@iann29/rastro 0.1.0-alpha.0 → 0.1.0-alpha.10

package/src/component/ingest.ts

@@ -5,25 +5,63 @@ import { mutation } from "./_generated/server.js";
 import { internal } from "./_generated/api.js";
 import {
   AGGREGATE_SHARDS,
+  MAX_BATCH_EVENT_GROUPS,
   DAY_MS,
+  HOURLY_AGGREGATE_SHARDS,
   DIMENSION_SLOTS,
   HOUR_MS,
+  LIVE_LEAVE_GRACE_MS,
   LIVE_SESSION_TTL_MS,
+  LIVE_SWEEP_INTERVAL_MS,
   MAX_BATCH_EVENTS,
   MAX_SITE_BYTES_PER_SHARD_WINDOW,
   MAX_SITE_EVENTS_PER_SHARD_WINDOW,
   MAX_BYTES_PER_SESSION_WINDOW,
   MAX_EVENTS_PER_SESSION_WINDOW,
   MAX_FUNNELS_PER_SITE,
+  MAX_FUNNEL_STEPS,
   MAX_GOALS_PER_SITE,
+  MAX_SITE_MAP_ROUTES_PER_DAY,
   RATE_LIMIT_WINDOW_MS,
+  REPORT_ROLLUP_SHARDS,
+  ROUTE_TRANSITION_SLOTS,
+  SECONDARY_AGGREGATE_SHARDS,
   SITE_INGEST_SHARDS,
+  SITE_MAP_OTHER_ROUTE,
+  VITAL_PAGE_SLOTS,
 } from "./constants.js";
+import {
+  addVisitor,
+  createVisitorSketch,
+  unionVisitorSketches,
+  type VisitorSketch,
+} from "./cardinality.js";
+import {
+  addToVitalHistogram,
+  emptyVitalHistogram,
+  mergeVitalHistograms,
+  vitalDevice,
+  VITAL_ALL,
+  VITAL_OTHER_PAGES,
+  type VitalMetric,
+} from "./vitals.js";
 import { fail } from "./errors.js";
+import { ensureAnalyticsControl } from "./coverage.js";
+import {
+  bufferEventBatch,
+  eventBatchKey,
+  eventBucketStart,
+  type EventBatchBuffer,
+  findTelemetry,
+  flushEventBatches,
+  insertEventBatch,
+  inspectTelemetryMinute,
+} from "./eventStore.js";
 import {
   cleanString,
-  ensureBatchBytes,
   isValidMoney,
+  measureBatchBytes,
+  normalizeRoute,
   originAllowed,
   sanitizeContext,
   sanitizeCurrency,
@@ -46,6 +84,7 @@ import {
 
 type DimensionType =
   | "source"
+  | "campaign"
   | "page"
   | "country"
   | "device"
@@ -67,11 +106,78 @@ type AggregateDelta = {
   outboundClicks: number;
   sessions: number;
   visitors: number;
+  visitorSketch: VisitorSketch;
   conversions: number;
   revenueCents: number;
   dimensions: Doc<"aggregateBuckets">["dimensions"];
 };
 type AggregateDeltas = Map<string, AggregateDelta>;
+type GoalRollupDelta = {
+  siteId: Id<"sites">;
+  goalId: Id<"goals">;
+  bucketStart: number;
+  shard: number;
+  goalKey: string;
+  completions: number;
+  valueCents: number;
+};
+type FunnelRollupDelta = {
+  siteId: Id<"sites">;
+  funnelId: Id<"funnels">;
+  bucketStart: number;
+  shard: number;
+  steps: Map<
+    number,
+    {
+      step: number;
+      entrants: number;
+      completions: number;
+    }
+  >;
+};
+type AffiliateRollupDelta = {
+  siteId: Id<"sites">;
+  affiliateId: Id<"affiliates">;
+  bucketStart: number;
+  shard: number;
+  affiliateSlug: string;
+  visits: number;
+  visitorSketch: VisitorSketch;
+  conversions: number;
+  revenueCents: number;
+  commissionCents: number;
+};
+type VitalRollupDelta = {
+  siteId: Id<"sites">;
+  bucketStart: number;
+  device: string;
+  page: string;
+  shard: number;
+  metrics: Map<
+    VitalMetric,
+    { count: number; sum: number; histogram: number[] }
+  >;
+};
+type RouteRollupDelta = {
+  siteId: Id<"sites">;
+  bucketStart: number;
+  shard: number;
+  route: string;
+  pageviews: number;
+  entries: number;
+  continued: number;
+  dwellMs: number;
+  dwellSamples: number;
+  visitorSketch: VisitorSketch;
+  outgoing: Map<string, number>;
+};
+type ReportRollupDeltas = {
+  goals: Map<string, GoalRollupDelta>;
+  funnels: Map<string, FunnelRollupDelta>;
+  affiliates: Map<string, AffiliateRollupDelta>;
+  vitals: Map<string, VitalRollupDelta>;
+  routes: Map<string, RouteRollupDelta>;
+};
 type AttributionSnapshot = {
   affiliateId: Id<"affiliates">;
   affiliateSlug: string;
@@ -85,6 +191,12 @@ type SessionState = {
   isNewVisitorDay: boolean;
   startedAt: number;
   source: string;
+  visitorKey: string;
+  utmCampaign?: string;
+  country?: string;
+  city?: string;
+  latitude?: number;
+  longitude?: number;
   pageviewCount: number;
   eventCount: number;
   affiliateId?: Id<"affiliates">;
@@ -93,6 +205,36 @@ type SessionState = {
   affiliateExpiresAt?: number;
 };
 
+type SessionRateWindowState = {
+  siteId: Id<"sites">;
+  sessionId: string;
+  window: Doc<"ingestWindows"> | null;
+  eventCount: number;
+  byteCount: number;
+  dirty: boolean;
+};
+
+type CachedRow<T> = {
+  current: T | null;
+  dirty: boolean;
+};
+
+type CachedDocument<
+  TableName extends keyof import("./_generated/dataModel.js").DataModel,
+> = Omit<Doc<TableName>, "_creationTime"> & { _creationTime?: number };
+
+type BatchState = {
+  sessions: Map<string, CachedRow<CachedDocument<"sessions">>>;
+  liveSessions: Map<string, CachedRow<CachedDocument<"liveSessions">>>;
+  visitorAttributions: Map<
+    string,
+    CachedRow<CachedDocument<"visitorAttributions">>
+  >;
+  funnelProgress: Map<string, CachedRow<CachedDocument<"funnelProgress">>>;
+  eventBatches: EventBatchBuffer;
+  liveSweepSites: Set<string>;
+};
+
 export const ingestBatch = mutation({
   args: {
     siteId: v.id("sites"),
@@ -110,46 +252,135 @@ export const ingestBatch = mutation({
       });
     }
     if (args.events.length < 1 || args.events.length > MAX_BATCH_EVENTS) {
-      fail("BATCH_TOO_LARGE", `events must contain 1-${MAX_BATCH_EVENTS} items`, {
-        limit: MAX_BATCH_EVENTS,
-      });
+      fail(
+        "BATCH_TOO_LARGE",
+        `events must contain 1-${MAX_BATCH_EVENTS} items`,
+        {
+          limit: MAX_BATCH_EVENTS,
+        },
+      );
     }
+    let eventByteCounts: number[];
     try {
-      ensureBatchBytes({ siteId: String(args.siteId), events: args.events });
+      ({ eventByteCounts } = measureBatchBytes(
+        String(args.siteId),
+        args.events,
+      ));
     } catch {
       fail("BATCH_TOO_LARGE", "batch exceeds 64 KiB", { limitBytes: 65_536 });
     }
 
     const now = Date.now();
-    const eventByteCounts = await consumeSiteRateLimit(
+    await ensureAnalyticsControl(ctx, args.siteId, now);
+    await consumeSiteRateLimit(
       ctx,
       args.siteId,
       args.events,
+      eventByteCounts,
       now,
     );
     const context = sanitizeContext(args.context);
     const aggregateDeltas: AggregateDeltas = new Map();
+    const reportRollupDeltas = createReportRollupDeltas();
+    const sessionRateWindows = new Map<string, SessionRateWindowState>();
+    const batchState: BatchState = {
+      sessions: new Map(),
+      liveSessions: new Map(),
+      visitorAttributions: new Map(),
+      funnelProgress: new Map(),
+      eventBatches: new Map(),
+      liveSweepSites: new Set(),
+    };
     let goals: Doc<"goals">[] | null = null;
     let funnels: Doc<"funnels">[] | null = null;
     let accepted = 0;
     let duplicates = 0;
     let rejected = 0;
-    for (let eventIndex = 0; eventIndex < args.events.length; eventIndex += 1) {
-      const input = args.events[eventIndex];
-      let event: TrackerEvent;
+    const events = args.events.map((input) => {
       try {
-        event = sanitizeEvent(input, now);
+        return sanitizeEvent(input, now);
       } catch {
+        return null;
+      }
+    });
+    const duplicateInBatch = new Set<number>();
+    const eventIdsByGroup = new Map<string, Set<string>>();
+    const requestGroups = new Set<string>();
+    const dedupeGroups = new Map<
+      string,
+      { sessionId: string; bucketStart: number; eventIds: Set<string> }
+    >();
+    for (let eventIndex = 0; eventIndex < events.length; eventIndex += 1) {
+      const event = events[eventIndex];
+      if (!event) continue;
+      const bucketStart = eventBucketStart(event.timestamp);
+      const groupKey = eventBatchKey(event.sessionId, bucketStart);
+      requestGroups.add(groupKey);
+      if (
+        event.type === "heartbeat" ||
+        event.type === "vital" ||
+        event.type === "leave"
+      )
+        continue;
+      const eventIds = eventIdsByGroup.get(groupKey) ?? new Set<string>();
+      if (eventIds.has(event.eventId)) {
+        duplicateInBatch.add(eventIndex);
+        continue;
+      }
+      eventIds.add(event.eventId);
+      eventIdsByGroup.set(groupKey, eventIds);
+      const group = dedupeGroups.get(groupKey) ?? {
+        sessionId: event.sessionId,
+        bucketStart,
+        eventIds: new Set<string>(),
+      };
+      group.eventIds.add(event.eventId);
+      dedupeGroups.set(groupKey, group);
+    }
+    if (requestGroups.size > MAX_BATCH_EVENT_GROUPS) {
+      fail(
+        "BATCH_TOO_LARGE",
+        `events may span at most ${MAX_BATCH_EVENT_GROUPS} session-minute groups`,
+        { limit: MAX_BATCH_EVENT_GROUPS },
+      );
+    }
+    const persistedEventIds = new Map<string, Set<string>>();
+    await Promise.all(
+      [...dedupeGroups].map(async ([key, group]) => {
+        const { eventIds, persistedCount } = await inspectTelemetryMinute(ctx, {
+          siteId: args.siteId,
+          sessionId: group.sessionId,
+          bucketStart: group.bucketStart,
+        });
+        const newCount = [...group.eventIds].filter(
+          (eventId) => !eventIds.has(eventId),
+        ).length;
+        if (
+          newCount > 0 &&
+          persistedCount + newCount > MAX_EVENTS_PER_SESSION_WINDOW
+        ) {
+          fail("RATE_LIMITED", "session event minute limit exceeded", {
+            limit: MAX_EVENTS_PER_SESSION_WINDOW,
+          });
+        }
+        persistedEventIds.set(key, eventIds);
+      }),
+    );
+
+    for (let eventIndex = 0; eventIndex < events.length; eventIndex += 1) {
+      const event = events[eventIndex];
+      if (!event) {
         rejected += 1;
         continue;
       }
 
