fix(infra): pods list live by default, and a terminated pod is "Ended" not "Down"

apps/api/src/routes/internal/query-engine.http.ts

@@ -1387,7 +1387,9 @@ export const HttpQueryEngineLive = HttpApiBuilder.group(MapleInternalApi, "query
 								saturation: Number(row.saturation) || 0,
 							})),
 							// The denominator has to match the predicate the list ran, or a scoped
-							// view reads "Top 17 of 541". The scope counts are already computed.
+							// view reads "Top 17 of 541". The scope counts are already computed
+							// within the requested lifecycle, so only the unscoped case has to
+							// pick which lifecycle total it wants.
 							totalCount:
 								Number(
 									payload.scope === "saturated"
@@ -1396,9 +1398,12 @@ export const HttpQueryEngineLive = HttpApiBuilder.group(MapleInternalApi, "query
 											? countRow?.elevatedPods
 											: payload.scope === "unbounded"
 												? countRow?.unboundedPods
-												: payload.scope === "stale"
-													? countRow?.stalePods
-													: countRow?.totalPods,
+												: payload.lifecycle === "ended"
+													? countRow?.endedPods
+													: payload.lifecycle === "all"
+														? Number(countRow?.livePods ?? 0) +
+															Number(countRow?.endedPods ?? 0)
+														: countRow?.livePods,
 								) ||
 								// A failed count must not render as "0 of 0" under a list with rows.
 								rows.length,
@@ -1410,11 +1415,11 @@ export const HttpQueryEngineLive = HttpApiBuilder.group(MapleInternalApi, "query
 						const tenant = yield* CurrentTenant.Context
 						const row = yield* runQueryFirst(Queries.podsSummary, tenant, payload)
 						return new PodsSummaryResponse({
-							totalPods: Number(row?.totalPods) || 0,
+							livePods: Number(row?.livePods) || 0,
+							endedPods: Number(row?.endedPods) || 0,
 							saturatedPods: Number(row?.saturatedPods) || 0,
 							elevatedPods: Number(row?.elevatedPods) || 0,
 							unboundedPods: Number(row?.unboundedPods) || 0,
-							stalePods: Number(row?.stalePods) || 0,
 						})
 					}),
 				)

apps/web/src/api/warehouse/infra.ts

@@ -58,7 +58,10 @@ export type PodSortKey = "saturation" | "cpuUsage" | "cpuLimitPct" | "memoryLimi
 export type SortDirection = "asc" | "desc"
 
 /** One-click fleet scopes from the summary band. */
-export type PodScope = "saturated" | "elevated" | "unbounded" | "stale"
+export type PodScope = "saturated" | "elevated" | "unbounded"
+
+/** Which slice of the window's pods to list — see the domain contract. */
+export type PodLifecycle = "live" | "ended" | "all"
 
 export interface InfraPresenceInput {
 	startTime: string
@@ -207,6 +210,8 @@ export interface ListPodsInput {
 	workloadKind?: WorkloadKind
 	workloadName?: string
 	scope?: PodScope
+	/** Server-side default is `live`. */
+	lifecycle?: PodLifecycle
 	sortBy?: PodSortKey
 	sortDir?: SortDirection
 	limit?: number
@@ -245,6 +250,7 @@ export function listPods({ data }: { data: ListPodsInput }) {
 					workloadKind: data.workloadKind,
 					workloadName: data.workloadName,
 					scope: data.scope,
+					lifecycle: data.lifecycle,
 					sortBy: data.sortBy,
 					sortDir: data.sortDir,
 					limit: data.limit,

apps/web/src/components/infra/format.test.ts

@@ -13,12 +13,12 @@ describe("deriveHostStatus", () => {
 		expect(deriveHostStatus("2026-04-24T11:58:30Z", now)).toBe("idle")
 	})
 
