Sign In
Create Account
0
Basically, the current project will to make the whole client more ''Human-like'' just like when we train yourself. Starting off with mouse clicks and logging different information for randomization. I will post information when #002 is posted in a few days/weeks.
Quote
Whenever you click a mouse button there is a small delay between the time your kernel registers the press and release events. Applications see this as the press duration or p.d. - not to be confused with potential difference in electrical circuits, although the abbreviation is aptly named since your mouse and keyboard operate on a matrix circuit (which on a side-node is why you can't press more than 3 keys on non-specialised keyboards at the same time).
Vulcan and Salvation discovered a pattern to the p.d.s after clicking in rapid succession. They hypothesised this was due to the nature of the hardware.
To put things into perspective, if you sample the p.d.s from the bot right now and plot a grouped histogram you would see something like this:
After several hundreds clicks in a few hour session your p.d. data will produce an unrealistic distribution which can be flagged up as a bot.
Introducing Heteromouse
The new mouse implementation I am working on to replace Timer's Homomouse uses a simple physics engine for hardware simulation.
We make the assumption that the better the hardware the lower the p.d., but since a linear gradient is too unnatural I use the following function:
Which visually produces:
To obtain the mean p.d. we integrate over the curve:
Where: k = quasi-randomised constant; ε = our matrix of adjustment factors; x = environmental hardware metrics.
We also need a standard deviation:
Then finally, we can randomise p.d.s about a normal distribution for more human like values:
A positive skew is then added to go a step further than Gaussian randomisation. For any number of clicks the end result would be almost identical to the same number of clicks you would do through manual playing with your physical mouse, eliminating at least one method of detection analysis.
Vulcan and Salvation discovered a pattern to the p.d.s after clicking in rapid succession. They hypothesised this was due to the nature of the hardware.
To put things into perspective, if you sample the p.d.s from the bot right now and plot a grouped histogram you would see something like this:
After several hundreds clicks in a few hour session your p.d. data will produce an unrealistic distribution which can be flagged up as a bot.
Introducing Heteromouse
The new mouse implementation I am working on to replace Timer's Homomouse uses a simple physics engine for hardware simulation.
We make the assumption that the better the hardware the lower the p.d., but since a linear gradient is too unnatural I use the following function:
Which visually produces:
To obtain the mean p.d. we integrate over the curve:
Where: k = quasi-randomised constant; ε = our matrix of adjustment factors; x = environmental hardware metrics.
We also need a standard deviation:
Then finally, we can randomise p.d.s about a normal distribution for more human like values:
A positive skew is then added to go a step further than Gaussian randomisation. For any number of clicks the end result would be almost identical to the same number of clicks you would do through manual playing with your physical mouse, eliminating at least one method of detection analysis.
Paris
Managing Director
Back to top
.png)
