I can't give you a reference, but it looks like Moser has written:
$$\lambda =\sum_{n=1}^{\infty}\frac{p_n}{10^{n(n+1)/2}}$$ and thus encoded the primes in the decimal digits.
Then defining:
$$D_n= 10^{n(n+1)/2}\lambda\\
C_n=\lfloor D_n\rfloor$$
We have: $$\begin{align} C_n&=p_n+10^{n}p_{n-1}+10^{n+n-1}p_{n-2}+\cdots +10^{n+(n-1)+\cdots+2}p_1\\&= p_n+10^nC_{n-1} \end{align} $$
So $p_n=C_n-10^nC_{n-1}.$
Without the floor functions, the expression is, of course, zero. We can replace the floor function with integer rounding, which will amount to the same thing since $2p_n<10^n.$
We could do this for any sequence of positive integers with $a_n<10^n-1.$ We'd get a $\lambda$ which encodes $\{a_n\}.$
This is one of those things which is showing how much information can be stuffed into a real number. It really shows nothing about the primes.
There is nothing particularly important about $f(n)=10^{n(n+1)/2},$ other than the fact that $f(n)/f(n-1)$ is always an integer and $f(0)=1$ and $p_n+1<\frac{f(n)}{f(n-1)}.$ Under those circumstances, if we define $\lambda =\sum_n \frac{p_n}{f(n)},$ we get:
$$p_n=\lfloor f(n)\lambda\rfloor -\frac{f(n)}{f(n-1)}\lfloor f(n-1)\lambda\rfloor$$