-      const existingSession = await ctx.db
-        .query("sessions")
-        .withIndex("by_siteId_and_sessionId", (range) =>
-          range.eq("siteId", args.siteId).eq("sessionId", event.sessionId),
-        )
-        .unique();
+      const sessionRow = await loadSession(
+        ctx,
+        args.siteId,
+        event.sessionId,
+        batchState,
+      );
+      const existingSession = sessionRow.current;
       if (existingSession && existingSession.visitorId !== event.visitorId) {
         rejected += 1;
         continue;
@@ -160,13 +391,14 @@ export const ingestBatch = mutation({
           rejected += 1;
           continue;
         }
-        const liveSession = await ctx.db
-          .query("liveSessions")
-          .withIndex("by_siteId_and_sessionId", (range) =>
-            range.eq("siteId", args.siteId).eq("sessionId", event.sessionId),
-          )
-          .unique();
-        if (liveSession?.lastHeartbeatEventId === event.eventId) {
+        const liveSession = (
+          await loadLiveSession(ctx, args.siteId, event.sessionId, batchState)
+        ).current;
+        if (
+          liveSession &&
+          (liveSession.lastHeartbeatEventId === event.eventId ||
+            liveSession.recentHeartbeatEventIds?.includes(event.eventId))
+        ) {
           duplicates += 1;
           continue;
         }
@@ -177,6 +409,7 @@ export const ingestBatch = mutation({
             event.sessionId,
             eventByteCounts[eventIndex],
             now,
+            sessionRateWindows,
           ))
         ) {
           rejected += 1;
@@ -189,18 +422,94 @@ export const ingestBatch = mutation({
           event,
           context,
           now,
+          batchState,
         );
         accepted += 1;
         continue;
       }
 
-      const duplicate = await ctx.db
-        .query("events")
-        .withIndex("by_siteId_and_eventId", (range) =>
-          range.eq("siteId", args.siteId).eq("eventId", event.eventId),
-        )
-        .unique();
-      if (duplicate) {
+      if (event.type === "leave") {
+        // A departure only shortens live presence: it stores no telemetry,
+        // touches no aggregates, and never extends anything, so a replayed
+        // or late leave is inert. The read side filters by expiresAt and the
+        // sweep deletes the row, so the dashboard sees the exit in seconds.
+        if (!existingSession) {
+          rejected += 1;
+          continue;
+        }
+        if (
+          !(await consumeRateLimit(
+            ctx,
+            args.siteId,
+            event.sessionId,
+            eventByteCounts[eventIndex],
+            now,
+            sessionRateWindows,
+          ))
+        ) {
+          rejected += 1;
+          continue;
+        }
+        const liveState = await loadLiveSession(
+          ctx,
+          args.siteId,
+          event.sessionId,
+          batchState,
+        );
+        const liveSession = liveState.current;
+        if (liveSession) {
+          const expiresAt = Math.min(
+            liveSession.expiresAt,
+            now + LIVE_LEAVE_GRACE_MS,
+          );
+          if (expiresAt < liveSession.expiresAt) {
+            await ctx.db.patch("liveSessions", liveSession._id, { expiresAt });
+            liveState.current = { ...liveSession, expiresAt };
+          }
+          await ensureLiveSweep(ctx, args.siteId, now, batchState);
+        }
+        accepted += 1;
+        continue;
+      }
+
+      if (event.type === "vital") {
+        // A vital annotates telemetry its session already produced; it never
+        // creates sessions, live presence, event batches, or aggregate counts.
+        if (!existingSession) {
+          rejected += 1;
+          continue;
+        }
+        if (
+          !(await consumeRateLimit(
+            ctx,
+            args.siteId,
+            event.sessionId,
+            eventByteCounts[eventIndex],
+            now,
+            sessionRateWindows,
+          ))
+        ) {
+          rejected += 1;
+          continue;
+        }
+        foldVitalRollups(
+          reportRollupDeltas,
+          args.siteId,
+          event,
+          vitalDevice(context.device),
+        );
+        accepted += 1;
+        continue;
+      }
+
+      if (
+        duplicateInBatch.has(eventIndex) ||
+        persistedEventIds
+          .get(
+            eventBatchKey(event.sessionId, eventBucketStart(event.timestamp)),
+          )
+          ?.has(event.eventId)
+      ) {
         duplicates += 1;
         continue;
       }
@@ -212,6 +521,7 @@ export const ingestBatch = mutation({
           event.sessionId,
           eventByteCounts[eventIndex],
           now,
+          sessionRateWindows,
         ))
       ) {
         rejected += 1;
@@ -231,22 +541,33 @@ export const ingestBatch = mutation({
           .withIndex("by_siteId", (range) => range.eq("siteId", args.siteId))
           .take(MAX_FUNNELS_PER_SITE);
       }
-      await accountEvent(ctx, {
+      const session = await accountEvent(ctx, {
         siteId: args.siteId,
         event,
         context,
         existingSession,
-        goals: canMatchDefinition ? goals ?? [] : [],
-        funnels: canMatchDefinition ? funnels ?? [] : [],
+        goals: canMatchDefinition ? (goals ?? []) : [],
+        funnels: canMatchDefinition ? (funnels ?? []) : [],
         aggregateDeltas,
+        reportRollupDeltas,
         recordTelemetry: true,
         financialConversion: false,
         trustedGoalKey: undefined,
         now,
+        batchState,
       });
+      bufferEventBatch(
+        args.siteId,
+        event,
+        batchState.eventBatches,
+        session.country ?? "unknown",
+      );
       accepted += 1;
     }
+    await flushBatchState(ctx, batchState);
+    await flushSessionRateLimits(ctx, sessionRateWindows, now);
     await flushAggregates(ctx, aggregateDeltas);
+    await flushReportRollups(ctx, reportRollupDeltas);
     return { accepted, duplicates, rejected };
   },
 });
@@ -258,17 +579,24 @@ export const trackConversion = mutation({
     const site = await ctx.db.get("sites", args.siteId);
     if (!site) fail("NOT_FOUND", "site not found");
     if (!isValidMoney(args.revenueCents)) {
-      fail("INVALID_ARGUMENT", "revenueCents must be an integer between 0 and 1000000000");
+      fail(
+        "INVALID_ARGUMENT",
+        "revenueCents must be an integer between 0 and 1000000000",
+      );
     }
     let event: TrackerEvent;
+    let trustedGoalKey: string | undefined;
     try {
+      trustedGoalKey = args.goalKey
+        ? sanitizeSlug(args.goalKey, "goalKey")
+        : undefined;
       event = sanitizeEvent(
         {
           eventId: sanitizeOpaqueId(args.eventId, "eventId"),
           sessionId: sanitizeOpaqueId(args.sessionId, "sessionId"),
           visitorId: sanitizeOpaqueId(args.visitorId, "visitorId"),
           type: "conversion",
-          name: args.goalKey ? sanitizeSlug(args.goalKey, "goalKey") : "conversion",
+          name: trustedGoalKey ?? "conversion",
           path: sanitizePath(args.path ?? "/conversion"),
           timestamp: args.timestamp,
           sequence: 0,
@@ -279,16 +607,13 @@ export const trackConversion = mutation({
         Date.now(),
       );
     } catch (error) {
-      fail("INVALID_ARGUMENT", error instanceof Error ? error.message : "invalid conversion");
+      fail(
+        "INVALID_ARGUMENT",
+        error instanceof Error ? error.message : "invalid conversion",
+      );
     }
 
     const siteCurrency = site.currency ?? "USD";
-    if (event.currency !== siteCurrency) {
-      fail("CURRENCY_MISMATCH", "conversion currency does not match site currency", {
-        siteCurrency,
-        conversionCurrency: event.currency ?? "USD",
-      });
-    }
     const priorConversion = await ctx.db
       .query("conversions")
       .withIndex("by_siteId_and_eventId", (range) =>
@@ -296,15 +621,27 @@ export const trackConversion = mutation({
       )
       .unique();
     if (priorConversion) {
+      if (!sameTrustedConversion(priorConversion, event, trustedGoalKey)) {
+        fail(
+          "CONFLICT",
+          "conversion eventId is already associated with a different payload",
+        );
+      }
       return { accepted: 0, duplicates: 1, rejected: 0 };
     }
+    if (event.currency !== siteCurrency) {
+      fail(
+        "CURRENCY_MISMATCH",
+        "conversion currency does not match site currency",
+        {
+          siteCurrency,
+          conversionCurrency: event.currency ?? "USD",
+        },
+      );
+    }
 
-    const existingTelemetry = await ctx.db
-      .query("events")
-      .withIndex("by_siteId_and_eventId", (range) =>
-        range.eq("siteId", args.siteId).eq("eventId", event.eventId),
-      )
-      .unique();
+    const telemetry = await findTelemetry(ctx, args.siteId, event);
+    const existingTelemetry = telemetry.match;
     if (
       existingTelemetry &&
       (existingTelemetry.sessionId !== event.sessionId ||
@@ -321,20 +658,19 @@ export const trackConversion = mutation({
     if (existingSession && existingSession.visitorId !== event.visitorId) {
       fail("CONFLICT", "session belongs to a different visitor");
     }
-    const conversionContext: IngestContext = existingTelemetry
-      ? {
-          country: existingTelemetry.country,
-          city: existingTelemetry.city,
-          latitude: existingTelemetry.latitude,
-          longitude: existingTelemetry.longitude,
-          browser: existingTelemetry.browser,
-          os: existingTelemetry.os,
-          device: existingTelemetry.device,
-        }
-      : sanitizeContext(undefined);
+    const conversionContext: IngestContext = {
+      country: existingTelemetry?.country ?? existingSession?.country,
+      city: existingTelemetry?.city ?? existingSession?.city,
+      latitude: existingTelemetry?.latitude ?? existingSession?.latitude,
+      longitude: existingTelemetry?.longitude ?? existingSession?.longitude,
+      browser: existingTelemetry?.browser ?? existingSession?.browser,
+      os: existingTelemetry?.os ?? existingSession?.os,
+      device: existingTelemetry?.device ?? existingSession?.device,
+    };
 
     const aggregateDeltas: AggregateDeltas = new Map();
-    await accountEvent(ctx, {
+    const reportRollupDeltas = createReportRollupDeltas();
+    const session = await accountEvent(ctx, {
       siteId: args.siteId,
       event,
       context: conversionContext,
@@ -342,34 +678,149 @@ export const trackConversion = mutation({
       goals: [],
       funnels: [],
       aggregateDeltas,
-      recordTelemetry: existingTelemetry === null,
+      reportRollupDeltas,
+      recordTelemetry: existingTelemetry === null && !telemetry.saturated,
       financialConversion: true,
-      trustedGoalKey: args.goalKey
-        ? sanitizeSlug(args.goalKey, "goalKey")
-        : undefined,
+      trustedGoalKey,
       now: Date.now(),
     });
+    if (!existingTelemetry && !telemetry.saturated) {
+      await insertEventBatch(
+        ctx,
+        args.siteId,
+        event,
+        session.country ?? "unknown",
+      );
+    }
     await flushAggregates(ctx, aggregateDeltas);
+    await flushReportRollups(ctx, reportRollupDeltas);
+    await ensureAnalyticsControl(ctx, args.siteId, Date.now());
     return { accepted: 1, duplicates: 0, rejected: 0 };
   },
 });
 
