diff --git a/backend/__tests__/unit/services/activityMentionNeedle.test.js b/backend/__tests__/unit/services/activityMentionNeedle.test.js
new file mode 100644
index 000000000..e9cd6dd16
--- /dev/null
+++ b/backend/__tests__/unit/services/activityMentionNeedle.test.js
@@ -0,0 +1,61 @@
+/**
+ * The pod-context cue (#1216/#1244) tells agents, in the frame they receive on
+ * every wake: "human attention is matched on the literal @handle, so 'Sam
+ * should decide this' is addressed to nobody."
+ *
+ * That is a claim about THIS function and nothing else. An agent cannot
+ * falsify it — it acts on the cue and never sees the code. Before this file,
+ * nothing did either: dropping the '@' from `mentionNeedle` left all 2033
+ * backend unit tests green (measured at origin/main e86a4a4a, twice, with an
+ * unmutated 2033/2033 control). The cue would have become a lie with its own
+ * text untouched and the suite still passing.
+ *
+ * So this pins the BEHAVIOUR, not the copy. The negative is the load-bearing
+ * case; each one is paired with the positive that proves the fixture reaches
+ * the branch at all.
+ */
+const ActivityService = require('../../../services/activityService');
+
+const flagsFor = (content, username) => ActivityService.computeFlags({
+  actor: { id: 'u1', type: 'user' },
+  type: 'message',
+  action: 'posted',
+  content,
+  username,
+});
+
+describe('human mentions are matched on the literal @handle', () => {
+  it('a bare name is NOT a mention — the cue\'s central claim', () => {
+    expect(flagsFor('Sam should decide this', 'sam').isMention).toBe(false);
+  });
+
+  it('CONTROL: the same sentence with the handle IS a mention', () => {
+    expect(flagsFor('@sam should decide this', 'sam').isMention).toBe(true);
+  });
+
+  it('the @ must be adjacent — "@ sam" does not address sam', () => {
+    expect(flagsFor('@ sam should decide this', 'sam').isMention).toBe(false);
+  });
+
+  it('matching is case-insensitive on both sides', () => {
+    expect(flagsFor('Ping @SAM about this', 'Sam').isMention).toBe(true);
+  });
+
+  it('an empty username never matches, however the content reads', () => {
+    expect(flagsFor('@ everyone @', '').isMention).toBe(false);
+  });
+
+  it('the handle is found mid-sentence, not just at the start', () => {
+    expect(flagsFor('I think @sam owns this one', 'sam').isMention).toBe(true);
+  });
+
+  /**
+   * Documented over-match, pinned so the next reader sees it is known rather
+   * than believing the handle is matched as a whole token. `includes` has no
+   * right boundary, so a longer handle contains a shorter one. Reported
+   * separately; this test records the behaviour, it does not endorse it.
+   */
+  it('KNOWN OVER-MATCH: a longer handle flags the shorter one it contains', () => {
+    expect(flagsFor('@sammy is on it', 'sam').isMention).toBe(true);
+  });
+});