diff --git a/src/server/request-log.ts b/src/server/request-log.ts
index c900542ede6..096a028bff0 100644
--- a/src/server/request-log.ts
+++ b/src/server/request-log.ts
@@ -1433,7 +1433,12 @@ export function addFinalRequestLog(
     ...(attempt.deliverySummary ? { deliverySummary: { ...attempt.deliverySummary } } : {}),
   }));
   const isCombo = logCtx.comboId !== undefined && (attempts?.length ?? 0) > 0;
-  const aggregate = isCombo ? aggregateAttemptUsage(attempts ?? []) : null;
+  // A key or account rotation splits the attempt and clears `logCtx.usage`, so a multi-attempt
+  // request totals from its attempts, combo or not. One attempt keeps the request-level usage
+  // (and its input estimate); a locally answered row sent nothing and keeps its own.
+  const aggregate = isCombo || ((attempts?.length ?? 0) > 1 && logCtx.localTerminalReason === undefined)
+    ? aggregateAttemptUsage(attempts ?? [])
+    : null;
   const loggedUsage = aggregate?.usage ?? existing.usage;
   const usageStatus = aggregate?.status ?? existing.status;
   const totalTokens = aggregate?.totalTokens ?? existing.totalTokens;
diff --git a/structure/dashboard-and-usage.md b/structure/dashboard-and-usage.md
index 8254b5a1de5..5eb03e6bd18 100644
--- a/structure/dashboard-and-usage.md
+++ b/structure/dashboard-and-usage.md
@@ -185,7 +185,7 @@ retains separate flat records when retries change keys, and isolates each record
 from parent combo totals and adapter-loop aggregation. Reported failure usage is retained;
 missing usage and historical identities remain unknown. Native wire snapshots replace only the
 current physical response contribution, preserving prior sends on the same key without counting
-repeated inspections twice. Consumers sum the flat attempts once and keep subscription quota
+repeated inspections twice. A row with two or more attempts, combo or not, takes its request total from those attempts, so a key or account rotation is not reported as its last record alone; a single-attempt or locally answered row keeps its request-level usage. Consumers sum the flat attempts once and keep subscription quota
 observations separate from token or API-equivalent cost totals.
 
 `src/server/hub-usage.ts` serves `GET /v1/usage` on hubs for an explicit configured data key. The authenticated key selects the aggregate; query parameters cannot select an API-key identity. Unscoped environment/admin credentials and loopback bypass are not admitted. The response projects only this client's numeric totals, provider/model/day rows and incomplete-history metadata through `src/remote/hub-usage.ts`; accounts, raw records and key IDs are omitted. Unknown fields are stripped at every object boundary and the serialized body is capped at 1 MiB.
diff --git a/tests/routing/routing-policy-dispatch-scope.test.ts b/tests/routing/routing-policy-dispatch-scope.test.ts
index 61fd25f0f2a..e96f0a36b40 100644
--- a/tests/routing/routing-policy-dispatch-scope.test.ts
+++ b/tests/routing/routing-policy-dispatch-scope.test.ts
@@ -377,8 +377,9 @@ describe("policy scope through real request preparation", () => {
     const rows: RequestLogEntry[] = [];
     addFinalRequestLog("policy-skipped-candidate", Date.now(), log, response.status, undefined, row => rows.push(row));
     expect(settled).toEqual([lastUsage]);
-    expect(rows[0]).toMatchObject({ provider: expected.at(-1), model: `served-${expected.at(-1)}`, usage: lastUsage,
-      status: response.status, spend: { sends: expected.length, settled: expected.length, unresolved: 0 },
+    // The row totals every physical attempt (10/1 + 20/2), not only the last one.