+function sameTrustedConversion(
+  existing: Doc<"conversions">,
+  event: TrackerEvent,
+  goalKey: string | undefined,
+): boolean {
+  return (
+    existing.source === "trusted" &&
+    existing.sessionId === event.sessionId &&
+    existing.visitorId === event.visitorId &&
+    existing.timestamp === event.timestamp &&
+    existing.path === event.path &&
+    existing.revenueCents === (event.revenueCents ?? 0) &&
+    existing.currency === (event.currency ?? "USD") &&
+    existing.goalKey === goalKey
+  );
+}
+
+async function loadSession(
+  ctx: MutationCtx,
+  siteId: Id<"sites">,
+  sessionId: string,
+  batchState: BatchState,
+): Promise<CachedRow<CachedDocument<"sessions">>> {
+  const cached = batchState.sessions.get(sessionId);
+  if (cached) return cached;
+  const current = await ctx.db
+    .query("sessions")
+    .withIndex("by_siteId_and_sessionId", (range) =>
+      range.eq("siteId", siteId).eq("sessionId", sessionId),
+    )
+    .unique();
+  const loaded = { current, dirty: false };
+  batchState.sessions.set(sessionId, loaded);
+  return loaded;
+}
+
+async function loadLiveSession(
+  ctx: MutationCtx,
+  siteId: Id<"sites">,
+  sessionId: string,
+  batchState: BatchState,
+): Promise<CachedRow<CachedDocument<"liveSessions">>> {
+  const cached = batchState.liveSessions.get(sessionId);
+  if (cached) return cached;
+  const current = await ctx.db
+    .query("liveSessions")
+    .withIndex("by_siteId_and_sessionId", (range) =>
+      range.eq("siteId", siteId).eq("sessionId", sessionId),
+    )
+    .unique();
+  const loaded = { current, dirty: false };
+  batchState.liveSessions.set(sessionId, loaded);
+  return loaded;
+}
+
+async function flushBatchState(ctx: MutationCtx, batchState: BatchState) {
+  for (const state of batchState.sessions.values()) {
+    if (state.dirty && state.current) {
+      await ctx.db.replace(
+        "sessions",
+        state.current._id,
+        withoutSystemFields(state.current),
+      );
+    }
+  }
+  for (const state of batchState.liveSessions.values()) {
+    if (state.dirty && state.current) {
+      await ctx.db.replace(
+        "liveSessions",
+        state.current._id,
+        withoutSystemFields(state.current),
+      );
+    }
+  }
+  for (const state of batchState.visitorAttributions.values()) {
+    if (state.dirty && state.current) {
+      await ctx.db.replace(
+        "visitorAttributions",
+        state.current._id,
+        withoutSystemFields(state.current),
+      );
+    }
+  }
+  for (const state of batchState.funnelProgress.values()) {
+    if (state.dirty && state.current) {
+      await ctx.db.replace(
+        "funnelProgress",
+        state.current._id,
+        withoutSystemFields(state.current),
+      );
+    }
+  }
+  await flushEventBatches(ctx, batchState.eventBatches);
+}
+
+function withoutSystemFields<
+  T extends { _id: unknown; _creationTime?: number },
+>(document: T): Omit<T, "_id" | "_creationTime"> {
+  const { _id, _creationTime, ...fields } = document;
+  void _id;
+  void _creationTime;
+  return fields;
+}
+
 async function accountEvent(
   ctx: MutationCtx,
   input: {
     siteId: Id<"sites">;
     event: TrackerEvent;
     context: IngestContext;
-    existingSession: Doc<"sessions"> | null;
+    existingSession: CachedDocument<"sessions"> | null;
     goals: Doc<"goals">[];
     funnels: Doc<"funnels">[];
     aggregateDeltas: AggregateDeltas;
+    reportRollupDeltas: ReportRollupDeltas;
     recordTelemetry: boolean;
     financialConversion: boolean;
     trustedGoalKey: string | undefined;
     now: number;
+    batchState?: BatchState;
   },
-) {
+): Promise<SessionState> {
   const { event } = input;
   const session = input.recordTelemetry
     ? await upsertSession(
@@ -379,72 +830,91 @@ async function accountEvent(
         input.context,
         input.existingSession,
         input.financialConversion,
+        input.reportRollupDeltas,
+        input.batchState,
       )
     : await accountTrustedSession(ctx, input.existingSession, event);
-
-  if (input.recordTelemetry) {
-    await ctx.db.insert("events", {
-      siteId: input.siteId,
-      eventId: event.eventId,
-      sessionId: event.sessionId,
-      visitorId: event.visitorId,
-      type: event.type,
-      ...(event.name ? { name: event.name } : {}),
-      path: event.path,
-      ...(event.referrer ? { referrer: event.referrer } : {}),
-      timestamp: event.timestamp,
-      sequence: event.sequence,
-      ...(event.target ? { target: event.target } : {}),
-      ...(event.href ? { href: event.href } : {}),
-      ...(event.properties ? { properties: event.properties } : {}),
-      ...(event.revenueCents !== undefined
-        ? { revenueCents: event.revenueCents }
-        : {}),
-      ...(event.currency ? { currency: event.currency } : {}),
-      ...(event.affiliateSlug ? { affiliateSlug: event.affiliateSlug } : {}),
-      ...(input.context.country ? { country: input.context.country } : {}),
-      ...(input.context.city ? { city: input.context.city } : {}),
-      ...(input.context.latitude !== undefined
-        ? { latitude: input.context.latitude }
-        : {}),
-      ...(input.context.longitude !== undefined
-        ? { longitude: input.context.longitude }
-        : {}),
-      browser: input.context.browser ?? "Other",
-      os: input.context.os ?? "Other",
-      device: input.context.device ?? "unknown",
-    });
+  if (input.recordTelemetry && event.type === "pageview") {
+    // `existingSession` is the snapshot from before `upsertSession`, so it
+    // still carries the route the visitor is leaving.
+    foldRouteRollups(
+      input.reportRollupDeltas,
+      input.siteId,
+      event,
+      session,
+      input.existingSession,
+    );
   }
 
   const matchedGoalKeys: string[] = [];
-  for (const goal of input.goals) {
-    if (!goal.active || !goalMatches(goal, event)) continue;
-    const prior = await ctx.db
-      .query("goalCompletions")
-      .withIndex("by_siteId_and_goalId_and_eventId", (range) =>
-        range
-          .eq("siteId", input.siteId)
-          .eq("goalId", goal._id)
-          .eq("eventId", event.eventId),
-      )
-      .unique();
-    if (prior) continue;
+  const matchingGoals = input.goals.filter(
+    (goal) => goal.active && goalMatches(goal, event),
+  );
+  if (matchingGoals.length > 1) {
+    fail("LIMIT_EXCEEDED", "active goals must use unique matchers");
+  }
+  for (const goal of matchingGoals) {
+    const completionEventId = JSON.stringify([
+      event.sessionId,
+      eventBucketStart(event.timestamp),
+      event.eventId,
+    ]);
+    const scopedPrior = (
+      await ctx.db
+        .query("goalCompletions")
+        .withIndex("by_siteId_and_goalId_and_eventId", (range) =>
+          range
+            .eq("siteId", input.siteId)
+            .eq("goalId", goal._id)
+            .eq("eventId", completionEventId),
+        )
+        .take(2)
+    ).find((completion) => completion.idempotencyVersion === 2);
+    const legacyPrior = scopedPrior
+      ? undefined
+      : (
+          await ctx.db
+            .query("goalCompletions")
+            .withIndex("by_siteId_and_goalId_and_eventId", (range) =>
+              range
+                .eq("siteId", input.siteId)
+                .eq("goalId", goal._id)
+                .eq("eventId", event.eventId),
+            )
+            .take(2)
+        ).find(
+          (completion) =>
+            completion.idempotencyVersion === undefined &&
+            completion.sessionId === event.sessionId &&
+            eventBucketStart(completion.timestamp) ===
+              eventBucketStart(event.timestamp),
+        );
+    if (scopedPrior || legacyPrior) continue;
     await ctx.db.insert("goalCompletions", {
       siteId: input.siteId,
       goalId: goal._id,
       goalKey: goal.key,
-      eventId: event.eventId,
+      eventId: completionEventId,
+      idempotencyVersion: 2,
       sessionId: event.sessionId,
       visitorId: event.visitorId,
       timestamp: event.timestamp,
       valueCents: goal.valueCents ?? 0,
     });
+    foldGoalRollup(input.reportRollupDeltas, input.siteId, event, goal);
     matchedGoalKeys.push(goal.key);
   }
 
   for (const funnel of input.funnels) {
     if (funnel.active) {
-      await advanceFunnel(ctx, input.siteId, funnel, event);
+      await advanceFunnel(
+        ctx,
+        input.siteId,
+        funnel,
+        event,
+        input.reportRollupDeltas,
+        input.batchState,
+      );
     }
   }
 
@@ -455,18 +925,23 @@ async function accountEvent(
       session,
       event,
       input.trustedGoalKey,
+      input.reportRollupDeltas,
     );
   }
 
+  const campaignDimensions: Dimension[] = session.utmCampaign
+    ? [{ type: "campaign", value: session.utmCampaign }]
+    : [];
   const dimensions: Dimension[] = [
     { type: "source", value: session.source },
+    ...campaignDimensions,
     { type: "page", value: event.path },
-    { type: "country", value: input.context.country ?? "unknown" },
+    { type: "country", value: session.country ?? "unknown" },
     { type: "device", value: input.context.device ?? "unknown" },
     { type: "browser", value: input.context.browser ?? "Other" },
     {
       type: "event",
-      value: event.type === "custom" ? event.name ?? "custom" : event.type,
+      value: event.type === "custom" ? (event.name ?? "custom") : event.type,
     },
     ...matchedGoalKeys.map((value): Dimension => ({ type: "goal", value })),
   ];
@@ -491,13 +966,15 @@ async function accountEvent(
       input.context,
       session,
       input.now,
+      input.batchState,
     );
   }
+  return session;
 }
 