-	it("returns down when last-seen is older than 10x scrape interval", () => {
-		expect(deriveHostStatus("2026-04-24T11:55:00Z", now)).toBe("down")
+	it("returns ended when last-seen is older than 10x scrape interval", () => {
+		expect(deriveHostStatus("2026-04-24T11:55:00Z", now)).toBe("ended")
 	})
 
-	it("returns down for a malformed ISO string", () => {
-		expect(deriveHostStatus("not-a-date", now)).toBe("down")
+	it("returns ended for a malformed ISO string", () => {
+		expect(deriveHostStatus("not-a-date", now)).toBe("ended")
 	})
 })
 

apps/web/src/components/infra/format.ts

@@ -3,7 +3,17 @@ import { toEpochMs } from "@maple/ui/lib/time-format"
 // Generic number/byte/percent formatting lives in `@maple/ui/lib/format`; only
 // infra-specific status policy stays here.
 
-export type HostStatus = "active" | "idle" | "down"
+/**
+ * Collector freshness, which is all a metrics window can honestly report.
+ *
+ * `ended` is deliberately not "down": a series that stops mid-window means the
+ * resource stopped reporting, and for a pod on an autoscaled fleet that is
+ * almost always a normal termination — scale-in, a rollout, a replaced Fargate
+ * task. Calling that an error painted the expected case red. A real down state
+ * needs an expectation signal (`k8s.pod.phase`, or a workload's available vs
+ * desired replicas), and belongs beside these rather than instead of them.
+ */
+export type HostStatus = "active" | "idle" | "ended"
 export type SeverityLevel = "ok" | "warn" | "crit"
 
 export function severityLevel(fraction: number): SeverityLevel {
@@ -17,11 +27,11 @@ const SCRAPE_INTERVAL_MS = 30_000
 
 export function deriveHostStatus(lastSeenIso: string, reference: number | string = Date.now()): HostStatus {
 	const lastSeen = toEpochMs(lastSeenIso)
-	if (!Number.isFinite(lastSeen)) return "down"
+	if (!Number.isFinite(lastSeen)) return "ended"
 	const referenceMs = typeof reference === "number" ? reference : toEpochMs(reference)
 	const ref = Number.isFinite(referenceMs) ? referenceMs : Date.now()
 	const age = ref - lastSeen
 	if (age < SCRAPE_INTERVAL_MS * 2) return "active"
 	if (age < SCRAPE_INTERVAL_MS * 10) return "idle"
-	return "down"
+	return "ended"
 }

apps/web/src/components/infra/severity-tokens.ts

@@ -40,20 +40,20 @@ export const BAR_VALUE_TONE: Record<SeverityLevel, string> = {
 export const STATUS_DOT: Record<HostStatus, string> = {
 	active: "bg-[var(--severity-info)]",
 	idle: "bg-muted-foreground/60",
-	down: "bg-[var(--severity-error)]",
+	ended: "bg-muted-foreground/40",
 } satisfies Record<HostStatus, string>
 
 /** Status dot ring. */
 export const STATUS_RING: Record<HostStatus, string> = {
 	active: "ring-[color-mix(in_oklab,var(--severity-info)_45%,transparent)]",
 	idle: "ring-border",
-	down: "ring-[color-mix(in_oklab,var(--severity-error)_45%,transparent)]",
+	ended: "ring-border",
 } satisfies Record<HostStatus, string>
 
 const STATUS_LABEL: Record<HostStatus, string> = {
 	active: "Active",
 	idle: "Idle",
-	down: "Down",
+	ended: "Ended",
 } satisfies Record<HostStatus, string>
 
