Merge upstream main (voice geometry: resolve_area, find_imagery, osm_query) into pr/street-level-show-action

docs/CURRENT-STATE.md

@@ -44,6 +44,27 @@ pinching traverses globe and street scales with fewer gestures. Cesium retains
 its distance-aware zoom, collision handling and minimum/maximum distance limits.
 Fine-pointer desktop mouse and trackpad sensitivity is unchanged.
 
+User zoom is capped at 40,000 km (or a stricter host limit). Explicit camera
+flight, share-link and story destinations can go farther; from there, user zoom
+can't climb any higher and zooms in gradually from the current height, with no
+jump to the cap. Cesium can still use the limit when it computes an automatic
+bounding-sphere framing distance.
+Above 300 km, outward movement from the controller's configured zoom bindings
+(wheel, pinch distance and unmodified right-drag by default) pitches the camera
+down towards 8° below the Earth's limb. Large steps reach the limb immediately;
+the remaining margin eases into place, including after zooming stops or reaches
+the ceiling. Continuous outward inertia remains guarded without a time cutoff.
+Two-finger tilts, modifier-specific tilts and other camera gestures cancel the
+correction. A two-finger gesture counts as a pinch zoom only while the finger
+spacing changes more than the fingers move up or down, so spacing drift during
+a tilt is ignored. Explicit navigation handoffs, including toolbar tilt,
+cancel it before animation begins. Zooming in, disabled zoom/input, non-3D modes, tracked
+transforms and flights also cancel it; external pose changes cancel stationary
+settling. Oblique views below 300 km are left alone.
+
+`QA_BASE_URL=http://localhost:4173 node scripts/qa-zoom-out-guard.mjs` exercises
+the real browser controller at desktop and phone widths with keyless fixtures.
+
 Analyst region queries resolve unique Natural Earth names offline without
 geocoding, so Gold Coast and Ivory Coast name the West African natural regions.
 Duplicate names require a geocoded containment anchor; when several entries for
@@ -697,8 +718,9 @@ ordinary tool-card limits are unchanged.
 Voice speech: `src/voice/speech.js` adds `{ say, display, referents?, schedule? }`
 to layer toggles, fly_to_location, frame_overhead, select_nearest_aircraft,
 get_entity_context, annotate_map and analyst_query. Analyst speech and the card
-share one deterministic count/scope headline; lower bounds use "At least" and
-partial answers name the unanswered layers. `get_current_view_state` lists enabled layers only.
+share one bounded, lossless count/scope headline; combined flight, military and
+local ADS-B answers use the shared noun "aircraft"; lower bounds use "At least"
+and partial answers name the unanswered layers. `get_current_view_state` lists enabled layers only.
 FIRMS and ALPR entity labels remain English in shared context and tool data;
 voice-card titles and pointer presentation translate those known labels at display
 time, while arbitrary names and callsigns remain unchanged.
@@ -828,7 +850,9 @@ Voice controls bind to a protocol-independent session factory. The default WebRT
 
 The standalone app offers OpenAI and Gemini Live through an explicit provider chooser. Gemini keeps its permanent API key on the local server and mints a short-lived, single-use credential with server-owned instructions and the standalone action catalog. Its native WebSocket adapter supports open microphone and push-to-talk input, streaming audio, captions and the existing action runner. In push-to-talk, a tool call that arrives with no input, reply or tool result outstanding is answered with a failure and never runs, and neither does any other call in the reply it starts; open-microphone turns start on Gemini's server-side activity detection, which reports no boundary to check. Switching providers ends the old conversation and releases its microphone and connection; only the input mode carries over. Gemini uses bounded shutdown draining and explicit manual restart, without automatic reconnection or a promise of uninterrupted session duration. Hosts that do not opt into the provider facade retain their existing default or custom adapter.
 
