diff --git a/include/aare/hist/PedestalTrackingPixelHistogram.hpp b/include/aare/hist/PedestalTrackingPixelHistogram.hpp
index 3e6c89c..c42b597 100644
--- a/include/aare/hist/PedestalTrackingPixelHistogram.hpp
+++ b/include/aare/hist/PedestalTrackingPixelHistogram.hpp
@@ -107,6 +107,7 @@ class PedestalTrackingPixelHistogram {
     void push_pedestal_no_update(const NDView<FrameType, 2> &frame);
     void update_mean();
     NDArray<AxisType, 2> pedestal_mean() const;
+    NDArray<AxisType, 2> pedestal_std() const;
 
     void fill_async(NDArray<FrameType, 2> &&image);
 
diff --git a/python/src/bind_PedestalTrackingPixelHistogram.hpp b/python/src/bind_PedestalTrackingPixelHistogram.hpp
index 13a7047..eedcbf3 100644
--- a/python/src/bind_PedestalTrackingPixelHistogram.hpp
+++ b/python/src/bind_PedestalTrackingPixelHistogram.hpp
@@ -119,6 +119,30 @@ void define_pedestal_tracking_pixel_histogram_bindings(py::module &m) {
                  containing the current cached pedestal mean.
              )")
 
+        .def(
+            "pedestal_std",
+            [](const PedestalTrackingPixelHistogram &self) {
+                // pedestal_std() flushes + locks + memcpys; do all of
+                // that without the GIL, only reacquire to wrap into a
+                // numpy array.
+                NDArray<PedestalTrackingPixelHistogram::AxisType, 2> *ptr =
+                    nullptr;
+                {
+                    py::gil_scoped_release release;
+                    ptr = new NDArray<PedestalTrackingPixelHistogram::AxisType,
+                                      2>(self.pedestal_std());
+                }
+                return return_image_data(ptr);
+            },
+            R"(
+             Snapshot the per-pixel pedestal std stitched together
+             from all shards.   
+
+             Returns:
+                 A 2D numpy array (rows x cols, dtype: float64)
+                 containing the current cached pedestal std.
+             )")
+
         .def(
             "fill_async",
             [](PedestalTrackingPixelHistogram &self,
diff --git a/src/hist/PedestalTrackingPixelHistogram.cpp b/src/hist/PedestalTrackingPixelHistogram.cpp
index f913e71..932b480 100644
--- a/src/hist/PedestalTrackingPixelHistogram.cpp
+++ b/src/hist/PedestalTrackingPixelHistogram.cpp
@@ -386,6 +386,37 @@ PedestalTrackingPixelHistogram::pedestal_mean() const {
     return data;
 }
 
+NDArray<PedestalTrackingPixelHistogram::AxisType, 2>
+PedestalTrackingPixelHistogram::pedestal_std() const {
+    // Drain in-flight async fills and serialise with all other fan-outs
+    // (Fill / PushPedestal / UpdateMean). m_std is overwritten wholesale
+    // by Pedestal::update_mean, so without the lock we could read torn
+    // rows mid-update.
+    flush();
+    std::lock_guard<std::mutex> lock(fill_mutex_);
+
+    NDArray<AxisType, 2> data(
+        {static_cast<ssize_t>(rows_), static_cast<ssize_t>(cols_)});
+
+    // Each partial pedestal stores its slice of m_std in C-order
+    // [local_rows x cols], identical in layout to the corresponding
+    // [first_row .. first_row + local_rows)[col] slice of `data`, so
+    // we can copy each shard with a single memcpy.
+    const size_t row_stride = static_cast<size_t>(cols_);
+    for (int t = 0; t < n_threads_; ++t) {
+        const auto first_row = static_cast<size_t>(row_start(t));
+        const auto local_rows = static_cast<size_t>(row_count(t));
+        if (local_rows == 0)
+            continue;
+
+        const auto view = partial_std_[t].view();
+        std::memcpy(data.data() + first_row * row_stride, view.data(),
+                    local_rows * row_stride * sizeof(AxisType));
+    }
+
+    return data;
+}
+
 void PedestalTrackingPixelHistogram::fill_with_threshold_batch_(
     std::vector<NDArray<FrameType, 2>> &batch) {
     // Called only by the coordinator thread on images already shape-checked