Last tick I fixed the golden path's token fetch and left the identical 403-fetching command in reference.md twice. Fixing the instance rather than the class is the same mistake as scrubbing a string whose generator still writes it. Both are corrected, along with a stale python3 -m http.server line that predates serve-strict. The doc had already MEASURED the restriction and kept prescribing its cause: it recorded that tiles 403 from 127.0.0.1 and treated a blank basemap as the cost of working here, while the command above it fetched exactly that token. My first rewrite then overstated the fix, and the correction is in the doc rather than hidden in a commit. Measured with the doxa.life token: curl gets 403 with no Referer, 403 with a 127.0.0.1 Referer, 200 with a doxa.life Referer -- so the restriction is real and does block the curl path. But a browser gate still fetched 9 /v4/ tiles with zero failures, because the page resolves its token through a ladder and the built bundle carries the working baked one. The fix is to the shell commands; the browser gates were never affected. Two things worth inheriting came out of checking it. The basemap is Mapbox Light, pale by design -- I looked at a screenshot and briefly called it "not painting" because I expected terrain and roads, when the labels and borders ARE the basemap. And verify:map-paint passes under both tokens, since it asserts a live GL context rather than loaded tiles, so it is not the signal to consult about whether tiles arrived.
48 KiB
Reference
Look-up tables and contracts for doxa-maps. For learning, see the tutorial; for tasks, how-to.md; for the why, why.md; for the short "what it is", explanation.md.
doxa-maps is a Vite bundler: one shared library, N independent embeddable bundles. Each
bundle is a single self-contained .js file that registers a custom HTML element
(<doxa-map>, <doxa-research-map>, …) you can drop on any web page.
The three zones
| Zone | Path | Alias | License | You edit it… |
|---|---|---|---|---|
| Shared library | library/ |
@map |
Apache-2.0 | …once; the fix reaches every bundle on next build |
| Your app-profile | plugin-designs/<name>/ |
— | yours | …freely; nothing here leaks to other maps |
| The build machine | internal/ |
— | Apache-2.0 | …never — it's the bundler's own scripts |
Terminology. "Profile" is used at two levels: a bundle / app-profile is a folder under
plugin-designs/; a profile screen is one .vue file sitting flat in that folder (no
profiles/ subfolder — the entry globs ./*.vue). A bundle can hold several screens; the host
picks which to render with the profile field of profile-config.
Two words first. An element is a custom HTML tag this repo invents, such as <doxa-map>: the browser
learns it once, then you can paste it into any page as often as you like; each pasted tag is one live copy, an
instance (people also say "web component"). An app profile (this repo also says mounting file) is one
flat .vue in plugin-designs/<bundle>/ that defines a whole map app — its main component; the settings on the
tag (profile-config) name which app profile to load and the shared ProfileLoader picks it. In class-and-object
terms: the element class is the class, each tag is an instance, the app profile is the screen that instance loads.
Most repos need one such file per app; doxa-maps builds hundreds of maps from one shared library/, so every map
gets its own app profile — and those files come in five shapes:
Profile types — the ladder, smallest first (head words are established terms; the -element compounds are
this repo's own; the coined names stay as a.k.a.):
single-element → preset-element → multi-view-element → composite-element → app-shell.
| Type (a.k.a.) | Machine key | What it is | How the build recognises it |
|---|---|---|---|
| single-element (single) | single |
one app profile, one map — paste its tag anywhere, set it up in profile-config; the default |
a flat .vue with no <profile>.meta.json and no *.instances.json in the folder |
| preset-element (parameterized) | parameterized |
one app profile + a small table of named presets; each row is a ready-made map with its settings baked in, with its own embed page and sidebar entry — same code, same tag | the bundle ships <name>.instances.json — { "instances": [{ "id", "title", "props" }] } or a bare array; the build writes one embed page per row at <bundle>/<profile>/<id>/index.html |
| multi-view-element (monolithic) | monolithic |
one app profile holding several screens you switch between — tabs or steps — in one .vue, shown one screen at a time |
declared: <profile>.meta.json → { "type": "monolithic" } (or "multi-view-element") |
| composite-element (nested-elements-mfe) | nested-elements-mfe |
an app profile whose template places other app profiles' maps together (stacked or side by side, e.g. a map + a poster wall) and ships the set as one new element — one tag embeds it all | declared: <profile>.meta.json → { "type": "nested-elements-mfe", "composes": [ … ] } (or "composite-element") |
| app-shell (nested-mfe-spa) | nested-mfe-spa |
the outer frame — sidebar, header, navigation — that loads any of the other kinds and switches between them without a reload; the top rung, the full doxa-maps app | a compiled Vue custom element today — plugin-designs/doxa-maps-spa/doxa-maps-spa.vue, built via defineCustomElement like every other rung (its shell layout is library/layouts/app-wrapper-material/). app-wrapper-material/index.html is now only the dev-time plugin-gallery scaffold, not the shipped shell |
Which bundle is the capability-N example right now? See reference-capability-exemplars.md — generated from
manifest.jsonbybun run capability-exemplarsand gated bybun run check:capability-exemplars. Cite that file; never cite memory. This one question has been answered wrongly in three separate places — a gate, this README's capability captions, and the/plugin-designerskill — because a folder's apparent simplicity does not decide its tier.doxa-vision-cardlooks like the lightest thing in the repo and is capability 2, since it ships an*.instances.json.
