At a 10% pair budget, filtered counts gave higher mean next-week bin coverage and lower pair recall for signals at least -80 dBm in one student network

We tested whether a signal filter improves coverage of future proximity observations at a fixed pair budget. We used the public four-week Copenhagen Networks Study Bluetooth record. Three forecasts used one week to select 10% of its observed pairs for the next week. One score counted all input observations. The other counted only signals at least -80 dBm. For target signals at least -80 dBm, the filter increased mean bin coverage from 59.65% to 60.82%. It reduced mean recall of distinct target pairs from 18.11% to 17.68%. All three weekly tests had these directions. Mean person coverage also increased. With all target signals included, bin coverage instead fell from 45.73% to 40.35%. A less strict -90 dBm input filter increased both means for the -80 dBm target. Threshold, budget and availability checks show the tested scope. Score ties use exact expected inclusion weights. An independent source scan checked the main counts, and a complete rebuild checked the numerical results. Prior work already showed that stronger Bluetooth ties recur more often. This contribution adds a reproducible fixed-budget benchmark that separates observation volume from pair identity. Five-minute samples do not prove continuous contact or a distance. One student network, missing observations and shared people limit wider inference. The result does not establish a social cause or an effect on infection. The package preserves source bytes, provenance, code, all results, figures and repeat instructions.

Bluetooth · proximity networks · contact prediction · signal threshold · pair recall · Copenhagen Networks Study · reproducible benchmark