Something based on cosmic ray strikes? There are optical plastics and photo muliplier rigs that have been used for decades ...

But like anything tuning the device that gathers the data and to generate the range of randomness is an issue.

Watch what happens with the CAT system on the days that you use the vacuum cleaner...

And how many randoms a second will be needed - of course simply scale isnt always so simpel a room full of CATS they tend to look at each other alot restricting how 'random' their head movements will be.

Random is hard.

Something that is statistically random can still have signals embedded in it. You might think the cat is randomly looking around, but they could have some patterns that are not easily detected.

All PRNG engines exhibit patterns under analysis and sometimes under regular use. Visible patterns emerge especially when you use them at particular frequencies that interact with their internal algorithms. When programmers first started using PRNGs to display simulated television static the results were often unusable because of the clear patterns that appeared. The search is for something that is good enough.

Even things like nuclear decay are both stochastic (on the individual atom) yet also constant (on average). Radiation sensors are common in gambling because of the very low probability of someone tampering with them or predicting them, even though over the long term they will act within a specific known distribution. Gamblers are unable to guess when the next winning hand will be, but regulation boards can determine the overall frequency of events.

Any time you need a source of randomness, you need to figure out how random is random enough. The randomness of cats is more accurately the randomness of sensor noise, which can be very non-random.