Composite kinds are always declared in <profile>.meta.json ({ "type": …, "label": … }, either spelling —
see TYPE_ALIASES in internal/plugin-gallery/discover.mjs) — the build never guesses them from source (tag-scanning
produced false positives). Discovery is one pure module,
internal/plugin-gallery/discover.mjs; bun run dev renders its result as the plugin gallery at /__gallery
(the same generated page the build writes to public/__gallery/index.html — see "Plugin gallery" under Build
scripts: a list grouped by bundle, every map deep-linkable at #/<bundle>/<profile>[/<instance>]), and
bun run smoke:gallery proves every element actually boots inside that archived gallery page — not in the
studio, which bun run smoke:studio covers in ~5s. / itself serves the plugin studio in both trees, not
the gallery.
Directory tree
doxa-maps/
├── library/ ← SHARED LIBRARY (alias @map, Apache-2.0)
│ ├── index.js ← thin barrel; prefer subpath imports for tree-shaking
│ ├── ProfileLoader.vue ← the ONE component that reads profile-config & loads a screen
│ ├── components/ ← map UI: legends/, map-controls/, poster/, dashboards/, drawers
│ ├── composables/ ← useMapInstance, useMapData, useMapLayers, … (+ legends/, poster/)
│ ├── api/ ← DataSourceManager (csv | api | rest-api) + sources.json
│ ├── colors/ ← one color strategy per file, auto-registered by _registry.js
│ ├── constants/ ← never-change values: mapDefaults.js, zoom.js, poster*.js
│ ├── i18n/locales/ ← 11 base locales: ar de en es fr hi it pt ro ru zh
│ ├── stores/ ← Pinia: dataStore, mapStore, uiStore (per-instance)
│ └── utils/ ← pure helpers (geoUtils, clustering)
│
├── plugin-designs/ ← ALL bundles live here (each subfolder = 1 bundle)
│ ├── <bundle>/
│ │ ├── index.js ← bundle ENTRY — globs ./*.vue, registers the web component
│ │ ├── index.html ← Vite html entry = the dev STAGING page
│ │ ├── <profile>.vue ← 1..N FLAT profile screens (no profiles/ subfolder)
│ │ └── src/ ← everything else the map needs (its own colors/, api/, …)
│ └── template-bundle/ ← minimal copy-me reference bundle
│ (folders starting with _ or . are SKIPPED by the build)
│
├── internal/ ← the bundler's own build scripts — never edited by map builders
├── deployment/capacitor/ ← native projects (android/, ios/)
├── public/ ← THE SERVED WEB-ROOT, committed:
│ ├── index.html the app shell
│ ├── doxa-maps/<bundle>/ built bundles + manifest.json + staging HTML
│ └── assets/ posters & data the maps fetch at runtime
├── vite.config.js ← multi-entry: discovers plugin-designs/<name>/ folders
└── vite.doxa-maps-build-config.json ← output tree name + copy destinations + token settings
The profile-config runtime contract
A bundle exposes ONE attribute, profile-config — a JSON string parsed by
ProfileLoader.vue at runtime. It is the entire wiring between host page and map, so the same
built bundle re-points to different data/colors/language with no rebuild.
<doxa-map profile-config='{
"profile": "doxa-simple-map", // which flat .vue screen to render (required)
"tk": "pk.eyJ...", // Mapbox token (required; host-supplied, never baked)
"instanceId": "my-map", // optional; isolates window events between embeds
"dataSource": "pray-tools", // or { "type":"rest-api","endpoint":"https://…" }
"colorSet": "default",
"locale": "en",
"provider": "mapbox", // map engine: "mapbox" (default) | "maplibre" (preview);
// also flippable per page via ?provider= or window.MAP_PROVIDER
"assetsBase": "/assets", // where tables/, boundaries/, previews/ are served (see below)
"shareUrl": "https://…", // optional; the WHOLE url the Share button links/embeds (beats everything)
"embedBaseUrl": "https://…/doxa-maps", // optional; served tree root the Share button joins embed pages to
"tabs": [ /* parameterized screens read this */ ]
}'></doxa-map>
-
assetsBase— data-driven profiles (connectivity-map, resources-map, cities-map, the poster walls and data tables) fetch${assetsBase}/tables/*.jsonand${assetsBase}/boundaries/*.geojsonbefore they create their map. Unset, it defaults to./assetsrelative to the host page; if the host doesn't serve that path the fetch gets an HTML fallback,JSON.parsethrows, and the map is never created (no console error, only the profile's status line). The generated pages and the shell set it for you (../../../assetsor the absoluteassetsBaseUrlfrom the build config); any other host must point it at a copy ofpublic/assets/. -
Share ladder (
library/utils/embedBase.js) — what the Share button links and iframes:shareUrl(a whole url, e.g. the plugin gallery's own deep link) > the profile's tree-relative embed page (<bundle>/<profile>/index.html, theembedPathprop orshareEmbedPath) joined to an embed base —embedBaseUrl>window.MAP_EMBED_BASE_URL> the root the built bundle derives from its own script url (two segments above<bundle>/<bundle>.js, so a dropped tree works on any host/CDN with no config) > the page's own location. Never hardcode/js/doxa-maps/…in a profile. -
profilemust match a flat.vuefilename in the bundle. Mismatch → an inline "Profile not found" error box listing the available names. -
tkis supplied at runtime so the token is never compiled in (dev uses.env'sVITE_MAPBOX_TOKEN). -
instanceIdscopes cross-instance window events so multiple embeds never interfere. -
dataSourceis handed toDataSourceManager(csv|api|rest-api). -
ProfileLoaderprovide()s to descendants:mapboxToken,dataSource(both computed refs —unrefbefore use),colorSet,profileConfig,instanceId,lang, and the instance-scoped Pinia storesuiStore/mapStore/dataStore. Descendants inject; never calluseXxxStore()directly (it would cross-bleed between embeds).
