RNG Audit

This repository contains scripts enabling anyone to quickly audit the random number generator (RNG) used in the Verifiable Draws project.

It is separated in 2 parts:

• Audit of the Chainlink VRF random numbers that Verifiable Draws uses as a source of randomness
• Audit of the observed drawing outcomes

For each part we collect a huge sample of randomness which can then be manually analyzed by simply looking at the data and realizing that it looks random. Alternatively, you can perform a more in-depth analysis by using the collected data as inputs for the Diehard tests to precisely measure the quality of the random number generator.

Setup

Clone this repository on your local machine:

git clone https://github.com/lancelot-c/rng-audit.git

Empirical testing on Chainlink VRF

This is the testing of the raw randomness received from Chainlink VRF requests.

A smart contract is needed to generate VRF randomness on-demand and make it available for export. Such a contract is available at src/OnlyRaw.sol and is deployed for your convenience at 0xBBCD0c8DBdC112dd29af8c57Ee8740bD9feE084B. The variable wordsCounter is the number of random values that the contract has already generated. At the time of writing, the contract holds a total of 3,125,210 random words, which we consider enough to perform a relevant analysis.

[Optional] - Generate additional VRF randomness

If you need more randomness, the easiest way is to registrer a time-based Upkeep for this contract on the function makeVrfRequest(110) with the CRON expression */1 * * * * for an execution every minute. Make sure the VRF subscription for this contract is sufficiently funded, otherwise add funds to the subscription with your own wallet: Connect Wallet > Actions dropdown > Fund subscription.

ℹ️ Even though Chainlink says you can generate a maximum number of 500 random words per VRF request, it turns out in practice that 110 is the maximum you can ask for (at least on Arbitrum Sepolia), any value greater than that will return an error.

Export the VRF randomness

Now we want to export all this raw randomness out of the smart contract.

The OnlyRaw script is doing precisely that:

cd scripts/OnlyRaw

Set the variables in the .env depending on what values you want to export, for example if you want the first 1,000,000 values:

START_AT=1
HOW_MANY=1000000

Then run the script:

npm install
node OnlyRaw.js

The output file is located at script-outputs/OnlyRaw-<START_AT>-<HOW_MANY>.txt.

If you need additionnal values, let's say 5,000,000 more values, you can run the same script again but this time by ignoring the values that you've already exported:

START_AT=1000001
HOW_MANY=5000000

ℹ️ Verifiable Draws is using 64-bit numbers as a source of randomness, however each random word received from Chainlink is 256-bit so each random word is splitted into 4 numbers of 64-bit each and these 64-bit numbers are the ones being exported. This means that when you are exporting 1,000,000 values with this script, you are actually only exporting 250,000 random words from Chainlink. It is important to know this when setting the value for the HOW_MANY variable. The maximum value of HOW_MANY is therefore the value of wordsCounter multiplied by 4.

Parallelization

For faster execution, you can run several instances of this script in parallel with different .env values.

Empirical testing on observed drawing outcomes

We would like to collect the following 20 datasets:

Range	Positions	Replacement	Draws
2	1	N/A	1,000,000
17	5	No	20,000,000
31	12	No	50,000,000
57	20	No	50,000,000
105	7	No	100,000,000
194	30	No	100,000,000
358	20	No	100,000,000
660	40	No	50,000,000
1,217	50	No	50,000,000
2,243	10	No	200,000,000
4,135	12	No	200,000,000
7,622	4	No	500,000,000
14,050	9	No	200,000,000
25,899	75	No	20,000,000
47,742	10	No	200,000,000
65,536	5	No	500,000,000
72,859	10	No	200,000,000
80,989	25	No	50,000,000
90,007	15	No	100,000,000
100,000	100	No	10,000,000

This is roughly 32 billion samples total. Now, this is quite a lot of data to collect so instead of using Chainlink VRF as a source of randomness we can use fuzz testing which is a built-in feature of Foundry.

Install Foundry by running the following command in your terminal, then follow the onscreen instructions:

curl -L https://foundry.paradigm.xyz | bash

Install the project dependencies:

cd rng-audit
forge install lancelot-c/verifiable-draws
forge install smartcontractkit/chainlink
forge install OpenZeppelin/openzeppelin-contracts

We will run the test file OnlyOutcomes.t.sol which contains an equivalent of the checkDrawWinners function of the Verifiable Draws smart contract adapted for fuzz testing.

Make sure your .env has the desired dataset parameters, for example if you want to collect dataset #2 your .env should have:

RANGE=17
POSITIONS=5
FOUNDRY_FUZZ_RUNS=20000000

Then run:

forge test

The output file is located at script-outputs/OnlyOutcomes-<RANGE>-<POSITIONS>-<FOUNDRY_FUZZ_RUNS>.txt.

On a regular consumer laptop the dataset #1 should take approximatively 90 seconds to generate.

Parallelization

For faster execution, each dataset can be generated in parallel if you launch several instances of this test on different machines with different .env values.

Getting Help

Our Discord is the best place to ask for help.