Anyone interested in helping with a new random number generator?

[ad1mt]

I would use MWC instead of LCG generators.

I've been thinking about your suggestion while looking at the MWC algorithm. I'm puzzled about why you think its better for cryptography.
It reveals more of its internal state in the output. The internal state does not have many variables. [EDIT] And the values that the variables are given is restricted to safe primes.
These things make it easier to brute-force.
I have 18 variables, which can take any odd number value, all of which take different values every time an encryption is performed; and these values will likely never be used more than once.
One advantage of MWC is that it is significantly faster. Another disadvantage of my algorithm might be that other algorithms probably take up less storage that 18 * 4 bytes.
I developed my idea as an experiment, and although I think it is fairly strong, I didn't expect anyone else to agree with me.

[Thaddy]

MWC is significantly faster than LCG

No, both Nitorami and I already explained MWC's are an implementation of an LCG, one of many. (See the wiki for the Marsalgia's again)
In Nitorami's proposal the state comes from the different parameters. Basically what you wanted in the first place, only with a more solid basis.
BTW: @Nitorami, 32 bit days. Will add a small 128 bit based version of a 64 bit MWC later. But it is about the principle.

[ad1mt]

Here's my simple password manager, in case anyone is interested...
   https://mark-taylor.me.uk/downloads/EncView_Source.zip

[ADMGNS]

hello

i suggest, you can use double generators from existing ones to produce a new generator that it works totally random but not reversible (ie, non-decryptable)

good luck

[Thaddy]

Sigh, we all knew that and is also shown in the discussion. You add nothing new.
(It is also wrong in the context of the question)

[ad1mt]

MWC's are an implementation of an LCG, one of many. (See the wiki for the Marsalgia's again)

Are you sure this is correct?
I looked at the MWC code: it did a 32bit-modulo operation; used the lower 32bits as the output; then added the high 32bits to the lower 32bits as the seed for the next round.
That seems significantly different to me.

[ad1mt]

In this thread, there have been lots of suggestions to use a different algorithm. But no reasons given as why my algorithm is bad, or why the alternatives are better.
I believe my algorithm has many significant advantages, which I think make it an interesting new idea.

[Thaddy]

Well, as we already discussed: if it passes most standard tests you are good.