Each bundle registers a shadow-DOM web component via defineCustomElement, with its own
ShadowRoot, styles, Pinia, and i18n. customElements.define() is one-shot per tag name, so
each bundle registers a distinct tag.
Build scripts & output
| Script | What it does |
|---|---|
bun run dev |
Vite dev server + HMR (default http://localhost:5173, or the next free port — read the terminal). / is the plugin gallery of every element — internal/generate-gallery.mjs in mode dev over internal/plugin-gallery/{discover,dev-plugin}.mjs; every bundle is also served at its own html path. Dev-only routes: /__plugin-gallery/manifest.json (the manifest-shaped model the page fetches), /__plugin-gallery/registry.json (the discovery registry the smoke tools read), /assets/* (served from public/assets/, else plugin-designs/*/data/). Writes nothing to public/ — no HMR across that boundary. |
bun run smoke:gallery |
Opens every element by deep link in Firefox on the ARCHIVED gallery page at /__gallery and proves it boots there: element defined, shadow DOM populated, no error card, a WebGL canvas for map profiles (checked through nested shadow roots), no HTML-for-JSON data responses. --json <file> writes the evidence; exit 1 on any failure. It is the only gate watching the archived surface — it found the tree base-join bug (404 /manifest.json) that no other gate could see, because the studio carries tree-base and the gallery did not. It does NOT cover the studio: that is smoke:studio (~5s), and / serves the studio in both trees. |
bun run smoke:share |
The share round-trip gate (Firefox): opens a host page (--plugin-gallery = every canvas element of the dev plugin gallery; --url <page> --tag <element> = the website or any CDN page), clicks Share, reads the URL, fetches it, reopens it in a fresh page and asserts the SAME custom element mounts with a WebGL canvas. --json <file>; exit 1 on any failure — proves REQ-1 ("links work wherever the tree is hosted"). |
bun run check:public-hygiene [<dir>] [--strict] |
Proves a served folder is fit to publish: scans for local paths, internal folder names, dev-server URLs, JWT-shaped tokens, .env/token files, source maps, documentation and staging/preview pages, and prints an inventory (bundles, leaf pages, assets by folder). Under --strict any root entry outside the drop-in contract is a finding; the allowed set is READ FROM vite.doxa-maps-build-config.json → dropin.include (index.html, __gallery/, assets/, doxa-maps/) rather than restated here, so the gate cannot disagree with the contract. Exit 1 on a finding. Run it on the website's drop-in before any public push. |
bun run sync:dropin <dest> [--from-config | --only a,b] [--prune] / bun run deploy:dev |
Syncs the built public/ into a host checkout (e.g. the website's public/embed/doxa-maps-build/). The drop-in contract lives in ONE place — vite.doxa-maps-build-config.json → dropin.include, four root entries: index.html (the plugin studio host page generated by internal/generate-tree.mjs, which mounts <plugin-studio>), __gallery/ (the archived bundle-index gallery generated by internal/generate-gallery.mjs, listing every bundle/profile/instance and mounting the chosen one), assets/ (data) and doxa-maps/ (the bundle folders + manifest.json). --from-config copies only those, --prune removes everything else in the destination, so a host carries exactly the served web-root and nothing else at its root. dropin.exclude (same file) lists built files inside the allowlist that no page fetches — the full people-groups CSV, the tables only internal scripts read, the table index; they stay in public/ for the scripts, are skipped by --from-config (counted as "excluded (not shipped)") and removed from the destination if a copy is there. Without an allowlist it is a plain full mirror. Never copies credentials, docs or staging pages; --protect a,b keeps host-owned files; --dry-run first; prints a copied/unchanged/deleted/pruned/excluded summary and the resulting top-level tree. deploy:dev = build → sync --from-config --prune into the website (DOXA_DROPIN_DEST overrides the path) — the only sanctioned way a map changes on the site; then commit in both repos. |
bun run build |
clean → one vite build per plugin-designs/*/ folder → tree → check:drop → check:parse → verify → gallery (writes public/__gallery/index.html, the archived bundle-index gallery, from the manifest). The root page public/index.html — the studio host — is written by tree. |
bun run build:bundle |
Builds ONE bundle (needs BUNDLE=<name>). |
bun run build:watch |
Rebuilds into public/ on every change (integration loop). |
bun run clean |
Wipes + recreates the public/<name>/ output tree. |
bun run tree |
Writes manifest.json + self-contained staging HTML, then copies the tree to every build-config paths destination. |
bun run verify |
Data-integrity + shipped-bundle + network-policy guards (fast, no browser). |
bun run verify:full |
+ rendered posters and live apps in Firefox. |
Plugin gallery. One generator, two modes. internal/generate-gallery.mjs exports
renderGallery(model, { mode }), a pure function from a manifest-shaped model (bundles → profiles →
instances, with scriptUrl / iframeUrl / embed / integrity) to the whole page: the list, the stage,
#/<bundle>/<profile>[/<instance>] deep links, Open page, Copy embed, the token ladder, lazy engine load,
the phone layout and the baked fallback. Mode build (bun run gallery) renders
public/doxa-maps/manifest.json into public/__gallery/index.html — ARCHIVED beside the root, which is now
the plugin studio. Mode dev is what bun run dev serves at /__gallery: the
same page over galleryModel(discoverSpaRegistry()) (internal/plugin-gallery/dev-plugin.mjs), differing
only in the loader (it imports /plugin-designs/<bundle>/index.js instead of adding the built IIFE), the live
model route (/__plugin-gallery/manifest.json), the /assets/* dev route and the composability ladder's
"?" popovers in the header — kept out of the built page to keep it lean. Nothing about the gallery is
written twice.