+    expect(rows[0]).toMatchObject({ provider: expected.at(-1), model: `served-${expected.at(-1)}`,
+      usage: expected.length === 1 ? lastUsage : { inputTokens: 30, outputTokens: 3 }, status: response.status, spend: { sends: expected.length, settled: expected.length, unresolved: 0 },
     });
     expect(rows[0]?.attempts?.map(attempt => attempt.provider)).toEqual(expected);
     expect(rows[0]?.attempts?.every(attempt => attempt.sendCount === 1 && attempt.usage !== undefined)).toBe(true);
diff --git a/tests/usage/key-attribution.test.ts b/tests/usage/key-attribution.test.ts
index 70dd335b7a0..960963885a9 100644
--- a/tests/usage/key-attribution.test.ts
+++ b/tests/usage/key-attribution.test.ts
@@ -1,6 +1,6 @@
 import { describe, expect, test } from "bun:test";
 import {
-  addFinalRequestLog, beginRequestAttempt, finishRequestAttempt, noteProviderAttemptSend,
+  addFinalRequestLog, aggregateAttemptUsage, beginRequestAttempt, finishRequestAttempt, noteProviderAttemptSend,
   recordKeyAttemptFailure, recordKeyAttemptUsage, applyResponseLogMetadata, sealRequestAttemptIdentity,
   inspectResponseLogSsePayload, type RequestLogContext, type RequestLogEntry,
 } from "../../src/server/request-log";
@@ -110,6 +110,52 @@ describe("key attempt accounting", () => {
     expect(roundTrip.usage).toMatchObject({ inputTokens: 400, outputTokens: 60 });
     expect(JSON.stringify(roundTrip)).not.toContain("test-key-");
   });
+  test("a rotated non-combo request totals every physical attempt, not only the last", async () => {
+    const key = (reference: string) => ({ adapter: "openai-chat" as const, authMode: "key" as const,
+      baseUrl: "https://example.test/v1", _apiKeyAttempt: { reference } });
+    const active = beginRequestAttempt(1, "test-provider", "test-model", "openai-chat");
+    const ctx: RequestLogContext = { provider: "test-provider", model: "test-model",
+      activeAttempt: active, activeAttemptStartedAt: Date.now(), attempts: [active] };
+    noteProviderAttemptSend(ctx, "test-provider", key("test-key-a"), undefined);
+    await recordKeyAttemptFailure(ctx, Response.json({ usage: { prompt_tokens: 100, completion_tokens: 20 } }, { status: 429 }));
+    noteProviderAttemptSend(ctx, "test-provider", key("test-key-b"), undefined, "key-429");
+    recordKeyAttemptUsage(ctx, { inputTokens: 300, outputTokens: 40 });
+    const rows: RequestLogEntry[] = [];
+    addFinalRequestLog("non-combo-rotation", Date.now(), ctx, 200, undefined, row => rows.push(row));
+    expect(rows[0].attempts?.map(attempt => attempt.usage?.inputTokens)).toEqual([100, 300]);
+    expect(rows[0]).toMatchObject({ provider: "test-provider", usageStatus: "reported", totalTokens: 460,
+      usage: { inputTokens: 400, outputTokens: 60 } });
+  });
+  test("a row's usage status is the aggregate of its non-combo attempts", async () => {
+    const key = (reference?: string) => ({ adapter: "openai-chat" as const, authMode: "key" as const,
+      baseUrl: "https://example.test/v1", _apiKeyAttempt: reference ? { reference } : undefined });
+    const active = beginRequestAttempt(1, "test-provider", "test-model", "openai-chat");
+    const ctx: RequestLogContext = { provider: "test-provider", model: "test-model",
+      activeAttempt: active, activeAttemptStartedAt: Date.now(), attempts: [active] };
+    noteProviderAttemptSend(ctx, "test-provider", key("test-key-a"), undefined);
+    await recordKeyAttemptFailure(ctx, Response.json({ usage: { prompt_tokens: 100, completion_tokens: 20 } }, { status: 429 }));
+    noteProviderAttemptSend(ctx, "test-provider", key("test-key-b"), undefined, "key-429");
+    await recordKeyAttemptFailure(ctx, Response.json({ error: "synthetic" }, { status: 401 }));
+    noteProviderAttemptSend(ctx, "test-provider", key(), undefined, "key-401");
+    recordKeyAttemptUsage(ctx, { inputTokens: 300, outputTokens: 40 });
+    const rows: RequestLogEntry[] = [];
+    addFinalRequestLog("non-combo-mixed-status", Date.now(), ctx, 200, undefined, row => rows.push(row));
+    expect(rows[0]).toMatchObject({ totalTokens: 460, usage: { inputTokens: 400, outputTokens: 60 } });
+    // Whatever status the no-usage refusal earns, the row reports the aggregate of its attempts,
+    // not the last attempt's status alone.
+    expect(rows[0].usageStatus).toBe(aggregateAttemptUsage(rows[0].attempts!).status);
+  });
+  test("a locally answered non-combo request keeps its own usage", () => {
+    const sent = beginRequestAttempt(1, "test-provider", "test-model", "openai-chat");
+    Object.assign(sent, { sendCount: 1, status: 429, usageStatus: "reported", usage: { inputTokens: 100, outputTokens: 10 } });
+    const local = beginRequestAttempt(2, "test-provider", "test-model", "openai-chat");
+    local.locallyAnswered = true;
+    const ctx: RequestLogContext = { provider: "test-provider", model: "test-model", activeAttempt: local,
+      attempts: [sent, local], localTerminalReason: "synthetic", usage: { inputTokens: 7, outputTokens: 3 } };
+    const rows: RequestLogEntry[] = [];
+    addFinalRequestLog("local-answer", Date.now(), ctx, 200, undefined, row => rows.push(row));
+    expect(rows[0]).toMatchObject({ totalTokens: 10, usage: { inputTokens: 7, outputTokens: 3 } });
+  });
   test.each([
     ["generic OAuth", "o111111", "o222222"],
     ["Codex pool", "paaaaa1", "pbbbbb2"],