- World timeline: a scrub counts as at now within a share of the track
(0.5%), now held to at most two minutes. A layer whose source covers
years made the window weeks wide, so most of the last month snapped to
live. Short tracks keep their window.
- Fires: card and detail ages count back from the end of the 24 hours the
layer shows when its snapshot ends at a set time, and from now only when
it runs to now. A window that ended months ago read "new 105d" on fires
three hours old at the time shown.
- Weather: a product's max gap and the delay after which its latest frame
reads late are at least one and a half of its feed's median frame
spacing (over its latest frames). A regional cloud feed with frames an
hour apart drew nothing for the second half of every hour while
scrubbing and read late after twenty minutes; feeds with frames minutes
apart keep their gaps.
formatAge and formatAgeLabel repeated the validity guard, the <1h case,
the rounding and the 48-hour switch, which is why the rounding fix had to
land twice. Both now read { unit, count } from a private ageParts, so
they cannot pick different units again. Output is unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
formatAge and formatAgeLabel picked the hour unit on the raw value and
rounded afterwards, so a detection between 47.5 and 48 hours old read
48h, while one exactly 48 hours old read 2d. Round to whole hours first,
then choose the unit.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The heatmap now takes over above 4,500 km instead of 3,000 km, with
the cross-fade between 3,600 and 5,400 km. The markers band caps at
40,000 detections, which covers the ~40,000 in the padded view over
Africa at the low end of the fade, so weak fires still stay as markers.
To keep that rebuild near its old cost, markers outside the close band
read the anchor already built for each fire instead of the ground floor
cache: only the close band warms floors, so the lookup there always
missed. A 40,000-marker rebuild now costs ~55 ms, down from ~120 ms.
About 90% of detections are under 20 MW, so the old blue and teal
stops painted most of the map in cold colors that do not read as fire.
The ramp now runs ember, red, flame, orange, gold and white-hot at the
same thresholds, each stop lighter than the one below, so the brightest
marks are the strongest fires. The heatmap uses the same colors.
Main moved the layer rows to Vue, so the fire confidence switch and
minimum-FRP slider now render through RowSwitch and RowSlider on the
kit's UiSegmented and UiSlider, with the same classes as before. The
fire filter strings are translated into Arabic.
The global and regional LOD bands drape the heat raster on the active
imagery host and draw no rectangle entities; the 2/1 degree aggregation now
only feeds the "N FIRES" cards, whose accent is the strongest fire's FRP
color. Markers below ~750 km hide the heatmap, as do disable and destroy.
A stale raster is rebuilt immediately on first paint and with a 150 ms
trailing debounce afterwards, so a slider drag does not re-encode on every
step. The shell hands the layer its imagery host, so the heatmap follows
globe and 3D-tiles map stacks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Catalog: refresh/updates wording, received SDR aircraft, playing states,
move-start button, impersonal layer note, live-data descriptions,
arrival steps, retry timing, tracking spelling, military chip, dollar
amounts, label: value statuses, upstream source, globe map, stage
recovery, comparison signs as words, ellipses and earlier round-2 fixes.
Code: FIRMS cell cards use the plural entry for their count (FIRES
pruned), translated locator pin titles are isolated in RTL, legend
stops keep their sign, voice timeout minutes agree with the number,
CCTV badges are translated, first-run and referenced-item alignment.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Fixes from the Arabic screenshot review:
- t() isolates every interpolated value in RTL locales (U+2068…U+2069), so
a time (10/28 09:00 UTC), a code or an English name keeps its order inside
an Arabic sentence; coordinates (30.326° ش) are isolated too. No-op in
left-to-right languages.
- The cockpit position readout reads right to left again now that its
coordinates are isolated; gauges, tapes and roll counters stay LTR.
- The start window's arrows point along the reading direction; FM seek and
the glyph-only mini transport stay left to right, the worded PREV / PLAY /
NEXT row follows the page; the Bhote Koshi clock stays a number.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Coloring each far-zoom cell with the ramp stop of its strongest detection
turned a burning region into a blue / teal / lime mosaic: neighbouring
cells flipped hue on one outlier and the cold stops did not read as fire.
Cells say where it burns, so they go back to the heat-normalized score
(density, night share, max FRP) on the warm end of the ramp — lime, amber,
red — with the cell card accent matching. Single detections keep the full
six-stop FRP ramp.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Color every FIRMS detection by fire radiative power on the Portolan
cold → hot ramp (blue · teal · lime · amber · orange · red at
0 · 5 · 20 · 50 · 100 · 300 MW), replacing the three warm stops that
made a 2 MW and a 40 MW fire look alike. Heat cells, glow sprites and
card accents all read the same ramp (src/data/firmsIntensity.js).
Add a minimum-FRP log slider whose track is painted with the ramp (so
it doubles as the legend), an All / Nominal+ / High confidence switch
and an "N of M shown" counter to the Active Fires row. The filter runs
once in ingestion.applyFilter(), so markers, cards, heat cells, the
context store, the top-fires query and the analyst records all see one
set. Default is 0 MW / All: many of the strongest fires are rated low.
