I'm just following up this post to add another graph to the data I reported in the first post.
Where "tenbyboy2" had G.INP retransmission happening in full flow, with some blocks being retransmitted more than once, there is obviously a knock-on impact to latency for *some* packets.
That impacts can be seen in a TBB BQM - where the yellow "max latency" bar is thicker by perhaps 5ms from 10pm through to 3pm. I guess that tells us the maximum additional latency that G.INP introduces.