 /** Human-readable status word, paired with color so it is never the sole signal. */

apps/web/src/components/infra/status-badge.tsx

@@ -6,7 +6,7 @@ import { statusLabel } from "./severity-tokens"
 const STATUS_TEXT: Record<HostStatus, string> = {
 	active: "text-[var(--severity-info)]",
 	idle: "text-muted-foreground",
-	down: "text-[var(--severity-error)]",
+	ended: "text-muted-foreground",
 } satisfies Record<HostStatus, string>
 
 interface HostStatusBadgeProps {

apps/web/src/routes/infra/index.tsx

@@ -37,7 +37,7 @@ const STATUS_FILTERS: ReadonlyArray<{ value: StatusFilter; label: string }> = [
 	{ value: "all", label: "All" },
 	{ value: "active", label: "Active" },
 	{ value: "idle", label: "Idle" },
-	{ value: "down", label: "Down" },
+	{ value: "ended", label: "Ended" },
 ]
 
 function InfraPage() {
@@ -162,7 +162,7 @@ function FleetView({
 	)
 
 	const counts = useMemo(() => {
-		const c: Record<HostStatus, number> = { active: 0, idle: 0, down: 0 } satisfies Record<
+		const c: Record<HostStatus, number> = { active: 0, idle: 0, ended: 0 } satisfies Record<
 			HostStatus,
 			number
 		>

apps/web/src/routes/infra/kubernetes/nodes/index.tsx

@@ -24,7 +24,7 @@ import {
 
 const DEFAULT_PRESET = "12h"
 
-const NodeStatusParam = Schema.optional(Schema.Literals(["active", "idle", "down"]))
+const NodeStatusParam = Schema.optional(Schema.Literals(["active", "idle", "ended"]))
 
 const nodesSearchSchema = Schema.Struct({
 	q: Schema.optional(Schema.String),
@@ -44,9 +44,10 @@ export const Route = createFileRoute("/infra/kubernetes/nodes/")({
 
 /**
  * The states are collector freshness, not Kubernetes conditions: a node is
- * "Down" here when no kubelet metric has arrived recently, which is a fact about
- * the collector as much as about the node. `k8s.node.condition_ready` is
- * collected but unqueried — when it lands, it belongs beside these, not instead.
+ * "Ended" here when no kubelet metric has arrived recently, which on an ASG or
+ * a Karpenter-managed fleet usually means the node was scaled in, not that it
+ * failed — so it reads neutral. `k8s.node.condition_ready` is collected but
+ * unqueried; when it lands it belongs beside these, not instead.
  */
 const STATUS_CELLS: ReadonlyArray<{
 	status: HostStatus
@@ -55,13 +56,13 @@ const STATUS_CELLS: ReadonlyArray<{
 }> = [
 	{ status: "active", hint: "reporting", tone: "info" },
 	{ status: "idle", hint: "quiet >1m", tone: "warn" },
-	{ status: "down", hint: "silent >5m", tone: "crit" },
+	{ status: "ended", hint: "silent >5m", tone: "neutral" },
 ]
 
 const STATUS_SEGMENT: Record<HostStatus, string> = {
 	active: "bg-[var(--severity-info)]",
 	idle: "bg-[var(--severity-warn)]",
-	down: "bg-[var(--severity-error)]",
+	ended: "bg-muted-foreground/40",
 } satisfies Record<HostStatus, string>
 
 function NodesPage() {
@@ -148,7 +149,7 @@ function NodesPage() {
 
 					// The band counts the whole scope as of the window's end, so it keeps
 					// saying what the search and the status cell just hid.
-					const counts = { active: 0, idle: 0, down: 0 } satisfies Record<HostStatus, number>
+					const counts = { active: 0, idle: 0, ended: 0 } satisfies Record<HostStatus, number>
 					for (const node of nodes) counts[deriveHostStatus(node.lastSeen, endTime)]++
 
 					const q = searchText.trim().toLowerCase()

apps/web/src/routes/infra/kubernetes/pods/index.tsx

@@ -33,14 +33,19 @@ import {
 const PAGE_SIZE = 50
 const DEFAULT_PRESET = "12h"
 
