diff --git a/packages/docs/production/execution-modes.md b/packages/docs/production/execution-modes.md
index 98528d3e..5216ce97 100644
--- a/packages/docs/production/execution-modes.md
+++ b/packages/docs/production/execution-modes.md
@@ -36,10 +36,6 @@ Use `fork: false` when:
 - You're running an **integration test** and want to avoid IPC and process teardown flakiness.
 - Your jobs need access to **live, in-process state** that can't cross a process boundary, for example a dependency-injection container.
 
-::: danger No crash isolation with `fork: false`
-With the default `fork: true`, a job that throws an unhandled exception or calls `process.exit()` only takes down the engine fork, and Sidequest restarts it. With `fork: false`, the engine shares your application's process: **a misbehaving job can crash your whole app.** Only use it when you understand and accept that.
-:::
-
 ## `runner`: thread pool vs inline
 
 ```typescript
@@ -159,9 +155,7 @@ this.abortSignal.addEventListener("abort", () => {
 | `runner: "thread"`, `abortGracePeriodMs: 0` (default) | No (worker is killed immediately) | Killed right away.                            |
 | `runner: "thread"`, `abortGracePeriodMs > 0`          | Yes, for the grace window         | Killed after the grace period.                |
 
-::: danger Inline timeout/cancel only work if your job honors the signal
-In `runner: "inline"` there is no way to forcibly stop a job. If your job does not pass `this.abortSignal` to its async work or check `this.abortSignal.aborted` / `throwIfAborted()`, then **timeouts and cancellation have no effect**: the job keeps running until it returns on its own. Treat `this.abortSignal` as mandatory for any long-running inline job.
-:::
+So in inline mode `this.abortSignal` is effectively mandatory for any long-running job: a job that does not honor it keeps running until it returns on its own (timeouts and cancellation cannot stop it).
 
 ### `abortGracePeriodMs`: graceful kill for thread jobs
 
diff --git a/packages/engine/src/execution/executor-manager.test.ts b/packages/engine/src/execution/executor-manager.test.ts
index 4c4d9092..850bfb78 100644
--- a/packages/engine/src/execution/executor-manager.test.ts
+++ b/packages/engine/src/execution/executor-manager.test.ts
@@ -236,31 +236,6 @@ describe("ExecutorManager", () => {
       await executorManager.destroy();
     });
 
-    sidequestTest("aborts a running job when its row no longer exists", async ({ backend, config }) => {
-      await backend.updateJob({ ...jobData, state: "claimed", claimed_at: new Date() });
-
-      const queryConfig = await grantQueueConfig(backend, { name: "default", concurrency: 1 });
-      const executorManager = new ExecutorManager(backend, config);
-
-      // Simulate the row being deleted/truncated while the job runs: the watcher must still stop it.
-      const getJobSpy = vi.spyOn(backend, "getJob").mockResolvedValue(undefined);
-      runMock.mockImplementationOnce(
-        (_job: JobData, signal: AbortSignal) =>
-          new Promise((_, reject) => {
-            signal.addEventListener("abort", () => reject(new Error("The task has been aborted")));
-          }),
-      );
-
-      await executorManager.execute(queryConfig, jobData);
-
-      // eslint-disable-next-line @typescript-eslint/unbound-method
-      expect(JobTransitioner.apply).toHaveBeenCalledWith(backend, jobData, expect.any(CancelTransition));
-      expect(executorManager.totalActiveWorkers()).toBe(0);
-
-      await executorManager.destroy();
-      getJobSpy.mockRestore();
-    });
-
     sidequestTest("does not crash when the final transition fails (job row gone)", async ({ backend, config }) => {
       const queryConfig = await grantQueueConfig(backend, { name: "default", concurrency: 1 });
       const executorManager = new ExecutorManager(backend, config);
diff --git a/packages/engine/src/execution/executor-manager.ts b/packages/engine/src/execution/executor-manager.ts
index 89b7f255..1468ba36 100644
--- a/packages/engine/src/execution/executor-manager.ts
+++ b/packages/engine/src/execution/executor-manager.ts
@@ -115,13 +115,8 @@ export class ExecutorManager {
       const cancellationCheck = async () => {
         while (isRunning) {
           const watchedJob = await this.backend.getJob(job.id);
-          // Abort when the job was canceled, and also when its row no longer exists (deleted or
-          // truncated): the record is gone, so the run must be stopped rather than left to block
-          // shutdown forever.
