diff --git a/datasets/concordia/README.md b/datasets/concordia/README.md
index dbcb203..e53b6e2 100644
--- a/datasets/concordia/README.md
+++ b/datasets/concordia/README.md
@@ -21,7 +21,7 @@ size_categories:
 
 ![B-mode reconstructions of the five SynthUS-FSA phantom classes](assets/classes.png)
 
-*The five phantom classes (anechoic, hypoechoic, hyperechoic, diverse, point-target), each reconstructed from `data/raw_data`: [`image_0005`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0005.hdf5), [`image_0470`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0470.hdf5), [`image_0640`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0640.hdf5), [`image_1100`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_1100.hdf5), [`image_1808`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_1808.hdf5).*
+*The five phantom classes (anechoic, hypoechoic, hyperechoic, diverse, point-target), one capture per column: [`image_0012`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0012.hdf5), [`image_0410`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0410.hdf5), [`image_0733`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0733.hdf5), [`image_0753`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_0753.hdf5), [`image_1808`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/concordia/data/image_1808.hdf5). Rows show the reference B-mode (`data/image`), the simulated scatterer cloud (`data/scatterers`) and the label map (`data/segmentation` for the lesion classes, `data/diverse_source_image` for diverse; point-target captures store none).*
 
 ## Dataset Description
 
diff --git a/datasets/concordia/assets/classes.png b/datasets/concordia/assets/classes.png
index 2048116..67fe09b 100644
Binary files a/datasets/concordia/assets/classes.png and b/datasets/concordia/assets/classes.png differ
diff --git a/datasets/concordia/assets/reference_capture.png b/datasets/concordia/assets/reference_capture.png
index a426cce..9bd38b5 100644
Binary files a/datasets/concordia/assets/reference_capture.png and b/datasets/concordia/assets/reference_capture.png differ
diff --git a/datasets/concordia/reconstruct.py b/datasets/concordia/reconstruct.py
index 6fbd5c1..faf5864 100644
--- a/datasets/concordia/reconstruct.py
+++ b/datasets/concordia/reconstruct.py
@@ -59,7 +59,7 @@
 # --- Inputs -----------------------------------------------------------------
 # Defaults stream straight from the published corpus. Swap any of these for a
 # local path to run against your own copy.
-ZEA_FILE = "hf://nvidia/OpenH-RF/concordia/data/image_1808.hdf5"
+ZEA_FILE = "hf://nvidia/OpenH-RF/concordia/data/image_0753.hdf5"
 FRAME = 0  # Frame index (default: 0)
 OUT = HERE / "assets" / "reference_capture.png"
 SAVE_YAML = None  # Optionally write the pipeline to a reusable pipeline.yaml