fix(desktop): hide the main cache rate when the main-only summary read fails

A failed narrow query left the blended summary in place, and the rate
then read the blend while presenting it as the main loop's — for 1M main
input tokens with 950k cached plus 100k uncached auxiliary input, 86.36%
shown where the main rate is 95%. The preload now marks the narrower
read as unavailable, and the overview hides the rate instead of letting
the blended number stand in. A summary that never asked the narrower
question still falls back to the blend.

Carries the #5691 review finding.

Generated-by: GLM-5.3-Flash (ZCode)

Author: ggbdpq

apps/desktop/src/main/__tests__/session-inspector-usage-stats.test.ts

@@ -383,3 +383,25 @@ test('reads the cache rate from the main loop alone when the main summary is ava
   assert.ok(refined);
   assert.equal(Math.round(refined * 100), 95);
 });
+
+test('shows no cache rate when the main-only read failed rather than blending', () => {
+  // A failed narrow query leaves the blended totals in place. Reporting the
+  // blended 86% as the main loop's rate is the same misreading the main
+  // summary exists to prevent (#5691 review): hide the rate instead.
+  const { cacheHitRate } = deriveInspectorOverviewModel(undefined, {
+    ...usageSummary({
+      totalTokens: {
+        input: 1_100_000,
+        output: 60_300,
+        cacheMiss: 150_000,
+        cacheRead: 950_000,
+        cacheWrite: 0,
+        reasoning: 0,
+        total: 1_160_300,
+      },
+    }),
+    mainSummaryUnavailable: true,
+  });
+
+  assert.equal(cacheHitRate, undefined);
+});

apps/desktop/src/preload/bridge-contract.d.ts

@@ -824,6 +824,8 @@ export interface DesktopSessionUsageSummary extends UsageSummaryV2 {
    * reads this: auxiliary prompts have their own cache prefix (#5691).
    */
   readonly mainSummary?: DesktopSessionUsageSummary;
+  /** The narrower read failed; the blended rate must not stand in for it. */
+  readonly mainSummaryUnavailable?: boolean;
 }
 
 export interface MakaBridge {

apps/desktop/src/preload/preload.ts

@@ -1329,10 +1329,11 @@ async function loadSessionUsageSummary(
   ])) as [Result<DesktopSessionUsageSummary>, Result<DesktopSessionUsageSummary>];
   if (!summary.ok) return summary;
   // The main-only read refines the cache rate; losing it must not lose the
-  // overview, so a failed auxiliary query just leaves the blended rate.
+  // overview — but it must also not pass the blended rate off as the main
+  // loop's, so the failure is marked and the rate hides itself (#5691).
   return main.ok
     ? { ...summary, data: { ...summary.data, mainSummary: main.data } }
-    : summary;
+    : { ...summary, data: { ...summary.data, mainSummaryUnavailable: true } };
 }
 
 async function updateDailyReviewConfig(

apps/desktop/src/renderer/application/contracts/session-inspector/session-inspector-overview-model.ts

@@ -28,6 +28,8 @@ type SessionUsageSummary = UsageSummaryV2 & {
   readonly provenance?: UsageProvenance;
   /** The agent loop's own calls, when the narrower read succeeded (#5691). */
   readonly mainSummary?: SessionUsageSummary;
+  /** The narrower read failed; the blended rate must not stand in for it. */
+  readonly mainSummaryUnavailable?: boolean;
 };
 
 /**
@@ -249,8 +251,12 @@ export function deriveInspectorOverviewModel(
 function usageCacheHitRate(usage: SessionUsageSummary | undefined): number | undefined {
   // Auxiliary calls keep their own prompt prefix, so blending them into the
   // rate reports the main loop's caching as worse than it is (#5691). The
-  // main-only summary is the rate's input; the blended summary is only the
-  // fallback when the narrower read is unavailable.
+  // main-only summary is the rate's input. When that narrower read failed,
+  // show no rate at all: the blended number is about a different set of calls
+  // and presenting it as the main loop's is the same lie in weaker type. The
+  // fallback to the blended summary only serves summaries that never asked
+  // the narrower question.
+  if (usage?.mainSummaryUnavailable && !usage.mainSummary) return undefined;
   const main = usage?.mainSummary ?? usage;
   if (!main || main.totalTokens.input === 0) return undefined;
   if (