docs: tighten the display-text notes

Name routeId among the canonical analyst IDs in CURRENT-STATE, say what the
changelog entry changes for users, and drop the comment that repeated the
canonical-field doc.

changelog.d/1280-text-issue-fix.md

@@ -1 +1,3 @@
-- Bound untrusted display text at output boundaries (asturksever, #1280).
+- Keep outside text in voice and MCP tool results to one line of 180
+  characters: ALPR tags, station names and other display fields (asturksever,
+  #1280).

docs/CURRENT-STATE.md

@@ -137,16 +137,13 @@ Stale fallback runs per response after the reservation is released, and disk
 writes finish asynchronously. Settled and rejected reservations are removed
 only while they still own their map entry.
 
-External ALPR tags, surveillance-tool display fields, voice context text and
+External ALPR tags, surveillance-tool fields, voice context text and compact
-compact analyst strings collapse controls and whitespace before a final
+analyst strings become one line of at most 180 characters (scene labels: 120).
-180-character cap (scene labels retain their 120-character cap). Empty ALPR
+Empty ALPR values stay null, and the cap never splits a surrogate pair.
-values remain null and Unicode text is preserved without splitting a surrogate
+Canonical analyst identifiers such as `id` and `routeId` stay unchanged, because
-pair. Only display fields are normalized: canonical analyst identifiers
+voice follow-ups match on them and bikeshare uses the raw station name as its
-(`id`, `layerKey`, `icao24`, `mmsi`, `noradId`, `registration`, `callsign`)
+ID. The text is still untrusted: normalization does not prevent prompt
-pass through unchanged, because bikeshare uses the raw station name as its ID
+injection.
-and numbered voice references resolve by that key. These fields remain
-untrusted source data in structured tool output; text normalization does not
-establish trust or eliminate prompt injection.
 
 Terrain height requests accept at most 64 points, matching the browser service's
 existing batch size and one bounded upstream chunk. Larger requests return HTTP

src/data/displayText.js

@@ -1,4 +1,4 @@
-/** Bounded, single-line external display text; it remains untrusted source data. */
+/** One-line text capped at `maxLength` without splitting a surrogate pair. Still untrusted. */
 export function normalizeDisplayText(value, maxLength = 180) {
   if (value == null || typeof value === 'object' || typeof value === 'function')
     return '';

src/voice/gevActions.js

@@ -5353,9 +5353,9 @@ const ANALYST_IDENTITY_FIELDS = Object.freeze([
 ]);
 
 /**
- * Analyst item keys that identify a record rather than describe it: the
+ * Keys that identify an analyst record. The referent registry, track_entity,
- * referent registry, track_entity and locateLayerEntity match on them, so
+ * locateLayerEntity and follow-up filters match on them, so they skip
- * display-text normalization must leave them byte-for-byte unchanged.
+ * display-text normalization.
  */
 const ANALYST_CANONICAL_FIELDS = new Set([
   'layerKey',
@@ -5405,9 +5405,7 @@ function compactAnalystItem(r) {
     } else compact[key] = r[key];
   }
   if (present(r.ageHours)) compact.ageHours = r.ageHours;
-  // Display text only: canonical keys are lookup identities (bikeshare uses
+  // Canonical keys stay raw: bikeshare's id is the raw station name.
-  // the raw station name as its id), so the referent registry and
-  // locateLayerEntity must see them exactly as the layer reports them.
   for (const key of Object.keys(compact)) {
     if (typeof compact[key] === 'string' && !ANALYST_CANONICAL_FIELDS.has(key))
       compact[key] = normalizeDisplayText(compact[key]);