From 4196a4cd76db434854b679035e4146c7f69689ce Mon Sep 17 00:00:00 2001
From: Dotta <bippadotta@protonmail.com>
Date: Thu, 24 Sep 2026 16:06:08 -0700
Subject: [PATCH] test(runner): decouple file evidence from admission timing

Pre-capture the real baseline for the file-change assertion so EC2 load cannot turn its 100 ms admission deadline into an unrelated evidence-test failure. Keep the real after-snapshot and exact file assertions.

Co-Authored-By: Paperclip <noreply@paperclip.ing>
---
 .../src/live/live-session.test.ts             | 39 ++++++++++++-------
 1 file changed, 25 insertions(+), 14 deletions(-)

diff --git a/packages/paperclip-runner/src/live/live-session.test.ts b/packages/paperclip-runner/src/live/live-session.test.ts
index 4f5ab051ded..999c5a96830 100644
--- a/packages/paperclip-runner/src/live/live-session.test.ts
+++ b/packages/paperclip-runner/src/live/live-session.test.ts
@@ -24,6 +24,7 @@ import {
 import { DurableCapabilityLiveSessionStore } from "./durable-live-session-store.js";
 import { defaultCapabilityRunnerdBinary } from "./runnerd-codex-transport.js";
 import { captureTurnRejection } from "../../test/capture-turn-rejection.js";
+import * as workspaceDiff from "./workspace-diff.js";
 
 class AsyncNotifications implements AsyncIterable<CodexRpcNotification> {
   #values: CodexRpcNotification[] = [];
@@ -753,20 +754,30 @@ describe("Capability live runnerd and Codex session", () => {
     const service = new CapabilityLiveSessionService({ transportFactory: fakeTransportFactory(state) });
     const session = await service.create({ workingDirectory: root });
 
-    const result = await session.sendMessage("edit the workspace");
-    expect(result.snapshot.workspaceDiffs).toHaveLength(1);
-    expect(result.snapshot.workspaceDiffs?.[0]).toMatchObject({
-      turnId: "turn-1",
-      diff: {
-        schema: "paperclip.workspace.diff.v1",
-        source: "runner_verified",
-        complete: true,
-        totals: { files: 2 },
-      },
-    });
-    expect(result.snapshot.workspaceDiffs?.[0]?.diff.files.map((file) => file.path))
-      .toEqual(["created.ts", "existing.txt"]);
-    await service.shutdown(session.id);
+    // Test the file-change evidence, not whether a loaded CI worker scans the
+    // fixture within the production admission deadline. Keep both snapshots
+    // real; make only the already-captured baseline immediately available.
+    const baseline = await workspaceDiff.capturePaperclipWorkspace(root);
+    const capture = vi.spyOn(workspaceDiff, "capturePaperclipWorkspace")
+      .mockResolvedValueOnce(baseline);
+    try {
+      const result = await session.sendMessage("edit the workspace");
+      expect(result.snapshot.workspaceDiffs).toHaveLength(1);
+      expect(result.snapshot.workspaceDiffs?.[0]).toMatchObject({
+        turnId: "turn-1",
+        diff: {
+          schema: "paperclip.workspace.diff.v1",
+          source: "runner_verified",
+          complete: true,
+          totals: { files: 2 },
+        },
+      });
+      expect(result.snapshot.workspaceDiffs?.[0]?.diff.files.map((file) => file.path))
+        .toEqual(["created.ts", "existing.txt"]);
+    } finally {
+      capture.mockRestore();
+      await service.shutdown(session.id);
+    }
   });
 
   it("suspends after every per-turn response and restores the same provider session", async () => {