Hourly recipient shuffles keep most message repeats within five minutes for 82 student senders over four weeks

We tested repeated recipients in four weeks of public Copenhagen student text message logs. We selected 82 senders with at least 20 consecutive outgoing message pairs at gaps of 1 to 300 seconds and at least two recipients. The mean repeat fraction was 94.86%, with equal weight for each sender. A shuffle that kept event times and recipient counts within each sender and hour had an exact mean of 93.51%. The excess was 1.35 percentage points. None of 9,999 hourly shuffles reached the observed score; the one-sided probability estimate was 0.0001 under uniform conditional shuffling. A shuffle over all four weeks gave an excess of 29.60 points. Thus, hourly constraints reduced the excess by 95.45%. Only 17.76% of pairs had a variable outcome under hourly shuffling. A small remaining excess does not show that social memory is absent. The local counts can themselves reflect conversations. These counts also use later events and do not form a real-time forecast. The call data had too few eligible senders for a channel comparison. The package preserves original data, hashes, code, exact formulas, shuffle scores, all fixed checks, figures, and reproduction steps.

temporal networks · text messages · recipient repetition · conditional randomization · Copenhagen Networks Study · reference models