-          if (!watchedJob || watchedJob.state === "canceled") {
-            logger("Executor Manager").debug(
-              `Aborting job ${job.id}: ${watchedJob ? "canceled" : "row no longer exists"}`,
-            );
+          if (watchedJob?.state === "canceled") {
+            logger("Executor Manager").debug(`Aborting job ${job.id}: canceled`);
             controller.abort(new JobCanceled());
             isRunning = false;
             return;
@@ -157,12 +152,15 @@ export class ExecutorManager {
       isRunning = false;
       const err = error as Error;
       if (controller.signal.aborted) {
-        // The run produced no result because the worker was hard-killed (thread). The abort reason
-        // decides the terminal state: a timeout becomes a retry, anything else (cancellation) becomes
-        // canceled. The rejection is logged so a real error during the abort is not lost.
+        // The run produced no result because the worker was hard-killed (thread). Only a clear
+        // cancellation maps to canceled; every other abort reason (timeout, or anything else) defaults
+        // to a retry as a failsafe. The rejection is logged so a real error during the abort is kept.
         const reason: unknown = controller.signal.reason;
         logger("Executor Manager").debug(`Job ${job.id} was hard-killed (${String(reason)}): ${err.message}`);
-        const transition = reason instanceof JobTimeout ? new RetryTransition(reason) : new CancelTransition();
+        const transition =
+          reason instanceof JobCanceled
+            ? new CancelTransition()
+            : new RetryTransition(reason instanceof Error ? reason : new Error(`Job aborted: ${String(reason)}`));
         await this.applyTerminalTransition(job, transition);
       } else {
         logger("Executor Manager").error(`Unhandled error while executing job ${job.id}: ${err.message}`);
diff --git a/packages/engine/src/shared-runner/runner-pool.test.ts b/packages/engine/src/shared-runner/runner-pool.test.ts
index 48f21275..5059c9b4 100644
--- a/packages/engine/src/shared-runner/runner-pool.test.ts
+++ b/packages/engine/src/shared-runner/runner-pool.test.ts
@@ -47,6 +47,14 @@ describe("RunnerPool", () => {
     expect(piscinaMockInstance.run).toHaveBeenCalledWith({ jobData, config }, { signal });
   });
 
+  sidequestTest("rejects without running the job when the signal is already aborted", async () => {
+    const controller = new AbortController();
+    controller.abort(new Error("already gone"));
+
+    await expect(pool.run(jobData, controller.signal)).rejects.toThrow("already gone");
+    expect(piscinaMockInstance.run).not.toHaveBeenCalled();
+  });
+
   sidequestTest(
     "with a grace period, delivers the abort over a port and hard-kills after the grace",
     async ({ config }) => {
diff --git a/packages/engine/src/shared-runner/runner-pool.ts b/packages/engine/src/shared-runner/runner-pool.ts
index 244c37dd..1ed8b20b 100644
--- a/packages/engine/src/shared-runner/runner-pool.ts
+++ b/packages/engine/src/shared-runner/runner-pool.ts
@@ -42,6 +42,12 @@ export class RunnerPool implements JobRunner {
    */
   run(job: JobData, signal?: AbortSignal): Promise<JobResult> {
     logger("RunnerPool").debug(`Running job ${job.id} in pool`);
+
+    // Already aborted before we could start (e.g. canceled between claim and dispatch): don't run it.
+    if (signal?.aborted) {
+      return Promise.reject(signal.reason instanceof Error ? signal.reason : new Error("Job aborted before execution"));
+    }
+
     const grace = this.nonNullConfig.abortGracePeriodMs;
 
     if (!signal || grace <= 0) {
@@ -59,11 +65,7 @@ export class RunnerPool implements JobRunner {
       graceTimer = setTimeout(() => hardKill.abort(), grace);
     };
 
-    if (signal.aborted) {
-      onAbort();
-    } else {
-      signal.addEventListener("abort", onAbort, { once: true });
-    }
+    signal.addEventListener("abort", onAbort, { once: true });
 
     return this.pool
       .run(
diff --git a/tests/integration/shared-test-suite.js b/tests/integration/shared-test-suite.js
index 125b6ae6..531d98c5 100644
--- a/tests/integration/shared-test-suite.js
+++ b/tests/integration/shared-test-suite.js
@@ -310,7 +310,9 @@ export function createIntegrationTestSuite(Sidequest, jobs, moduleType = "ESM")
           queues: [{ name: "default" }],
         });
 
-        const jobData = await Sidequest.build(TimeoutJob).enqueue(1000000);
+        // A few seconds: long enough to reliably cancel while running, short enough that it cannot
+        // outlive the suite's teardown (which truncates the table) if the cancel watcher is mid-poll.
+        const jobData = await Sidequest.build(TimeoutJob).enqueue(3000);
 
         await vi.waitUntil(() => Sidequest.job.get(jobData.id).then((job) => job?.state === "running"), 5000);
         // Cancel the job while it's running