 async function accountTrustedSession(
   ctx: MutationCtx,
-  existing: Doc<"sessions"> | null,
+  existing: CachedDocument<"sessions"> | null,
   event: TrackerEvent,
 ): Promise<SessionState> {
   if (!existing) {
@@ -515,6 +992,12 @@ async function accountTrustedSession(
     isNewVisitorDay: false,
     startedAt: existing.startedAt,
     source: existing.source,
+    utmCampaign: existing.utmCampaign,
+    visitorKey: existing.visitorKey ?? existing.visitorId,
+    country: existing.country,
+    city: existing.city,
+    latitude: existing.latitude,
+    longitude: existing.longitude,
     pageviewCount: existing.pageviewCount,
     eventCount: existing.eventCount,
     affiliateId: existing.affiliateId,
@@ -529,29 +1012,65 @@ async function upsertSession(
   siteId: Id<"sites">,
   event: TrackerEvent,
   context: IngestContext,
-  existing: Doc<"sessions"> | null,
+  existing: CachedDocument<"sessions"> | null,
   financialConversion: boolean,
+  reportRollupDeltas: ReportRollupDeltas,
+  batchState?: BatchState,
 ): Promise<SessionState> {
-  const source = sourceFromReferrer(event.referrer);
+  // A host-supplied id is the identity; anonymous sessions count through the
+  // daily key derived by the HTTP action, or once per session without one.
+  const visitorKey =
+    existing?.visitorKey ??
+    (event.visitorId !== event.sessionId
+      ? event.visitorId
+      : (context.visitorKey ?? event.sessionId));
   const utmSource = stringProperty(event.properties, "utm_source");
+  // Campaign traffic names its own source; the referrer host is the fallback.
+  const source = utmSource?.toLowerCase() ?? sourceFromReferrer(event.referrer);
   const utmMedium = stringProperty(event.properties, "utm_medium");
   const utmCampaign = stringProperty(event.properties, "utm_campaign");
-  const attribution = await attributionForEvent(ctx, siteId, event, existing);
+  const attribution = await attributionForEvent(
+    ctx,
+    siteId,
+    event,
+    existing,
+    batchState,
+  );
   if (existing) {
+    if (attribution && existing.affiliateId !== attribution.affiliateId) {
+      foldAffiliateVisitRollup(
+        reportRollupDeltas,
+        siteId,
+        event,
+        attribution,
+        visitorKey,
+      );
+    }
     const startedAt = Math.min(existing.startedAt, event.timestamp);
     const lastSeenAt = Math.max(existing.lastSeenAt, event.timestamp);
-    await ctx.db.patch("sessions", existing._id, {
+    const updates = {
       startedAt,
       lastSeenAt,
-      entryPath: event.timestamp < existing.startedAt ? event.path : existing.entryPath,
-      exitPath: event.timestamp >= existing.lastSeenAt ? event.path : existing.exitPath,
+      entryPath:
+        event.timestamp < existing.startedAt ? event.path : existing.entryPath,
+      exitPath:
+        event.timestamp >= existing.lastSeenAt ? event.path : existing.exitPath,
       eventCount: existing.eventCount + 1,
-      pageviewCount: existing.pageviewCount + (event.type === "pageview" ? 1 : 0),
+      pageviewCount:
+        existing.pageviewCount + (event.type === "pageview" ? 1 : 0),
+      ...(event.type === "pageview"
+        ? {
+            lastPageviewAt: Math.max(
+              existing.lastPageviewAt ?? 0,
+              event.timestamp,
+            ),
+          }
+        : {}),
       durationMs: Math.max(0, lastSeenAt - startedAt),
       conversionCount: existing.conversionCount + Number(financialConversion),
       revenueCents:
         existing.revenueCents +
-        (financialConversion ? event.revenueCents ?? 0 : 0),
+        (financialConversion ? (event.revenueCents ?? 0) : 0),
       ...(financialConversion && event.currency
         ? { currency: event.currency }
         : {}),
@@ -562,11 +1081,21 @@ async function upsertSession(
       ...(context.country ? { country: context.country } : {}),
       ...(context.city ? { city: context.city } : {}),
       ...(context.latitude !== undefined ? { latitude: context.latitude } : {}),
-      ...(context.longitude !== undefined ? { longitude: context.longitude } : {}),
+      ...(context.longitude !== undefined
+        ? { longitude: context.longitude }
+        : {}),
       browser: context.browser ?? existing.browser,
       os: context.os ?? existing.os,
       device: context.device ?? existing.device,
-    });
+    };
+    if (batchState) {
+      const state = batchState.sessions.get(event.sessionId);
+      if (!state) throw new Error("session cache was not initialized");
+      state.current = { ...existing, ...updates };
+      state.dirty = true;
+    } else {
+      await ctx.db.patch("sessions", existing._id, updates);
+    }
     return {
       isNewSession: false,
       isNewVisitorHour:
@@ -577,6 +1106,12 @@ async function upsertSession(
         Math.floor(event.timestamp / DAY_MS),
       startedAt,
       source: existing.source,
+      utmCampaign: existing.utmCampaign,
+      visitorKey,
+      country: context.country ?? existing.country,
+      city: context.city ?? existing.city,
+      latitude: context.latitude ?? existing.latitude,
+      longitude: context.longitude ?? existing.longitude,
       pageviewCount:
         existing.pageviewCount + (event.type === "pageview" ? 1 : 0),
       eventCount: existing.eventCount + 1,
@@ -584,17 +1119,31 @@ async function upsertSession(
     };
   }
 
-  const priorVisit = await ctx.db
+  const priorVisits = await ctx.db
     .query("sessions")
     .withIndex("by_siteId_and_visitorId_and_lastSeenAt", (range) =>
       range.eq("siteId", siteId).eq("visitorId", event.visitorId),
     )
     .order("desc")
     .take(1);
-  await ctx.db.insert("sessions", {
+  let priorVisit: CachedDocument<"sessions"> | null = priorVisits[0] ?? null;
+  if (batchState) {
+    for (const state of batchState.sessions.values()) {
+      const candidate = state.current;
+      if (
+        candidate?.visitorId === event.visitorId &&
+        candidate.sessionId !== event.sessionId &&
+        (!priorVisit || candidate.lastSeenAt > priorVisit.lastSeenAt)
+      ) {
+        priorVisit = candidate;
+      }
+    }
+  }
+  const fields: Omit<Doc<"sessions">, "_id" | "_creationTime"> = {
     siteId,
     sessionId: event.sessionId,
     visitorId: event.visitorId,
+    visitorKey,
     startedAt: event.timestamp,
     lastSeenAt: event.timestamp,
     entryPath: event.path,
@@ -607,46 +1156,71 @@ async function upsertSession(
     ...(context.country ? { country: context.country } : {}),
     ...(context.city ? { city: context.city } : {}),
     ...(context.latitude !== undefined ? { latitude: context.latitude } : {}),
-    ...(context.longitude !== undefined ? { longitude: context.longitude } : {}),
+    ...(context.longitude !== undefined
+      ? { longitude: context.longitude }
+      : {}),
     browser: context.browser ?? "Other",
     os: context.os ?? "Other",
     device: context.device ?? "unknown",
     eventCount: 1,
     pageviewCount: event.type === "pageview" ? 1 : 0,
+    ...(event.type === "pageview" ? { lastPageviewAt: event.timestamp } : {}),
     durationMs: 0,
     conversionCount: Number(financialConversion),
-    revenueCents: financialConversion ? event.revenueCents ?? 0 : 0,
+    revenueCents: financialConversion ? (event.revenueCents ?? 0) : 0,
     ...(financialConversion && event.currency
       ? { currency: event.currency }
       : {}),
     ...(attribution ?? {}),
-  });
+  };
+  const sessionId = await ctx.db.insert("sessions", fields);
+  if (attribution) {
+    foldAffiliateVisitRollup(
+      reportRollupDeltas,
+      siteId,
+      event,
+      attribution,
+      visitorKey,
+    );
+  }
+  if (batchState) {
+    const state = batchState.sessions.get(event.sessionId);
+    if (!state) throw new Error("session cache was not initialized");
+    state.current = {
+      ...fields,
+      _id: sessionId,
+    };
+  }
   return {
     isNewSession: true,
     isNewVisitorHour:
-      priorVisit.length === 0 ||
-      Math.floor(priorVisit[0].lastSeenAt / HOUR_MS) <
+      priorVisit === null ||
+      Math.floor(priorVisit.lastSeenAt / HOUR_MS) <
         Math.floor(event.timestamp / HOUR_MS),
     isNewVisitorDay:
-      priorVisit.length === 0 ||
-      Math.floor(priorVisit[0].lastSeenAt / DAY_MS) <
+      priorVisit === null ||
+      Math.floor(priorVisit.lastSeenAt / DAY_MS) <
         Math.floor(event.timestamp / DAY_MS),
     startedAt: event.timestamp,
     source,
+    utmCampaign,
+    visitorKey,
+    country: context.country,
+    city: context.city,
+    latitude: context.latitude,
+    longitude: context.longitude,
     pageviewCount: event.type === "pageview" ? 1 : 0,
     eventCount: 1,
     ...(attribution ?? {}),
   };
 }
 
-function attributionSnapshot(
-  value: {
-    affiliateId?: Id<"affiliates">;
-    affiliateSlug?: string;
-    affiliateAttributedAt?: number;
-    affiliateExpiresAt?: number;
-  },
-): AttributionSnapshot | undefined {
+function attributionSnapshot(value: {
+  affiliateId?: Id<"affiliates">;
+  affiliateSlug?: string;
+  affiliateAttributedAt?: number;
+  affiliateExpiresAt?: number;
+}): AttributionSnapshot | undefined {
   if (
     !value.affiliateId ||
     !value.affiliateSlug ||
@@ -663,7 +1237,6 @@ function attributionSnapshot(
   };
 }
 
-
 function attributionApplies(
   attribution: AttributionSnapshot,
   timestamp: number,
@@ -680,7 +1253,8 @@ async function attributionForEvent(
   ctx: MutationCtx,
   siteId: Id<"sites">,
   event: TrackerEvent,
-  existingSession: Doc<"sessions"> | null,
+  existingSession: CachedDocument<"sessions"> | null,
+  batchState?: BatchState,
 ): Promise<AttributionSnapshot | undefined> {
   const sessionAttribution = existingSession
     ? attributionSnapshot(existingSession)
@@ -696,12 +1270,25 @@ async function attributionForEvent(
     return sessionAttribution;
   }
 
-  const current = await ctx.db
-    .query("visitorAttributions")
-    .withIndex("by_siteId_and_visitorId", (range) =>
-      range.eq("siteId", siteId).eq("visitorId", event.visitorId),
-    )
-    .unique();
+  let attributionState = batchState?.visitorAttributions.get(event.visitorId);
+  if (batchState && !attributionState) {
+    const current = await ctx.db
+      .query("visitorAttributions")
+      .withIndex("by_siteId_and_visitorId", (range) =>
+        range.eq("siteId", siteId).eq("visitorId", event.visitorId),
+      )
+      .unique();
+    attributionState = { current, dirty: false };
+    batchState.visitorAttributions.set(event.visitorId, attributionState);
+  }
+  const current = attributionState
+    ? attributionState.current
+    : await ctx.db
+        .query("visitorAttributions")
+        .withIndex("by_siteId_and_visitorId", (range) =>
+          range.eq("siteId", siteId).eq("visitorId", event.visitorId),
+        )
+        .unique();
   if (current) {
     const snapshot: AttributionSnapshot = {
       affiliateId: current.affiliateId,
@@ -733,22 +1320,30 @@ async function attributionForEvent(
 
