On Fri, 27 Feb 2026 at 05:49, Simon Schröder via Std-Proposals <
std-proposals_at_[hidden]> wrote:

>
>
> > On Feb 26, 2026, at 10:20 PM, Jens Maurer via Std-Proposals <
> std-proposals_at_[hidden]> wrote:
> >
> > You can select the seed value to get different sequences.
>
> But only a handful of seed values for the Mersenne Twister have been
> proven to generate true random sequences. So, don’t rely on seeding for
> true randomness (in areas where you want true randomness, e.g. Monte Carlo
> methods, you might want to have reproducibility anyways). Using random
> number generators is still hard to do correctly if you don’t know about
> them.


Which is why users should use the predefined typedefs for the engines, and
so enabling new specializations with very poor statistical properties
doesn't seem sensible.

Even allowing them to be instantiated with uint16_t seems like a mistake in
retrospect.

Using smaller sizes for distributions is OK though.



> The documentation of Tina RNG is a good read to understand RNGs.