diff --git a/docs/frontend-auth.md b/docs/frontend-auth.md
index 30bc764..ff60186 100644
--- a/docs/frontend-auth.md
+++ b/docs/frontend-auth.md
@@ -61,7 +61,7 @@ The `/dashboard` page reads from the metrics endpoints described in [metrics.md]
 
 **Default range:** last 7 days, inclusive of today. Computed at mount in UTC. There's no date-range picker yet.
 
-**Refetch behavior:** mount + `visibilitychange` (when the tab returns to the foreground). A 1-second debounce skips redundant refetches when the visibility flips rapidly. **No polling timer** — the ETL is server-side, refreshes every ~5 minutes, and the dashboard's job is to ask for the latest only when the user is actually looking.
+**Refetch behavior:** mount + `visibilitychange` (when the tab returns to the foreground, debounced 1s) + a 60-second polling timer that calls `refetch()` on every card to stay in lockstep with the backend ETL cadence (`METRICS_ETL_INTERVAL_SECONDS`, configured to 60s in production). Pre-2026-05 builds had no timer; this was added when the ETL was tightened from 300s to 60s for live-feel demos.
 
 **Card states.** Each metric card renders one of four states:
 - `loading` — skeleton placeholder
diff --git a/docs/metrics.md b/docs/metrics.md
index 2052dbf..60a3e3b 100644
--- a/docs/metrics.md
+++ b/docs/metrics.md
@@ -48,7 +48,7 @@ curl -s "http://localhost:3000/api/v1/metrics/context-switching?from=2026-05-01&
 
 ## ETL — how the aggregates are produced
 
-The periodic job [src/services/metrics-etl-scheduler.js](../src/services/metrics-etl-scheduler.js) ticks every `METRICS_ETL_INTERVAL_SECONDS` (default 300). On each tick:
+The periodic job [src/services/metrics-etl-scheduler.js](../src/services/metrics-etl-scheduler.js) ticks every `METRICS_ETL_INTERVAL_SECONDS` (code default 300, production runs at 60 — the dashboard's refresh timer is matched to this). On each tick:
 
 1. Reads up to `METRICS_ETL_BATCH_SIZE` (default 5000) new `activities` rows since the watermark stored in `etl_jobs.last_processed_activity_id`.
 2. Filters to `text_change` and `editor_switch` events; ignores all others (but still advances the watermark past them).
@@ -83,7 +83,7 @@ This uses `sequelize.sync()` (no `force: true`) so it will NOT touch existing ta
 
 | Variable | Default | Purpose |
 | --- | --- | --- |
-| `METRICS_ETL_INTERVAL_SECONDS` | `300` | Seconds between scheduler ticks. |
+| `METRICS_ETL_INTERVAL_SECONDS` | `300` | Seconds between scheduler ticks. Set to `60` in the cluster Secret so the dashboard's 60s auto-refresh sees fresh aggregates. |
 | `METRICS_ETL_BATCH_SIZE` | `5000` | Max `activities` rows consumed per pass. |
 | `METRICS_ETL_ENABLED` | `true` | Set to `false` or `0` to disable the scheduler (useful for tests / local debugging). |
 | `ADMIN_USER_IDS` | _(empty)_ | Comma-separated UUIDs allowed to call `POST /metrics/etl/run`. |
diff --git a/src/services/insight-trigger.service.js b/src/services/insight-trigger.service.js
index 9145cde..8211cdc 100644
--- a/src/services/insight-trigger.service.js
+++ b/src/services/insight-trigger.service.js
@@ -135,8 +135,11 @@ function evaluateRules({ metrics, session }) {
   if (veryLongSession) {
     return { triggered: true, rule: 'very_long_session' };
   }
-  if (longSession && highChurn) {
-    return { triggered: true, rule: 'long_session_high_churn' };
+  if (highChurn) {
+    return { triggered: true, rule: 'high_churn' };
+  }
+  if (longSession) {
+    return { triggered: true, rule: 'long_session' };
   }
   if (rapidSwitching) {
     return { triggered: true, rule: 'rapid_context_switching' };
@@ -225,7 +228,23 @@ export async function evaluateUser(userId) {
   const metrics = await getTodayMetrics(userId);
   const rule = evaluateRules({ metrics, session });
   if (!rule.triggered) {
-    return { skipped: true, reason: 'no_rule_fired' };
+    const diagnostics = {
+      duration_minutes: session?.duration_minutes ?? 0,
+      lines_added: metrics.lines_added,
+      lines_deleted: metrics.lines_deleted,
+      churn_ratio: metrics.churn_ratio,
+      switch_count: metrics.switch_count,
+      rapid_switch_count: metrics.rapid_switch_count,
+      thresholds: {
+        very_long_session_min: VERY_LONG_SESSION_MIN(),
+        long_session_min: LONG_SESSION_MIN(),
+        high_churn_ratio: HIGH_CHURN_RATIO(),
+        rapid_switch_count: RAPID_SWITCH_COUNT(),
+        delete_heavy_total: DELETE_HEAVY_TOTAL(),
+      },
+    };
+    logger.info('insight-trigger: no rule fired', { user_id: userId, ...diagnostics });
+    return { skipped: true, reason: 'no_rule_fired', diagnostics };
   }
 
   let llmOutput;