diff --git a/src/renderer/src/lib/worktree-runtime-owner-index.detected-perf.test.ts b/src/renderer/src/lib/worktree-runtime-owner-index.detected-perf.test.ts
index be7389f9d31..cba36ebf6de 100644
--- a/src/renderer/src/lib/worktree-runtime-owner-index.detected-perf.test.ts
+++ b/src/renderer/src/lib/worktree-runtime-owner-index.detected-perf.test.ts
@@ -129,7 +129,7 @@ describe('detected worktree index performance', () => {
     expect(walkedReads).toBeGreaterThan(100_000)
   })
 
-  it('keeps the lookup path an order of magnitude cheaper than the catalog walk', () => {
+  it('records lookup-path timing without making wall clock a CI gate', () => {
     const walkPool = freshCatalogPool()
     const walkMs = measureSweeps((sweep) => {
       const catalog = walkPool[sweep]!
@@ -194,9 +194,6 @@ describe('detected worktree index performance', () => {
       `${JSON.stringify(report, null, 2)}\n`
     )
 
-    // Only the warm ratio is asserted: a republished catalog pays one rebuild that costs about
-    // what a single sweep's walk did, so the fresh leg is parity and is recorded, not gated.
-    // The measured warm margin is ~100x, so this bound cannot flake under CI load.
-    expect(walkWarmMs / warmMs).toBeGreaterThan(10)
+    // Why: shared-runner timing is noisy; the read-count test above is the deterministic CI gate.
   })
 })