diff --git a/backend/__tests__/unit/services/agentMentionService.frameBudget.test.js b/backend/__tests__/unit/services/agentMentionService.frameBudget.test.js
new file mode 100644
index 000000000..3c40633af
--- /dev/null
+++ b/backend/__tests__/unit/services/agentMentionService.frameBudget.test.js
@@ -0,0 +1,120 @@
+/**
+ * The wake frame has a size, and nobody was paying for it (TASK-074).
+ *
+ * Every clause added to `formatPodContextFrame` and its companions is prepended
+ * to `chat.mention.payload.content` for EVERY mention to EVERY agent, forever.
+ * That is the correct place for a kernel affordance — the inline cue is the one
+ * thing a model will not deprioritize — and it is also the reason the cost is
+ * invisible: each clause is a few hundred characters in a diff nobody measures,
+ * and the total is paid per wake, fleet-wide.
+ *
+ * Measured while gating #1216/#1244: the pod-context frame alone went 1,639 →
+ * 2,269 (+38%) → ~2,697 (+65%) across one two-PR stack. No review comment
+ * mentioned the size, because there was no number to compare against.
+ *
+ * This is not a cap. It is a budget: raising it is a one-line change, and that
+ * one line is the entire point — it turns an invisible spend into a deliberate
+ * one that a reviewer can see and argue with. A PR that legitimately needs more
+ * room should raise the ceiling and say why in its body.
+ *
+ * Two-sided on purpose. A ceiling alone is satisfied by deleting the frame, and
+ * the copy assertions elsewhere in this suite pin individual sentences rather
+ * than the whole. The floor catches a frame that silently lost a section.
+ */
+
+jest.mock('../../../services/agentEventService', () => ({ enqueue: jest.fn() }));
+jest.mock('../../../models/AgentRegistry', () => ({
+  AgentInstallation: { find: jest.fn(), findOne: jest.fn() },
+}));
+jest.mock('../../../models/AgentProfile', () => ({ find: jest.fn() }));
+jest.mock('../../../models/Pod', () => ({ findById: jest.fn(), find: jest.fn() }));
+jest.mock('../../../models/User', () => ({ find: jest.fn(), findById: jest.fn() }));
+jest.mock('../../../services/chatSummarizerService', () => ({
+  constructor: { getLatestPodSummary: jest.fn() },
+  summarizePodMessages: jest.fn(),
+}));
+jest.mock('../../../models/AgentEvent', () => ({ countDocuments: jest.fn() }));
+jest.mock('../../../services/welcomeWakeService', () => ({ maybeFireWelcomeWake: jest.fn() }));
+jest.mock('../../../models/pg/Message', () => ({ findById: jest.fn(async () => null) }));
+jest.mock('../../../models/pg/ThreadUserState', () => ({
+  followByParticipation: jest.fn().mockResolvedValue(true),
+}));
+
+const AgentMentionService = require('../../../services/agentMentionService');
+const AgentEventService = require('../../../services/agentEventService');
+const { AgentInstallation } = require('../../../models/AgentRegistry');
+const AgentProfile = require('../../../models/AgentProfile');
+const Pod = require('../../../models/Pod');
+const User = require('../../../models/User');
+const AgentEvent = require('../../../models/AgentEvent');
+
+// The reference wake: a plain chat pod, one installed seat, an explicit
+// @mention, no thread, no wake-on-message opt-in. Deliberately the SMALLEST
+// real frame — a collaborative pod and a wake-on-message seat both add more, so
+// a budget measured here is a floor on what the fleet actually pays.
+const BUDGET_MAX = 3000;
+const BUDGET_MIN = 2600;
+
+const referenceWake = async () => {
+  AgentInstallation.find.mockReturnValue({
+    lean: jest.fn().mockResolvedValue([
+      { agentName: 'seat-a', instanceId: 'default', displayName: 'Seat A' },
+    ]),
+  });
+  AgentProfile.find.mockReturnValue({ lean: jest.fn().mockResolvedValue([]) });
+  User.find.mockReturnValue({ select: jest.fn().mockReturnThis(), lean: jest.fn().mockResolvedValue([]) });
+  User.findById.mockImplementation(() => ({
+    select: jest.fn().mockReturnThis(),
+    lean: jest.fn().mockResolvedValue({ _id: 'user-1', isBot: false }),
+  }));
+  Pod.findById.mockReturnValue({
+    select: jest.fn().mockReturnValue({
+      lean: jest.fn().mockResolvedValue({ type: 'chat', members: ['user-1', 'bot-1'] }),
+    }),
+    lean: jest.fn().mockResolvedValue({ _id: 'pod-1', type: 'chat', members: ['user-1', 'bot-1'] }),
+  });
+  AgentEvent.countDocuments.mockResolvedValue(0);
+
+  await AgentMentionService.enqueueMentions({
+    podId: 'pod-1',
+    userId: 'user-1',
+    username: 'alice',
+    message: { id: 'm-1', content: 'hi @seat-a' },
+  });
+  const call = AgentEventService.enqueue.mock.calls.find(([a]) => a.type === 'chat.mention');
+  return call[0].payload.content;
+};
+
+beforeEach(() => { jest.clearAllMocks(); });
+
+describe('wake frame size budget', () => {
+  test('the reference wake stays inside its character budget', async () => {
+    const content = await referenceWake();
+
+    // If this fails you added a clause. That is allowed — raise BUDGET_MAX in
+    // the same diff and say in the PR body what the fleet is buying, so the
+    // trade is on the record instead of inside a paragraph.
