Overview of the Microsoft driver measures
Microsoft distributes thousands of drivers via the Windows Update service, servicing millions of machines and users each month. Safely delivering the right driver at scale requires evaluating driver quality through real-world verification during distribution.
This document is a reference to publishers and authors of Windows device drivers. Publishers and authors can better understand the criteria Microsoft uses in evaluating driver quality during the driver flighting process. Becoming familiar with the driver quality criteria will help driver publishers understand how Microsoft reached a decision about releasing their driver.
Keywords in bold have corresponding definitions in the glossary.
This content contains three sections:
- Using measures: defines what measure are, the types of measures, and how measures evaluate quality.
- Driver measure attributes: defines the various attributes that each measure has.
- Driver measures dictionary: provides a definition for each driver measure, whether systemic or device class, with a description, attribute values, and calculation logic.
Microsoft defines a measure as a quantifiable metric to gauge the quality of products delivered by the company. Driver measures aggregate telemetry produced by customer machines, processing any events that are related to a driver. Each measure is scoped to a use case of the driver’s functions, ensuring that the end user can experience the component’s capabilities.
Types of measures
To evaluate the quality of drivers, Microsoft has two distinct types of measures: systemic measures and device-class measures.
Systemic Measures ensure that a driver installs without error and the machine continues to be reliable; Microsoft applies these measures to every driver submitted. Device-Class measures monitor specific capabilities of the driver to ensure the hardware component behaves as intended; each Device-Class has a set of distinct measures applied or only uses Systemic Measures for evaluation.
All drivers submitted to Microsoft Approval undergo systemic quality evaluation. Systemic measures evaluate quality and the status of the machine without needing to understand the specific functionality of the driver. The current systemic measures monitor the success of the driver installation and the reliability of the machine. Driver installation measures monitor the success of installation within the audience and detect any post-installation errors.
When a partner submits a driver to Microsoft, the driver is associated with a device-class that indicates which component the driver is for. Each device-class has a distinct set of measures used to evaluate a driver’s behavior on the component or only use Systemic measures for evaluation.
How measures assess driver quality
Each measure has its own calculation logic, which is an algorithm that parses telemetry for driver-related events and aggregates the results into a percent, ratio, or histogram of failures & successes. This result is the measure’s current value; the current value is evaluated against a minimum bar of quality, known as the measure’s passing criteria.
A measure is failing when its Current Value does not meet its Passing Criteria, triggering an investigation that may result in remediation, such as a flight rejection or an in-market expiration.
Evaluating by Targeting Cohort
A driver can be developed to support multiple systems and devices. It is not always sufficient or accurate to evaluate a driver’s quality aggregating its measure results across all of its targeting devices (See definition of targeting cohorts below). To ensure there is not a low performing targeting cohort, we analyze the cohorts to find any that fail to meet the measure requirements. All driver measures will be used to evaluate driver quality by targeting cohorts, as they are enabled to support evaluation by targeting cohorts. Refer to each measure definition page for the new measure attribute 'cohort-capable'. If a measure is marked as 'cohort-capable', the measure is enabled to support evaluation by targeting cohorts.
Targeting Cohorts/Clusters Definition
A targeting cohort/cluster is defined as a set of Windows systems and devices that a shipping label specifies and shares the same targeting attributes, including HWID, CHID, and OS version.
Cohort Evaluation Pass/Fail Criteria
If one or more driver measures fail to meet its passing criteria, the targeting cohort does not pass (it’s fails). A driver can be rejected if one or more failures is detected on one or more targeting cohorts. The minimum instances for cohorts will be added to the measure definition page upon enablement.
Data sources for measures
To evaluate driver quality, measures incorporate data from machines running in two distinct customer groups: Windows Insider Program (WIP) and Retail.
WIP data is vital to flighting scenarios, as users have opted-in to providing Microsoft with increased levels of telemetry for use in real-world verification. Retail data is collected from the general Windows Ecosystem and allows Microsoft to monitor quality issues on released drivers.
Count differences between measures
Microsoft constructs each measure differently, with a unique calculation logic, set of attributes, sampling percentages, and evaluation criteria. As a result, a set of measures applied to a distinct driver can have inconsistent counts reported; Microsoft expects these discrepancies.