Topic: How does latency stack in a chain of MADI devices?

I'm setting up a device chain like this:

UFX+ > UFX+ > M1610Pro > M32 AD PRO > M32 DA > (then back to the master UFX+)

How is sample latency going to work here? Does order matter?
(For example, does the MADI output from the slave UFX+ go directly to the Master UFX+ or does it have to go through all other devices first?)

If each devices adds 4 samples of latency, then how do they stack? (I'm waiting for a cable to arrive to integrate the M1610 and do measurements of the whole system).

Re: How does latency stack in a chain of MADI devices?

I'm still waiting for the correct cabling to be able to use the M1610 pro, but I decided to take readings with my current set up to see if I could make some kind of sense of the sample inaccuracies. The current chain is:

UFX+1 > UFX+2 > M32 AD Pro > M32 DA > (back to UFX+1)

Using Reapers "ReaInsert" plugging, which will send a ping to detect round trip latency, I was able to gather these latency readings (Which are added after the reported latency of 1105 samples from the master UFX+)

            UFX+1 > UFX+1:  0 samples
            UFX+1 > UFX+2: 14 samples
            UFX+1 > M32AD: -4 samples
            UFX+2 > UFX+1: 14 samples
            UFX+2 > UFX+2: 20 samples
            UFX+2 > M32AD:  4 samples

            M32DA > UFX+1: 56 samples
            M32DA > UFX+2: 70 samples
            M32DA > M32AD: 53 samples

Can someone explain why I'm getting these variations?

Why is signal coming from the UFX+1 going into the M32AD ahead of a signal coming UFX+1 back into itself?
Why is the UFX+2 coming in to the M32AD sooner than it comes back in to the UFX+1?
Is the MADI chain cumulative?
Is each unit adding a different amount of latency in each direction?