+    expect(content.length).toBeLessThanOrEqual(BUDGET_MAX);
+  });
+
+  test('and has not silently lost a section', async () => {
+    const content = await referenceWake();
+
+    // The other half of a budget. A ceiling on its own is satisfied by an
+    // empty frame, and the copy assertions in this suite pin sentences one at
+    // a time — none of them notices a whole section going missing.
+    expect(content.length).toBeGreaterThanOrEqual(BUDGET_MIN);
+  });
+
+  test('the sections that make up the cost are all present', async () => {
+    // Named so a budget failure is diagnosable: the number alone says the
+    // frame grew, not where. These are the four bracketed blocks a reference
+    // wake carries.
+    const content = await referenceWake();
+
+    expect(content).toContain('[Pod context:');
+    expect(content).toContain('[Trigger:');
+    expect(content).toContain('[Collaboration:');
+    expect(content).toContain('[Reply mechanics:');
+  });
+});
diff --git a/backend/__tests__/unit/services/agentMentionService.humansAreNotWoken.test.js b/backend/__tests__/unit/services/agentMentionService.humansAreNotWoken.test.js
new file mode 100644
index 000000000..061c95fa6
--- /dev/null
+++ b/backend/__tests__/unit/services/agentMentionService.humansAreNotWoken.test.js
@@ -0,0 +1,162 @@
+/**
+ * A human @handle routes nothing (TASK-070a / TASK-074, follow-up to #1244).
+ *
+ * #1244 added a paragraph to the pod-context frame telling every agent that a
+ * human's handle "is necessary and not sufficient — it flags the message in a
+ * mentions filter the human pulls; nothing pushes". That is an assertion about
+ * the kernel, made to a reader who cannot check it: an agent acts on the cue
+ * and has no view of `enqueueMentions`.
+ *
+ * #1244's three tests pin the SENTENCE (`toContain('nothing pushes')`). They go
+ * red if someone rewords the cue and stay green if someone makes it false. This
+ * file pins the other half: the behaviour the sentence describes.
+ *
+ * Two mechanisms, stated separately because they can break separately:
+ *
+ *  1. A handle that resolves to a human enqueues no `AgentEvent` of any type.
+ *     `enqueueMentions` has no human delivery branch at all — the handle is
+ *     filtered into `humanMentionHandles` and never reaches an enqueue. The
+ *     realistic future edit is TASK-070b (should a bare name route?): an
+ *     implementer who answers "yes, and push it" makes the cue a lie taught on
+ *     every wake, fleet-wide, and nothing in the suite objects.
+ *
+ *  2. Even the thread-follow half is narrower than the cue's readers assume:
+ *     `resolveHumanMentionUserIds` is called only inside `if (threadRootId)`,
+ *     so a plain channel post materialises no state for the mentioned human.
+ *     Hoisting that call out of the guard is the consistency fix that looks
+ *     correct and quietly widens what a handle does.
+ *
+ * Every negative here is paired with a control, per the house rule: an
+ * assertion that nothing was enqueued is worthless from a harness that cannot
+ * enqueue anything.
+ */
+
+jest.mock('../../../services/agentEventService', () => ({ enqueue: jest.fn() }));
+jest.mock('../../../models/AgentRegistry', () => ({
+  AgentInstallation: { find: jest.fn(), findOne: jest.fn() },
+}));
+jest.mock('../../../models/AgentProfile', () => ({ find: jest.fn() }));
+jest.mock('../../../models/Pod', () => ({ findById: jest.fn(), find: jest.fn() }));
+jest.mock('../../../models/User', () => ({ find: jest.fn(), findById: jest.fn() }));
+jest.mock('../../../services/chatSummarizerService', () => ({
+  constructor: { getLatestPodSummary: jest.fn() },
+  summarizePodMessages: jest.fn(),
+}));
+jest.mock('../../../models/AgentEvent', () => ({ countDocuments: jest.fn() }));
+jest.mock('../../../services/welcomeWakeService', () => ({ maybeFireWelcomeWake: jest.fn() }));
+jest.mock('../../../models/pg/Message', () => ({ findById: jest.fn(async () => null) }));
+jest.mock('../../../models/pg/ThreadUserState', () => ({
+  followByParticipation: jest.fn().mockResolvedValue(true),
+}));
+
+const AgentMentionService = require('../../../services/agentMentionService');
+const AgentEventService = require('../../../services/agentEventService');
+const { AgentInstallation } = require('../../../models/AgentRegistry');
+const AgentProfile = require('../../../models/AgentProfile');
+const Pod = require('../../../models/Pod');
+const User = require('../../../models/User');
+const AgentEvent = require('../../../models/AgentEvent');
+const ThreadUserState = require('../../../models/pg/ThreadUserState');
+
+const SEAT = { agentName: 'seat-a', instanceId: 'default', displayName: 'Seat A' };
+const HUMAN = { _id: 'user-sam', username: 'sam' };
+
+const mockInstallations = (installations) => {
+  AgentInstallation.find.mockReturnValue({ lean: jest.fn().mockResolvedValue(installations) });
+  AgentProfile.find.mockReturnValue({ lean: jest.fn().mockResolvedValue([]) });
+};
+
+// The human-handle resolver. MOCKED DELIBERATELY and load-bearing: `sam` is a
+// real non-bot row inside the pod, so anything that hands this handle to a
+// delivery path WILL find a user to deliver to. A harness where the lookup
+// returns nothing would pass every negative below for the wrong reason.
