Brings in range ring visibility and sonar pulses (#1518, #1519), the
drone airspace layer (#1086), the Houston CCTV pack (#1125), typed chat
(#1507, #1545), in-view ground sampling for infrastructure (#1464) and
the CCTV model specs refresh (#1319).
What main added that reached Cesium or the viewer:
- Drone airspace draws its zones as clamped fills and draped boundary
lines in draw sets and takes clicks from the input dispatcher, as fire
perimeters does.
- Infrastructure ground sampling tests the view frustum and the horizon
from the camera port's pose (geo/frustum, geo/horizon) and reads loaded
terrain through the height queries.
- The awareness lifecycle reads the canvas, frames and whether the camera
controls take input through the ports. The sonar echo field measures
with the geo helpers.
- The range ring's pulses need a disc whose fill is computed per pixel
from values set each frame, and an outlined line whose opacity changes
without a rebuild. The draw port has neither, so that code keeps to one
module, src/layers/awareness/rangeRingCesium.js, which is on the Cesium
allow-list until the draw port can draw them. It draws only on a viewer
with a Cesium primitive collection, and takes its primitives out of the
collection it drew them in. On the stub the layer draws the
ring without pulses as a haloed line through the draw port.
Also from review of the previous merge: tests that the scene's viewer
hooks install the zoom-out guard, and that an ambiguous area name is
settled by the ground at mid-screen, else where the camera stands.
Brings in radio country coverage (#1206), the restored format scope
(#1527), the zoom-out guard (#1522), voice geometry and OSM places (#846)
and paired traffic flow density (#1495).
What main added that reached Cesium or the viewer now goes through the
renderer:
- The zoom-out guard reads the Cesium camera controller's own input, so
it moves into the adapter (src/render/cesium/zoomOutGuard.js). The
application installs it through the scene's viewer hooks, and a scene
whose renderer gives no guard starts without one.
- OSM places draws its points as a draw set, clamped to the ground.
- The voice area actions read the view box and the view centre from the
camera port.
- The voice workflow harness is test support, named as such where the
recent-imagery doubles are.
The camera verbs size their keyhole fit by the renderer's canvas and listen
on it for the user's input, and the voice actions size their view samples
by it, rather than reaching through `viewer.scene`; a viewer with no
renderer bound answers with its own canvas, the same element. The test
fakes give their viewers a canvas, and a fake camera binding's renderer
reports the fake viewer's.
Both stack parents (#837, #842) are on main, so the voice geometry branch
now builds on main directly. Conflict resolutions keep main's behaviour and
add the geometry work on top: osm_query gating on every provider, restored
admin identity for Nominatim outlines, merged analyst follow-up memory, and
the cancelled Globe View reset instruction ahead of the generic cancellation.
- turn named places and mapped landmarks into reusable bounded areas
- count loaded records inside resolved, drawn and annotated geometry
- find recent imagery for an area, the view or the pointer snapshot
- keep osm place search behind an operator-configured overpass endpoint
- preserve cancellation, ownership and follow-up memory across every action
- cover geometry, provider lifecycle, cache isolation and voice routing
- 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.