docs: link reproducible evaluation tools

README.md

@@ -43,6 +43,15 @@ Maintainers can reproduce the no-model hosted check with
 [`hosted_smoke_test.py`](hosted_smoke_test.py), and use
 [`TROUBLESHOOTING.md`](TROUBLESHOOTING.md) for failure-stage guidance.
 
+For quantitative work, use [`evaluate_depth.py`](evaluate_depth.py) and
+[`evaluate_depth_dataset.py`](evaluate_depth_dataset.py) for depth arrays,
+[`evaluate_pointcloud.py`](evaluate_pointcloud.py) and
+[`evaluate_pointcloud_dataset.py`](evaluate_pointcloud_dataset.py) for exported
+3D/Gaussian positions, and validate calibrated inputs with
+[`validate_capture_manifest.py`](validate_capture_manifest.py). The copyable
+manifest format is documented in
+[`CAPTURE_MANIFEST_SCHEMA.md`](CAPTURE_MANIFEST_SCHEMA.md).
+
 ## Important limitations
 
 This Space does not claim metric-scale reconstruction, learned Gaussian