+const mockUserLookup = () => {
+  User.find.mockReturnValue({
+    select: jest.fn().mockReturnThis(),
+    lean: jest.fn().mockResolvedValue([HUMAN]),
+  });
+  User.findById.mockImplementation(() => ({
+    select: jest.fn().mockReturnThis(),
+    lean: jest.fn().mockResolvedValue({ _id: 'user-1', isBot: false }),
+  }));
+};
+
+const enqueued = () => AgentEventService.enqueue.mock.calls.map(([a]) => a);
+const mentions = () => enqueued().filter((a) => a.type === 'chat.mention');
+
+const send = (message, extra = {}) => AgentMentionService.enqueueMentions({
+  podId: 'pod-1', userId: 'user-1', username: 'alice', message, ...extra,
+});
+
+beforeEach(() => {
+  jest.clearAllMocks();
+  mockUserLookup();
+  Pod.findById.mockReturnValue({
+    select: jest.fn().mockReturnValue({
+      lean: jest.fn().mockResolvedValue({ type: 'chat', members: ['user-1', 'user-sam', 'bot-1'] }),
+    }),
+    lean: jest.fn().mockResolvedValue({
+      _id: 'pod-1', type: 'chat', members: ['user-1', 'user-sam', 'bot-1'],
+    }),
+  });
+  AgentEvent.countDocuments.mockResolvedValue(0);
+  mockInstallations([SEAT]);
+});
+
+describe('a human handle is not a delivery target', () => {
+  test('@handle for a human enqueues no event of any kind', async () => {
+    await send({ id: 'm-1', content: 'can you decide this @sam' });
+
+    // Not "no chat.mention" — no event at all. A future human-push branch is
+    // as likely to invent a type as to reuse this one.
+    expect(enqueued()).toHaveLength(0);
+  });
+
+  test('CONTROL: the same sentence addressed to an installed agent DOES enqueue', async () => {
+    await send({ id: 'm-2', content: 'can you decide this @seat-a' });
+
+    const got = mentions();
+    expect(got).toHaveLength(1);
+    expect(got[0]).toMatchObject({ agentName: 'seat-a', type: 'chat.mention' });
+  });
+
+  test('one message naming both routes to the agent only', async () => {
+    // The discriminating case. A human-push branch added beside the agent one
+    // leaves every single-target fixture above intact — it only shows up when
+    // both appear at once and the count stops being 1.
+    await send({ id: 'm-3', content: '@seat-a please answer, @sam to press' });
+
+    expect(enqueued()).toHaveLength(1);
+    expect(enqueued()[0]).toMatchObject({ agentName: 'seat-a' });
+  });
+});
+
+describe('the thread-follow half is guarded by threadRootId', () => {
+  test('a plain channel post materialises no thread state for the mentioned human', async () => {
+    await send({ id: 'm-4', content: 'over to you @sam' });
+
+    expect(ThreadUserState.followByParticipation).not.toHaveBeenCalled();
+    // And it declined to make the lookup at all, rather than making it and
+    // finding nobody — the guard is on the call, not on the result.
+    expect(User.find).not.toHaveBeenCalled();
+  });
+
+  test('CONTROL: the same message inside a thread DOES follow that human', async () => {
+    await send({
+      id: 'm-5', content: 'over to you @sam', thread_root_id: 101, threadRootId: 101,
+    });
+
+    expect(ThreadUserState.followByParticipation).toHaveBeenCalledWith(101, 'user-sam', 'pod-1');
+  });
+
+  test('a follow is not a wake — the threaded case still enqueues no event', async () => {
+    // Both halves of the cue's ceiling in one assertion: the handle bought a
+    // pull-surface row, and nothing was pushed.
+    await send({
+      id: 'm-6', content: 'over to you @sam', thread_root_id: 101, threadRootId: 101,
+    });
+
+    expect(ThreadUserState.followByParticipation).toHaveBeenCalled();
+    expect(enqueued()).toHaveLength(0);
+  });
+});