How a build works. vite.config.js is multi-entry by discovery: at build time it
registers every plugin-designs/ subfolder that (a) doesn't start with _/. and (b) has
both index.html and index.js. One vite build builds exactly one bundle, selected
by BUNDLE=<name>, so the output is a self-contained IIFE (inlineDynamicImports: true).
bun run build loops every folder. Each bundle only bundles what its screens import, so sizes
differ (simple ≈ 420 KB, research ≈ 1.3 MB).
Where output goes. One bundle folder → one file: public/doxa-maps/<bundle>/<bundle>.js.
CSS is injected by JS (cssCodeSplit: false), so there is no separate .css. The tree root
name comes from vite.doxa-maps-build-config.json name (default doxa-maps).
emptyOutDir: false keeps sibling bundles intact when you rebuild one.
Shadow-style comments are stripped from production bundles. Component CSS is injected
from useShadowStyles(\…`)template literals that ship verbatim, sointernal/vite-plugin-strip-shadow-css-comments.mjs(build only,enforce: 'pre') removes the /* … */blocks inside those literals and collapses the blank lines they leave; a literal with${…}` is left untouched with a warning, and dev mode keeps every comment.
What a built bundle folder contains. Beyond the IIFE, bun run tree generates an
index.html per bundle (staging/showcase), per profile, and per instance —
each profile/instance page is a standalone, directly linkable/iframe-able map — plus
public/doxa-maps/manifest.json with ready-made embed.iframe/embed.script snippets for
every profile. Full design: architecture/embeddability.md.
Preset instances. A <profile>.instances.json beside a parameterized profile
({ profile, apiBaseUrl?, instances: [{ id, title?, props }] }) makes the build emit one
standalone page per instance at <bundle>/<id>/index.html. Recipe:
how-to.md.
Bun is required. bun.lock is the committed lockfile; bun x vite runs project-local
Vite. npm run … resolves the same scripts but Bun must be on PATH.
Standalone gates — the SILENT failure classes
These are not in build:unlocked, deliberately: each checks something a fresh build would mask,
so running it inside the build would make it unfalsifiable. Run them before publishing.
| Script | Catches | Why it cannot live in the build |
|---|---|---|
bun run check:composite |
a composite (cap-4/cap-5) still shipping an OLD copy of a child it nests — a one-bundle rebuild does NOT recompile its parents | a full build makes every composite fresh by construction, so it could never fail there |
bun run check:shots |
a README screenshot that predates the bundle it shows — a stale screenshot looks exactly like a current one | needs a browser to repair (bun run shots), which CI does not have |
bun run check:library-index |
documentation/reference-library-index.md fallen behind library/ — its header promises it "can't drift", which is only true if regenerated (bun run library-index) |
a fresh checkout gives every file the same mtime, so mtime comparison always passes in CI (CI regenerates + diffs instead) |
bun run check:done-evidence |
a backlog card marked done with no bluf and no fields — a status word the next team cannot reconcile against the code |
reads _qa-cycle/, which is not in any repo CI clones |
bun run verify:gate1 |
the Gate-1 chain end-to-end against LIVE data: header filter → composite props → real grid re-filter → count back up | needs network; it is an acceptance test, not a build step |
bun run check:catalog-identity |
the live server serving something other than the catalog — a stale CDN copy, a half-finished publish, or the login wall. Hashes each live bundle and compares sha384 against the manifest's own integrity |
needs network egress, which CI cannot guarantee; it SKIPs when unreachable, and only "reached and wrong" fails |
bun run check:catalog-currency |
the live catalog being an older build than yours (identity passes on a catalog three publishes behind), and a locally-built bundle that is absent from the catalog — the "nothing was lost in the merge" guarantee | needs network egress; SKIPs when the catalog cannot be reached at all, because an offline box must not fail a build |
bun run smoke:embeds |
an element that builds, lists in the manifest, and renders nothing — a thrown setup() removes the subtree silently, so every build-time gate still passes. Mounts every profile page and asserts real shadow content |
needs a browser; CI has none |
bun run verify:map-paint |
a map that lays out but never paints — Mapbox creates a full-size <canvas> and only then tries for a GL context, so "canvas is 1440×900" is evidence of layout, never of paint. Asserts a live GL context + attribution |
needs a browser launched with the SwiftShader flags (the MCP/default browser here has no GL at all) |
bun run verify:5-caps |
a capability tier that no longer has a working exemplar — the 5×5 product claim going stale | needs a browser + a Mapbox token |
bun run check:quoted-scripts |
a doc that quotes a command which has since changed — an omitted step in a quoted chain is invisible: there is no wrong character to notice, just a check the reader believes is in the chain | compares docs against package.json; cheap, but it guards documentation rather than build output |
bun run check:capability-exemplars |
documentation/reference-capability-exemplars.md drifting from the manifest's real tiers (regenerate with bun run capability-exemplars) |
the table is generated from the manifest, so running it mid-build would compare a file against the thing that just wrote it |
bun run check:dep-names |
a DECLARED DEPENDENCY THAT IS NOT A REAL PACKAGE NAME. In build:unlocked. Checks two things offline: the name resolves under node_modules/, AND the installed package's own name field equals the declared key — the second catches a name that resolves to a DIFFERENT package, which a path-only check misses. SKIPs honestly when node_modules/ is absent. Found the hard way 2026-09-22: a folder rename rewrote @capacitor/{android,cli,core} to @deployment/capacitor/*, so bun install failed in a fresh clone while everything here passed, because the real packages were already installed from before the rename. |
|
bun run check:no-abs-paths |
a MACHINE-SPECIFIC PATH in tracked source — a /home/<user>/ prefix, its macOS equivalent, or the Windows user-profile form. It is in build:unlocked, listed here because the failure it prevents is a HAND-OFF failure: an absolute path works perfectly for whoever wrote it and breaks only on the inheritor's machine, so no test run by its author can catch it. Scans git ls-files (the set a clone receives), which excludes node_modules and build output without an ignore list. Measured 2026-09-22: the repo was clean BY ACCIDENT — a planted /home/… line passed check:public-hygiene, check:parse and check:quoted-scripts, because public-hygiene scans the SERVED tree, not source. |
|
bun run check:settled |
measuring or publishing a tree mid-build — a build's clean step rm -rfs public/doxa-maps, so samples taken seconds apart give opposite answers |
it exists to be run around builds, not inside one |
Before you run the list — three gates need a PRECONDITION, or they look broken
A bare bun run <gate> is not always meaningful, and the failure text does not always say so.
