Selected CollegeMsg sessions have more one-message internal turns than a fixed-transition reference; Copenhagen SMS sessions have fewer

We tested the order of turns in two public message records. A turn is a run of messages from the same sender. We selected pair sessions with at least six messages, at least two turns in each direction, and some freedom to change turn lengths. Sessions ended at gaps above one hour. The reference fixed each session's event times, direction counts, endpoint directions, and all first-order direction-transition counts. It assigned equal probability to the permitted orders. Internal turns exclude the first and last turn. CollegeMsg had 521 selected sessions and 4,679 internal turns. The share with one message was 83.18%, compared with an exact reference mean of 82.01%. Copenhagen SMS had 874 sessions and 8,578 internal turns. Its share was 72.90%, compared with 74.51%. The differences were +1.17 and -1.62 percentage points. All seven fixed checks kept these signs. The CollegeMsg time halves differed: +1.60 points early and -1.89 points late. Both SMS halves were negative. Planned tests with 9,999 draws gave a Holm-adjusted reference probability of 0.0002 for each record. An added convolution checked the full discrete law. Independent counts, exhaustive short-sequence checks, and a fresh-download repeat passed. Prior work already found temporal reciprocity in these sources. This study adds a test that fixes sender changes as well. The results do not show one common direction of turn-length departure. They do not identify causal memory or a channel effect. The package preserves code, a proof, aggregate counts, reference draws, figures, source hashes, and repeat instructions. Source event bytes are excluded; CollegeMsg identity is limited to a pinned mirror.

human communication · turn taking · temporal networks · conditional Markov test · reciprocity · CollegeMsg · Copenhagen Networks Study · reference models