I'd class this website as good if you want to make a visual impact, but giving subtly bad advice if you want to visualise data.
Looking at the selection for proportions [0] there are things like circle packing or bubble maps presented as options. These are terrible ways to encode visual data if the intent is for the audience to actually take an interest in the data itself. Humans are really bad at interpreting the relative size of areas, especially circles. They really only make sense if the intent is to wow the audience that you have data and design skills, then everyone moves on without asking tough questions. Fair enough, sometimes that is the right thing to do, but it isn't a good way to visualise data. Use bars.
I came across another one a while back I wish I had bookmarked, it was old style html just from a greybeard dataviz expert, and it was basically an exhaustive description of what to use in what situations, and it was chock full of advice that was both counterintuitive and obvious in hindsight... what was great about it is it focused on the data type first, and only recommended the viz styles at the end.
Not sure about the website, but the guy was probably Edward Tufte from the description. He and Stephen Few were kind of the OGs of data visualization for the modern era.
Ads were way too aggressive on iOS, pop up videos interfered with the use. I have a genuine interest in the topic but the ads just made my trust dissipate under a minute.
It'd be a better site as a single page listing all the charts types (like seaborn docs) but then you wouldn't have to click past two ads to see what a box-and-whisker plot looks like
i disagree that this is better because the OP is about why you would choose a given viz and what it actually expresses usefully, where this example is just pictures people can fit their data to for slides.
direction matters. most sankey diagrams could be pie charts (eg the birthday sankey) and dumb usages discredit useful viz techniques.
nested boxes are usually trees, and mixed nested boxes with lines without symmetry are almost always lazy reasoning or bullshit. start with symmetry and intent and it will be apparent.
Temporal graph visualization is quite lacking. It probably depends what one time slice of your network looks like.
If you have 10 to 20 nodes without a ton of edges you could use a fixed circular layout for the nodes on a timeline that just the edges change over time.
More than this and you start getting into hairball territory even without the changes over time. A hairball changing over time is even more useless than a hairball on its own.
The standard force directed layout is really quite useless other than seeing the global structure of the graph. The uselessness is more obvious when you try to change it over time. I suspect the layout is so standard because the visualization looks so cool.
Most data visualization though has this same problem. There is almost this property that the cooler and more beautiful the visualization is, the more useless it is as far as containing any insight about the data itself.
Picture musical notation, 5 horizontal lines where the y-axis is the note (higher on the staff, higher note) and the x-axis is position in time.
Say you have 5 event types, which occur with various frequencies over time. Plot them as dots on the staff.
Draw lines between events which link to each other.
If you draw the lines for the staff, they should be faint, only to help classify the event.
And you can of course have more than 5 event types, just add more lines. Hopefully less than 20 lines, otherwise this starts to get visually very messy.
For print/static graphic? An Ishikawa diagram paired with a, probably keyed, network diagrams(s). For video or interactive there would be more flexibility. Your delivery method ultimately makes many choices for you.
Looking at the selection for proportions [0] there are things like circle packing or bubble maps presented as options. These are terrible ways to encode visual data if the intent is for the audience to actually take an interest in the data itself. Humans are really bad at interpreting the relative size of areas, especially circles. They really only make sense if the intent is to wow the audience that you have data and design skills, then everyone moves on without asking tough questions. Fair enough, sometimes that is the right thing to do, but it isn't a good way to visualise data. Use bars.
[0] https://datavizcatalogue.com/search/proportions.html