Measured 2026-09-22, by running each one:
| Gate | Needs | What a bare run does |
|---|---|---|
smoke:gallery (and smoke:playground, its alias) |
a dev server: bun x vite on :5173 (bun run dev's own port) |
exits non-zero with "cannot read http://localhost:5173/__plugin-gallery/registry.json — is bun x vite running?" — a precondition, not a defect. It audits the ARCHIVED /__gallery page; --page re-aims it, --url stays at the server root where the registry lives |
smoke:share |
--url <page> arguments, a served tree, a token, and --browser chromium on this box |
prints usage: and exits non-zero; with default Firefox it stops with "Verified on a BLANK page, so it is the engine/box, not the product" and names the chromium flag itself |
★ WRITING A NEW GATE: you cannot import() plugin or library source into plain Node
This is the trap that will bite the next gate author, so it is written here rather than left to be rediscovered. Plugin and library source is written for Vite, not for Node, and it fails two different ways — counted 2026-09-22, not estimated:
| Vite-only feature | Where | What plain node does |
|---|---|---|
the @map alias (@map/ProfileLoader.vue) |
resolved by vite.config, not by Node |
Cannot find package '@map/ProfileLoader.vue' — this is what you hit first, before any glob |
import.meta.glob |
18 files: 11 of the 20 plugin index.js, plus library/ProfileLoader.vue, library/i18n/index.js, library/composables/useShadowStyles.js, library/colors/_registry.js, and 3 more |
TypeError: (intermediate value).glob is not a function |
import.meta.glob is a build-time feature: Vite statically replaces it with real imports while
bundling. Plain node leaves it undefined, so the module throws the instant it is evaluated:
TypeError: (intermediate value).glob is not a function
at library/colors/_registry.js:26
The glob failure is the nastier of the two because it can arrive transitively, from a module you never named — you import a config, the config imports a colour strategy, and the registry throws.
Measured 2026-09-22: a new check:facet-cascade gate did await import() on the dashboard config,
which transitively reached the colour registry, and crashed — while already wired into
build:unlocked, so bun run build failed for a reason that had nothing to do with the code being
built. The library file was correct and pre-existing; the defect was entirely in the gate reaching it.
So when a gate needs to exercise real plugin logic, pick one:
| Approach | When | Cost |
|---|---|---|
| Import only the leaf module under test | the logic is in a plain .js with no Vite features (most composables are) |
cheapest and most reliable — keep the dependency cone small |
Test against the BUILT bundle over bun run serve |
the logic only exists after bundling, or needs the DOM | needs a browser + a served tree, so it belongs in verify:browser-all, not build:unlocked |
| Parse the source text instead of executing it | you are asserting a structural property (a rule exists, a field is declared) | cannot catch runtime behaviour — say so in the gate's header |
Two hard rules that follow. A gate goes in build:unlocked only if it runs clean in plain Node
with no browser and no network — everything else is a standalone or browser gate, the way
check:a11y-button-names and check:composite are handled. And run the gate before wiring it in:
a broken gate inside the build chain is worse than no gate, because it goes red for the wrong reason
and trains whoever hits it to skip the step.
Other runtime constraints on the build box, all of which have produced a wrong answer at least
once: Node is v20 (no fs.glob, which is 22+); find is bfs, which rejects a relative
-newermt '-45 seconds' and errors to stderr while printing nothing — use bun run check:settled,
or an absolute $(date -u -d '-45 seconds' +%Y-%m-%dT%H:%M:%SZ); there is no hardware WebGL (judge
maps with verify:map-paint's SwiftShader path, never by expecting a painted basemap); and bunx
does not exist here — use the spaced bun x.
