Commit Graph
2 Commits
Author SHA1 Message Date
Sameh Khamis 02d492a079 Take pointer input through one dispatcher per scene
Every layer and tool bound its own ScreenSpaceEventHandler and picked the
same click again: with the usual layers on, one click ran about 26
scene.pick and 2 drillPick passes (120-215 ms in NYC with 18 layers). Pointer
input now comes in through src/input/dispatcher.js, one handler per scene.
A click is picked lazily and at most once per kind of query, and every
subscriber reads the same ClickEvent (pick, hit, id, ownerId, gesture,
drill, ground). The same click now costs 3 picks and 2 drills, 17-70 ms.

- Ambient onClick subscribers run only while no tool holds the pointer,
  checked once per event; tools subscribe with their lease. Moves and
  releases always reach ambient subscribers, so hover and drags end cleanly.
- Camera input and the viewer's stock select/track clicks are ref-counted
  holds (suspendCamera, suppressStockClicks). The dispatcher owns the
  controller's enableInputs, so a Cesium flight that ends no longer turns
  input back on under a drag or the cockpit.
- The draw tool, imagery box, CCTV gizmo, cockpit, directions and every
  selecting layer use it; bindTrackingClickGesture is gone (the gesture is
  on the event). A source scan forbids new handlers, enableInputs writes and
  stock-click removal outside the dispatcher.
- The sensor-style pick correction transit had now applies to every layer.
- Radio's 8-pick near-miss ring is one 17x17 windowed drill.
- A missed pick passed to pickContact is no longer picked again, and a
  touch tap's contact search shares the click's drill.
- A gizmo drag whose release Cesium never reported ends on the next press
  instead of leaving the camera held.

The dispatcher is exported as ./input with the pointer-lease functions, and
documented in docs/APPLICATION.md.
2026-11-04 08:48:51 -08:00
Bilawal Sidhu a19897ad15 feat(voice): voice core — honest layer reach, spoken summary with on-screen card, narration, point-and-ask
- 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  follow-ups and never lets a superseded query replace follow-up memory.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.
2026-09-28 21:19:23 -05:00