Persist both options in the layer state (token w: minFrp `f`,
confidence `c`), including the short share codec, and give the agent's
get_active_fires tool matching min_frp_mw / confidence arguments.
Layer panel gains generic `slider` and `segmented` row descriptors
(src/ui/rowFilterControls.js). New strings in es / ru / zh-CN.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Preserve Chinese unit casing and spacing in the HUD and cockpit, localize FIRMS cache ages, CCTV health and radio counts, and keep imagery resolution and route modes readable. Add regression coverage for localized output and unchanged English.
Co-Authored-By: Codex (GPT-6) <[email protected]>
The selected fire card read "4.9 MW" and "8.389°S 44.162°W" in every
language; it now uses the locale's decimal separator and hemisphere
words ("4,9 МВт", "8,389° ю. ш."). The data labels stay as they were.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A final review of both PDFs found text the earlier checks missed:
- Map cards: fire confidence (high/nominal/low) and age (24h), vessel
types (TOWING, TANKER) and installation classes (BASE, AIRFIELD) showed
in English. They now go through the catalog; the data stays English.
- The wind readout kept "10-14 14:00 UTC" and "29.93°N 97.74°W"; it now
uses the locale's date, decimal separator and hemisphere letters.
- Catalog: one term per concept (label/подпись, basemap/подложка,
PROVIDER CONTROLS, HEARD, zoom in, REGIONAL NEWS), calques and broken
grammar in the scene editor, provider settings and Bhote Koshi notes,
and Spanish wording that read wrong ("restringidas por el proveedor",
"Incluye … no se incluyen", "justo ahora", "Sal").
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Replace the semantic keys with the English source text, so code reads as the
UI does and English needs no catalog: t('Save keys'), tn(count, one, other)
for plurals, msg() for text stored in data, and bare data-i18n* markup that
translates the element's own English. Other languages are one flat JSON each.
- Simplify src/i18n: drop namespaces, the configurable locale pair
(GEV_DEFAULT_LOCALE / GEV_SECONDARY_LOCALE) and its build defines. Locale
comes from ?lang=, the stored choice, then browser languages.
- Add scripts/i18n.mjs (npm run i18n): extracts every message from t/tn/msg
and markup, and fails on missing, unused, placeholder-mismatched or
wrongly-shaped plural entries. npm test runs it via catalog.test.mjs.
- Add Russian (src/i18n/locales/ru.json, 833 messages), started from the
community catalog in #312 and reviewed against the current English.
- Fix data-i18n placed on parents that would wipe their children (cockpit
LEVEL and local weather cells, first-run EXPLORE MANUALLY tile), and a
doubled degree sign in the TRACK ACQUIRED signal.
- Rewrite docs/TRANSLATORS.md for the new workflow; drop sections the earlier
merge resurrected in docs/CURRENT-STATE.md.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Route every application-owned presentation string through the i18n
core: t() call sites across the app/layers/ui/data/voice surfaces,
data-i18n* attributes in the src/ui/templates/*.html markup (applied at
boot by applyDocumentTranslations from src/standalone/application.js),
and the dock locale selector (runtime-rendered from the offered pair —
a single EN button while only English ships).
Rebased onto current main, where the extraction follows the module
splits: the shell wiring now lands in the owner modules
(panelChrome, visualSettings, locationNavigation, shareRestoration,
shellFeedback, applicationShell) and the voice status corpus in the
realtime* modules; scene status lives in director.js +
playbackAdapter.js.
Modules shared with the node runtime (loadingFeedback,
installationFeedback, regionalModel, keySetupCore, vessels ingestion)
stay environment-neutral English and localize at their browser render
sites (chip maps in shellFeedback/installations controls, the weather
table in cockpitBriefing, keySetupRequirementEnvVars compositions), so
the package-boundary graph stays clean; the browser boundary groups
now own the i18n modules they reach (scripts/package-boundaries.json).
- Voice layer manifest (src/voice/layerManifest.js) generates the layer enums,
aliases and analyst field hints; a test fails when a catalog layer has no entry.
- analyst_query validates fields, operators, units, values and scopes, counts up
to 250,000 loaded records per layer, reports complete/partial/unanswered, tells
LAYER_OFF / NOT_READY / FEED_UNAVAILABLE apart from zero, refuses mismatched
follow-ups and never lets a superseded query replace follow-up memory. Seven
more layers expose analyst records.
- Tool results carry a code-built say line and a display card; the voice card in
the mic control shows captions, plan steps and the result
(src/voice/resultDisplay.js is the one adapter). Slow tools narrate their
current step out of band; holding Space while the assistant talks cuts it off.
Turn spans go to the debug log; caption transcription is metered; the log
omits what was said unless GEV_VOICE_LOG_CONTENT=1.
- Point and ask: the Space-down pointer snapshot resolves "this", "that one" and
"around here" through 'pointer' targets, and "the second one" through
referent:n from the last numbered list.
- Prompt states each policy once (instructions + tools 48.5k -> 46.7k chars).
- Voice benchmark harness (scripts/qa-voice-bench.mjs, scripts/voice-bench/)
shares its phrase table and grading with scripts/qa-voice-routing.mjs.