Time is Money�Strategic Timing Games in PoS

1

Barnabé Monnot

Robust Incentives Group (RIG), Ethereum Foundation

CryptoEconDay Paris, 2023

Short intro

Research scientist at the Robust Incentives Group,�an Ethereum Foundation research team for mechanism design

Joined EF in 2020, since then work on:

• EIP-1559
• Proof-of-Stake consensus
• MEV / Proposer-Builder Separation
• Collaborations with grantees, events etc.

2

Outline

1. What are timing games? Where do they appear?
2. A game-theoretic model of timing games.
3. Empirical measurements from MEV-Boost.

3

arXiv

What are timing games?

4

PoS Ethereum 101

Deposit 32 ETH to the staking contract,� get activated as a validator (today, ~675,000 validators)

Once activated, expected to perform some duties:

• Block proposer: 1 validator sampled every 12 seconds (slot)
• Attester: All validators do it once every 6.4 mins (epoch)

​

5

Anatomy of a slot

• Proposer expected to release block at t=0s
• Attesters vote as soon as they see the block or at t=4s (whichever is earliest)

6

The value of time

Proposer fill their blocks with incoming transactions� => The longer they wait, the more value they can pack

7

Slippery slope

Proposers may want to be as late as possible…� But then would attesters also be late?

Attesters care about two things:

• Correctness: They want to vote on the correct head
• Freshness: They want to release their vote before the next proposer

Late => More info, more correct, but risk of missing the slot…

8

Game-theoretic model of timing games

9

Slippery slope model

We were sort of stumped, couldn’t get a handle on the model.

Would there just be an unstoppable “fuite en avant?”

​

10

First approach: Dynamic model

• Player sequences: Proposers 1:N, attester committees 1:N
• Proposer chooses when to release
• Attesters choose when to release, vote for the block of their slot, or vote “empty” (no block-tree)
• No latency (wlog?)
• Strategies are dynamic: Function mapping time and history to some action
• Hard to reason about + measurability problems…

11

Second approach: Static model

• Player sequences: Proposers 1:N, Attester committees 1:N
• Proposer chooses when to release
• Attesters choose when to release, vote for the block of their slot, or vote “empty” (no block-tree)
• Player-specific latency from Attesters to Proposer
• Strategies are static: Players choose their schedule of actions at the start of the game
• “Game time” ≠ “Clock time”

12

Equilibrium in the static model

In the static model, we find an equilibrium� (in fact, many, but shifted by a constant amount of time)

High-level takeaways of equilibrium analysis:

• Attester committees “force” timely release of the Proposer.
• Proposer will still release at the last possible moment.
• Next Proposer would receive the rewards of a missing Proposer.

13

Empirical measurements from MEV-Boost

14

Empirical questions

• Are timing games being played now?
• What is the incentive for proposers to play them?

We can answer these questions partially with data from the builder market!

Dataset: ~150,000 slots (~1 month)� ~150,000,000 bids (~800 bids/slot)

15

Proposer-Builder Separation

Today, many Proposers delegate block construction to specialised entities known as Builders.

16

Are timing games played?

Seems like not…

• receivedAt�Relay receives bid
• eligibleAt�Relay validates bid
• signedAt�Relay receives signed bid from Proposer

17

Is this unique to Ethereum?

Probably not… We don’t assume much about the structure of Ethereum in the game.

In Proof-of-Work, could always get your slot “stolen” by another Proposer who solved the puzzle� In PoS, you don’t have this exogenous randomness…

Generally, review consensus algorithms in the light of incentives and rational participant model!

18

Economists!�There is so much data!

Thank you!

​

https://tinyurl.com/ethrig

​

Get in touch! barnabe@ethereum.org

20