-Voice action argument schemas have one portable owner under `src/voice/actionSchemas.js`. The Realtime provider builds 32 tools using separate description-only metadata. Description customization cannot replace argument types, enum values or required fields.
+Voice action argument schemas have one portable owner under `src/voice/actionSchemas.js`. The Realtime provider builds 34 tools using separate description-only metadata; a session is offered 33 of them unless `OVERPASS_UPSTREAMS` is set, which adds `osm_query` and its one prompt sentence (`server/providers/openai/tools.js`). Description customization cannot replace argument types, enum values or required fields.
+
+Voice areas live in a store owned by the action runner for the page (`src/data/areaStore.js`); the model gets an `areaId`, a rounded bounding box and a summary, never geometry. `resolve_area` (`src/voice/areaActions.js`, `src/data/areaResolver.js`) checks the store, bundled Natural Earth regions, then bundled countries and states by name — a name several prominent units share returns up to four candidates unless the view is inside exactly one — and then the annotation outline ladder (`resolveAnnotationTarget`: bundled counties, Census places and neighborhoods, the guarded `/api/geocode/outline`, and Overpass only when configured). Administrative geometry is stored only when its source identity agrees with the requested unit, by name or by a form the bundled matcher reads as that unit at its level ("State of Texas", "Province of Bagmati", "USA", "UK"; not "Great Britain" or "Korea"); stale, mismatched, synthesized or otherwise unverified administrative boundaries are refused rather than relabelled. Clarification IDs use the bounded `ne:<kind>:<slug>` grammar and are accepted only from the current resolver's candidate memory. A lookup past 8 s answers `AREA_TIMEOUT` and keeps filling the caches. `around=true` asks the same ladder for the grounds enclosing a landmark (Nominatim, then the OpenFreeMap area holding it) and otherwise stores a 250 m buffer marked approximate. `analyst_query` scopes add `area`, `drawn` (newest draw-tool outline) and `annotation` (newest or named outline); region scopes use the same resolver within 3 s. Drawn outlines are a simplified copy (≤1,200 vertices); counting reads the stored geometry (`src/data/areaGeometry.js`). `find_imagery` (`src/voice/imageryActions.js`) sets the Recent Imagery box from an area, the view or the turn's pointer snapshot and pins the best acquisition; any hand change to the panel or explicit layer OFF while it searches ends the request (`DISPLACED`) without pinning or success narration. `osm_query` asks `/api/osm/features` for one curated preset in a bounded box; the route answers `OVERPASS_NOT_CONFIGURED` without an operator Overpass. Results go to the session-only `osm-places` layer and become the analyst's last answer; a list capped at its limit stays incomplete in follow-ups, which keep its "listed the first N" caveat and report counts as floors. An explicit OSM Places OFF while the provider is pending aborts the work and prevents re-enable, result, credit and follow-up-memory commits. The voice answer deadline also aborts the client provider request, so the caller does not leave a timed-out request without a cancellation owner. The server's whole search (22 s across every configured endpoint, with Overpass told 20 s) ends before that 25 s deadline, because a browser that stops waiting cancels the upstream query and leaves nothing cached for a retry.
 
 `createGevActionRunner` routes `control_cctv` enable/disable through the shared
 layer visibility lifecycle. Results distinguish successful changes from
@@ -896,10 +920,9 @@ voice disposes the runner.
 Analyst result memory is committed only after any Contacts-window projection
 finishes and the call is still current. Cancelled or superseded queries cannot
 return success or replace follow-up memory; Contacts follow-ups filter the same
-cohort that was shown, and partial/unanswered coverage remains attached when a
-remembered result is narrowed.
-The narrowed answer retains the remembered scope label and detail, including
-the Contacts-window attribution, unless the follow-up supplies a new scope.
+cohort that was shown and retain its displayed scope and loaded-data note unless
+the follow-up supplies a new scope. Partial/unanswered coverage remains attached
+when a remembered result is narrowed.
 