   const attributedAt = event.timestamp;
   const expiresAt = attributedAt + affiliate.attributionWindowDays * DAY_MS;
+  const fields = {
+    siteId,
+    visitorId: event.visitorId,
+    affiliateId: affiliate._id,
+    affiliateSlug: affiliate.slug,
+    attributedAt,
+    expiresAt,
+  };
   if (current) {
-    await ctx.db.patch("visitorAttributions", current._id, {
-      affiliateId: affiliate._id,
-      affiliateSlug: affiliate.slug,
-      attributedAt,
-      expiresAt,
-    });
+    if (attributionState) {
+      attributionState.current = { ...current, ...fields };
+      attributionState.dirty = true;
+    } else {
+      await ctx.db.patch("visitorAttributions", current._id, fields);
+    }
   } else {
-    await ctx.db.insert("visitorAttributions", {
-      siteId,
-      visitorId: event.visitorId,
-      affiliateId: affiliate._id,
-      affiliateSlug: affiliate.slug,
-      attributedAt,
-      expiresAt,
-    });
+    const attributionId = await ctx.db.insert("visitorAttributions", fields);
+    if (attributionState) {
+      attributionState.current = {
+        ...fields,
+        _id: attributionId,
+        _creationTime: event.timestamp,
+      };
+    }
   }
   return {
     affiliateId: affiliate._id,
@@ -761,28 +1356,38 @@ async function attributionForEvent(
 async function accountHeartbeat(
   ctx: MutationCtx,
   siteId: Id<"sites">,
-  session: Doc<"sessions">,
+  session: CachedDocument<"sessions">,
   event: TrackerEvent,
   context: IngestContext,
   now: number,
+  batchState?: BatchState,
 ) {
   const startedAt = Math.min(session.startedAt, event.timestamp);
   const lastSeenAt = Math.max(session.lastSeenAt, event.timestamp);
-  await ctx.db.patch("sessions", session._id, {
+  const updates = {
     startedAt,
     lastSeenAt,
-    exitPath: event.timestamp >= session.lastSeenAt
-      ? event.path
-      : session.exitPath,
+    exitPath:
+      event.timestamp >= session.lastSeenAt ? event.path : session.exitPath,
     durationMs: Math.max(0, lastSeenAt - startedAt),
     ...(context.country ? { country: context.country } : {}),
     ...(context.city ? { city: context.city } : {}),
     ...(context.latitude !== undefined ? { latitude: context.latitude } : {}),
-    ...(context.longitude !== undefined ? { longitude: context.longitude } : {}),
+    ...(context.longitude !== undefined
+      ? { longitude: context.longitude }
+      : {}),
     browser: context.browser ?? session.browser,
     os: context.os ?? session.os,
     device: context.device ?? session.device,
-  });
+  };
+  if (batchState) {
+    const state = batchState.sessions.get(event.sessionId);
+    if (!state) throw new Error("session cache was not initialized");
+    state.current = { ...session, ...updates };
+    state.dirty = true;
+  } else {
+    await ctx.db.patch("sessions", session._id, updates);
+  }
   await updateLiveSession(
     ctx,
     siteId,
@@ -794,11 +1399,17 @@ async function accountHeartbeat(
       isNewVisitorDay: false,
       startedAt,
       source: session.source,
+      visitorKey: session.visitorKey ?? session.visitorId,
+      country: context.country ?? session.country,
+      city: context.city ?? session.city,
+      latitude: context.latitude ?? session.latitude,
+      longitude: context.longitude ?? session.longitude,
       pageviewCount: session.pageviewCount,
       eventCount: session.eventCount,
       affiliateSlug: session.affiliateSlug,
     },
     now,
+    batchState,
   );
 }
 
@@ -806,17 +1417,18 @@ async function consumeSiteRateLimit(
   ctx: MutationCtx,
   siteId: Id<"sites">,
   events: TrackerEvent[],
+  eventByteCounts: number[],
   now: number,
-): Promise<number[]> {
+): Promise<void> {
   const windowStart =
     Math.floor(now / RATE_LIMIT_WINDOW_MS) * RATE_LIMIT_WINDOW_MS;
-  const increments = new Map<number, { eventCount: number; byteCount: number }>();
-  const eventByteCounts: number[] = [];
-  const encoder = new TextEncoder();
+  const increments = new Map<
+    number,
+    { eventCount: number; byteCount: number }
+  >();
   for (let index = 0; index < events.length; index += 1) {
     const event = events[index];
-    const byteCount = encoder.encode(JSON.stringify(event)).byteLength;
-    eventByteCounts.push(byteCount);
+    const byteCount = eventByteCounts[index];
     const shard = stableHash(event.sessionId) % SITE_INGEST_SHARDS;
     const current = increments.get(shard) ?? { eventCount: 0, byteCount: 0 };
     current.eventCount += 1;
@@ -830,15 +1442,16 @@ async function consumeSiteRateLimit(
     window: Doc<"siteIngestWindows"> | null;
   }> = [];
   for (const [shard, increment] of increments) {
-    const window = await ctx.db
+    const rows = await ctx.db
       .query("siteIngestWindows")
       .withIndex("by_siteId_and_shard_and_windowStart", (range) =>
-        range
-          .eq("siteId", siteId)
-          .eq("shard", shard)
-          .eq("windowStart", windowStart),
+        range.eq("siteId", siteId).eq("shard", shard),
       )
-      .unique();
+      .order("desc")
+      .take(1);
+    const storedWindow = rows[0] ?? null;
+    const window =
+      storedWindow?.windowStart === windowStart ? storedWindow : null;
     if (
       (window?.eventCount ?? 0) + increment.eventCount >
         MAX_SITE_EVENTS_PER_SHARD_WINDOW ||
@@ -850,20 +1463,25 @@ async function consumeSiteRateLimit(
         windowMs: RATE_LIMIT_WINDOW_MS,
         eventLimitPerShard: MAX_SITE_EVENTS_PER_SHARD_WINDOW,
         byteLimitPerShard: MAX_SITE_BYTES_PER_SHARD_WINDOW,
-        siteEventLimit:
-          SITE_INGEST_SHARDS * MAX_SITE_EVENTS_PER_SHARD_WINDOW,
-        siteByteLimit:
-          SITE_INGEST_SHARDS * MAX_SITE_BYTES_PER_SHARD_WINDOW,
+        siteEventLimit: SITE_INGEST_SHARDS * MAX_SITE_EVENTS_PER_SHARD_WINDOW,
+        siteByteLimit: SITE_INGEST_SHARDS * MAX_SITE_BYTES_PER_SHARD_WINDOW,
       });
     }
-    windows.push({ shard, increment, window });
+    windows.push({ shard, increment, window: storedWindow });
   }
 
   for (const { shard, increment, window } of windows) {
     if (window) {
       await ctx.db.patch("siteIngestWindows", window._id, {
-        eventCount: window.eventCount + increment.eventCount,
-        byteCount: window.byteCount + increment.byteCount,
+        windowStart,
+        eventCount:
+          window.windowStart === windowStart
+            ? window.eventCount + increment.eventCount
+            : increment.eventCount,
+        byteCount:
+          window.windowStart === windowStart
+            ? window.byteCount + increment.byteCount
+            : increment.byteCount,
         updatedAt: now,
       });
     } else {
@@ -877,7 +1495,6 @@ async function consumeSiteRateLimit(
       });
     }
   }
-  return eventByteCounts;
 }
 
 async function consumeRateLimit(
@@ -886,43 +1503,71 @@ async function consumeRateLimit(
   sessionId: string,
   byteCount: number,
   now: number,
+  states: Map<string, SessionRateWindowState>,
 ): Promise<boolean> {
-  const windowStart = Math.floor(now / RATE_LIMIT_WINDOW_MS) * RATE_LIMIT_WINDOW_MS;
-  const window = await ctx.db
-    .query("ingestWindows")
-    .withIndex("by_siteId_and_sessionId_and_windowStart", (range) =>
-      range
-        .eq("siteId", siteId)
-        .eq("sessionId", sessionId)
-        .eq("windowStart", windowStart),
-    )
-    .unique();
+  const windowStart =
+    Math.floor(now / RATE_LIMIT_WINDOW_MS) * RATE_LIMIT_WINDOW_MS;
+  let state = states.get(sessionId);
+  if (!state) {
+    const rows = await ctx.db
+      .query("ingestWindows")
+      .withIndex("by_siteId_and_sessionId_and_windowStart", (range) =>
+        range.eq("siteId", siteId).eq("sessionId", sessionId),
+      )
+      .order("desc")
+      .take(1);
+    const window = rows[0] ?? null;
+    state = {
+      siteId,
+      sessionId,
+      window,
+      eventCount: window?.windowStart === windowStart ? window.eventCount : 0,
+      byteCount: window?.windowStart === windowStart ? window.byteCount : 0,
+      dirty: false,
+    };
+    states.set(sessionId, state);
+  }
   if (
-    window &&
-    (window.eventCount >= MAX_EVENTS_PER_SESSION_WINDOW ||
-      window.byteCount + byteCount > MAX_BYTES_PER_SESSION_WINDOW)
+    state.eventCount >= MAX_EVENTS_PER_SESSION_WINDOW ||
+    state.byteCount + byteCount > MAX_BYTES_PER_SESSION_WINDOW
   ) {
     return false;
   }
-  if (window) {
-    await ctx.db.patch("ingestWindows", window._id, {
-      eventCount: window.eventCount + 1,
-      byteCount: window.byteCount + byteCount,
-      updatedAt: now,
-    });
-  } else {
-    await ctx.db.insert("ingestWindows", {
-      siteId,
-      sessionId,
-      windowStart,
-      eventCount: 1,
-      byteCount,
-      updatedAt: now,
-    });
-  }
+  state.eventCount += 1;
+  state.byteCount += byteCount;
+  state.dirty = true;
   return true;
 }
 
+async function flushSessionRateLimits(
+  ctx: MutationCtx,
+  states: Map<string, SessionRateWindowState>,
+  now: number,
+) {
+  const windowStart =
+    Math.floor(now / RATE_LIMIT_WINDOW_MS) * RATE_LIMIT_WINDOW_MS;
+  for (const state of states.values()) {
+    if (!state.dirty) continue;
+    if (state.window) {
+      await ctx.db.patch("ingestWindows", state.window._id, {
+        windowStart,
+        eventCount: state.eventCount,
+        byteCount: state.byteCount,
+        updatedAt: now,
+      });
+    } else {
+      await ctx.db.insert("ingestWindows", {
+        siteId: state.siteId,
+        sessionId: state.sessionId,
+        windowStart,
+        eventCount: state.eventCount,
+        byteCount: state.byteCount,
+        updatedAt: now,
+      });
+    }
+  }
+}
+
 async function updateLiveSession(
   ctx: MutationCtx,
   siteId: Id<"sites">,
@@ -930,45 +1575,73 @@ async function updateLiveSession(
   context: IngestContext,
   session: SessionState,
   now: number,
+  batchState?: BatchState,
 ) {
-  const existing = await ctx.db
-    .query("liveSessions")
-    .withIndex("by_siteId_and_sessionId", (range) =>
-      range.eq("siteId", siteId).eq("sessionId", event.sessionId),
-    )
-    .unique();
-  const renewsExpiry =
-    !existing ||
-    existing.expiresAt <= now ||
-    event.type === "pageview" ||
-    event.type === "heartbeat";
-  const deadline = renewsExpiry
-    ? now + LIVE_SESSION_TTL_MS
-    : existing.expiresAt;
+  await ensureLiveSweep(ctx, siteId, now, batchState);
+  const liveState = batchState
+    ? await loadLiveSession(ctx, siteId, event.sessionId, batchState)
+    : undefined;
+  const existing = liveState
+    ? liveState.current
+    : await ctx.db
+        .query("liveSessions")
+        .withIndex("by_siteId_and_sessionId", (range) =>
+          range.eq("siteId", siteId).eq("sessionId", event.sessionId),
+        )
+        .unique();
+  const deadline = now + LIVE_SESSION_TTL_MS;
+  const country = context.country ?? session.country;
+  const city = context.city ?? session.city;
+  const latitude = context.latitude ?? session.latitude;
+  const longitude = context.longitude ?? session.longitude;
   if (existing) {
-    await ctx.db.patch("liveSessions", existing._id, {
+    const recentHeartbeatEventIds =
+      event.type === "heartbeat"
+        ? [
+            ...(existing.recentHeartbeatEventIds ??
+              (existing.lastHeartbeatEventId
+                ? [existing.lastHeartbeatEventId]
+                : [])),
+            event.eventId,
+          ].slice(-8)
+        : undefined;
+    const updates = {
       visitorId: event.visitorId,
       path: event.path,
       lastSeenAt: Math.max(existing.lastSeenAt, event.timestamp),
       expiresAt: deadline,
-      ...(context.country ? { country: context.country } : {}),
-      ...(context.city ? { city: context.city } : {}),
-      ...(context.latitude !== undefined ? { latitude: context.latitude } : {}),
-      ...(context.longitude !== undefined ? { longitude: context.longitude } : {}),
+      ...(country ? { country } : {}),
+      ...(city ? { city } : {}),
+      ...(latitude !== undefined ? { latitude } : {}),
+      ...(longitude !== undefined ? { longitude } : {}),
       browser: context.browser ?? existing.browser,
       os: context.os ?? existing.os,
       source: session.source,
       device: context.device ?? existing.device,
       pageviewCount: session.pageviewCount,
       eventCount: session.eventCount,
+      // The live map animates a visitor from the route they left to the one
+      // they reached; a heartbeat or click on the same page is not a move.
+      ...(event.type === "pageview" && event.path !== existing.path
+        ? { previousPath: existing.path, movedAt: event.timestamp }
+        : {}),
       ...(event.type === "heartbeat"
-        ? { lastHeartbeatEventId: event.eventId }
+        ? {
+            lastHeartbeatEventId: event.eventId,
+            recentHeartbeatEventIds,
+          }
         : {}),
-    });
+    };
+    if (liveState) {
+      liveState.current = { ...existing, ...updates };
+      liveState.dirty = true;
+    } else {
+      await ctx.db.patch("liveSessions", existing._id, updates);
+    }
     return;
   }
 
