First, watch Garmin’s YouTube videos. There are just a few and they aren’t long or very detailed but they give you a decent overview. But in short you are correct about no car data inputs. There is also no built-in information about tracks other than the approximate location of start/finish lines. Uploads to Garmin Connect at this time are of little use. They are a highly condensed 1 hz GPS path with speed, nothing more. Essentially the same data for people using their other gear hiking, biking etc.
Data analysis is limited to the screens and stats provided by the tablet and the raw data is not made available for export other than the above mentioned condensed GPS track. Viewed within the system the video does include speed, running lap time and best lap but that’s about it. You can pull the full session video and separate stitched together “optimal lap” video off the SD card but they do not include any overlay. (Yes, this is a common bitching point along with no OBD or CAN inputs, data export, etc.) The simple answer to why is that they add nothing to making you faster. This is a VERY focused single-minded version 1.0 of a genuinely new device. I admire them for that but honestly, not having an instant upload to YouTube, FB ... with the basic overlay was marketing malpractice. But put that distraction aside.
What’s not obvious is that the AI claims aren’t just marketing BS. The core of what makes this truly different is the camera (which also holds the accelerometers) which must be carefully aligned during setup. That camera supplies a view of the track so the AI can look for the edges of the track surface (think self-driving car tech). As you do your initial laps it uses that input to draw the actual track shape as well as plotting your path on it. Other systems draw your trajectory and may be able to superimpose that on a map or aerial of the track so you can tweak their alignment, but you need to be really careful trusting any representation of your line relative to the pavement. In theory, this learns/draws the track surface and your lateral position on it using the camera, and the GPS provides your progress and speed along that path. With that information the AI knows whether you apex early or late, or miss by 5ft one lap compared to another, etc.
But, it doesn’t know what’s possible until you do it and won’t assume T3 has the same grip as T2, or anything else for that matter. In theory, if you do every lap exactly the same way every time, it will never recommend a change. But nobody does that and those with the most potential to improve do it less. As you do laps it keeps track of what line, what brake points, rate and duration work best then presents that analysis to you in simple live voice prompts during a session and in up to three “Opportunities” for improvement in post-session review. It also provides some consistency stats and their version of theoretical best lap which they call True Optimal lap. The name is easily misconstrued but what makes it different is that the AI works out which fastest segments really can be accomplished in succession and blends them to come up with that best-case lap time and stitches together a video with the segments they have used. Other systems simply add fastest sectors which may not even be achievable as a full lap.
That’s actually a very short and incomplete summary. Watch the videos, read the FB group posts but keep in mind most the people are wrong about a lot so pay attention to the one Garmin guy who sometimes chimes in.