 Application startup constructs fresh layer instances from explicit source objects.
 Standalone composition selects the existing providers. Controls, launch orbit
@@ -1197,12 +1220,12 @@ addition to the locally accumulated trail, never a promise of complete coverage.
 Snapshot coverage, completeness and freshness are separate fields. A partially
 admitted aircraft snapshot retains absent contacts for up to five minutes before
 the usual missed-poll eviction. An OpenSky row with an ICAO24 address but null
-longitude and latitude is a contact without a position, not a malformed row: it
-leaves the snapshot complete and is listed in `retainIds`, so civil flights
-retain that aircraft under the same five-minute bound, unless it is likely
-landed, while other absent aircraft fade and leave on the missed-poll limits. A
-nonempty snapshot with no positioned rows is still malformed. Military readsb
-admission is unchanged. Unknown snapshot times remain unknown in stats. An invalid nonempty
+longitude and latitude, or a readsb row with a hex but no `lat`/`lon`, is a
+contact without a position, not a malformed row: it leaves the snapshot complete
+and is listed in `retainIds`, so civil and military flights retain that aircraft
+under the same five-minute bound, unless it is likely landed, while other absent
+aircraft fade and leave on the missed-poll limits. A nonempty snapshot with no
+positioned rows is still malformed. Unknown snapshot times remain unknown in stats. An invalid nonempty
 snapshot retains the previous display. Empty vessel refreshes retain the existing
 first-connect grace and warm-data behavior. Known source failures have bounded
 messages; arbitrary HTTP response bodies are not surfaced as diagnostics.
@@ -3268,9 +3291,15 @@ down | auth-failed` (plus the unchanged `missing-key`/`unsupported`) with
 >   first later successful fetch. Smaller chunks reduce timeout risk; complete
 >   camera batches still wait for their sequential requests, and unresolved
 >   placement continues to use the existing prior until real heights arrive.
-> - **Overpass cache admission:** `/api/overpass` refuses cross-site and forwarded
->   POSTs through the shared local-request gate, then limits requests to 300/min
->   per socket peer and 900/min globally before reading or validating bodies.
+> - **Overpass cache admission:** `/api/overpass` refuses cross-site POSTs
+>   through the keyless same-site gate (reverse-proxy forwarding headers are
+>   admitted, as for `/api/terrain/heights`), then limits requests to 300/min
+>   per peer — IPv6 peers keyed by /64 — and 900/min across all remote peers
+>   before reading or validating bodies. Direct loopback (no forwarding
+>   headers) has its own 300/min window outside that global; a local reverse
+>   proxy's requests carry forwarding headers and are counted as remote, keyed
+>   `proxied:<socket peer>`, so they cannot spend the owner's window. Upstream
+>   quota stays shared, so owner cache misses can still hit 429.
 >   Tag-filter brackets are scanned once; nested or unmatched brackets return 400.
 >   It then checks fresh memory, identical in-flight work, and fresh disk entries
 >   before invoking its separate local 90/min upstream limiter. Cache and
@@ -4775,8 +4804,9 @@ programmatic activation do not trigger this reveal.
 
 Radio directory admission is atomic on both sides of the proxy boundary. A
 refresh is healthy only when it reaches the minimum accepted-query and station
-coverage; schema-valid responses with zero normalized stations count as failed
-queries rather than inflating refresh health. Each specialist query also needs
+coverage from broad/specialist queries; their schema-valid responses with zero
+normalized stations count as failed queries rather than inflating refresh health.
+Optional priority-country outcomes do not gate catalog health. Each specialist query also needs
 an accepted station whose normalized tags match that requested category; rows
 tagged only for another category remain usable catalog data but do not earn
 specialist health credit. A partial cold result remains
@@ -5769,7 +5799,7 @@ pins and share links already supply complete boxes.
 - Fleet model eligibility is distance-based with on-screen priority and hard caps (`MODEL_MAX`, `MODEL_MAX_ALL`) to avoid draw-call explosions. Each model owns its own `modelMatrix`; shared scratch matrices are forbidden because they caused stacking/flicker.
 - Tracked aircraft use standalone model primitives driven from the already-settled dead-reckoned display position, while the tracked Cesium entity remains billboard-backed so `viewer.trackedEntity` always has a ready bounding sphere.
 - Flights and military layers mirror the same tracking invariants: no warm-up freeze/jump, altitude-scaled framing, trail head glued to the displayed plane, no pull-out when switching targets, and no cross-layer orphan when switching between commercial and military tracks.
-- Regression surface: `npm run test:track` drives the real app headless with synthetic aircraft feeds and asserts the tracking invariants without depending on live OpenSky/adsb data. Arrival-rotation checks observe the first reveal within a bounded wall-clock window before the one-second idle fallback, rather than assuming the 80 ms fleet cadence fits inside four rendered frames; an incorrect first orientation fails immediately. `src/data/trackedModelRegime.test.mjs` pins the tracked contact's threshold math, the enter/exit asymmetry, and the default-on / cockpit / TR-3B / deselect wiring in both layers.
+- Regression surface: `npm run test:track` drives the real app headless with synthetic aircraft feeds and asserts the tracking invariants without depending on live OpenSky/adsb data. Arrival-rotation checks observe the settle frame and then the first reveal frame, each within a 500 ms wall-clock budget that sits inside the one-second idle fallback, rather than assuming the 80 ms fleet cadence fits inside four rendered frames; an incorrect first orientation fails immediately. Only an inconclusive timeout is retried (once); a timeout keeps every angle it measured (the settle frame before a reveal timeout, and the late frame itself, which is read before the deadline check), so a measured wrong angle fails without retry. `src/data/trackedModelRegime.test.mjs` pins the tracked contact's threshold math, the enter/exit asymmetry, and the default-on / cockpit / TR-3B / deselect wiring in both layers.
 
