Could you please detail in what part of the spec it describes the exact way of doing point 2)?
Also in what part of the spec it says what part of the content get lost in the final alt attribute? (in this case “uri1”)
In my opinion your interpretation is one possible interpretation of different ones that would be not opposed to the spec.
My proposal is to either remove this example from the spec or either to detail in much more precision all the steps in order to calculate the “alt” text content.