diff --git a/lectures/user-experience/slides.qmd b/lectures/user-experience/slides.qmd
index fce6d2e702a580e9a273b8efc8ef237575340a05..5c14b3232ade592fa050a4caecaede33557f49e2 100644
--- a/lectures/user-experience/slides.qmd
+++ b/lectures/user-experience/slides.qmd
@@ -249,7 +249,9 @@ A result is there in
 
 * the data model behind the interaction design
 
-# Example for UX design: redesigning ICON output
+# redesigning ICON output
+*(as a UX design example)*
+
 ## 1. User research: what are typical plots that scientists make?
 
 ![Analysis of ICON output typically loads way too much data](static/icon_plot_examples.png){width="60%"}
@@ -276,12 +278,12 @@ $ ls *.nc | wc -l
 * fast selection in time and space dimension
 
 ```python
-ds = cat.ICONrun.to_dask()
-t_hamburg = ds.tas.sel(...
+ds = cat.ICONrunID.to_dask()
+t_hamburg = ds.tas.sel(...)
 ```
 
 ## 2. Interaction design / Usability
-*new coherent output as a single dataset*
+*build on existing user knowledge: a single **dataset***
 ![](static/icon_new_output.png){width="60%"}
 
 ::: {.notes}
@@ -291,14 +293,15 @@ t_hamburg = ds.tas.sel(...
 
 * bad example: WALES flight altitude is time-dependent, but was a constant reference height. Bad variable name and misleading info
 * -> the structure and type of the data / array holds important information
-* best practice: limit only to the absolute necessary structure/info. 
+* best practice: limit only to the absolute necessary structure/info.
 * e.g. a dataset with u, v arrays is easy to understand. If there is also ws (redundant), it is unclear whether this is exactly the same as (u**2 + v**2)**0.5
 :::
 
 ## 3. Information architecture
-* chunked dataset
-* on HEALPix grid
+* dataset holds pieces together
+* chunks increase performance
 * hierarchical dataset ("denormalized" to increase efficiency)
+* HEALPix grid supports chunks & hierarchy<br/>(but users had to learn something new)
 
 :::{.samller}
 *Further info on redesigned ICON output in [Tobi's EGU24 talk](https://tobi.pages.gwdg.de/egu2024/slides.html)*