 ### TR-3B conversion Easter egg (August 2026)
 
@@ -6377,6 +6407,46 @@ Existing catalog validation, category filters, tuning and voice playback behavio
 remain unchanged. Audio connects directly to the broadcaster after an explicit
 play action; the source does not relay or record streams.
 
+### Radio priority-country coverage
+
+Optional `RADIO_PRIORITY_COUNTRIES` is a server-owned comma-separated list of
+up to eight ISO alpha-2 codes (for example `IN,PL`). Restart after changing it.
+The `UK` alias is canonicalized to `GB` before deduplication and querying.
+It needs no API key and adds no POWER UP entry. An empty list preserves the
+existing global/tag-based directory. Invalid configuration fails at startup.
+
+Each country adds one Radio Browser search capped at 200 rows, with existing
+HTTPS/geolocation/codec validation and an additional matching-country check.
+Up to 40 unique accepted stations per country are seeded before the specialist
+tag queries and global popularity fill, within the unchanged 750-station cap.
+No country gets exhaustive coverage. A valid, in-limit country response with
+zero eligible stations succeeds with zero selected stations.
+The response's `coverage.priorityCountries` reports accepted query status and
+selected counts when configured. The existing broad/specialist health policy
+still applies separately. Country outcomes are advisory: failed, malformed or
+oversized country queries report unsuccessful coverage but cannot degrade or
+block admission of an otherwise healthy global catalog. Queries share the existing
+three-worker pool, single-flight refresh and 45-minute cache.
+Structurally malformed rows are skipped and counted in the upstream
+`rejectedStationCount` diagnostic; mixed country responses retain their valid
+rows, while an entirely malformed nonempty country response reports unsuccessful
+coverage. The diagnostic does not block catalog admission.
+
+For every installation, including those with no priority countries configured,
+at most three mirrors are tried per request; fallback mirrors beyond the first
+three candidates are not attempted. Failed or partial catalog refreshes retry
+after one minute, so a cold-start failure can continue returning 503 for that
+minute even if upstream recovers sooner. Healthy catalogs retain the 45-minute
+cache. The seven-day last-good expiry and known-station-only click reporting
+remain in force.
+
+Validation: `node --test src/data/radioProxy.test.mjs` covers configuration,
+country quotas under global popularity pressure, duplicate records, concurrency,
+cache reuse, country mismatch/oversize/empty/ineligible responses, failure isolation on cold
+and warm refreshes beyond seven days, and retry/backoff recovery both with and
+without configured countries. `node scripts/qa-radio.mjs --url http://localhost:4173` covers the existing
+750-station browser presentation and playback contract.
+
 ## Local RTL-SDR and Local ADS-B
 
 The Radio panel holds a Local RTL-SDR card below internet radio. One browser