-  const liveSessionId = await ctx.db.insert("liveSessions", {
+  const fields: Omit<Doc<"liveSessions">, "_id" | "_creationTime"> = {
     siteId,
     sessionId: event.sessionId,
     visitorId: event.visitorId,
@@ -976,11 +1649,10 @@ async function updateLiveSession(
     startedAt: event.timestamp,
     lastSeenAt: event.timestamp,
     expiresAt: deadline,
-    expiryChainDeadline: deadline,
-    ...(context.country ? { country: context.country } : {}),
-    ...(context.city ? { city: context.city } : {}),
-    ...(context.latitude !== undefined ? { latitude: context.latitude } : {}),
-    ...(context.longitude !== undefined ? { longitude: context.longitude } : {}),
+    ...(country ? { country } : {}),
+    ...(city ? { city } : {}),
+    ...(latitude !== undefined ? { latitude } : {}),
+    ...(longitude !== undefined ? { longitude } : {}),
     browser: context.browser ?? "Other",
     os: context.os ?? "Other",
     source: session.source,
@@ -988,13 +1660,46 @@ async function updateLiveSession(
     pageviewCount: session.pageviewCount,
     eventCount: session.eventCount,
     ...(event.type === "heartbeat"
-      ? { lastHeartbeatEventId: event.eventId }
+      ? {
+          lastHeartbeatEventId: event.eventId,
+          recentHeartbeatEventIds: [event.eventId],
+        }
       : {}),
-  });
-  await ctx.scheduler.runAt(deadline, internal.live.expire, {
-    liveSessionId,
-    scheduledDeadline: deadline,
-  });
+  };
+  const liveSessionId = await ctx.db.insert("liveSessions", fields);
+  if (liveState) {
+    liveState.current = {
+      ...fields,
+      _id: liveSessionId,
+    };
+  }
+}
+
+async function ensureLiveSweep(
+  ctx: MutationCtx,
+  siteId: Id<"sites">,
+  now: number,
+  batchState?: BatchState,
+) {
+  const key = String(siteId);
+  if (batchState?.liveSweepSites.has(key)) return;
+  const existing = await ctx.db
+    .query("liveSweepStates")
+    .withIndex("by_siteId", (range) => range.eq("siteId", siteId))
+    .unique();
+  if (!existing || existing.scheduledAt <= now) {
+    const scheduledAt = now + LIVE_SWEEP_INTERVAL_MS;
+    if (existing) {
+      await ctx.db.patch("liveSweepStates", existing._id, { scheduledAt });
+    } else {
+      await ctx.db.insert("liveSweepStates", { siteId, scheduledAt });
+    }
+    await ctx.scheduler.runAt(scheduledAt, internal.live.sweep, {
+      siteId,
+      scheduledAt,
+    });
+  }
+  batchState?.liveSweepSites.add(key);
 }
 
 function goalMatches(goal: Doc<"goals">, event: TrackerEvent): boolean {
@@ -1009,20 +1714,39 @@ async function advanceFunnel(
   siteId: Id<"sites">,
   funnel: Doc<"funnels">,
   event: TrackerEvent,
+  reportRollupDeltas: ReportRollupDeltas,
+  batchState?: BatchState,
 ) {
-  const progress = await ctx.db
-    .query("funnelProgress")
-    .withIndex("by_siteId_and_funnelId_and_visitorId", (range) =>
-      range
-        .eq("siteId", siteId)
-        .eq("funnelId", funnel._id)
-        .eq("visitorId", event.visitorId),
-    )
-    .unique();
+  const cacheKey = JSON.stringify([funnel._id, event.visitorId]);
+  let progressState = batchState?.funnelProgress.get(cacheKey);
+  if (batchState && !progressState) {
+    const current = await ctx.db
+      .query("funnelProgress")
+      .withIndex("by_siteId_and_funnelId_and_visitorId", (range) =>
+        range
+          .eq("siteId", siteId)
+          .eq("funnelId", funnel._id)
+          .eq("visitorId", event.visitorId),
+      )
+      .unique();
+    progressState = { current, dirty: false };
+    batchState.funnelProgress.set(cacheKey, progressState);
+  }
+  const progress = progressState
+    ? progressState.current
+    : await ctx.db
+        .query("funnelProgress")
+        .withIndex("by_siteId_and_funnelId_and_visitorId", (range) =>
+          range
+            .eq("siteId", siteId)
+            .eq("funnelId", funnel._id)
+            .eq("visitorId", event.visitorId),
+        )
+        .unique();
   const matchesFirst = stepMatches(funnel.steps[0], event);
   if (!progress) {
     if (!matchesFirst) return;
-    await ctx.db.insert("funnelProgress", {
+    const fields: Omit<Doc<"funnelProgress">, "_id" | "_creationTime"> = {
       siteId,
       funnelId: funnel._id,
       visitorId: event.visitorId,
@@ -1030,33 +1754,65 @@ async function advanceFunnel(
       currentStep: 1,
       startedAt: event.timestamp,
       lastStepAt: event.timestamp,
-    });
+    };
+    const progressId = await ctx.db.insert("funnelProgress", fields);
+    if (progressState) {
+      progressState.current = {
+        ...fields,
+        _id: progressId,
+      };
+    }
     await incrementFunnelStep(ctx, siteId, funnel._id, event, 0, false);
+    foldFunnelRollup(reportRollupDeltas, siteId, funnel._id, event, 0, false);
     return;
   }
-  if (progress.completedAt !== undefined || event.timestamp < progress.lastStepAt) return;
+  if (
+    progress.completedAt !== undefined ||
+    event.timestamp < progress.lastStepAt
+  )
+    return;
   if (event.timestamp - progress.startedAt > funnel.conversionWindowMs) {
     if (!matchesFirst) return;
-    await ctx.db.patch("funnelProgress", progress._id, {
+    const updated = {
+      ...progress,
       sessionId: event.sessionId,
       currentStep: 1,
       startedAt: event.timestamp,
       lastStepAt: event.timestamp,
-      completedAt: undefined,
-    });
+    };
+    delete updated.completedAt;
+    if (progressState) {
+      progressState.current = updated;
+      progressState.dirty = true;
+    } else {
+      await ctx.db.patch("funnelProgress", progress._id, {
+        sessionId: event.sessionId,
+        currentStep: 1,
+        startedAt: event.timestamp,
+        lastStepAt: event.timestamp,
+        completedAt: undefined,
+      });
+    }
     await incrementFunnelStep(ctx, siteId, funnel._id, event, 0, false);
+    foldFunnelRollup(reportRollupDeltas, siteId, funnel._id, event, 0, false);
     return;
   }
   const nextStep = funnel.steps[progress.currentStep];
   if (!nextStep || !stepMatches(nextStep, event)) return;
   const currentStep = progress.currentStep + 1;
   const completed = currentStep === funnel.steps.length;
-  await ctx.db.patch("funnelProgress", progress._id, {
+  const updates = {
     sessionId: event.sessionId,
     currentStep,
     lastStepAt: event.timestamp,
     ...(completed ? { completedAt: event.timestamp } : {}),
-  });
+  };
+  if (progressState) {
+    progressState.current = { ...progress, ...updates };
+    progressState.dirty = true;
+  } else {
+    await ctx.db.patch("funnelProgress", progress._id, updates);
+  }
   await incrementFunnelStep(
     ctx,
     siteId,
@@ -1065,6 +1821,14 @@ async function advanceFunnel(
     progress.currentStep,
     completed,
   );
+  foldFunnelRollup(
+    reportRollupDeltas,
+    siteId,
+    funnel._id,
+    event,
+    progress.currentStep,
+    completed,
+  );
 }
 
 function stepMatches(
@@ -1085,7 +1849,7 @@ async function incrementFunnelStep(
   completed: boolean,
 ) {
   const bucketStart = Math.floor(event.timestamp / DAY_MS) * DAY_MS;
-  const shard = stableHash(event.visitorId) % AGGREGATE_SHARDS;
+  const shard = stableHash(event.visitorId) % SECONDARY_AGGREGATE_SHARDS;
   const bucket = await ctx.db
     .query("funnelStepBuckets")
     .withIndex(
@@ -1123,6 +1887,7 @@ async function accountConversion(
   session: SessionState,
   event: TrackerEvent,
   trustedGoalKey: string | undefined,
+  reportRollupDeltas: ReportRollupDeltas,
 ) {
   const revenueCents = event.revenueCents ?? 0;
   const currency = event.currency ?? "USD";
@@ -1154,8 +1919,17 @@ async function accountConversion(
     commissionCents,
   });
   if (commissionAffiliate && attribution) {
+    foldAffiliateConversionRollup(
+      reportRollupDeltas,
+      siteId,
+      event,
+      commissionAffiliate,
+      attribution.affiliateSlug,
+      revenueCents,
+      commissionCents,
+    );
     const bucketStart = Math.floor(event.timestamp / DAY_MS) * DAY_MS;
-    const shard = stableHash(event.sessionId) % AGGREGATE_SHARDS;
+    const shard = stableHash(event.sessionId) % SECONDARY_AGGREGATE_SHARDS;
     const stats = await ctx.db
       .query("affiliateDailyStats")
       .withIndex(
@@ -1189,6 +1963,558 @@ async function accountConversion(
   }
 }
 
