A five-day contact rule covered 69.6% and 40.3% of next-day contact time between departments in two office records

We tested group selection from past contacts in two public office records. Each target used the prior five nonempty days. The budget selected 10 of 92 people in 2013 and 24 of 232 people in 2015. A greedy rule selected people with the largest uncovered historical contact time between departments. The groups covered 69.6% and 40.3% of next-day external contact time, averaged over five target days per record. A contact counted as covered when at least one person was selected. Exact random groups had mean coverage of 20.7% and 19.7%. Random groups with the same department counts had mean coverage of 20.2% and 23.5%. Selection from the target day gave 98.1% and 91.6%. The result depended on the outcome. When each distinct pair counted once, total contact time performed better than the coverage rule in 2015. External contact time also performed better at the 5% budget in both records. These results do not establish a generally best rule. The contribution is a reproducible test set with fixed budgets, past-only selection, exact reference scores, and all selected groups. Independent scalar checks rebuilt all 40 primary selections and all 800 coverage scores. All four source files matched fresh HTTP checks. The two short records do not establish workplace-wide prediction, sensor performance, or disease prevention.

temporal networks · office contacts · department mixing · past-only prediction · group selection · contact duration · reproducible benchmark