Copy-paste: the full share round-trip over every plugin that has a Share button. Assembled and run
2026-09-22 — 8/8 pass. Without this recipe an inheritor has to work out the server, the token and the
eight --url flags from scratch, which is why the gate had never been run across all of them.
cd doxa-maps-plugin-studio
( node internal/serve-strict.mjs --port 4950 & ) # serve the built tree, LOOPBACK ONLY
# ★ Take the token from the BUILD CONFIG, not from doxa.life/api/maps/token — see the note below.
export MAPBOX_TOKEN=$(python3 -c "import json;print(json.load(open('vite.doxa-maps-build-config.json'))['token'])")
B=http://127.0.0.1:4950/doxa-maps
ARGS=""
for p in demo-map/demo-map cities-internet/connectivity-map doxa-research-map/research-map \
country-posters/countries-map doxa-simple-map/doxa-simple-map \
gospel-resources/resources-map my-upg-100-list/upg-100-list my-upg-100-list/upg-100-mfe; do
ARGS="$ARGS --url $B/$p/index.html?tk=$MAPBOX_TOKEN"
done
node internal/share-roundtrip-smoke.mjs $ARGS --browser chromium
The pass count alone proves nothing — check the reopened TAGS. The failure this gate exists to catch (an SPA fallback, or a host that strips the
#deep-link) renders a map for every share URL, so every target passes while every link is wrong. The discriminating evidence is that the run returned six distinct custom-element tags —demo-map,cities-internet-map,doxa-research-map,country-posters-map,doxa-map,gospel-resources-map— each matching its own plugin, withcanvas:yon all eight. (The twomy-upg-100-listprofiles legitimately share one tag.) Derive the plugin list the same way it was derived here rather than pasting it stale:grep -rln ShareButton plugin-designs/*/*.vue. |check:dxlife| a siblingroot-dx-life/clone beside this repo | exits 2 — "clone not found — nothing to check". That is a deliberate THIRD value (0unchanged ·1drift ·2no clone), because this gate exists after a check that returned the PASS value BECAUSE IT FAILED. Butbun runand any&&chain see only zero/non-zero, so it surfaces as FAIL. Measured 2026-09-22 inside a fresh clone: every other browser-free gate passed and only this one went red, for want of a directory an inheritor has no reason to have. It is NOT inbuild:unlocked, so a clone still builds. | |smoke:embeds·verify:map-paint·verify:5-caps|MAPBOX_TOKENin the environment | does not fail — it DEGRADES. Without a token map profiles cannot mount, andsmoke:embedsreports them as TIMEOUTS. It once read 27/40 for this reason and 40/40 with a token. A timeout here means "no token" far more often than "broken plugin" |
# The token every map gate wants. Read it from the build config: measured 2026-09-23, the config
# token serves tiles (200) from any origin, while the doxa.life endpoint's token returns 403 from
# anywhere but doxa.life. Both are 90-odd-char pk. public keys; only one of them works here.
export MAPBOX_TOKEN=$(python3 -c "import json;print(json.load(open('vite.doxa-maps-build-config.json'))['token'])")
THIS LIMITATION WAS SELF-INFLICTED, and it is fixed — corrected 2026-09-23. This paragraph used to say a blank basemap was simply the cost of working here. It measured the right thing about the WRONG TOKEN: the one served by
https://doxa.life/api/maps/tokenis URL-restricted todoxa.life, returning 403 from127.0.0.1,localhost,nuxtbase.appor no Referer. That is a property of that token, not of this box — and the command above used to fetch it, so the doc diagnosed the blank basemap and then kept prescribing its cause.The
tokenfield invite.doxa-maps-build-config.jsonreturns 200 on the identical request from any origin. It is apk.public key baked deliberately (Driver ruling 2026-09-21), protected by URL restrictions rather than secrecy;check:public-hygieneallowspk.and fails the build onsk./tk.. So the SHELL COMMANDS above — the ones that curl a token and hand it to a gate asMAPBOX_TOKEN— now work where they previously fetched a token that 403s.★ BE PRECISE ABOUT WHAT THIS FIXES, because an earlier draft of this very paragraph overstated it. Measured 2026-09-23 with the doxa.life token: curl returns 403 with no Referer, 403 with
Referer: http://127.0.0.1:4173/, and 200 withReferer: https://doxa.life/. So the restriction is real and it does block the curl path. But a browser gate loading a leaf still fetched 9/v4/tiles with 0 failures — because the page resolves its token through a LADDER and the built bundle carries the working baked token, so the?tk=rung is not necessarily what the map ends up using. The fix is to the SHELL commands; the browser gates were never the ones affected.Two corrections that came out of measuring this, both worth inheriting. First, the basemap here is Mapbox Light — pale land, grey water, minimal detail so the choropleth dominates. Looking at a screenshot I briefly called it "not painting" because I expected terrain and roads; the country labels and borders ARE the basemap. Judge tile loading by the
/v4/response count, not by whether the image looks like a road map. Second,verify:map-paintpasses under BOTH tokens, because it asserts a live GL context and attribution rather than loaded tiles — so it is not the signal to consult about whether tiles arrived.
★ AN EXIT CODE IS NOT A VERDICT UNTIL YOU HAVE READ THE MESSAGE. Two of these read as FAIL in a batch run and are neither passing nor failing — they never started. A checklist that turns a precondition into a red line trains its reader to ignore red lines, which is how a real failure gets skipped.
