1 in 7 Devices Loses Its Data Configuration Every Month. Here Is the Evidence.

Twelve months of device-level telemetry from roughly 26,000 monthly reporting Android devices in Mexico, and what it reveals about the configuration layer no probe can see.

Mobile networks are monitored from the infrastructure side: probes, drive tests, crowdsourced speed samples. All of them stop at the radio link. None of them can see the layer where connectivity actually succeeds or fails for a subscriber: the device's carrier configuration. APN profiles, IMS registration settings, VoLTE state, network selection mode. When that layer breaks, the network looks healthy and the customer has no data.

Between July 2025 and June 2026, SDKCX, our on-device agent operating with Android Carrier Privileges, recorded every carrier configuration change event on an average of approximately 26,000 monthly reporting Android devices in Mexico. This report covers what that layer looks like when you can finally measure it. We define configuration drift as the loss or alteration of working carrier configuration after a device's initial provisioning. All figures below are steady-state: each device's first seven days are excluded, so onboarding activity does not inflate drift. How often devices arrive misconfigured in the first place is measured separately.

Finding 1: APN configuration loss is chronic, not incidental

Over twelve months we recorded 281,832 steady-state changes to the internet APN configuration state. In a typical month, around 3,600 devices, roughly one in seven monthly reporting devices, experienced at least one. The median affected device lost it once; the 90th percentile device up to ten times. Critically, 78 percent of all APN configuration events occur after the first week of a device's life. This is not setup noise. It is configured, working devices losing their data configuration in the field, continuously, invisible to any infrastructure-side monitoring.

Finding 2: most losses are corrected before anyone calls support

Because SDKCX both detects and remediates, and every configuration write is attributed in device telemetry, we can measure the full loop with no inference. Across the same twelve months we observed 67,628 loss-to-restore sequences across 7,129 distinct devices over the year, where a device lost its internet APN configuration and SDKCX's deterministic remediation engine, operating through Carrier Privileges, restored it. The median time from loss to restore was 12 seconds, with 78.5 percent of losses corrected within five minutes, 90.4 percent within an hour, and 96 percent within 24 hours. Users fixed the configuration themselves in about 6 percent of restore events; the rest never needed to know anything was wrong. The remediation is rule-based and auditable end to end: the engine validates device state against the expected operator profile and applies defined corrections. No AI sits in that decision path.

Finding 3: voice configuration drifts silently

Roughly 2,100 devices per month, about 8 percent of monthly reporting devices, change VoLTE state after initial setup, and about 1,900 change their IMS APN configuration. Subscribers do not report this as a settings problem; they report degraded call quality or failed calls, and the network-side investigation finds nothing. Worth stating plainly: the majority of VoLTE configuration events, 59 percent, happen during onboarding, which is a finding in itself. Out-of-box voice configuration is incomplete or incorrect on a large share of enrolled devices before drift even begins.

Finding 4: instability concentrates in a small chronic population

Manual network selection changes involve only about 600 devices in a typical month, but those devices are intense: a median of 3 changes each, with the 90th percentile at 47. Ninety-four percent of this activity is steady-state. The concentration is extreme: the top 10 percent of these devices generate 66 percent of all manual network selection changes. A small, identifiable population of chronically unstable devices generates a disproportionate share of configuration instability, which means targeted remediation of a few devices removes a large share of the problem.

About the data

All figures come from deduplicated, validated device telemetry. Events with no actual state change were excluded, rapid repeated reports were collapsed into single events, anomalous reporting cohorts were removed, and each device's first seven days were excluded to separate provisioning from drift. Percentages are expressed against devices reporting at least one configuration event in the month, a conservative lower bound of the active fleet.

Why this data exists at all

Every dataset in this report comes from measurement at the only layer where carrier configuration lives: the device itself. SDKCX operates with Android Carrier Privileges, the mechanism that allows an operator-signed application to read and correct carrier configuration state. Probe-based and crowdsourced tools, whatever their scale, end where this data begins.