diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index fbde6cd2..d43233ed 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -3,12 +3,12 @@
 ### 1.0.0-rc.162
 This is an UNSTABLE release. It includes many experimental features, as well as many AI generated fixes. We recommend using rc.152 for production use.
 
-* CUDA builds made with a CUDA 12 toolkit now also contain device code for Volta, so the RHEL 8 packages and the portable Linux `.tgz` run on a V100; CUDA 13 dropped Volta, so the RHEL 9, Ubuntu and Windows artefacts remain Turing and newer.
+* CUDA 12 builds now contain device code for Volta, so the RHEL 8 packages and the portable Linux `.tgz` run on a V100; the CUDA 13 artefacts (RHEL 9, Ubuntu, Windows) remain Turing and newer.
 * Documentation: the supported GPU generations and the minimum NVIDIA driver version of every released artefact.
-* `bit_depth_readout` now reports the bit depth of the stored image rather than the detector's electronic readout depth, which is what NXmx readers expect - DIALS derives its masking markers from that field and cannot read a 32-bit image without it.
-* NXmx `underload_value` (lowest valid value) is now written, so a reader masks the error-pixel marker instead of treating it as an intensity.
-* A DECTRIS detector sending signed images is no longer declared unsigned in the outgoing stream and in HDF5.
-* rugnux: `_process.h5` describes the pixel format of the images it links to, instead of the signed 32-bit container rugnux processes in - its virtual dataset was typed `int32` over unsigned 16- or 32-bit data, so any reader going through it saw converted values.
+* HDF5 and image stream: `bit_depth_readout` now reports the bit depth of the stored image, not the detector's electronic readout depth.
+* HDF5: NXmx `underload_value`, the lowest valid pixel value, is now written.
+* A DECTRIS detector sending signed images is no longer declared unsigned in the image stream and in HDF5.
+* rugnux: `_process.h5` now describes the pixel format of the images it links to, instead of the container rugnux processes in.
 
 ### 1.0.0-rc.161
 This is an UNSTABLE release. It includes many experimental features, as well as many AI generated fixes. We recommend using rc.152 for production use.
diff --git a/docs/RUGNUX.md b/docs/RUGNUX.md
index 3d40cb57..85ec13f8 100644
--- a/docs/RUGNUX.md
+++ b/docs/RUGNUX.md
@@ -51,6 +51,10 @@ reusing them would hide the spot-finding settings from the lattice search.
   integration, azimuthal integration, per-image statistics). See
   [HDF5 / NeXus data format](HDF5.md) for the layout. Written by default only when **not** merging
   (i.e. under `--no-merge`); add `--write-process-h5` to also write it when merging.
+  It does not copy the images: `/entry/data/data` is a virtual dataset over the *input* files, so
+  the input has to stay where it was for the pictures to be readable, and the pixel metadata
+  (`bit_depth_readout`, `underload_value`, the dataset type) describes those files rather than the
+  signed 32-bit container rugnux processes in.
 - Merging is **on by default** (`--no-merge` disables it). The merged reflections are written in
   **three** formats — each has its uses downstream:
   - `<prefix>.mtz` — CCP4 MTZ (`IMEAN`/`I(+)`/`I(-)`, French–Wilson `F`, `FreeR_flag`) for the CCP4 /