Gaining access to high quality randomness on-chain requires a solution like Chainlink's VRF, but it also requires you to understand some of the ways that miners or validators can potentially manipulate randomness generation. Here are some of the top security considerations you should review in your project.

• Use requestId to match randomness requests with their fulfillment in order
• Choose a safe block confirmation time, which will vary between blockchains
• Do not re-request randomness, even if you don't get an answer right away
• Don't accept bids/bets/inputs after you have made a randomness request
• The fulfillRandomWords function must not revert
• Use VRFConsumerBaseV2 in your contract to interact with the VRF service

Use requestId to match randomness requests with their fulfillment in order

If your contract could have multiple VRF requests in flight simultaneously, you must ensure that the order in which the VRF fulfillments arrive cannot be used to manipulate your contract's user-significant behavior.

Blockchain miners/validators can control the order in which your requests appear on-chain, and hence the order in which your contract responds to them.

For example, if you made randomness requests A, B, C in short succession, there is no guarantee that the associated randomness fulfillments will also be in order A, B, C. The randomness fulfillments might just as well arrive at your contract in order C, A, B or any other order.

We recommend using the requestID to match randomness requests with their corresponding fulfillments.

Choose a safe block confirmation time, which will vary between blockchains

In principle, miners/validators of your underlying blockchain could rewrite the chain's history to put a randomness request from your contract into a different block, which would result in a different VRF output. Note that this does not enable a miner to determine the random value in advance. It only enables them to get a fresh random value that might or might not be to their advantage. By way of analogy, they can only re-roll the dice, not predetermine or predict which side it will land on.

You must choose an appropriate confirmation time for the randomness requests you make. Confirmation time is how many blocks the VRF service waits before writing a fulfillment to the chain to make potential rewrite attacks unprofitable in the context of your application and its value-at-risk.

On Ethereum, rewrites are very expensive due to the very high rate of work performed by Ethereum's proof-of-work. The hashrate of the Ethereum network is currently 630 trillion hashes per second, and any attacker would have to control at least 51% of that for the duration of the attack. Therefore, major centralized exchanges consider a 20-block confirmation time as highly secure for deposit confirmation times. The block confirmation time required from one use case to the next may differ.

Do not re-request randomness, even if you don't get an answer right away

Doing so would give the VRF service provider the option to withhold a VRF fulfillment, if it doesn't like the outcome, and wait for the re-request in the hopes that it gets a better outcome, similar to the considerations with block confirmation time.

Don't accept bids/bets/inputs after you have made a randomness request

Consider the example of a contract that mints a random NFT in response to a user's actions.

The contract should:

1. Record whatever actions of the user may affect the generated NFT.
2. Stop accepting further user actions that might affect the generated NFT and issue a randomness request.
3. On randomness fulfillment, mint the NFT.

Generally speaking, whenever an outcome in your contract depends on some user-supplied inputs and randomness, the contract should not accept any additional user-supplied inputs after it submits the randomness request.

Otherwise, the cryptoeconomic security properties may be violated by an attacker that can rewrite the chain.

fulfillRandomWords must not revert

If your fulfillRandomWords() implementation reverts, the VRF service will not attempt to call it a second time. Make sure your contract logic does not revert. Consider simply storing the randomness and taking more complex follow-on actions in separate contract calls made by you, your users, or a keeper.

Use VRFConsumerBaseV2 in your contract, to interact with the VRF service

VRFConsumerBaseV2 includes a check to ensure the randomness is fulfilled by VRFCoordinatorV2. For this reason, it is a best practice to inherit from VRFConsumerBaseV2. Similarly, don't override rawFulfillRandomness.