Matching department counts reduced the next-day coverage gain from prior-day contact time selection in two office data sets

We tested selection by prior-day contact time in two French office data sets from 2013 and 2015. Each supplied ten workdays and nine test transitions. We selected the ceiling of 10% of people observed on the training day. The sets covered 23.00% and 28.14% of next-day pair contact time. Uniform random means were 16.19% and 16.68%. Random sets with the same department counts covered 19.07% and 19.43%. Department matching thus reduced the apparent gains by 2.88 and 2.75 percentage points. A diagnostic also matched selected counts by next-day observed status within departments. Its means were 20.41% and 20.18%. The remaining gaps were 2.59 and 7.97 points. The 2013 result was inside the matched random policy range; the 2015 result was above it. The ranges describe random selection on fixed records. They are not population confidence intervals. At 10% selection, neither deployment showed between-department time coverage above its matched 95% random range. Score and set-size checks gave different patterns across deployments. The study adds exact random-reference code and a reproducible coverage benchmark. It does not establish infection prevention or general use in other offices. The package contains original CC0 inputs, source hashes, code, all selections, results, two figures, validation, and repeat instructions.

human contact networks · temporal prediction · contact coverage · department structure · random reference · reproducible benchmark · SocioPatterns