+function createReportRollupDeltas(): ReportRollupDeltas {
+  return {
+    goals: new Map(),
+    funnels: new Map(),
+    affiliates: new Map(),
+    vitals: new Map(),
+    routes: new Map(),
+  };
+}
+
+function reportRollupCoordinates(
+  siteId: Id<"sites">,
+  definitionId: string,
+  timestamp: number,
+  visitorId: string,
+) {
+  const bucketStart = Math.floor(timestamp / DAY_MS) * DAY_MS;
+  const shard = stableHash(visitorId) % REPORT_ROLLUP_SHARDS;
+  return {
+    bucketStart,
+    shard,
+    key: `${siteId}:${definitionId}:${bucketStart}:${shard}`,
+  };
+}
+
+function foldGoalRollup(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  event: TrackerEvent,
+  goal: Doc<"goals">,
+) {
+  const coordinates = reportRollupCoordinates(
+    siteId,
+    String(goal._id),
+    event.timestamp,
+    event.visitorId,
+  );
+  const current = deltas.goals.get(coordinates.key) ?? {
+    siteId,
+    goalId: goal._id,
+    bucketStart: coordinates.bucketStart,
+    shard: coordinates.shard,
+    goalKey: goal.key,
+    completions: 0,
+    valueCents: 0,
+  };
+  current.completions += 1;
+  current.valueCents += goal.valueCents ?? 0;
+  deltas.goals.set(coordinates.key, current);
+}
+
+function foldFunnelRollup(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  funnelId: Id<"funnels">,
+  event: TrackerEvent,
+  step: number,
+  completed: boolean,
+) {
+  const coordinates = reportRollupCoordinates(
+    siteId,
+    String(funnelId),
+    event.timestamp,
+    event.visitorId,
+  );
+  const rollup = deltas.funnels.get(coordinates.key) ?? {
+    siteId,
+    funnelId,
+    bucketStart: coordinates.bucketStart,
+    shard: coordinates.shard,
+    steps: new Map(),
+  };
+  const current = rollup.steps.get(step) ?? {
+    step,
+    entrants: 0,
+    completions: 0,
+  };
+  current.entrants += 1;
+  current.completions += Number(completed);
+  rollup.steps.set(step, current);
+  deltas.funnels.set(coordinates.key, rollup);
+}
+
+function foldAffiliateVisitRollup(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  event: TrackerEvent,
+  attribution: AttributionSnapshot,
+  visitorKey: string,
+) {
+  const coordinates = reportRollupCoordinates(
+    siteId,
+    String(attribution.affiliateId),
+    event.timestamp,
+    event.visitorId,
+  );
+  const current = deltas.affiliates.get(coordinates.key) ?? {
+    siteId,
+    affiliateId: attribution.affiliateId,
+    bucketStart: coordinates.bucketStart,
+    shard: coordinates.shard,
+    affiliateSlug: attribution.affiliateSlug,
+    visits: 0,
+    visitorSketch: createVisitorSketch(),
+    conversions: 0,
+    revenueCents: 0,
+    commissionCents: 0,
+  };
+  current.visits += 1;
+  current.visitorSketch = addVisitor(current.visitorSketch, visitorKey);
+  deltas.affiliates.set(coordinates.key, current);
+}
+
+function foldAffiliateConversionRollup(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  event: TrackerEvent,
+  affiliate: Doc<"affiliates">,
+  affiliateSlug: string,
+  revenueCents: number,
+  commissionCents: number,
+) {
+  const coordinates = reportRollupCoordinates(
+    siteId,
+    String(affiliate._id),
+    event.timestamp,
+    event.visitorId,
+  );
+  const current = deltas.affiliates.get(coordinates.key) ?? {
+    siteId,
+    affiliateId: affiliate._id,
+    bucketStart: coordinates.bucketStart,
+    shard: coordinates.shard,
+    affiliateSlug,
+    visits: 0,
+    visitorSketch: createVisitorSketch(),
+    conversions: 0,
+    revenueCents: 0,
+    commissionCents: 0,
+  };
+  current.conversions += 1;
+  current.revenueCents += revenueCents;
+  current.commissionCents += commissionCents;
+  deltas.affiliates.set(coordinates.key, current);
+}
+
+function foldVitalRollups(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  event: TrackerEvent,
+  device: string,
+) {
+  const metric = event.name as VitalMetric;
+  const value = event.value ?? 0;
+  const bucketStart = Math.floor(event.timestamp / DAY_MS) * DAY_MS;
+  const shard = stableHash(event.sessionId) % REPORT_ROLLUP_SHARDS;
+  // Each measurement lands twice: on its page (across devices) and on its
+  // device class (across pages), so both breakdowns stay exact and bounded.
+  for (const target of [
+    { device: VITAL_ALL, page: event.path },
+    { device, page: VITAL_ALL },
+  ]) {
+    const key = JSON.stringify([
+      siteId,
+      bucketStart,
+      target.device,
+      target.page,
+      shard,
+    ]);
+    const row = deltas.vitals.get(key) ?? {
+      siteId,
+      bucketStart,
+      device: target.device,
+      page: target.page,
+      shard,
+      metrics: new Map<
+        VitalMetric,
+        { count: number; sum: number; histogram: number[] }
+      >(),
+    };
+    const entry = row.metrics.get(metric) ?? {
+      count: 0,
+      sum: 0,
+      histogram: emptyVitalHistogram(),
+    };
+    entry.count += 1;
+    entry.sum += value;
+    entry.histogram = addToVitalHistogram(entry.histogram, value);
+    row.metrics.set(metric, entry);
+    deltas.vitals.set(key, row);
+  }
+}
+
+function foldRouteDelta(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  route: string,
+  timestamp: number,
+  visitorId: string,
+  apply: (delta: RouteRollupDelta) => void,
+) {
+  const coordinates = reportRollupCoordinates(
+    siteId,
+    `route:${route}`,
+    timestamp,
+    visitorId,
+  );
+  const current = deltas.routes.get(coordinates.key) ?? {
+    siteId,
+    bucketStart: coordinates.bucketStart,
+    shard: coordinates.shard,
+    route,
+    pageviews: 0,
+    entries: 0,
+    continued: 0,
+    dwellMs: 0,
+    dwellSamples: 0,
+    visitorSketch: createVisitorSketch(),
+    outgoing: new Map<string, number>(),
+  };
+  apply(current);
+  deltas.routes.set(coordinates.key, current);
+}
+
+/**
+ * One pageview feeds the site map three ways: the route it lands on gains a
+ * view (and an entry when it opens the session), the route it leaves gains
+ * the transition and the time spent there, and the session's entry route is
+ * marked as continued the first time its visitor moves on — which is what
+ * lets the report derive bounces and exits without a session-end hook.
+ */
+function foldRouteRollups(
+  deltas: ReportRollupDeltas,
+  siteId: Id<"sites">,
+  event: TrackerEvent,
+  session: SessionState,
+  existing: CachedDocument<"sessions"> | null,
+) {
+  const route = normalizeRoute(event.path);
+  const priorPageviews = existing?.pageviewCount ?? 0;
+  const firstPageview = existing === null || priorPageviews === 0;
+  foldRouteDelta(
+    deltas,
+    siteId,
+    route,
+    event.timestamp,
+    event.visitorId,
+    (delta) => {
+      delta.pageviews += 1;
+      if (firstPageview) delta.entries += 1;
+      delta.visitorSketch = addVisitor(delta.visitorSketch, session.visitorKey);
+    },
+  );
+  // Out-of-order telemetry keeps its view but cannot say where it came from.
+  if (existing === null || firstPageview) return;
+  if (event.timestamp < existing.lastSeenAt) return;
+  const previous = normalizeRoute(existing.exitPath);
+  const lastPageviewAt = existing.lastPageviewAt;
+  foldRouteDelta(
+    deltas,
+    siteId,
+    previous,
+    event.timestamp,
+    event.visitorId,
+    (delta) => {
+      delta.outgoing.set(route, (delta.outgoing.get(route) ?? 0) + 1);
+      if (lastPageviewAt !== undefined && event.timestamp >= lastPageviewAt) {
+        delta.dwellMs += event.timestamp - lastPageviewAt;
+        delta.dwellSamples += 1;
+      }
+    },
+  );
+  if (priorPageviews === 1) {
+    foldRouteDelta(
+      deltas,
+      siteId,
+      normalizeRoute(existing.entryPath),
+      existing.startedAt,
+      event.visitorId,
+      (delta) => {
+        delta.continued += 1;
+      },
+    );
+  }
+}
+
+async function upsertRouteRollup(
+  ctx: MutationCtx,
+  delta: RouteRollupDelta,
+  dayRowCounts: Map<string, number>,
+) {
+  const existing = await ctx.db
+    .query("routeDailyRollups")
+    .withIndex("by_siteId_and_route_and_bucketStart_and_shard", (range) =>
+      range
+        .eq("siteId", delta.siteId)
+        .eq("route", delta.route)
+        .eq("bucketStart", delta.bucketStart)
+        .eq("shard", delta.shard),
+    )
+    .unique();
+  if (!existing && delta.route !== SITE_MAP_OTHER_ROUTE) {
+    // A new route row spends one bounded read per batch to respect the daily
+    // cap; past it, the day's remaining routes fold into "(other)".
+    const capRows = MAX_SITE_MAP_ROUTES_PER_DAY * REPORT_ROLLUP_SHARDS;
+    const dayKey = `${delta.siteId}:${delta.bucketStart}`;
+    let dayRows = dayRowCounts.get(dayKey);
+    if (dayRows === undefined) {
+      dayRows = (
+        await ctx.db
+          .query("routeDailyRollups")
+          .withIndex("by_siteId_and_bucketStart", (range) =>
+            range
+              .eq("siteId", delta.siteId)
+              .eq("bucketStart", delta.bucketStart),
+          )
+          .take(capRows + 1)
+      ).length;
+    }
+    if (dayRows >= capRows) {
+      await upsertRouteRollup(
+        ctx,
+        { ...delta, route: SITE_MAP_OTHER_ROUTE },
+        dayRowCounts,
+      );
+      return;
+    }
+    dayRowCounts.set(dayKey, dayRows + 1);
+  }
+  const outgoing = existing?.outgoing.map((entry) => ({ ...entry })) ?? [];
+  for (const [target, count] of delta.outgoing) {
+    let entry = outgoing.find((candidate) => candidate.route === target);
+    if (!entry) {
+      const named = outgoing.filter(
+        (candidate) => candidate.route !== SITE_MAP_OTHER_ROUTE,
+      ).length;
+      const route =
+        target !== SITE_MAP_OTHER_ROUTE && named >= ROUTE_TRANSITION_SLOTS
+          ? SITE_MAP_OTHER_ROUTE
+          : target;
+      entry = outgoing.find((candidate) => candidate.route === route);
+      if (!entry) {
+        entry = { route, count: 0 };
+        outgoing.push(entry);
+      }
+    }
+    entry.count += count;
+  }
+  const fields = {
+    siteId: delta.siteId,
+    bucketStart: delta.bucketStart,
+    shard: delta.shard,
+    generation: 1 as const,
+    route: delta.route,
+    pageviews: (existing?.pageviews ?? 0) + delta.pageviews,
+    entries: (existing?.entries ?? 0) + delta.entries,
+    continued: (existing?.continued ?? 0) + delta.continued,
+    dwellMs: (existing?.dwellMs ?? 0) + delta.dwellMs,
+    dwellSamples: (existing?.dwellSamples ?? 0) + delta.dwellSamples,
+    visitorSketch: unionVisitorSketches(
+      existing?.visitorSketch ?? createVisitorSketch(),
+      delta.visitorSketch,
+    ),
+    outgoing,
+  };
+  if (existing) {
+    await ctx.db.replace("routeDailyRollups", existing._id, fields);
+  } else {
+    await ctx.db.insert("routeDailyRollups", fields);
+  }
+}
+
+async function upsertVitalRollup(ctx: MutationCtx, delta: VitalRollupDelta) {
+  const existing = await ctx.db
+    .query("vitalRollups")
+    .withIndex("by_key", (range) =>
+      range
+        .eq("siteId", delta.siteId)
+        .eq("bucketStart", delta.bucketStart)
+        .eq("device", delta.device)
+        .eq("page", delta.page)