The rest, grouped by what they guard — same rule (outside the build, run before publishing):
| Script | Catches |
|---|---|
bun run smoke · bun run smoke:gallery · bun run smoke:share |
the built leaf pages, the ARCHIVED /__gallery page, and share-link round-tripping — each needs a browser. The studio at / is bun run smoke:studio |
bun run check:a11y-button-names |
an icon-only control with no accessible name (needs a browser to read the computed name) |
bun run check:capability-citations |
the README claiming a capability the manifest does not back |
bun run check:dxlife |
any edit to the read-only DX-life clone — the standing law that tree is never written |
bun run check:mobile · bun run check:tauri · bun run check:wp-embed |
the Capacitor sync, the Tauri config contract, and the WordPress PHP embed — deploy methods 2, 5 and 4 |
bun run verify:full |
the research corpus end-to-end |
★ A STANDALONE GATE ONLY FIRES IF SOMEONE REMEMBERS IT — so this table IS the mechanism. Measured 2026-09-22: 26 gates live outside
build:unlocked; only nine were listed here. Two of the unlisted ones (check:quoted-scripts,check:capability-exemplars) were already FAILING, and had been for two sessions, because a build step and a bundle were added without their docs being regenerated. Nothing was broken in the product — the gates were doing their job — but nobody ran them, because an inheritor reads this table as the list of what to run before publishing. An undocumented standalone gate is indistinguishable from one that does not exist.
Why SRI and not an HTTP status. https://nuxtbase.app/app/doxa-maps/manifest.json returns 200
and is a 2,872-byte HTML sign-in page — the app-shell login wall. A status check reads that as
healthy. The public door is /plugin-assets/doxa-maps/. Every profile already carries an integrity
hash the build computed, so comparing it against the live bytes is a cryptographic identity proof
that costs nothing extra to maintain. Re-measured 2026-09-22: 18 of 18 bundles identity-proven, and
the live manifest byte-identical to platform/mfe-index. (It read "17 of 17" on 2026-09-21; the catalog
has since gained golden-path-check.)
A dated measurement is not a stale claim — but it still misleads once the underlying number moves. "17 of 17" was true when written and the gate reads the live manifest, so nothing was broken. The problem is what a READER concludes: seeing 17 beside a catalog of 18, they cannot tell whether a bundle went missing or the sentence simply aged. ★ If you quote a count here, re-run the gate and restate it —
bun run check:catalog-identityprints the current figure in one line.
The screenshot gates — three different lies a picture can tell
A README image can fail in three ways that all look identical to a passing shot run. bun run shots
reports OK for every one of them, because it proves a page loaded — never that it looks right.
| Script | Catches | In CI? |
|---|---|---|
bun run check:shots |
a PNG older than the bundle it shows — a stale screenshot looks exactly like a current one | report-only (repair needs a browser) |
bun run check:shot-distinctness |
two slots capturing the same view — shipped as "List — page 1"/"page 2", the same picture under two captions | blocking (parses one file) |
bun run check:shot-fidelity |
this machine cannot render what the shots must show (no emoji font ⇒ every icon a tofu box) | not yet — reports locally |
The distinctness case is the instructive one: the shot chased a "Next" control and a scroll that
do not exist (the list has no pagination), every attempt sat inside try/catch so the miss was
silent, and the two PNGs differed by 0.16% render noise — so they were not byte-identical and no
checksum would have caught them. Meanwhile a real third profile (poster-wall) had no shot at all.
A slot faking a view it cannot reach, while a genuine artifact goes unshot, is the recurring
shape — the same one that captioned the product SPA as the plugin-studio.
bun run check:quoted-scripts is the exception that proves the rule above — it is the one
doc-integrity gate that DOES run in CI (blocking), because it compares two files already in the
checkout: it needs no browser, no network, and no mtime. It asserts that the scripts block
quoted verbatim in documentation/README-SECTION-3-cli-commands.md still matches package.json. That doc
carries its own hand-written plea to re-transcribe after every edit, and drifted anyway — twice.
The second drift omitted check:public-hygiene (the secret scan) from the build chain it
documents, which is the dangerous shape: an omission in a quoted list is invisible, since
there is no wrong character to notice. package.json is what runs, so the fix is always to update
the doc.
bun run verify:5-caps proves all five capability tiers mount and render. It needs a token —
MAPBOX_TOKEN=$(curl -s <token-endpoint>) bun run verify:5-caps — and it sets both ?tk=
(the top-level element) and window.MAP_TOKEN (a composite's nested children read the global,
not the query param). Without a token it reports the token outage rather than the capabilities.
vite.doxa-maps-build-config.json fields
Edit this one file, then bun run build.
| Field | What it does |
|---|---|
name |
Output folder name (default doxa-maps). |
paths |
Extra folders to copy the whole built tree into. Point at a host repo's public/ → that host serves the maps and becomes the CDN. Empty (the default) ⇒ written to public/<name>/ only, since public/ is already the committed web-root. |
embedBaseUrl |
The public URL the tree is served from, used to generate absolute, paste-anywhere embed snippets. Empty ⇒ relative URLs (position-independent; keeps the drop-portability gate green; required for Capacitor, which loads from https://localhost). |
tokenUrl |
Endpoint each generated page fetches its Mapbox token from — a URL, not a secret. Set absolute (https://…/api/maps/token) so a CDN folder and the mobile app both work; endpoint must send CORS for the embedding origins. window.MAP_TOKEN_URL overrides at runtime. |
token |
Opt-in literal token baked into every page. Leave empty — prefer tokenUrl. |
assetsBaseUrl |
Where the shared asset tree (previews, posters, boundaries) is served from. Empty ⇒ a relative path per page depth (self-contained under one web-root). Set absolute for partner embeds and mobile, where there is no shared web-root to climb to. |
Token resolution ladder: per-map tk prop > ?tk= > baked token > window.MAP_TOKEN_URL > tokenUrl > same-origin /api/maps/token.
