I have been thinking about what this specification can do to minimize fingerprinting. Highly-fingerprint-able values are both unique and stable; when working with numeric values, adding jitter (to prevent stability) and rounding (to prevent uniqueness) are good ideas. We currently name, but do not explicitly recommend jitter and rounding, and do not get into any details.
The only current implementation (Chromium) does not currently apply any jitter or rounding.
The amount of jitter/rounding that will effectively mitigate fingerprinting without ruining the utility of the hints seems to be somewhat hint-specific, requiring some knowledge of both the current user-agent landscape (to understand how unique/stable the resulting values are likely to be) and current use cases (to understand the loss in utility).
recommend some specific (and hint-specific) methods for how jitter and rounding should be applied, based on modeling of the current device landscape, usage patterns, and knowledge of current use cases.
I have been thinking about what this specification can do to minimize fingerprinting. Highly-fingerprint-able values are both unique and stable; when working with numeric values, adding jitter (to prevent stability) and rounding (to prevent uniqueness) are good ideas. We currently name, but do not explicitly recommend jitter and rounding, and do not get into any details.
The only current implementation (Chromium) does not currently apply any jitter or rounding.
The amount of jitter/rounding that will effectively mitigate fingerprinting without ruining the utility of the hints seems to be somewhat hint-specific, requiring some knowledge of both the current user-agent landscape (to understand how unique/stable the resulting values are likely to be) and current use cases (to understand the loss in utility).
So! I think it might be useful to: