IPLD is a system for understanding and working with data.
Firstly, we define a "Data Model" which says what the domain and range of data is. This is important because its the foundation of everything else we will build.
(Broadly, we can say the Data Model is "like JSON", and you've probably got the right idea -- maps, strings, lists, etc.)
Thereafter, we define "Codecs", which say how it can be parsed from serial messages and emitted as serial messages.
IPLD has lots of Codecs. You can choose to use different codecs based on what other applications you'd like to interface with, or just based on what fits the performance vs human-readability that you prefer for your own applications.
A key part of IPLD is its ability to link together documents.
IPLD linking isn't like some other forms of linking, like URLs (which refer to "locations" of data) -- instead, it's based on content-addressing (which means refering to data by a hash of its content). IPLD uses a format called CIDs for this, to be specific.
What's neat about content-addressing is that because it does not involve talking about data location, it's inherently friendly to decentralization. This means large graphs of documents can be linked together, and once you have some of the document graph, you don't need to go look online to some specific server to get the rest of the linked documents; you can get them anywhere that content can be found.
Content-addressing also separates document identity from discussion of authority. This again contributes to decentralization-friendly systems: once you get part of a document graph, and have decided it's the document you're looking for, you can get all the related documents without having to bounce through some other system to re-determine what the authoratitive document is for every related document -- you already know what that is, because you already have its content ID.
Then we provide a couple other ways to handle data via the Data Model: Schemas, which can describe the structure of data, and be used for validation, detecting structure, and some kinds of basic data transformation; and Advanced Data Layouts, which let us do things like assemble complex data structures to be presented as simpler ones (so you can work on them "like basic Data Model", even if they have more power, such as sharded, or encryption, or etc).(click to enlarge)
For the next step up in level of details, you can continue with The Brief Primer.
Otherwise, you can jump anywhere throughout the docs and specs using the nav menu at the left. (If you're on a mobile screen, tap the menu icon at the top left of the page to open the nav menu.)