Chronic Devices: The 10 Percent That Cause Most of the Instability

A small population of devices generates most configuration instability, stays unstable month after month, suffers more configuration loss across the board, and looks completely ordinary in any inventory system. Profiling the chronic device.

In our measurements of configuration drift, one pattern kept returning: instability is not evenly distributed. The top 10 percent of devices generate 66 percent of all manual network selection changes. This report asks the follow-up questions: are they the same devices over time, do they suffer other problems too, and can they be identified in advance. We define a chronic device as one that remains in the top decile of configuration instability across consecutive months.

Finding 1: chronic is the right word

Month over month, between 40 and 83 percent of top-decile devices, typically more than half, remain in the top decile the following month. Instability is not a passing event that visits random devices. It is a persistent condition concentrated in a small, largely stable population.

Finding 2: instability travels across parameters

Devices in the chronic cohort are not just unstable in one dimension. Compared with the rest of the fleet, they are more likely to experience internet APN configuration loss (57.2 percent vs 45.2 percent incidence) and suffer twice the average loss volume (12.5 vs 6.3 losses per device over twelve months). A device that churns network selection is disproportionately a device that also loses its data configuration. Instability is a device condition, not a parameter condition.

Finding 3: they look completely ordinary on paper

Here is the part that matters operationally. The chronic cohort is statistically indistinguishable from the rest of the fleet on every inventory attribute we compared: Carrier Privileges coverage (98 percent vs 97 percent), device age (about 300 days vs 270), Android version (median 15 vs 14). No CRM field, no device database, no procurement record identifies them. The only way to find a chronic device is to observe its behavior, continuously, at the device layer.

Why this matters

If most instability comes from a small, persistent, identifiable-only-by-behavior population, the operational conclusion is direct: fleet-wide averages hide the problem, and targeted intervention on a few hundred devices removes a disproportionate share of configuration failures, support contacts, and degraded experience. Finding those devices requires measuring where the instability lives.

About the data

The chronic cohort comprises 442 devices, the top decile by steady-state manual network selection changes between July 2025 and June 2026, after deduplication, removal of no-op events, collapse of sub-minute repeats, and exclusion of implausible reporting cohorts. Persistence is measured as month-over-month top-decile retention. Comparison metrics use identically cleaned data for both cohorts.