diff --git a/.github/workflows/pull_request.yaml b/.github/workflows/pull_request.yaml
index 0b8e9d645..3055afa29 100644
--- a/.github/workflows/pull_request.yaml
+++ b/.github/workflows/pull_request.yaml
@@ -32,6 +32,7 @@ jobs:
     runs-on: ubuntu-latest
     env:
       HUGGING_FACE_TOKEN: ${{ secrets.HUGGING_FACE_TOKEN }}
+      PYTHONUNBUFFERED: "1"
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
diff --git a/.github/workflows/push.yaml b/.github/workflows/push.yaml
index 5ca271170..84e774cc9 100644
--- a/.github/workflows/push.yaml
+++ b/.github/workflows/push.yaml
@@ -35,6 +35,7 @@ jobs:
       id-token: "write"
     env:
       HUGGING_FACE_TOKEN: ${{ secrets.HUGGING_FACE_TOKEN }}
+      PYTHONUNBUFFERED: "1"
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
diff --git a/changelog.d/ci-progress-heartbeats.fixed.md b/changelog.d/ci-progress-heartbeats.fixed.md
new file mode 100644
index 000000000..5d15ff48b
--- /dev/null
+++ b/changelog.d/ci-progress-heartbeats.fixed.md
@@ -0,0 +1 @@
+Make long-running dataset builds emit plain CI heartbeat logs so release workflows are less likely to die silently during calibration.
diff --git a/policyengine_uk_data/tests/test_progress.py b/policyengine_uk_data/tests/test_progress.py
new file mode 100644
index 000000000..25c655ba0
--- /dev/null
+++ b/policyengine_uk_data/tests/test_progress.py
@@ -0,0 +1,38 @@
+from policyengine_uk_data.utils.progress import ProcessingProgress
+
+
+def test_track_dataset_creation_logs_in_ci(monkeypatch, capsys):
+    monkeypatch.setenv("GITHUB_ACTIONS", "true")
+
+    progress = ProcessingProgress()
+
+    with progress.track_dataset_creation(["Build base", "Save final"]) as (
+        update_dataset,
+        nested_progress,
+    ):
+        assert nested_progress is None
+        update_dataset("Build base", "processing")
+        update_dataset("Build base", "completed")
+        update_dataset("Save final", "processing")
+        update_dataset("Save final", "completed")
+
+    output = capsys.readouterr().out
+    assert "[dataset] starting: Build base" in output
+    assert "[dataset] completed (1/2): Build base" in output
+    assert "[dataset] completed (2/2): Save final" in output
+
+
+def test_track_calibration_logs_heartbeats_in_ci(monkeypatch, capsys):
+    monkeypatch.setenv("CI", "true")
+
+    progress = ProcessingProgress()
+
+    with progress.track_calibration(12) as update_calibration:
+        for iteration in range(1, 13):
+            update_calibration(iteration, calculating_loss=True)
+            update_calibration(iteration, loss_value=iteration / 10)
+
+    output = capsys.readouterr().out
+    assert "[calibration] epoch 1/12: calculating loss" in output
+    assert "[calibration] epoch 10/12: loss=1.000000" in output
+    assert "[calibration] epoch 12/12: loss=1.200000" in output
diff --git a/policyengine_uk_data/utils/progress.py b/policyengine_uk_data/utils/progress.py
index e6a70f89d..539cd0657 100644
--- a/policyengine_uk_data/utils/progress.py
+++ b/policyengine_uk_data/utils/progress.py
@@ -6,8 +6,8 @@
 """
 
 from contextlib import contextmanager
+import os
 from typing import Any, Dict, List, Optional, Union
-import time
 
 from rich.console import Console
 from rich.progress import (
@@ -187,6 +187,12 @@ def __init__(self, console: Optional[Console] = None):
         """
         self.console = console or Console()
         self.progress_manager: Optional[RichProgress] = None
+        self._plain_output = (
+            os.environ.get("GITHUB_ACTIONS") == "true" or os.environ.get("CI") == "true"
+        )
+
+    def _emit(self, message: str):
+        print(message, flush=True)
 
     @contextmanager
     def track_dataset_creation(self, datasets: List[str]):
@@ -198,6 +204,23 @@ def track_dataset_creation(self, datasets: List[str]):
         Yields:
             Tuple of (update_dataset function, progress manager for nested tasks).
         """
+        if self._plain_output:
+            completed_count = 0
+
+            def update_dataset(dataset_name: str, status: str = "processing"):
+                nonlocal completed_count
+
+                if status == "processing":
+                    self._emit(f"[dataset] starting: {dataset_name}")
+                elif status == "completed":
+                    completed_count += 1
+                    self._emit(
+                        f"[dataset] completed ({completed_count}/{len(datasets)}): {dataset_name}"
+                    )
+
+            yield update_dataset, None
+            return
+
         with create_progress(self.console) as progress:
             # Main dataset creation progress
             main_task = progress.add_task(
@@ -265,6 +288,27 @@ def track_calibration(self, iterations: int, nested_progress=None):
         Yields:
             Function to update calibration progress.
         """
+        if self._plain_output:
+
+            def update_calibration(
+                iteration: int,
+                loss_value: Optional[float] = None,
+                calculating_loss: bool = False,
+            ):
+                if calculating_loss:
+                    self._emit(
+                        f"[calibration] epoch {iteration}/{iterations}: calculating loss"
+                    )
+                elif loss_value is not None and (
+                    iteration == 1 or iteration == iterations or iteration % 10 == 0
+                ):
+                    self._emit(
+                        f"[calibration] epoch {iteration}/{iterations}: loss={loss_value:.6f}"
+                    )
+
+            yield update_calibration
+            return
+
         if nested_progress:
             # Add calibration as a nested task in existing progress
             calibration_task = nested_progress.add_task(