I think it may have been someone else (7LM?) /joint decision that 30 mins was to err on the side of caution so I cant claim full credit for that.
iirc 7LM also found other documentation about BTs DLM but I cant recall now what that was, but in his credit he discovered a few things too which we were previously unaware of.
I'm flattered that you remember that, my own investigations into DLM were around 8 years ago and these days I have trouble remembering 8 days ago.
So far as I recall, what I found was some BT patents that described an implementation of a DLM. We might reasonably assume that they described BT's own DLM, but that is just an assumption.
These patents suggested that there might be different 'Penalties' for different degrees of instability. Crossing the threshold of error rates might get your target margin reduced by 3dB, but extreme instability, defined in these patents as 10 retrains in one hour, got you an immediate 6dB reduction.
The other interesting thing (to me) about these patents was there was an element of hysteresis. Once you had been penalised for instability, in order to remove that penalty, you had to demonstrate significantly greater stability than had existed before the penalty. Just reducing the error rate to previous levels wasn't enough, they had to be far before previous levels.
But that was all for the old 20cn DLM, I know very little at all about current DLM and above is probably of little relevance or interest, so I shall hush myself.