Appendix C — Five Deployments Across the Randomization Transition (2019–2024)
The validation in Section B.3 uses data from 2019 and 2020, collected while the default-on transition was underway. This appendix records how the same quantities look under current operating systems: we recomputed the randomization statistics for three later deployments by the authors, the last collected when iOS 17 and Android 14 were current. Randomized addresses are identified by the locally administered bit of the MAC address (in 2024, by an equivalent firmware flag), and each probe-request record counts as one detection; the 2019 and 2020 figures were re-derived from the raw captures under the same definition.
| Deployment | Randomized detections1 | Randomized addresses2 | Non-randomized devices3 |
|---|---|---|---|
| UNIST campus, Ulsan (2019) | 4.6% | 92.4% | 176,555 |
| Commercial district near the University of Ulsan (2020) | 22.3% | 96.7% | 144,968 |
| Old-downtown commercial district, Ulsan (2022) | 60.7% | 98.1% | 13,030 |
| Pedestrian shopping street, Jeonju (2023) | 60.1% | 98.2% | 920 |
| Privately owned public space, Ulsan (2024) | 47.6% | 97.1% | 4,117 |
| 1 Share of probe-request records broadcasting a randomized address; each record counts as one detection. 2 Share of unique MAC addresses that are randomized. 3 Count of devices with non-randomized addresses; reflects each deployment's duration and extent (from 26 days to a 19-hour window) and is not comparable across sites. Counts are derived from the raw captures before the toolkit's preprocessing (stationary-device and minimum-detection filtering) and therefore exceed the analysis-dataset counts in Appendix A. Deployment details are given in Table C.2. | |||
These are opportunistic snapshots of different sites and populations, not a time series. Two regularities hold throughout. First, randomization saturated early at the address level: 92.4% of unique addresses were already randomized in 2019, and the share has stayed between 96% and 98% since. Independent longitudinal measurements report the same saturation, with the randomized share rising from 40% to 94% between 2016 and 2020 (Ribeiro et al. 2022). Second, every deployment retains non-randomized devices in meaningful numbers, from 176,555 over the 26-day campus deployment to 920 within a 19-hour window in Jeonju, consistent with laboratory evidence that many handsets in wide use still do not randomize effectively (Fenske et al. 2021). The detection-level share varies with site type and visitor mix (47.6% to 60.7% after the transition); the 2024 value is lower than in 2022 and 2023 because a public open space draws a different device population, not because randomization receded.
In the three later deployments, hourly counts of non-randomized devices continue to co-vary with the randomized group (median sensor-day \(r\) of 0.84 to 0.90 by deployment), reproducing the validation in Section B.3 under current operating systems. Capture windows of 19 to 26 hours make these correlations indicative rather than confirmatory.
C.1 The Five Sites
The five deployments span the settings in which passive WiFi sensing is most often proposed for pedestrian research, and their visitor populations differ accordingly (see the site descriptions below). These population differences are one reason the detection-level randomization shares in Table C.1 vary across sites.
| Panel | Site | Period | Sensors | Coordinates1 |
|---|---|---|---|---|
| (a) | UNIST campus, Ulsan | Oct–Nov 2019 (26 days) | 24 | 35.5743 N, 129.1892 E |
| (b) | Commercial district, Ulsan | Jul 2020 (9 days) | 17 | 35.5426 N, 129.2604 E |
| (c) | Old-downtown district, Ulsan | Nov 2022 (3 days) | 8 | 35.5560 N, 129.3215 E |
| (d) | Pedestrian street, Jeonju | Jul 2023 (19 hours) | 3 | 35.8208 N, 127.1449 E |
| (e) | Public open space, Ulsan | Apr 2024 (2 days) | 3 | 35.5400 N, 129.3263 E |
| 1 Centroid of each sensor array (WGS 84). Full site descriptions and street addresses are given below. Panel letters refer to Figure C.1. | ||||
(a) UNIST campus, Ulsan. A self-contained science-and-technology campus in a greenbelt on the western fringe of Ulsan, with on-campus dormitories and no adjacent residential district; its pedestrian population is almost entirely students and staff. 50 UNIST-gil, Eonyang-eup, Ulju-gun, Ulsan.
(b) Commercial district near the University of Ulsan. The university-front quarter of Mugeo-dong, where restaurants, cafés, and student housing line a designated pedestrian-priority street; students and young residents dominate. 8 Daehak-ro 94beon-gil, Nam-gu, Ulsan.
(c) Old-downtown commercial district, Ulsan. The main commercial street of Ulsan’s original center in Jung-gu (the Seongnam-dong area), a traditional retail core undergoing publicly funded regeneration near the newly opened Ulsan Art Museum; it draws a general urban public. 277-1 Okgyo-dong, Jung-gu, Ulsan (legal-lot address).
(d) Pedestrian shopping street, Jeonju. Runs beside the Jeonju Gaeksa, a Joseon-era state guesthouse in Jeonju’s historic center; revitalized since the mid-2010s and locally nicknamed “Gaekridan-gil,” it is strongly visitor- and tourist-oriented. 40-1 Jeonjugaeksa 3-gil, Wansan-gu, Jeonju.
(e) Privately owned public space, Ulsan. A small plaza at the base of a high-rise mixed-use building on the Beonyeong-ro arterial in Nam-gu, Ulsan, provided under the Korean Building Act, which requires large developments to dedicate part of their lot as publicly accessible open space (gonggae gongji); its users are chiefly nearby residents and passers-by. 165 Beonyeong-ro, Nam-gu, Ulsan.
Addresses are romanized Korean road-name addresses (legal-lot address for site c).
What these patterns imply for each of the five metrics is discussed in Section B.4.