When Conflicting Reports Collide

The WILD Attack's decisive variable is not simply whether a provider receives a false report. It is what the provider does when that report collides with legitimate observations for the same BSSID. Comparing Google WPS, Apple WPS, A-Map, and WiGLE on the same dimensions—how they classify spatial disagreement, how they update state, and what users experience—shows why no single number called “accuracy” captures the security of a location database.

Nearby agreement, distant conflict

A provider has to distinguish ordinary measurement variation from a report that points a stationary access point to a different place. That distinction cannot be made from the identifier alone. It depends on the spatial relationship among reports and on assumptions about what an access point is: a fixed landmark, a mobile hotspot, or a signal that has been misread.

The WILD study tested this boundary by submitting reports for the same set of previously unused BSSIDs from two locations while keeping the reporting procedures otherwise comparable. When the locations were close enough, the reports behaved as if they were collaborative evidence. In Google's experiment, two reporting sites 183 meters apart caused five test entries to appear in about eight days, compared with roughly 13 days when reports came from either site alone. The resulting coordinates fell between the two sites. In that range, adding a second location increased the amount of evidence the provider treated as mutually compatible.

When the sites were moved farther apart, the behavior changed. At a separation of 392 meters in the same study, the Google entries took about 23 days to appear, and their recorded coordinates aligned with one site or the other rather than landing between them. The provider was no longer treating the observations as a single spatial cluster. It was treating them as competing accounts of where the BSSID belonged.

The numbers illustrate a design boundary, not a universal radius. The authors use approximately 200 meters as an example threshold for Google, but they explicitly did not attempt to establish the exact thresholds enforced by each provider. Conditions such as signal coverage, reporting location, provider policy, and the access point's history can affect the result. More importantly, the study does not claim that Apple, A-Map, and WiGLE classify distance exactly as Google does.

The distinction still matters even without a universal cutoff. If reports from nearby places are averaged or jointly counted, a falsified observation may strengthen the update rather than challenge it. If reports from separated places are treated as competing, the provider has to choose among at least three broad responses: favor one side, reject both sides, or let the record move as more reports arrive. That decision is where the WILD Attack's data-quality problem becomes a policy problem.

The word “conflict” also describes a change in the provider's reasoning, not necessarily an immediate error. A provider may initially create a record from one location and later receive reports from elsewhere. It may invalidate the old record before accepting a new one, or it may keep a blended estimate. The visible outcome depends on the sequence of state transitions as well as on the final coordinate. The same false input can therefore produce different timing and different symptoms even when the BSSID and reporting locations are the same.

Spatial classification is a proxy for a harder question: could these observations plausibly describe the same stationary access point?