diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 16543d5b3..3abbcce00 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -7,6 +7,8 @@ See the CHANGELOG rule in [AGENTS.md](../AGENTS.md).
 
 - Kit 14fa729c (core 2.6.0) is the scaffold, tooling and catalog pin, so a new product's bot registry
   takes a text handler and inline callbacks besides commands.
+- `POST /api/runs/{id}/start` stamps `started_at` on the first accepted start, so deploy and QA runs
+  show a duration on the Journey timeline instead of "not recorded".
 
 ## 2026-10-16
 
diff --git a/services/api/src/routers/runs.py b/services/api/src/routers/runs.py
index 86ae74342..c6d3c95a5 100644
--- a/services/api/src/routers/runs.py
+++ b/services/api/src/routers/runs.py
@@ -957,6 +957,8 @@ async def start_run(
 
     Starting a run that is already running is the same answer, so a worker
     retrying after a lost response is not refused its own start.
+
+    The first accepted start stamps `started_at`; a refused start never does.
     """
     run, fenced = await _lock_launch_context(run_id, db)
     if fenced or run.status in _TERMINAL_RUN_STATUSES:
@@ -965,6 +967,8 @@ async def start_run(
         return DeployRunStart(run_id=run_id, started=False, run_status=RunStatus(run.status))
 
     run.status = RunStatus.RUNNING.value
+    if run.started_at is None:
+        run.started_at = datetime.now(UTC)
     await db.commit()
     return DeployRunStart(run_id=run_id, started=True, run_status=RunStatus.RUNNING)
 
diff --git a/services/api/tests/service/test_admin_console.py b/services/api/tests/service/test_admin_console.py
index ce77d9896..204c2fff3 100644
--- a/services/api/tests/service/test_admin_console.py
+++ b/services/api/tests/service/test_admin_console.py
@@ -6,8 +6,9 @@
 import pytest
 from sqlalchemy import delete
 
-from shared.contracts.dto.run_result import DeployRunResult
+from shared.contracts.dto.run_result import DeployRunResult, QARunResult
 from shared.contracts.queues.deploy import DeployOutcome
+from shared.contracts.queues.qa import QAOutcome
 from shared.models import (
     Application,
     Deployment,
@@ -272,6 +273,40 @@ async def test_journey_detail_lays_out_the_story_and_its_passport(async_client,
     ]
 
 
+async def test_a_started_and_completed_qa_run_shows_its_duration_on_the_journey(
+    async_client, db_session, product
+):
+    run_id = f"qa-{uuid.uuid4().hex[:8]}"
+    db_session.add(
+        Run(
+            id=run_id,
+            type="qa",
+            status="queued",
+            project_id=product["project"].id,
+            story_id=product["story"].id,
+        )
+    )
+    await db_session.commit()
+
+    started = await async_client.post(f"/api/runs/{run_id}/start")
+    assert started.json()["started"] is True
+    completed = await async_client.patch(
+        f"/api/runs/{run_id}",
+        json={
+            "status": "completed",
+            "result": QARunResult(qa_outcome=QAOutcome.PASSED).model_dump(mode="json"),
+        },
+    )
+    assert completed.status_code == 200, completed.text
+
+    response = await async_client.get(f"/api/admin/v2/journeys/{product['story'].id}")
+    assert response.status_code == 200, response.text
+    [qa] = [s for s in response.json()["timeline"]["stages"] if s["kind"] == "qa"]
+    assert (qa["label"], qa["state"]) == ("QA 1", "done")
+    assert qa["duration_seconds"] is not None
+    assert qa["duration_seconds"] >= 0
+
+
 MODULE_PR = "https://github.com/vladmesh/codegen-product-kit/pull/42"
 
 
diff --git a/services/api/tests/service/test_deploy_dispatch_boundary.py b/services/api/tests/service/test_deploy_dispatch_boundary.py
index 4727f1ca5..ef3eb5fe8 100644
--- a/services/api/tests/service/test_deploy_dispatch_boundary.py
+++ b/services/api/tests/service/test_deploy_dispatch_boundary.py
@@ -83,6 +83,7 @@ async def test_a_cancelled_run_cannot_be_started(async_client: AsyncClient):
     run = await async_client.get(f"/api/runs/{run_id}")
     assert run.json()["status"] == "cancelled"
     assert run.json()["completed_at"] is not None
+    assert run.json()["started_at"] is None
     claimed = await async_client.post(f"/api/runs/{run_id}/dispatch-claim")
     assert claimed.json()["granted"] is False
 
@@ -106,12 +107,26 @@ async def test_starting_a_live_run_twice_is_the_same_answer(async_client: AsyncC
     run_id = await _deploy_run(async_client, run_status="queued")
 
     first = await async_client.post(f"/api/runs/{run_id}/start")
+    first_stamp = (await async_client.get(f"/api/runs/{run_id}")).json()["started_at"]
     second = await async_client.post(f"/api/runs/{run_id}/start")
 
     assert first.json()["started"] is True
     assert second.json()["started"] is True
     run = await async_client.get(f"/api/runs/{run_id}")
     assert run.json()["status"] == "running"
+    assert run.json()["started_at"] == first_stamp
+
+
+@pytest.mark.asyncio
+async def test_the_first_start_records_when_the_run_started(async_client: AsyncClient):
+    run_id = await _deploy_run(async_client, run_status="queued")
+    assert (await async_client.get(f"/api/runs/{run_id}")).json()["started_at"] is None
+
+    before = datetime.now(UTC)
+    await async_client.post(f"/api/runs/{run_id}/start")
+
+    started_at = (await async_client.get(f"/api/runs/{run_id}")).json()["started_at"]
+    assert before <= datetime.fromisoformat(started_at) <= datetime.now(UTC)
 
 
 @pytest.mark.asyncio