-/** A one-click scope from the band. `undefined` means the whole fleet. */
-type PodScope = "saturated" | "elevated" | "unbounded" | "stale"
+/**
+ * A one-click scope from the band. `undefined` means the live fleet, which is
+ * what the list shows by default — see `PodLifecycle` in the domain contract.
+ * `ended` is a lifecycle rather than a saturation bucket, but it rides in the
+ * same URL param so the band stays one uniform row of cells.
+ */
+type PodScope = "saturated" | "elevated" | "unbounded" | "ended"
 
 const PodSortKeyParam = Schema.optional(
 	Schema.Literals(["saturation", "cpuUsage", "cpuLimitPct", "memoryLimitPct", "podName", "lastSeen"]),
 )
 const SortDirParam = Schema.optional(Schema.Literals(["asc", "desc"]))
-const ScopeParam = Schema.optional(Schema.Literals(["saturated", "elevated", "unbounded", "stale"]))
+const ScopeParam = Schema.optional(Schema.Literals(["saturated", "elevated", "unbounded", "ended"]))
 
 const podsSearchSchema = Schema.Struct({
 	q: Schema.optional(Schema.String),
@@ -83,7 +88,7 @@ const SCOPE_LABEL: Record<PodScope, string> = {
 	saturated: "at or above 90% of a limit",
 	elevated: "at or above 60% of a limit",
 	unbounded: "running with no limits set",
-	stale: "whose collector has gone quiet",
+	ended: "no longer running",
 } satisfies Record<PodScope, string>
 
 function PodsPage() {
@@ -127,14 +132,17 @@ function PodsPage() {
 	const searchText = search.q ?? ""
 	const debouncedSearch = useDebouncedValue(searchText, 300)
 
+	// "Ended" is the lifecycle dial, not a saturation bucket: the other three
+	// scopes narrow the live fleet, this one swaps which fleet is on screen.
 	const podsResult = useAtomValue(
 		listPodsResultAtom({
 			data: {
 				startTime,
 				endTime,
 				...filters,
 				search: debouncedSearch.trim() || undefined,
-				scope,
+				scope: scope === "ended" ? undefined : scope,
+				lifecycle: scope === "ended" ? "ended" : "live",
 				sortBy,
 				sortDir,
 				limit: PAGE_SIZE,
@@ -182,6 +190,13 @@ function PodsPage() {
 		patchSearch({ sortBy: key, sortDir: key === "podName" ? "asc" : "desc" })
 	}
 
+	// An unfiltered page can come back empty because the fleet is entirely gone
+	// rather than never instrumented — a job that finished, an environment scaled
+	// to zero. The band already knows, so the empty state can stop guessing.
+	const endedPods = Result.builder(summaryResult)
+		.onSuccess((counts) => counts.endedPods)
+		.orElse(() => 0)
+
 	const hasStructuredFilter = Object.values(filters).some((v) => (v?.length ?? 0) > 0)
 	const hasAnyNarrowing = hasStructuredFilter || Boolean(searchText.trim()) || Boolean(scope)
 
@@ -253,22 +268,22 @@ function PodsPage() {
 								tone: "warn",
 							},
 							{
-								scope: "stale",
-								label: "Stale collector",
-								hint: ">5m",
-								value: counts.stalePods,
+								scope: "ended",
+								label: "Ended",
+								hint: "not running",
+								value: counts.endedPods,
 								tone: "neutral",
 							},
 						]
 						const healthy = Math.max(
-							counts.totalPods - counts.saturatedPods - counts.elevatedPods,
+							counts.livePods - counts.saturatedPods - counts.elevatedPods,
 							0,
 						)
 						return (
 							<FleetBand
-								total={counts.totalPods}
-								noun="pod"
-								caption="share of the fleet by peak utilization"
+								total={counts.livePods}
+								noun="live pod"
+								caption="share of the live fleet by peak utilization"
 								segments={[
 									{ key: "healthy", count: healthy, className: "bg-muted-foreground/35" },
 									{
@@ -305,12 +320,26 @@ function PodsPage() {
 										<EmptyMedia variant="icon">
 											<FolderIcon size={16} />
 										</EmptyMedia>
-										<EmptyTitle>No pods reporting yet</EmptyTitle>
+										<EmptyTitle>
+											{endedPods > 0
+												? "Nothing running right now"
+												: "No pods reporting yet"}
+										</EmptyTitle>
 										<EmptyDescription>
-											Install the Maple Kubernetes Helm chart so the kubelet stats
-											receiver can start collecting per-pod CPU and memory metrics.
+											{endedPods > 0
+												? `Every pod that reported in this window has since ended. Open the Ended scope to see the ${endedPods.toLocaleString()} that ran.`
+												: "Install the Maple Kubernetes Helm chart so the kubelet stats receiver can start collecting per-pod CPU and memory metrics."}
 										</EmptyDescription>
 									</EmptyHeader>
+									{endedPods > 0 ? (
+										<Button
+											variant="outline"
+											size="sm"
+											onClick={() => patchSearch({ scope: "ended" })}
+										>
+											Show ended pods
+										</Button>
+									) : null}
 								</Empty>
 							)
 						}

