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  follow-ups and never lets a superseded query replace follow-up memory.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.

src/voice/voiceCost.js

@@ -154,6 +154,69 @@ export function resolveVoiceModelById(modelId) {
   };
 }
 
+/* ------------------------------------------------------------------ *
+ * INPUT TRANSCRIPTION (voice-card captions)
+ * ------------------------------------------------------------------ */
+
+/**
+ * ⚠️ VERIFY AT RELEASE — external prices, USD per 1M tokens, same shape as
+ * the realtime rate tables. Transcription is billed separately from the
+ * session's responses and arrives as `usage` on
+ * `conversation.item.input_audio_transcription.completed`. An unknown model
+ * bills at the most expensive entry, as for session models.
+ */
+const transcriptionTable = ({ textInput, audioInput, textOutput }) =>
+  Object.freeze({
+    textInput,
+    audioInput,
+    textOutput,
+    // Transcription output is text; usage without output details would
+    // otherwise be read as audio and priced at zero. Cached input is priced
+    // as uncached so the meter never under-counts.
+    audioOutput: textOutput,
+    textCachedInput: textInput,
+    audioCachedInput: audioInput,
+  });
+
+export const TRANSCRIPTION_MODEL_RATES = Object.freeze({
+  'gpt-4o-mini-transcribe': transcriptionTable({
+    textInput: 1.25,
+    audioInput: 3,
+    textOutput: 5,
+  }),
+  'gpt-4o-transcribe': transcriptionTable({
+    textInput: 2.5,
+    audioInput: 6,
+    textOutput: 10,
+  }),
+});
+
+/** Per-second rate for duration-billed transcription (whisper-style). */
+export const TRANSCRIPTION_USD_PER_SECOND = 0.006 / 60;
+
+/** Rate table for a transcription model; unknown ids bill at the priciest. */
+export function transcriptionRates(modelId) {
+  const id = typeof modelId === 'string' ? modelId.trim() : '';
+  if (Object.prototype.hasOwnProperty.call(TRANSCRIPTION_MODEL_RATES, id))
+    return TRANSCRIPTION_MODEL_RATES[id];
+  return Object.values(TRANSCRIPTION_MODEL_RATES).reduce((worst, rates) =>
+    rates.audioInput > worst.audioInput ? rates : worst,
+  );
+}
+
+/**
+ * USD for one transcription's usage: token usage priced with the model's
+ * rates, or duration usage priced per second.
+ */
+export function estimateTranscriptionCostUsd(usage, modelId) {
+  if (!usage) return 0;
+  if (usage.type === 'duration') {
+    const usd = nonNegative(usage.seconds) * TRANSCRIPTION_USD_PER_SECOND;
+    return Number.isFinite(usd) ? usd : 0;
+  }
+  return estimateUsageCostUsd(usage, transcriptionRates(modelId));
+}
+
 /* ------------------------------------------------------------------ *
  * SPEND GUARD CONFIG
  * ------------------------------------------------------------------ */
@@ -429,7 +492,14 @@ export function createVoiceCostTracker(options = {}) {
      * @param {object} usage - `response.usage`
      */
     record(usage) {
-      const usd = estimateUsageCostUsd(usage, model.rates);
+      return this.recordUsd(estimateUsageCostUsd(usage, model.rates));
+    },
+    /**
+     * Fold an already-priced cost (e.g. input transcription) into the same
+     * session total, so the warning and the cap cover it.
+     * @param {number} usd
+     */
+    recordUsd(usd) {
       if (usd > 0) {
         totalUsd += usd;
         responses += 1;