UPG-100 first; it now asserts all seven
published rows in order (still relative, with tab 1 positional). It needs a browser: rega runs it.
Each production tab reads mapInfo.tab.<tabId>.lead and .takeaway, and Regions also its note, with
the India/China reasoning beside it. Only files already in the tree are listed as examples (the
UPG-100 poster); the rest join when their shots land. A tab's button renders only when its lead
key resolves in the bundled catalogue, so a key that has not landed yet hides the button instead of
opening a popup of raw key names.
A key in a map's own area either renders blank in the gallery or, named literally, pulls that map's
whole catalogue into the gallery bundle (internal/i18n-areas.mjs). Measured with areasForBundle:
development's doxa-maps-spa carries buttons, grid, messages; this branch carries buttons, grid,
mapInfo, messages, and no upgMap. The UPG-100 entry reads mapInfo.tab.upg-100.lead and .takeaway,
which upg100 adds with every tab's strings.
The library MapInfo (an info button, a one-time pointer and a popup of real, pre-generated output)
is mounted in each pane whose tab has an entry in src/tabInfo.js, with the gallery's live locale and
its assets base. UPG-100 is the first entry, with its poster; its packet pages and card deck join
when their files are in the tree, because a missing file would show "not available" to the audience.
Regions and the other tabs follow as their owners' keys land. profile-config info:false removes the
buttons.
The gallery re-drove the visible tab by re-setting its profile-config, which works only for a map
that re-reads its whole config live. Two production maps do not: Big Picture resolves its tStatic
strings once (2 of 352 changed on en to ar, measured) and Simple Research copies its config into its
panes once (0 of 43, 0 of 105). A fresh element reads the new locale at mount, so the visible tab is
now rebuilt in the new language whatever the map does inside, the way a provider change rebuilds.
View state is lost on a language switch; that is accepted.
Hidden tabs that are still mounted (a light tab stays mounted when hidden, and select() never
remounts a mounted tab) are reset too, so their next open mounts in the new language. The old comment
said they picked it up lazily; for light tabs they did not. Tabs are reset in place, keeping the slot
ref, because a light pane renders once and releaseTab's deleted ref would never come back.
DX Life publishes the gallery shell on a hidden page named Deep Research, so that is now the
default heading (profile-config.heading or the heading prop still override it), and the meta,
instance and demo page say the same.
profile-config experimental:false (or "hidden") hides the Experimental folder: the manifest is not
mined for experimental tabs, the folder has no rows, and a tab= request for one falls back to the
first visible tab. It is off only on request, so every other host keeps the folder as before.
The shell's comments carried 23 of the reference baseline's lines, one more written in capitals that
the baseline's pattern never saw, and four internal card ids. Each is reworded to keep the engineering
reason: the 404s, the grey pane, the scroll trap, the device rule and the curated order all stay
explained. Quotes became the rule they stated.
Comments only. The script (esbuild, whitespace and syntax only, no renaming), the template (compiled
without comments) and the styles (comments removed) are identical before and after. The baseline
shrinks by the 23 lines that are gone.