Data sources
★ Persecution-safety — a rule about WHICH FIELDS a profile may carry
Read this before adding a field or a link to any profile. Some profiles are deliberately kept free of gospel/church/believer data, and a profile failing a "should have resource links" check may be OBEYING that rule rather than missing a feature.
The binding instance, stated in the code it governs (plugin-designs/cities-internet/connectivity-list.vue):
PERSECUTION-SAFETY: this profile carries connectivity + population + language only. It has NO gospel-resource, church or believer field and must not gain one.
Measured 2026-09-22: all three cities-internet profiles have zero href bindings, and the live leaf
renders 1,567 nodes with 0 anchors. That is correct. Its source field —
{name: "World Bank Open Data", url: "https://api.worldbank.org/v2"} — is API provenance, not a
user-facing destination, and it is surfaced as a citation paragraph when a region is expanded, which is the
separate promise that header makes ("every figure carries its source, so the numbers can be audited").
★ Why this is written down here and not left in one file header. A standing work item reads
"gospel-resources + cities-internet maps must render with REAL resource links" — one line batching two
profiles whose contracts are opposite. gospel-resources genuinely renders 155 real eBible links; taking
the same line literally for cities-internet means adding a gospel-resource field to a profile whose
header forbids exactly that. A requirement that names two subjects deserves a per-subject check before
anyone drains it.
"Persecution-safe" is used for three different things in this repo — don't assume one rule. Here it means FIELD SEPARATION. In
key-cities/cities-list.vueit means placename precision ("a wrong placename in a persecution-sensitive dataset is the dangerous kind of wrong"). Inresearch-dashboard/.../DashboardDataTable.vueit means a dependency-free table (no third-party grid). Only the first governs what data a profile may hold.
Maps read data through @map/api/ driven by library/api/sources.json. Three source types:
| Type | What | Points at |
|---|---|---|
rest-api |
versioned REST endpoint (default) | the live people-group API; base URL resolves at runtime (window.MAP_APP_API_URL → VITE_API_BASE_URL → same-origin) |
csv |
a local CSV parsed at load | the Joshua Project dataset (+ photo fallbacks) |
api |
a single JSON URL | any plain JSON endpoint |
Country/region polygons are not a sources.json connection — they render from Mapbox's built-in
country-boundaries-v1 tileset plus a lazy-loaded WAGF-regions GeoJSON.
Adding or changing a source? Full contract + checklist: architecture/data-sources.md.
Colors
Reused by more than one map → shared library/colors/. Only your map uses it → your bundle's
src/colors/ (merges over the shared set, that bundle only). Filename derives the mode key
(language-family.js → languageFamily).
Writing or overriding a strategy? Full contract + checklist: architecture/color-strategies.md · two homes in depth: ../contributing/COLORS.md.
Importing from @map
Everything under library/ is importable via the @map/… alias (vite.config.js maps @map
→ library/). The import path mirrors the file path: a file at library/<sub>/<file> imports
as @map/<sub>/<file>. Examples:
import SemanticTreeLegend from '@map/components/legends/SemanticTreeLegend.vue'
import { useMapInstance } from '@map/composables/useMapInstance.js'
import { useMapData } from '@map/composables/useMapData.js'
import langFamilyByIso from '@map/data/langFamilyByIso.json'
Prefer subpath imports over the thin @map barrel (which exports MAP_CORE_VERSION only) so
tree-shaking works. The per-file catalog in architecture/file-map.md is the
authoritative menu — every library/ file listed is importable this way. A generated one-line import index can also be produced with
node internal/generate-library-index.mjs.
Going deeper — the architecture folder
Simple lookups live here; the full contracts, catalogs, and wiring detail live in
architecture/:
| Doc | What it holds |
|---|---|
| file-map.md | every file, one line each, with the gotcha each carries |
| data-sources.md | the api/ seam: connection contract + checklist |
| color-strategies.md | the strategy contract + legend wiring |
| components.md | shared component catalog + conventions |
| i18n.md | the locale contract + add-a-language checklist |
| embeddability.md | generated per-map pages + manifest — embeddable by default |
| app-wrappers.md | the SPA shell: nav.js + PROFILES contracts |
| poster-generation.md | the poster/preview generator pipeline |
| patterns-and-provenance.md | the two multi-map patterns + provenance |
The Semantic Tree Legend component (a reusable hierarchical legend) is documented in file-map.md.
Quick reference — "I want to…"
| I want to… | Do this |
|---|---|
| Add a new map | copy plugin-designs/template-bundle/ → rename folder + tag in index.js |
| Add a screen to an existing bundle | drop a .vue flat in that bundle's folder, request it via profile-config.profile |
| Fix a legend / control for ALL maps | edit it once in library/ (@map); rebuild bundles |
| Add a color strategy | new file in @map/colors/ (shared) or the bundle's src/colors/ (map-private) |
| Build everything | bun run build → public/doxa-maps/<bundle>/<bundle>.js |
| Build just one | BUNDLE=<name> bun run build:bundle |
| Preview in isolation | bun run dev → http://localhost:5173/plugin-designs/<name>/index.html |
| See a change on the served site | bun run build, then refresh (NOT dev) |
| Repoint the API without a rebuild | set window.MAP_APP_API_URL before the bundle loads |
| Repoint the Mapbox token | pass tk in profile-config, or set window.MAP_TOKEN_URL / build-config tokenUrl |