Make metrics reflect autoscaling policies (Refs #137)

src/routes/metrics/+page.svelte

@@ -6,7 +6,7 @@
     import { timeZone } from '$ui/stores/time-zone.js';
     import { createTimeFormatter } from '$lib/time-zone.js';
     import { pollVisible } from '$ui/poll.js';
-    import { metricDuration } from '$ui/metrics.js';
+    import { metricDuration, metricNumber } from '$ui/metrics.js';
     import { createMetricsRefresh } from '$ui/metrics-refresh.js';
     import SectionHeader from '$ui/components/primitives/SectionHeader.svelte';
     import Card from '$ui/components/primitives/Card.svelte';
@@ -301,8 +301,8 @@
             </p>
         {/if}
         {#if report.summary.observed === 0}<p class="notice">
-                All captured server status is unknown. A dash means occupancy could not be
-                calculated; it does not mean servers were empty.
+                All captured lobby status is unknown. Running counts include these lobbies, but
+                their occupancy and availability could not be determined.
             </p>{/if}
         <MetricsChart points={report.points} days={report.days} timeZone={report.scope.timeZone} />
     {/if}
@@ -316,7 +316,7 @@
             tabs={[
                 { id: 'maps', label: 'Maps' },
                 { id: 'instances', label: 'Instances' },
-                { id: 'insights', label: 'Hosting insights' }
+                { id: 'insights', label: 'Autoscaling signals' }
             ]}
             value={view}
             onchange={changeView}
@@ -364,15 +364,28 @@
         <p>
             Previous window: {format.period(report.previous.since)} – {format.period(
                 report.previous.until
-            )}. Known time: {metricDuration(report.previous.summary.observed)} previous / {metricDuration(
-                report.summary.observed
-            )} current. Percentage points show the difference between two percentages. Equal windows do
-            not guarantee equal observation coverage.
+            )}. Average lobbies running: {report.previous.running.average === null
+                ? '—'
+                : metricNumber(report.previous.running.average)} previous /
+            {report.running.average === null ? '—' : metricNumber(report.running.average)} current. Captured
+            wall time: {metricDuration(report.previous.running.capturedHours)} previous /
+            {metricDuration(report.running.capturedHours)} current. Equal windows do not guarantee equal
+            observation coverage.
         </p>
         <p>
-            Time with players is occupied server time divided by known server time. Known time
-            includes occupied and idle observations; unknown status and collection gaps are excluded
-            from that percentage.
+            Lobbies running averages the concurrent lobby count over captured wall time and includes
+            lobbies with unknown status. Peak is the largest captured concurrent count. Empty spare
+            share divides lobby time with zero reported players by all captured lobby-hours,
+            including unknown status. Empty spares are expected when an autoscaling policy keeps
+            lobbies ready.
+        </p>
+        <p>
+            No empty lobby measures map time when at least one lobby has known occupancy and no
+            tracked lobby reports zero players. If every lobby is unknown, map availability is
+            unknown and excluded from the percentage. Player limits are not required. This
+            availability signal does not establish failed joins or unmet demand. Autoscaling signals
+            compare observations with the current policy; historical policy settings are not
+            recorded.
         </p>
         <p>
             Each periodic observation covers at most {ONLINE_MAX_MS / 1000} seconds. Longer collection
@@ -384,8 +397,10 @@
                 '_',
                 ' '
             )}. Daily buckets are 24-hour periods, not local calendar days. The first and latest
-            buckets can be partial. Percentage bars divide occupied, idle, and unknown time by
-            captured time; the headline occupancy percentage uses known time only.
+            buckets can be partial. Average bars divide occupied, empty spare, and unknown
+            lobby-hours by captured wall time within each bucket, so a partial period is not a lower
+            lobby count merely because it is shorter. Percentage bars use all captured lobby time.
+            Instance time with players uses known lobby time only.
         </p>
         <p>
             Historical server rows are grouped by name and may include multiple registrations using