From 6df4cc6f734b87792a98af3bbc21388931c76165 Mon Sep 17 00:00:00 2001
From: luvs01 <27862058+luvs01@users.noreply.github.com>
Date: Sun, 16 Aug 2026 19:28:09 +0900
Subject: [PATCH] fix(usage): recompute normalized estimated totals

---
 src/server/request-log.ts              | 14 +++++--
 tests/request-log-estimate-cap.test.ts | 51 ++++++++++++++++++++++++++
 2 files changed, 62 insertions(+), 3 deletions(-)

diff --git a/src/server/request-log.ts b/src/server/request-log.ts
index f4c4df6626..c69ea2a110 100644
--- a/src/server/request-log.ts
+++ b/src/server/request-log.ts
@@ -1004,10 +1004,14 @@ function finalizedUsage(
   const usageFallback = !finalUsage && estimate !== undefined
     ? { inputTokens: estimate, outputTokens: 0, estimated: true }
     : undefined;
-  const loggedUsage = finalUsage && estimate !== undefined
+  const combinedInputTokens = finalUsage && estimate !== undefined
+    ? Math.max(finalUsage.inputTokens, estimate)
+    : undefined;
+  const loggedUsage = finalUsage && combinedInputTokens !== undefined
     ? {
         ...finalUsage,
-        inputTokens: Math.max(finalUsage.inputTokens, estimate),
+        inputTokens: combinedInputTokens,
+        totalTokens: combinedInputTokens + finalUsage.outputTokens,
         estimated: true,
       }
     : finalUsage
@@ -1017,7 +1021,11 @@ function finalizedUsage(
       // ESTIMATE via capEstimateAtContextWindow, and Math.max preserves a real
       // provider-reported count, so it needs no further reduction.
       ? (finalUsage.estimated && contextWindow !== undefined && finalUsage.inputTokens > contextWindow
-          ? { ...finalUsage, inputTokens: contextWindow }
+          ? {
+              ...finalUsage,
+              inputTokens: contextWindow,
+              totalTokens: contextWindow + finalUsage.outputTokens,
+            }
           : finalUsage)
       : usageFallback;
   const totalTokens = usageTotalTokens(loggedUsage);
diff --git a/tests/request-log-estimate-cap.test.ts b/tests/request-log-estimate-cap.test.ts
index b51f7aeb6b..27b7abed2d 100644
--- a/tests/request-log-estimate-cap.test.ts
+++ b/tests/request-log-estimate-cap.test.ts
@@ -31,6 +31,57 @@ describe("request log token-estimate context-window cap (codex-router PR #140)",
     expect(attempt.usage?.inputTokens).toBe(100_000);
   });
 
+  test("capping adapter-estimated input recomputes an explicit total", () => {
+    const attempt = beginRequestAttempt(1, "kiro", "deepseek-3.2", "kiro");
+    finishRequestAttempt(attempt, 200, 10, {
+      inputTokens: 200_000,
+      outputTokens: 4,
+      totalTokens: 200_004,
+      estimated: true,
+    });
+    expect(attempt.usage).toEqual({
+      inputTokens: DEEPSEEK_WINDOW,
+      outputTokens: 4,
+      totalTokens: DEEPSEEK_WINDOW + 4,
+      estimated: true,
+    });
+    expect(attempt.totalTokens).toBe(DEEPSEEK_WINDOW + 4);
+  });
+
+  test("combining a local estimate recomputes the nested and outer totals", () => {
+    const attempt = beginRequestAttempt(1, "kiro", "deepseek-3.2", "kiro");
+    noteAttemptSend(attempt, 200_000);
+    finishRequestAttempt(attempt, 200, 10, {
+      inputTokens: 50,
+      outputTokens: 4,
+      totalTokens: 54,
+    });
+    expect(attempt.usage).toEqual({
+      inputTokens: DEEPSEEK_WINDOW,
+      outputTokens: 4,
+      totalTokens: DEEPSEEK_WINDOW + 4,
+      estimated: true,
+    });
+    expect(attempt.totalTokens).toBe(DEEPSEEK_WINDOW + 4);
+  });
+
+  test("caps an estimated Cursor checkpoint without double-adding output", () => {
+    const attempt = beginRequestAttempt(1, "cursor", "claude-4.6-opus-high", "cursor");
+    finishRequestAttempt(attempt, 200, 10, {
+      inputTokens: 499_994,
+      outputTokens: 6,
+      totalTokens: 500_000,
+      estimated: true,
+    });
+    expect(attempt.usage).toEqual({
+      inputTokens: CURSOR_CLAUDE_WINDOW,
+      outputTokens: 6,
+      totalTokens: CURSOR_CLAUDE_WINDOW + 6,
+      estimated: true,
+    });
+    expect(attempt.totalTokens).toBe(CURSOR_CLAUDE_WINDOW + 6);
+  });
+
   test("a positive provider-reported input count is never reduced by the cap", () => {
     const attempt = beginRequestAttempt(1, "kiro", "deepseek-3.2", "kiro");
     noteAttemptSend(attempt, 200_000); // estimate would exceed the window