apps/web/src/routes/infra/kubernetes/workloads/index.tsx

@@ -64,7 +64,7 @@ const KIND_OPTIONS = [
  * pods view is where a spike shows.
  */
 function scopeOf(workload: WorkloadRow, referenceTime: string): WorkloadScope | null {
-	if (deriveHostStatus(workload.lastSeen, referenceTime) === "down") return "stale"
+	if (deriveHostStatus(workload.lastSeen, referenceTime) === "ended") return "stale"
 	const level = severityLevel(Math.max(workload.avgCpuLimitPct, workload.avgMemoryLimitPct))
 	if (level === "crit") return "saturated"
 	if (level === "warn") return "elevated"

packages/domain/src/http/query-engine.ts

@@ -1207,7 +1207,14 @@ const PodSortKeyLiteral = Schema.Literals([
 const SortDirectionLiteral = Schema.Literals(["asc", "desc"])
 
 /** One-click fleet scopes from the browse summary band. */
-const PodScopeLiteral = Schema.Literals(["saturated", "elevated", "unbounded", "stale"])
+const PodScopeLiteral = Schema.Literals(["saturated", "elevated", "unbounded"])
+
+/**
+ * Which slice of the window's pods to return. A windowed list is the union of
+ * everything that reported at any point in it, so on an autoscaled fleet most
+ * of those pods no longer exist. Defaults to `live` server-side.
+ */
+const PodLifecycleLiteral = Schema.Literals(["live", "ended", "all"])
 
 export class ListPodsRequest extends Schema.Class<ListPodsRequest>("ListPodsRequest")({
 	startTime: TinybirdDateTime,
@@ -1236,6 +1243,7 @@ export class ListPodsRequest extends Schema.Class<ListPodsRequest>("ListPodsRequ
 	workloadKind: Schema.optional(WorkloadKindLiteral),
 	workloadName: Schema.optional(Schema.String),
 	scope: Schema.optional(PodScopeLiteral),
+	lifecycle: Schema.optional(PodLifecycleLiteral),
 	sortBy: Schema.optional(PodSortKeyLiteral),
 	sortDir: Schema.optional(SortDirectionLiteral),
 	limit: Schema.optional(Schema.Number),
@@ -1286,11 +1294,13 @@ export class PodsSummaryRequest extends Schema.Class<PodsSummaryRequest>("PodsSu
 }) {}
 
 export class PodsSummaryResponse extends Schema.Class<PodsSummaryResponse>("PodsSummaryResponse")({
-	totalPods: Schema.Number,
+	/** Still reporting at the window's end — the fleet as it stands. */
+	livePods: Schema.Number,
+	/** Reported earlier in the window and stopped: scale-in, a rollout, a cycled task. */
+	endedPods: Schema.Number,
 	saturatedPods: Schema.Number,
 	elevatedPods: Schema.Number,
 	unboundedPods: Schema.Number,
-	stalePods: Schema.Number,
 }) {}
 
 // Containers (Docker) — docker_stats receiver rows, identity (container.name,