We will build change that in the contract, instead of being calculated the winner the block it is called, it will calculate it from the 15th block after it start rolling (the last block users can bet) and block the contract untill the funtion to roll a winner it's called.
I belive in that way we will surpasse that flaw.

We can redeply the contract with that funtion in public, anyway the bot will call it as soon as possible so it would make users spend unnecessary gas.

again as stated, if for some reason the bot don't call the function, after the 15th block that can be called by any player that send a transaction with enough gas.

That's a prevenstion mesure if someone tries to send a gasfee higher than the bot has, it will make them calculate the winner but thats not how it regulary procedes, that just to avoid users to bet after the block hash is public. The funtion is called my the bot at the 14th block after atleast 2 players are in, this way it will be confirmed at 15th (or try to). under this sircunstances how can the outcome be predicted?

function endLottery() internal {
        uint sum = 0;
        uint winningNumber = uint(block.blockhash(block.number - 1)) % totalBets;

        for (uint i = 0; i < nbUsers; i++) {
            sum += users[i].amount;

            if (sum >= winningNumber) {
                // destroy this contract and send the balance to users[i]
                withrawWin(users[i].addPlayer,winningNumber);
                return;
            }
        }
    }

But PRNG is weak. So owner will get your money.

Allowing people to reduce their kelly multiplier for free re-enables 'whale dodging" behavior, which the dilution fee was designed to prevent.

If even the entire network is captured by one miner, then nothing will happen, since randomness in addition to the time of the block gives a hash function Keccak

The project looks good.I don't have enough time right now to check the algorithm for randomly selecting winners on your main page but as soon as I will,I'll update this thread with my feedback.Maybe you should do a couple of test draws with an example to prove the algorithm works as it's described in the code.
Watching this closely.

I don't see any. I think it's a very good idea. I think "ideally" the site should probably have 4 bankrolls, one for 0.5x, 1x, 2x and 4x.  (and then combine them into 1 "virtualized" bankroll).  And allow people to invest into any leveraged bankroll they want. I think combined with offsite, it would be a pretty ideal system.

I kind of ran out of steam at that point. I can't imagine that "real leverage" is better than "offsite investing" in every case. Shouldn't there be a trade-off?

Back to the example:

A has 10 onsite and 90 offsite.
C has 10 onsite with 10x "real" leverage.

Suppose a whale plays for a long period with 50% chance of winning each bet. All bets are the maximum, aiming to win 1% of the bankroll. The payout when he wins is 1.98x (1% house edge), and he wins and loses the same amount of bets (as he is expected to do, long term). Let's call the effective bankroll "B".

50% of the time the player wins 1% of the bankroll: B/100. When this happens, A's effective bankroll is multiplied by 99/100, and C's is multiplied by 90/100.

The other 50% of the time the player loses his stake. He's aiming to profit by B/100, with a payout multiplier of 1.98x, so he's risking and losing B/98. When this happens, A's effective bankroll is multiplied by 99/98, and C's is multiplied by 108/98.

Since the whale wins and loses the same number of bets, we can pair these bets. Each pair consists of one win and one loss.

For each pair, A's effective bankroll is multiplied by 99/100 and 99/98, for a net growth factor of 99/100 * 99/98 = 9801/9800 ~= 1.0001x

And C's effective bankroll is multiplied by 90/100 and 108/98, for a net growth factor of 90/100 * 108/98 = 9720 / 9800 ~= 0.9918x

And so we see that my intuition was correct, and that "real" leverage is worse than this "offsite investment" thing. The "offsite investment" has a (small but) positive expected bankroll growth whereas the "real leverage" expects to lose almost 1% of the bankroll for every pair of (1 win + 1 lose) bets.

We used this to calculate the house edge.
51 Numbers.

1 x 50.9
24 x 2.13
16 x 3.07
10 x 5.08

This gives 98.99019608%, so house edge becomes 1.00980392%.

Please let me know how did you get your digits?

Regards