Rework the profitability calculation/test

[jackhack00]

• On each timestep (of the duration), the lead Sequencer tests: 1) The gas required for the L1 transaction (taken from DEFAULT_DETERMINISTIC_GAS_ESTIMATOR = L1GasEstimators ) 2) The current gas price (taken from GAS_FEE_L1_TIME_SERIES_LIST ) 3) Multiplies these costs with a safety buffer (currently 2) (need rework on functional form) 4) Computes their expected rewards (taken from expected_rewards = state['cumm_block_rewards'] - history[-1][0]['cumm_block_rewards'] ) 5) From the rewards, subtracts a constant "op_cost" param, the fee, and the safety buffer (2fee) 6) aka Rewards (our assumption) - op_cost (our assumption) - 3*fee (multiplier our assumption, fee our choice of shock) 7) If the result is positive, they will be able to make the transaction

Tasks:

• Rework functional form to be more realistic
• Potentially rework op_cost to vary depending on agent

[SeanMcOwen]

160

[SeanMcOwen]

@jackhack00 Is the constant gas assumption normal, or is there any variance in the gas?

[SeanMcOwen]

@jackhack00 Should I be investigating validity for GAS_FEE_L1_TIME_SERIES_LIST or do we feel good on it

[SeanMcOwen]

@jackhack00 What is the safety buffer meant to capture? And is there empirical data to support it?

[jackhack00]

@SeanMcOwen Gas should be constant for DEFAULT_DETERMINISTIC_GAS_ESTIMATOR = L1GasEstimators Gas Price (in Gwei) should fluctuate: Steady state gives basis, shock adds on top. Both SS and Shock are pulled from timeseries.

Checking validity of GAS_FEE_L1_TIME_SERIES_LIST Yes please, that sounds great.

• I have pulled a couple years of daily average values.
• Might swap those out for block by block values for more realizm
• GAS_FEE_L1_TIME_SERIES_LIST was tuned by hand, needs validation to fit with actual values
• Want to also tune shock better, currently goes up by ~100, which brings total ~135 Gwei.
• Past two years, max daily average was 474, so got some room to tune

Safety Buffer Since rewards and costs are kinda unknown, we have to find reasonable guesses.

• Reward: we can somewhat guesstimate.
• Costs: Reflected through op_costs param. Meant to capture the fact that aside from gas fees, the rewards also need to keep the lights on. Idea was to get more insights into computational work / costs of operations, and plug that in.
• Safety Buffer: Originally intended to capture part of what is now in op_costs. Agent needs some profit to come away with. Arbitrarily set at 2x gas costs at time when op_cost did not exist. Empirical Evidence:
• We do know that these operations cost money in most cases
• We do not have data on real costs, real profit margins, etc.
• We can either deviate from profitability test, due to too many unknowns, but then need a different conditional on why someone would not be willing to make an l1 transaction (re-focussing on the gas thresholds for example)
• Or we define better op_costs and safety buffer, with reasonable estimates and knowledge that this needs iteration

[jackhack00]

An additional fix to this might include testing more realistic costs and rewards. Currently, an agent tests total rewards for the block (instead of their share of the rewards) against costs of the current transaction.

Realistically, an agent would check their own reward share vs. the remaining transactions at current gas price (since a singular transaction is likely under the total block rewards)

gas: Gas = params["gas_estimators"].commitment_bond(state) #Note: checking against only the current transaction might be simplifying a little too much
fee = gas * state["gas_fee_l1"]
SAFETY_BUFFER = 2 * fee  # HACK:

expected_rewards = params["reward_per_block"] #NOTE: reward per block is total reward
expected_costs: float = params["op_costs"] + fee + SAFETY_BUFFER

# Translate to ETH
expected_costs = expected_costs * 1e-9
payoff_reveal = expected_rewards - expected_costs