+        .eq("shard", delta.shard),
+    )
+    .unique();
+  if (
+    !existing &&
+    delta.device === VITAL_ALL &&
+    delta.page !== VITAL_OTHER_PAGES
+  ) {
+    // A new page row spends one bounded read to respect the daily page slots;
+    // past the cap, the day's remaining pages fold into "(other)".
+    const pageRows = await ctx.db
+      .query("vitalRollups")
+      .withIndex("by_key", (range) =>
+        range
+          .eq("siteId", delta.siteId)
+          .eq("bucketStart", delta.bucketStart)
+          .eq("device", VITAL_ALL),
+      )
+      .take(VITAL_PAGE_SLOTS * REPORT_ROLLUP_SHARDS + 1);
+    if (pageRows.length > VITAL_PAGE_SLOTS * REPORT_ROLLUP_SHARDS) {
+      await upsertVitalRollup(ctx, { ...delta, page: VITAL_OTHER_PAGES });
+      return;
+    }
+  }
+  const metrics = existing?.metrics.map((entry) => ({ ...entry })) ?? [];
+  for (const [metric, addition] of delta.metrics) {
+    const entry = metrics.find((candidate) => candidate.metric === metric);
+    if (entry) {
+      entry.count += addition.count;
+      entry.sum += addition.sum;
+      entry.histogram = mergeVitalHistograms(
+        entry.histogram,
+        addition.histogram,
+      );
+    } else {
+      metrics.push({
+        metric,
+        count: addition.count,
+        sum: addition.sum,
+        histogram: [...addition.histogram],
+      });
+    }
+  }
+  const fields = {
+    siteId: delta.siteId,
+    bucketStart: delta.bucketStart,
+    device: delta.device,
+    page: delta.page,
+    shard: delta.shard,
+    metrics,
+  };
+  if (existing) {
+    await ctx.db.replace("vitalRollups", existing._id, fields);
+  } else {
+    await ctx.db.insert("vitalRollups", fields);
+  }
+}
+
+async function flushReportRollups(
+  ctx: MutationCtx,
+  deltas: ReportRollupDeltas,
+) {
+  for (const delta of deltas.goals.values()) {
+    const existing = await ctx.db
+      .query("goalDailyRollups")
+      .withIndex("by_siteId_and_goalId_and_bucketStart_and_shard", (range) =>
+        range
+          .eq("siteId", delta.siteId)
+          .eq("goalId", delta.goalId)
+          .eq("bucketStart", delta.bucketStart)
+          .eq("shard", delta.shard),
+      )
+      .unique();
+    const fields = {
+      siteId: delta.siteId,
+      goalId: delta.goalId,
+      bucketStart: delta.bucketStart,
+      shard: delta.shard,
+      generation: 1 as const,
+      goalKey: delta.goalKey,
+      completions: (existing?.completions ?? 0) + delta.completions,
+      valueCents: (existing?.valueCents ?? 0) + delta.valueCents,
+    };
+    if (existing) {
+      await ctx.db.replace("goalDailyRollups", existing._id, fields);
+    } else {
+      await ctx.db.insert("goalDailyRollups", fields);
+    }
+  }
+  for (const delta of deltas.funnels.values()) {
+    const existing = await ctx.db
+      .query("funnelDailyRollups")
+      .withIndex("by_siteId_and_funnelId_and_bucketStart_and_shard", (range) =>
+        range
+          .eq("siteId", delta.siteId)
+          .eq("funnelId", delta.funnelId)
+          .eq("bucketStart", delta.bucketStart)
+          .eq("shard", delta.shard),
+      )
+      .unique();
+    const steps = existing?.steps.map((step) => ({ ...step })) ?? [];
+    for (const addition of delta.steps.values()) {
+      const current = steps.find((step) => step.step === addition.step);
+      if (current) {
+        current.entrants += addition.entrants;
+        current.completions += addition.completions;
+      } else {
+        if (steps.length >= MAX_FUNNEL_STEPS) {
+          fail("LIMIT_EXCEEDED", "daily funnel rollup step limit exceeded");
+        }
+        steps.push(addition);
+      }
+    }
+    const fields = {
+      siteId: delta.siteId,
+      funnelId: delta.funnelId,
+      bucketStart: delta.bucketStart,
+      shard: delta.shard,
+      generation: 1 as const,
+      steps,
+    };
+    if (existing) {
+      await ctx.db.replace("funnelDailyRollups", existing._id, fields);
+    } else {
+      await ctx.db.insert("funnelDailyRollups", fields);
+    }
+  }
+  for (const delta of deltas.affiliates.values()) {
+    const existing = await ctx.db
+      .query("affiliateDailyRollups")
+      .withIndex(
+        "by_siteId_and_affiliateId_and_bucketStart_and_shard",
+        (range) =>
+          range
+            .eq("siteId", delta.siteId)
+            .eq("affiliateId", delta.affiliateId)
+            .eq("bucketStart", delta.bucketStart)
+            .eq("shard", delta.shard),
+      )
+      .unique();
+    const fields = {
+      siteId: delta.siteId,
+      affiliateId: delta.affiliateId,
+      bucketStart: delta.bucketStart,
+      shard: delta.shard,
+      generation: 1 as const,
+      affiliateSlug: delta.affiliateSlug,
+      visits: (existing?.visits ?? 0) + delta.visits,
+      visitorSketch: unionVisitorSketches(
+        existing?.visitorSketch ?? createVisitorSketch(),
+        delta.visitorSketch,
+      ),
+      conversions: (existing?.conversions ?? 0) + delta.conversions,
+      revenueCents: (existing?.revenueCents ?? 0) + delta.revenueCents,
+      commissionCents: (existing?.commissionCents ?? 0) + delta.commissionCents,
+    };
+    if (existing) {
+      await ctx.db.replace("affiliateDailyRollups", existing._id, fields);
+    } else {
+      await ctx.db.insert("affiliateDailyRollups", fields);
+    }
+  }
+  const routeDayRowCounts = new Map<string, number>();
+  for (const delta of deltas.routes.values()) {
+    await upsertRouteRollup(ctx, delta, routeDayRowCounts);
+  }
+  for (const delta of deltas.vitals.values()) {
+    await upsertVitalRollup(ctx, delta);
+  }
+}
+
 function foldAggregates(
   aggregateDeltas: AggregateDeltas,
   siteId: Id<"sites">,
@@ -1198,16 +2524,16 @@ function foldAggregates(
   recordTelemetry: boolean,
   financialConversion: boolean,
 ) {
-  const shard = stableHash(event.sessionId) % AGGREGATE_SHARDS;
-  const revenueCents = financialConversion ? event.revenueCents ?? 0 : 0;
-  const dimensionDeltas = dimensions.map(
-    (dimension): DimensionDelta => ({
-      ...dimension,
-      count: Number(recordTelemetry),
-      revenueCents,
-    }),
-  );
+  const revenueCents = financialConversion ? (event.revenueCents ?? 0) : 0;
+  const dimensionDeltas = dimensions.map((dimension): DimensionDelta => ({
+    ...dimension,
+    count: Number(recordTelemetry),
+    revenueCents,
+  }));
   for (const granularity of ["hour", "day"] as const) {
+    const shard =
+      stableHash(event.sessionId) %
+      (granularity === "hour" ? HOURLY_AGGREGATE_SHARDS : AGGREGATE_SHARDS);
     const interval = granularity === "hour" ? HOUR_MS : DAY_MS;
     const bucketStart = Math.floor(event.timestamp / interval) * interval;
     const key = `${siteId}:${granularity}:${bucketStart}:${shard}`;
@@ -1225,6 +2551,7 @@ function foldAggregates(
         outboundClicks: 0,
         sessions: 0,
         visitors: 0,
+        visitorSketch: createVisitorSketch(),
         conversions: 0,
         revenueCents: 0,
         dimensions: [],
@@ -1245,6 +2572,9 @@ function foldAggregates(
         ? Number(session.isNewVisitorHour)
         : Number(session.isNewVisitorDay)
       : 0;
+    if (recordTelemetry) {
+      delta.visitorSketch = addVisitor(delta.visitorSketch, session.visitorKey);
+    }
     delta.conversions += Number(financialConversion);
     delta.revenueCents += revenueCents;
     delta.dimensions = mergeDimensions(delta.dimensions, dimensionDeltas);
@@ -1272,6 +2602,13 @@ async function flushAggregates(
       bucket?.dimensions ?? [],
       delta.dimensions,
     );
+    const priorSketch = bucket?.visitorSketch;
+    const visitorSketch = priorSketch
+      ? unionVisitorSketches(priorSketch, delta.visitorSketch)
+      : delta.visitorSketch;
+    const visitorSketchComplete = bucket
+      ? bucket.visitorSketchComplete === true && priorSketch !== undefined
+      : true;
     if (bucket) {
       await ctx.db.patch("aggregateBuckets", bucket._id, {
         events: bucket.events + delta.events,
@@ -1281,6 +2618,8 @@ async function flushAggregates(
         outboundClicks: bucket.outboundClicks + delta.outboundClicks,
         sessions: bucket.sessions + delta.sessions,
         visitors: bucket.visitors + delta.visitors,
+        visitorSketch,
+        visitorSketchComplete,
         conversions: bucket.conversions + delta.conversions,
         revenueCents: bucket.revenueCents + delta.revenueCents,
         dimensions,
@@ -1298,6 +2637,8 @@ async function flushAggregates(
         outboundClicks: delta.outboundClicks,
         sessions: delta.sessions,
         visitors: delta.visitors,
+        visitorSketch,
+        visitorSketchComplete,
         conversions: delta.conversions,
         revenueCents: delta.revenueCents,
         dimensions,
@@ -1325,11 +2666,8 @@ function mergeDimensions(
     const typeSlots = result.filter(
       (candidate) => candidate.type === addition.type,
     );
-    if (typeSlots.length < DIMENSION_SLOTS) {
-      const occupied = Array.from(
-        { length: DIMENSION_SLOTS },
-        () => false,
-      );
+    if (typeSlots.length < DIMENSION_SLOTS - 1) {
+      const occupied = Array.from({ length: DIMENSION_SLOTS }, () => false);
       for (const candidate of typeSlots) occupied[candidate.slot] = true;
       let slot = stableHash(addition.value) % DIMENSION_SLOTS;
       while (occupied[slot]) slot = (slot + 1) % DIMENSION_SLOTS;
@@ -1342,15 +2680,36 @@ function mergeDimensions(
       });
       continue;
     }
-    if (addition.count === 0) continue;
+    const overflow = typeSlots.find(
+      (candidate) => candidate.value === "(other)",
+    );
+    if (overflow) {
+      overflow.count += addition.count;
+      overflow.revenueCents += addition.revenueCents;
+      continue;
+    }
+    if (addition.count === 0 && addition.revenueCents === 0) continue;
+    if (typeSlots.length < DIMENSION_SLOTS) {
+      const occupied = new Set(typeSlots.map((candidate) => candidate.slot));
+      let slot = stableHash("(other)") % DIMENSION_SLOTS;
+      while (occupied.has(slot)) slot = (slot + 1) % DIMENSION_SLOTS;
+      result.push({
+        type: addition.type,
+        slot,
+        value: "(other)",
+        count: addition.count,
+        revenueCents: addition.revenueCents,
+      });
+      continue;
+    }
 
     let replacement = typeSlots[0];
     for (const candidate of typeSlots.slice(1)) {
       if (candidate.count < replacement.count) replacement = candidate;
     }
-    replacement.value = addition.value;
+    replacement.value = "(other)";
     replacement.count += addition.count;
-    replacement.revenueCents = addition.revenueCents;
+    replacement.revenueCents += addition.revenueCents;
   }
   return result;
 }
@@ -1360,5 +2719,7 @@ function stringProperty(
   key: string,
 ): string | undefined {
   const value = properties?.[key];
-  return typeof value === "string" ? cleanString(value, 120) || undefined : undefined;
+  return typeof value === "string"
+    ? cleanString(value, 120) || undefined
+    : undefined;
 }