How it all works, in a nutshell The web app get a request and decides what user it's for, and what type of resource it's for. The web application loads the user's style for that resource and then loads each layer for that style, unless it's already loaded. (layers are selectively cached) The S2 system builds a context which contains the property values set by the top-most layer for each property, as well as a function vtable computed the same way. Next the web application invokes the global prop_init() function which the layout layer might've implemented. The job of this function is to initialize the constructional properties. Constructional properties are properties which the web app looks at to influence its subsequent data structure population. For example, FotoBilder examines the value of $*gallery_max_page_size to determine how many thumbnails to load and return for a gallery. However, a layout wouldn't always want to directly expose this property to the users. What if the layout was a grid and the user wanted a page size that was a prime number? How do you make an even grid out of 37 pictures? This sort of layout should ask the user for the number of rows and columns they want, then compute $*gallery_max_page_size in prop_init() by multiplying $*rows and $*cols. The web app looks at the current context and determines both the version number of the core layer and the values of all the constructional properties, then builds an appropriate data-structure for the resource requested. This instantiated object is likely pretty deep, containing dozens of instantiated objects all hanging off it. The web app invokes the entry point method (probably called ->print()) on the object it just instantiated. From here, S2 code runs, printing what it wants, and invoking methods on other objects in its huge data structure which may also print.