simulations.org

Simulations

Statistics that I desire

• How many superusers, users, and lurkers are?
• How many in the signature?
• Difference in histogram of messages.
• Given a user with little messages, can you guarantee that all the messages are his?
• Given a user with a lot of messages, can you guarantee that it is a superuser?
• What about normal users?

On histogram attack

It has to be defined the 3 groups that users have, and a statistical way gather them.

1-9-90 Rule

The way is simple: count the occurences of each user in the list of messages, as well as the final list. This is already being done in the users.

Gather it by user and sort it by the number of messages. Then make the sets of users of each of them. They should be a range if left unchanged in the list messages, but we need to get it as there will be different simulations with a change of user bahaviors.

The idea is that after the signature, although super-user may remain superusers or users, all the super-users in the signature will not be close to superuser. For lurkers, if the average privacy is kept, then you may be able to analyze if they are lurkers, but a high anonymity may remain.

The design

Review needs a more porwerful approach, as it should gather more data per user. As such, it will be converted from user to a DTO with different fields:

• user id
• messages in list of messages, i.e. messages
• messages in signatures, i.e. signatures

Then both statistics can be explained. There should be one additional method, to gather the proportion of messages that have been sent by the user, as it needs to accuretely describe how the curve bends the encryption, making it more harder to statistically analyze the user patterns.

Design on Review

I feel like review may become a monstruous class that I fear. I know that doing this:

import dataclasses as dto

@dto.dataclass
class MyClass:
    a: int


def add_two(self):
    """
    Some help
    """
    self.a += 2

MyClass.add_two = add_two

c = MyClass(3)
c.add_two()
print(help(c.add_two))

Would definetely help, but it may be a little bit trickier. I don’t know if this should be the standard or offer a base object and functions that take said object.

Problems with parameters

There is a delicacy between which parameters should be used for this encryption system. There are different kind of attacks that can be done:

• Super-user attacks: how much certinety can you pick the super-users? How much privacy do they have?
• User attacks: how much certain are you to pick a user? how much anonymity do they have on average?
• Lurkers attacks: all lurkers should have more than x messages.

  There seems to be no good parameters to feed all of them. Maybe a different approach is necesserly.

PROJ The Random approach

No message-initial-ration weight is good to solve all the problems. Lurkers will be easy to spot with heavy-weight ratio, and super-users will be easy to spot with light-ratios.

The idea is that every message has a completely random ratio, maybe it would solve the problem, as lurkers will have with some messages good chances to be used and super-users will have more chances to appear quite normally.

Analysing first message in a window

How many probabilities have the first message in the window be the real message?

The idea is that maybe the window favors way too much that the first message that appears is the real one, and the subsequential of them is just the increased probabilities of that one. So, studying how it works would be a better way to check if this forum is secure enough.

How to compute it

0 arrays for the users. Add one for every new message, substract one when it leaves the window.

When a new window arrives, for every user in the window:

if it has appeared

skip

if the user has not appeared and its user has send the message

add one to broken and total

if the user has not appeared and has not send the message

add one to total

broken / total is the probability of a message being sent by it’s user in the window. To keep it comparable with the anonymity, we will use the inverse of the probability.

reviewer	anonymity	medium dev	mean	median	min	q0.25	q0.75	max	#messages	#signatures
normal	0	19.044009081392986	25.166891788486186	14.454545454545455	2.6823529411764704	7.0	36.0	134.0	5397	43176
window	0	23.90631659601147	39.713108471151486	33	2.5194805194805197	15	57.0	164.0	5397	43176

Results

Window is better, by far. Proceed to write thesis about that.